diff options
48 files changed, 30956 insertions, 407 deletions
@@ -1,6 +1,6 @@ # sketchbook -A small Zig 0.16 platform for sketching games and toys, in the spirit of +A small Zig 0.17 platform for sketching games and toys, in the spirit of Processing and pygame: write plain data and two functions, save, and the running program picks up the new code with its state intact. @@ -29,25 +29,75 @@ An app is an ordinary module. Every declaration is optional: | Declaration | Meaning | | --- | --- | | `name`, `width`, `height` | Window title and initial canvas | -| `State` | Persistent state: saved to a file, snapshotted, replayed. No pointers (compile error). | -| `Session` | Survives reloads, never saved: caches, loaded handles, audio buffers. May point into host memory. | +| `State` | Saved to a file, snapshotted, replayed: what `update` changes. No pointers and no host handles (compile errors). | +| `Session` | This process: survives reloads, never saved. Caches, image and buffer handles, playback cursors. | +| `Store` | On disk, at the same address in every run: survives reloads and restarts. Big data that is not replayed (a recording, a document). May point into itself; no host handles. | | `init(state, host)` | Setup for brand-new state, after the declared defaults | | `update(state, input)` | **Deterministic**: the host replays recorded input through it | -| `draw(state, session, frame)` | Pictures and sound, as commands; may change only `Session` | -| `command(state, session, console, line) !bool` | Console lines the host does not know | +| `draw(state, session, store, frame)` | Pictures and sound, as commands; may change Session and Store | +| `command(state, session, store, console, line) !bool` | Console lines the host does not know | | `commands` | Console completions | +Functions take the parameters they name, matched by type and in any order: +`draw(s: *const State, f: *sk.Frame)` and `draw(s: *const State, store: +*Store, ss: *Session, f: *sk.Frame)` both work. `update` can only ask for +`State` and `Input`; asking for more is a compile error, since a replay could +not reproduce it. + +Memory comes by lifetime, one allocator each: + +| Allocator | Lifetime | Released | +| --- | --- | --- | +| `f.scratch` | this frame | all at once, after the frame is drawn | +| `f.session` | this process | one allocation at a time; the rest when Session's layout changes, along with the images and buffers the session made | +| `f.store` | on disk, `<state>.store` | one allocation at a time; `--fresh` empties it | + +A Store layout change migrates field by field like State: unchanged fields, +pointers included, keep their values, since the store never moves. +`command` gets the session and store allocators on its `Console`. + +`update` sees only State, so when a key should do something to Session or +Store, State carries a request: `seek: sk.Request(f32)`, sent with +`s.seek.send(0.5)`. Session keeps an `sk.Handled`, and `draw` runs +`if (ss.seek.take(s.seek)) |to| ...` once per request. A fresh Session +counts earlier requests as handled, so nothing re-fires after a restart. + +`input.width`/`height` are the canvas, which a sketch sets with +`f.canvas(w, h)` and the host scales to fit its window. +`input.window_width`/`window_height` are the room it has (the window, in CSS +pixels on the web): a sketch that sets its canvas to them, and lays out by +the canvas, is responsive, at its own scale on a phone as on a desktop. + `update` sees `input.events`: pointer, wheel, key, text, `cancel` (focus loss, -console) and one `tick` per frame carrying `dt`. Recorded ticks make +console) and one `tick` per frame carrying `dt`. Pointer and wheel events +carry the modifier keys held as they happened (`e.modifiers()`, `Key.Mod` +bits): read those, not `input.held`, for ctrl+wheel and the like, since a +trackpad's or a touch screen's pinch arrives as ctrl+wheel with no key +events at all. Recorded ticks make time-driven sketches replay exactly. `input.started` is true on the first update after the program starts, never after a reload: state is saved, so this is where to stop what should not resume on its own. `draw` gets a `Frame`: `clear`, `fill`, `image`, `text`, `print`, `clip`, -`canvas`, `cursor`, `voice`, `audio`, `pixels`, `quit`, plus `f.arena` (frame -lifetime), `f.gpa` (session lifetime) and `f.host` (`loadImage`, -`decodeImage`, `newImage`, `loadSound`, `decodeSound`, `readFile`, `env`, -`log`). +`canvas`, `cursor`, `voice`, `audio`, `microphone`, `pixels`, `gpu`, +`compute`, `upload`, `download`, `openFile`, `run`, `quit`, plus the three allocators and +`f.host` (`loadImage`, `decodeImage`, `newImage`, `newBuffer`, `loadSound`, +`decodeSound`, `samples`, `readFile`, `env`, `log`). Commands may be built +from locals: the methods copy what they keep (text, GPU descriptions, +uniforms) into scratch. Bulk data (vertices, samples, pixels, uploads) is +not copied and must outlive the frame: scratch, session or store memory. + +`f.openFile("open", .{ .pattern = "wav;flac" })` shows the system's file +dialog; the answer comes back as the console line `open <path>`, so it +reaches the sketch's `command` like anything typed there. `f.run(line)` +does the same with any line, after the frame: `draw` cannot write State, +so this is how it tells State something only it learns (how many takes a +Store holds, say). + +`f.download(buffer, .{ .size = n, .key = k })` reads a GPU buffer back +without stalling: the bytes arrive in a later frame's `in.downloads`, a +frame or two behind. For measurements a shader made (a pitch, a level), +not for pixels. Geometry and colors are duck-typed. Anything with `x`/`y` is a point, anything that also has `w`/`h` (or `width`/`height`) is a rect, in any number type, and @@ -58,6 +108,14 @@ helpers return the caller's own type: - `sk.Camera`: `toScreen`, `toWorld`, `pan`, `zoomAt` - `sk.tween`: `lerp` over any plain data, `lerpFields(a, b, t, .{ .x, .zoom })`, `toward(a, .{ .x = 100 }, t)`, and `ease` +**Sound scopes**, in the `scope` module (import `sb.modules.scope`): a +`scope.Scope` kept in Session takes samples with `push` and, once a frame, +`draw`s the views given rects for: the last seconds as a scrolling waveform +and spectrogram, the smoothed spectrum of the newest samples, and an +oscilloscope. All analysed and drawn on the GPU, new columns only; its +shaders swap to WGSL for the web like any others. Lifted from micscope's +live view; spheres uses it. + **Vector graphics**, in the `vector` module (import `sb.modules.vector`): SVG parsing (the static subset Typst emits, and more) into cubic Bézier data; flattening, stroking, dashing, trimming, clipping and morphing as @@ -66,21 +124,38 @@ backend computes exact coverage and runs anywhere, wasm included. The GPU backend describes stencil-then-cover passes (8× MSAA, gradients in a shader). Both are tested against resvg (see `sketches/vecanim` in the game repository). -**GPU work is data.** A sketch describes pipelines (SPIR-V, vertex layout, -blend, stencil) and passes (`f.gpu(.{ .image, .pipelines, .draws, ... })`); -the host runs them with SDL_GPU. It caches shaders and pipelines by content -hash, so an edited shader is just a new pipeline after a reload. No sketch -touches a graphics API, and another host (Vulkan, WebGPU) would be another -translation of the same data. `abi.gpu.pipeline(Vertex, ...)` derives the +**GPU work is data.** A sketch describes pipelines (shader code, vertex +layout, blend, stencil) and passes (`f.gpu(.{ .image, .pipelines, .draws, +... })`); the native host runs them with SDL_GPU, the web host with WebGPU. +Both cache shaders and pipelines by content hash, so an edited shader is +just a new pipeline after a reload. No sketch touches a graphics API. The +shaders are GLSL; the build turns them into SPIR-V for native and WGSL for +the web, and each build imports the form its host reads (see Build API). `abi.gpu.pipeline(Vertex, ...)` derives the vertex layout from a struct, and `abi.gpu.uniform` checks std140 at -compile time. +compile time. A pipeline without attributes makes its own vertices from +`gl_VertexIndex` (a full-screen triangle, for shader-toy style views). + +Compute is data too. `host.newBuffer(size)` makes a GPU storage buffer, +`f.upload(buffer, offset, bytes)` fills it, `f.compute(.{ .kernel, .readonly, +.readwrite, .uniforms, .groups })` runs a compute shader over buffers, and a +draw reads buffers through `.vertex_storage` / `.fragment_storage`. Results +stay on the GPU: a compute shader's output is a later draw's input, with no +readback. Each dispatch is its own compute pass, so it sees everything the +commands before it wrote. Sound is declarative. `f.voice(.{ .key, .audio, .serial, .loop, .volume })` declares a voice that should be sounding. A new key or serial starts it, and an undeclared key stops it. A finished one-shot stays finished while it is still declared, so replays and seeks never re-fire sounds. `f.audio(key, samples, .{})` streams PCM, and `input.stream(key)` reports what the host has -queued and played. +queued and played. `f.microphone()` keeps the default recording device open +while declared, and the next frame's `input.microphone` holds what it heard +(to `draw` only: it is a device, not an event, so it does not replay). +`host.samples(audio)` gives a loaded sound's samples, mono at +`input.sample_rate`. + +Sounds decode with single-file C libraries in `vendor/` (dr_wav, dr_flac, +stb_vorbis): WAV, FLAC and Ogg Vorbis, chosen by content. ## The host @@ -90,13 +165,18 @@ queued and played. shared via `MAP_SHARED`. Its header carries the reflected layout, so a layout change migrates every slot field by field, with enums matched by tag name, and `sketchbook-probe` can print any sketch's state. +- **Store file.** `<name>.state.store` is a heap mapped at a fixed address + (`0x2000_0000_0000`, 64 GiB reserved), grown in place, locked to one + process. Power-of-two size classes with free lists in the file. Snapshots + and replays do not cover it (only `draw` and `command` change it). - **Time travel.** Recordings, replay, seek and step work on any sketch, because they only need state bytes and events. - **Console and overlays.** `get`/`set <field path>`, snapshots, recordings and the memory and scope overlays come from the host; unknown lines go to the sketch. - **Scripts** (`--headless --fresh --script file`) drive the same routing and - frames as a player: `click`, `key`, `type`, `console`, `expect field value`, + frames as a player: `click`, `key`, `hold`/`release` (a key down until + released, for modifiers), `wheel`, `type`, `console`, `expect field value`, `await field value` (frames until it holds, a minute at most), `expect-audio`, `screenshot`, `load`, `reload`, `await-reload`. A failed `expect`, or a console line that errors (a sketch command returning an @@ -104,9 +184,43 @@ queued and played. with `SDL_VIDEO_DRIVER` set, and GPU work still runs, into a texture. - **GPU.** The host's 2D renderer runs on an SDL GPU device, which also runs sketches' GPU passes. Without a GPU it falls back to any renderer. + `--profile-gpu` times every upload, dispatch and pass alone on an idle + GPU (SDL_GPU has no timestamp queries) and prints count, mean, worst and + per-frame time by `label` at exit. - **Memory.** SDL allocations go through a leak-checking allocator; a leak fails the exit. +## The web host + +`web/host.js` runs the same sketch, compiled to wasm32-freestanding, in a +page. Draw commands go to a 2D canvas; GPU passes, dispatches, uploads and +downloads to WebGPU, whose renders are shown through the canvas; streams and +the microphone to Web Audio (at 48 kHz where the browser agrees, sound +starting at the first click or key, as browsers require); `f.openFile` to a +file picker, whose sound the browser decodes and hands over by name before +the console line arrives. `f.run` and the sketch's `command` work as on +native. + +Touch: one finger is the mouse's left button. Two fingers pan, as the wheel +(60 px a notch), and pinch, as ctrl+wheel (a notch per 20 % of spread), both +around the point between them: what a trackpad sends, so a sketch that +handles a trackpad handles a touch screen. Keys have no touch equivalent; +draw buttons for what matters. + +Without WebGPU, the page says why the first time the sketch asks for it: +not a secure context (WebGPU and the microphone need https, or localhost), +WebGPU turned off (Firefox on Linux: `dom.webgpu.enabled`), or no usable +adapter. + +`.web = .{ .pwa = .{ .icons = ..., .theme = ... } }` makes the page an +installable app that opens offline: a manifest, icons, and `web/sw.js`, a +service worker that keeps a copy of everything the page fetches, network +first, so each new build shows as soon as there is a network. Service +workers need a secure context, like WebGPU. + +For tests, `window.sketch` reads state (`field(path)`), runs console lines +(`command(line)`) and hands over decoded sound (`sound(name, samples)`). + ## Build API ```zig @@ -119,38 +233,72 @@ pub fn build(b: *std.Build) void { } ``` -`addSteps` installs the executable and adds `run`, `lib`, `web`, `probe`, -`e2e` and `test`. Every piece is also available on its own and returns plain -`std.Build` objects: +`addSteps` installs the executable and adds `run`, `lib`, `probe`, `e2e` and +`test`, and with `.web = .{}` also `web` (the page into `zig-out/web`) and +`serve` (it, over https on port 8042, to the local network too; `zig build +web --watch` alongside rebuilds it for the next reload). Every piece is also available on its own +and returns plain `std.Build` objects: | Call | Returns | | --- | --- | | `app.addLibrary()` | the hot-reloadable shared library | | `app.addExecutable()` | host and sketch linked into one binary | -| `app.addWeb(.{ .assets = ... })` | the wasm module and a ready-to-serve directory; tests can read its state by path with `sk_field` (numbers) and `sk_tag` (enum tag names) | +| `app.addWeb(.{ .assets = ... })` | the wasm module and a ready-to-serve directory, built from the app's modules with every shader module swapped for its WGSL twin; tests can read its state by path with `sk_field` (numbers) and `sk_tag` (enum tag names) | +| `app.addServe(install)` | `miniserve` serving an installed web directory over https, on every interface, with a self-signed certificate made once by `openssl`: https makes the page a secure context on any device, which WebGPU and the microphone need. Its defaults are environment variables, so miniserve options after `--` replace them | | `app.addRun(lib, .{})` | the generic host running the installed library | | `app.addScript(exe_or_lib, file, .{})` | a headless, fresh-state scripted run, of the executable or of the library through the host (so the script can `reload` it) | | `app.addProbe(state)` | the probe on a state file | | `app.addTest(.{})` | tests with the app importable as `app` | -Shaders, in the same style: +Shaders: ```zig -const layouts = sb.addGlslLayouts(.{ .module = my_layouts, .types = &.{ "Vertex", "Paint" } }); -const shaders = sb.addSpirvModule(&.{ - .{ .name = "fill_vert", .spirv = sb.addGlslc(.{ .source = b.path("fill.vert"), .include_dirs = &.{layouts.dir} }).spirv }, +const shaders = sb.addShaders(.{ + .layouts = .{ .module = my_layouts, .types = &.{ "Vertex", "Paint" } }, + .sources = &.{ b.path("shaders/fill.vert"), b.path("shaders/fill.frag") }, }); +const app = sb.app(.{ ..., .imports = &.{.{ .name = "shaders", .module = shaders.module }} }); +// in the sketch: .{ .code = shaders.fill_vert } ``` -`addGlslLayouts` writes `<Name>.glsl` declarations generated from Zig -structs (tagged with `pub const glsl = .vertex` or -`.{ .uniform = .{ .set, .instance } }`), so the Zig struct is the one -definition. `addGlslc` compiles one file. `addSpirvModule` embeds the -results, aligned for Vulkan. +The layouts are Zig structs (tagged with `pub const glsl = .vertex` or +`.{ .uniform = .{ .set, .instance } }`) written out as `<Name>.glsl` for the +sources to `#include`, so the Zig struct is the one definition. Each source +is compiled by `glslc` into `shaders.module`, as SPIR-V under its file name +(`fill.vert` is `fill_vert`). For the web it is compiled again with +`SKETCHBOOK_WEB` defined and translated by `naga` into `shaders.web`, the +same names in WGSL; `addWeb` puts that twin in place of the SPIR-V module +wherever the app's imports reach it, however deep (the `vector` module's +shaders included). Only the web build runs `naga`, and a missing tool +fails only the steps that need it, listing where it looked. WGSL has no +write-only storage buffers and no `coherent`: write them plain, or under +`#ifndef SKETCHBOOK_WEB`. Sources may `#include` files beside them; +`glslc`'s depfiles tell the build, so editing an included file recompiles +what includes it. `addGlslLayouts` and `addGlslc` are the pieces, for +anything else. + +Tools: `glslc` (shaderc, or the Vulkan SDK's) for shaders, for web builds +with shaders `naga` (`cargo install naga-cli`; found on PATH or in cargo's +directory), and for `serve`, `miniserve` and `openssl`. `sb.modules`, `sb.host_exe` and `sb.probe_exe` are exposed for anything else. Add `.sketchbook = .{ .path = "..." }` to your `build.zig.zon`. +## Known limits + +- The console line holds 96 characters, so long paths cannot be typed; + `f.openFile` answers with a line of any length. +- Every upload, compute dispatch and GPU pass is its own command buffer and + submit. Fine at a handful per frame; batch them per frame if that grows. +- The store is not crash-safe: a crash in the middle of an allocation can + leave it inconsistent, and `--fresh` starts it over. +- Handles inside Store containers (`std.ArrayList(sk.Image)`) are not caught + at compile time: the check does not follow pointers. +- On the web: nothing persists (Store lasts as long as the page), + synthesised voices are silent, `loadSound` knows only sounds the page + decoded, multisampling is 4× at most, and mixing runs on the main thread + (a ScriptProcessor), so a busy page can crackle. + ## Designs - [Regions in source](docs/regions-in-source.md): editing layout from the @@ -6,14 +6,21 @@ //! //! const sb = sketchbook.init(b, .{ .target = target, .optimize = optimize }); //! const app = sb.app(.{ .name = "demo", .root_source_file = b.path("src/main.zig") }); -//! _ = app.addSteps(.{}); // run, lib, web, probe, e2e, test, install +//! _ = app.addSteps(.{}); // run, lib, probe, e2e, test, install; +//! // web and serve with `.web = .{}` //! //! or pick the pieces: `app.addLibrary()`, `app.addExecutable()`, //! `app.addWeb(...)`, `app.addRun(...)`, `app.addScript(...)`, -//! `app.addProbe(...)`, `app.addTest(...)`. For GPU shaders: -//! `sb.addGlslLayouts(...)`, `sb.addGlslc(...)`, `sb.addSpirvModule(...)`. -//! Each returns the ordinary `std.Build` objects, so anything can be -//! extended or rewired. +//! `app.addProbe(...)`, `app.addTest(...)`. GPU shaders are GLSL, built by +//! `sb.addShaders(...)` into SPIR-V for native hosts and WGSL for the web; +//! `sb.addGlslLayouts(...)` and `sb.addGlslc(...)` are its pieces. Each +//! returns the ordinary `std.Build` objects, so anything can be extended or +//! rewired. +//! +//! Tools, found on PATH when a step needs them: `glslc` (shaderc or the +//! Vulkan SDK) for any shader, `naga` (`cargo install naga-cli`) only for +//! the web build of a sketch with shaders, and `miniserve` and `openssl` +//! only for `serve`. const std = @import("std"); const Build = std.Build; const translate_c = @import("translate_c"); @@ -59,26 +66,52 @@ pub fn build(b: *Build) void { }); b.installArtifact(probe); - var sb: Sketchbook = .{ .b = b, .target = target, .optimize = optimize, .modules = .{ .abi = abi, .sketch = sketch, .host = host, .vector = undefined }, .host_exe = host_exe, .probe_exe = probe, .web = b.path("web"), .glsl = b.path("src/glsl.zig") }; + var sb: Sketchbook = .{ .b = b, .target = target, .optimize = optimize, .modules = .{ .abi = abi, .sketch = sketch, .host = host, .vector = undefined, .scope = undefined }, .host_exe = host_exe, .probe_exe = probe, .web = b.path("web"), .glsl = b.path("src/glsl.zig"), .shared = .create(b) }; - // Vector graphics: built with the same shader step apps use. + // Vector graphics: built with the same shader step apps use. Both forms + // of its shaders are exported, so apps' web builds can swap them too. const vector_layouts = b.createModule(.{ .root_source_file = b.path("src/vector/layouts.zig") }); - const glsl = sb.addGlslLayouts(.{ .module = vector_layouts, .types = &.{ "Vertex", "Viewport", "Paint" } }); - const vector_shaders = sb.addSpirvModule(&.{ - .{ .name = "path_vert", .spirv = sb.addGlslc(.{ .source = b.path("src/vector/shaders/path.vert"), .include_dirs = &.{glsl.dir} }).spirv }, - .{ .name = "stencil_frag", .spirv = sb.addGlslc(.{ .source = b.path("src/vector/shaders/stencil.frag") }).spirv }, - .{ .name = "cover_frag", .spirv = sb.addGlslc(.{ .source = b.path("src/vector/shaders/cover.frag"), .include_dirs = &.{glsl.dir} }).spirv }, + const vector_shaders = sb.addShaders(.{ + .layouts = .{ .module = vector_layouts, .types = &.{ "Vertex", "Viewport", "Paint" } }, + .sources = &.{ b.path("src/vector/shaders/path.vert"), b.path("src/vector/shaders/stencil.frag"), b.path("src/vector/shaders/cover.frag") }, }); + b.modules.put(b.graph.arena, "vector_shaders", vector_shaders.module) catch @panic("OOM"); + b.modules.put(b.graph.arena, "vector_shaders_web", vector_shaders.web) catch @panic("OOM"); const vector_imports: []const Build.Module.Import = &.{ .{ .name = "abi", .module = abi }, .{ .name = "vector_layouts", .module = vector_layouts }, - .{ .name = "vector_shaders", .module = vector_shaders }, + .{ .name = "vector_shaders", .module = vector_shaders.module }, }; sb.modules.vector = b.addModule("vector", .{ .root_source_file = b.path("src/vector/root.zig"), .imports = vector_imports }); + // Sound scopes: waveform, spectrogram, spectrum and oscilloscope of a + // stream of samples, analysed and drawn on the GPU. + const scope_layouts = b.createModule(.{ .root_source_file = b.path("src/scope/layouts.zig") }); + const scope_shaders = sb.addShaders(.{ + .layouts = .{ .module = scope_layouts, .types = &.{ "SpectrumParams", "WaveParams", "WaveView", "SpectrumView", "SpectrogramParams", "SpectrogramView" } }, + .sources = &.{ + b.path("src/scope/shaders/view.vert"), + b.path("src/scope/shaders/waveform.comp"), + b.path("src/scope/shaders/waveform.frag"), + b.path("src/scope/shaders/spectrogram.comp"), + b.path("src/scope/shaders/spectrogram.frag"), + b.path("src/scope/shaders/spectrum.comp"), + b.path("src/scope/shaders/spectrum.frag"), + }, + }); + b.modules.put(b.graph.arena, "scope_shaders", scope_shaders.module) catch @panic("OOM"); + b.modules.put(b.graph.arena, "scope_shaders_web", scope_shaders.web) catch @panic("OOM"); + const scope_imports: []const Build.Module.Import = &.{ + .{ .name = "sketch", .module = sketch }, + .{ .name = "scope_layouts", .module = scope_layouts }, + .{ .name = "scope_shaders", .module = scope_shaders.module }, + }; + sb.modules.scope = b.addModule("scope", .{ .root_source_file = b.path("src/scope/root.zig"), .imports = scope_imports }); + const test_step = b.step("test", "Run unit tests and the reload end-to-end test"); for ([_]*Build.Module{ b.createModule(.{ .root_source_file = b.path("src/vector/root.zig"), .target = target, .optimize = optimize, .imports = vector_imports }), + b.createModule(.{ .root_source_file = b.path("src/scope/root.zig"), .target = target, .optimize = optimize, .imports = scope_imports }), b.createModule(.{ .root_source_file = b.path("src/glsl.zig"), .target = target, .optimize = optimize }), b.createModule(.{ .root_source_file = b.path("src/abi.zig"), .target = target, .optimize = optimize }), b.createModule(.{ .root_source_file = b.path("src/sketch/root.zig"), .target = target, .optimize = optimize, .imports = &.{.{ .name = "abi", .module = abi }} }), @@ -112,13 +145,20 @@ fn cModule(b: *Build, target: Build.ResolvedTarget, optimize: std.builtin.Optimi .c_source_file = b.addWriteFiles().add("c.h", \\#define SDL_DISABLE_OLD_NAMES \\#include <SDL3/SDL.h> - \\#include <sndfile.h> + \\#include "codecs.h" + \\#include "text.h" ), .target = target, .optimize = optimize, }); translator.linkLibrary(sdl); - translator.linkSystemLibrary("sndfile", .{}); + translator.addIncludePath(b.path("vendor")); + translator.mod.addCSourceFile(.{ .file = b.path("vendor/codecs.c"), .flags = &.{"-fno-sanitize=undefined"} }); + if (target.result.os.tag == .linux) { + translator.mod.addCSourceFile(.{ .file = b.path("vendor/text.c"), .flags = &.{"-std=c23"} }); + translator.mod.linkSystemLibrary("pangocairo-1.0", .{ .use_pkg_config = .force }); + translator.mod.linkSystemLibrary("pangoft2-1.0", .{ .use_pkg_config = .force }); + } return translator.mod; } @@ -168,24 +208,51 @@ pub const Options = struct { pub fn init(b: *Build, options: Options) Sketchbook { const dep = b.dependency(options.dependency, .{ .target = options.target, .optimize = options.optimize }); + const shared: *Shared = .create(b); + for ([_][]const u8{ "vector_shaders", "scope_shaders" }) |name| { + shared.web_twins.put(b.graph.arena, dep.module(name), dep.module(b.fmt("{s}_web", .{name}))) catch @panic("OOM"); + } return .{ .b = b, .target = options.target, .optimize = options.optimize, - .modules = .{ .abi = dep.module("abi"), .sketch = dep.module("sketch"), .host = dep.module("host"), .vector = dep.module("vector") }, + .modules = .{ .abi = dep.module("abi"), .sketch = dep.module("sketch"), .host = dep.module("host"), .vector = dep.module("vector"), .scope = dep.module("scope") }, .host_exe = dep.artifact("sketchbook"), .probe_exe = dep.artifact("sketchbook-probe"), .web = dep.path("web"), .glsl = dep.path("src/glsl.zig"), + .shared = shared, }; } +/// What every copy of a `Sketchbook` value shares. +const Shared = struct { + /// SPIR-V shader modules and their WGSL twins (see `addShaders`). + web_twins: std.AutoArrayHashMapUnmanaged(*Build.Module, *Build.Module) = .empty, + /// Lookups of external tools, made once, run when a step needs them. + glslc: ?Build.LazyPath = null, + naga: ?Build.LazyPath = null, + miniserve: ?Build.LazyPath = null, + openssl: ?Build.LazyPath = null, + /// A self-signed certificate and its key, for `addServe`. + certificate: ?Certificate = null, + + const Certificate = struct { cert: Build.LazyPath, key: Build.LazyPath }; + + fn create(b: *Build) *Shared { + const shared = b.graph.arena.create(Shared) catch @panic("OOM"); + shared.* = .{}; + return shared; + } +}; + pub const Sketchbook = struct { b: *Build, target: Build.ResolvedTarget, optimize: std.builtin.OptimizeMode, - /// `vector`: SVG, cubic geometry, and CPU and GPU renderers. - modules: struct { abi: *Build.Module, sketch: *Build.Module, host: *Build.Module, vector: *Build.Module }, + /// `vector`: SVG, cubic geometry, and CPU and GPU renderers. `scope`: + /// waveform, spectrogram, spectrum and oscilloscope views of a sound. + modules: struct { abi: *Build.Module, sketch: *Build.Module, host: *Build.Module, vector: *Build.Module, scope: *Build.Module }, /// The generic host: loads and hot-reloads a sketch library. host_exe: *Build.Step.Compile, /// Prints any sketch's state file. @@ -194,6 +261,7 @@ pub const Sketchbook = struct { web: Build.LazyPath, /// The GLSL declaration generator (see `addGlslLayouts`). glsl: Build.LazyPath, + shared: *Shared, pub const GlslLayouts = struct { run: *Build.Step.Run, @@ -238,36 +306,160 @@ pub const Sketchbook = struct { spirv: Build.LazyPath, }; - /// One GLSL file compiled to SPIR-V by `glslc` (from the Vulkan SDK or - /// shaderc, on PATH). + /// One GLSL file compiled to SPIR-V by `glslc`. pub fn addGlslc(sb: Sketchbook, options: struct { source: Build.LazyPath, include_dirs: []const Build.LazyPath = &.{}, + /// Macros defined for the source, as `NAME` or `NAME=VALUE`. + defines: []const []const u8 = &.{}, target_env: []const u8 = "vulkan1.1", optimize: bool = true, }) Glslc { const b = sb.b; - const run = b.addSystemCommand(&.{ "glslc", b.fmt("--target-env={s}", .{options.target_env}) }); + const name = std.fs.path.basename(options.source.getDisplayName()); + const run = Build.Step.Run.create(b, b.fmt("glslc {s}", .{name})); + run.addFileArg(sb.glslc()); + run.addArg(b.fmt("--target-env={s}", .{options.target_env})); if (options.optimize) run.addArg("-O"); + for (options.defines) |define| run.addArg(b.fmt("-D{s}", .{define})); for (options.include_dirs) |dir| run.addPrefixedDirectoryArg("-I", dir); + // The files the source #includes, for the cache: editing one + // recompiles every shader that includes it. + run.addArgs(&.{ "-MD", "-MF" }); + _ = run.addDepFileOutputArg(b.fmt("{s}.d", .{name})); run.addArg("-o"); - const spirv = run.addOutputFileArg(b.fmt("{s}.spv", .{std.fs.path.basename(options.source.getDisplayName())})); + const spirv = run.addOutputFileArg(b.fmt("{s}.spv", .{name})); run.addFileArg(options.source); return .{ .run = run, .spirv = spirv }; } - /// A module declaring each SPIR-V file as `pub const <name>: []const u8`, - /// 4-byte aligned as Vulkan reads it. - pub fn addSpirvModule(sb: Sketchbook, shaders: []const struct { name: []const u8, spirv: Build.LazyPath }) *Build.Module { + pub const Shaders = struct { + /// SPIR-V, one `pub const <name>: []const u8` per shader, 4-byte + /// aligned as Vulkan reads it: the module to import. + module: *Build.Module, + /// The same declarations in WGSL, translated from the SPIR-V by + /// `naga`. The web build puts it in place of `module` wherever + /// `module` is imported (see `addWeb`), so sketches import one + /// module and never choose. + web: *Build.Module, + }; + + /// GLSL shaders for every host: compiled to SPIR-V, and for the web + /// compiled again with `SKETCHBOOK_WEB` defined and translated to WGSL, + /// which only the web build waits for. Each source is declared under its + /// file name with dots as underscores: `tuner.comp` is `tuner_comp`. + /// + /// WGSL lacks a little GLSL has: no write-only storage buffers (declare + /// them `buffer`), and no `coherent` (WGSL's barriers make storage + /// writes visible to the workgroup anyway, so leave it out under + /// `#ifdef SKETCHBOOK_WEB`). + pub fn addShaders(sb: Sketchbook, options: struct { + sources: []const Build.LazyPath, + /// Uniform and vertex layouts generated from Zig types, which the + /// sources `#include` as `<Type>.glsl` (see `addGlslLayouts`). + layouts: ?struct { module: *Build.Module, types: []const []const u8 } = null, + /// More `#include` directories. + include_dirs: []const Build.LazyPath = &.{}, + }) Shaders { const b = sb.b; - const files = b.addWriteFiles(); - var source: std.ArrayList(u8) = .empty; - for (shaders) |shader| { - const file = b.fmt("{s}.spv", .{shader.name}); - _ = files.addCopyFile(shader.spirv, file); - source.print(b.allocator, "const {s}_words align(4) = @embedFile(\"{s}\").*;\npub const {s}: []const u8 = &{s}_words;\n", .{ shader.name, file, shader.name, shader.name }) catch @panic("OOM"); + const include_dirs = b.graph.arena.alloc(Build.LazyPath, options.include_dirs.len + 1) catch @panic("OOM"); + @memcpy(include_dirs[0..options.include_dirs.len], options.include_dirs); + var dirs: []const Build.LazyPath = include_dirs[0..options.include_dirs.len]; + if (options.layouts) |layouts| { + include_dirs[options.include_dirs.len] = sb.addGlslLayouts(.{ .module = layouts.module, .types = layouts.types }).dir; + dirs = include_dirs; + } + + const native = b.addWriteFiles(); + const web = b.addWriteFiles(); + var native_source: std.ArrayList(u8) = .empty; + var web_source: std.ArrayList(u8) = .empty; + for (options.sources) |source| { + const file = std.fs.path.basename(source.getDisplayName()); + const name = std.mem.replaceOwned(u8, b.graph.arena, file, ".", "_") catch @panic("OOM"); + const spirv = sb.addGlslc(.{ .source = source, .include_dirs = dirs }).spirv; + _ = native.addCopyFile(spirv, b.fmt("{s}.spv", .{name})); + native_source.print(b.graph.arena, "const {s}_words align(4) = @embedFile(\"{s}.spv\").*;\npub const {s}: []const u8 = &{s}_words;\n", .{ name, name, name, name }) catch @panic("OOM"); + + const translate = Build.Step.Run.create(b, b.fmt("naga {s}", .{file})); + translate.addFileArg(sb.naga()); + // SDL_GPU flips Vulkan's viewport, so the SPIR-V's clip space is + // already WebGPU's (y up): naga must not flip it again. + translate.addArg("--keep-coordinate-space"); + translate.addFileArg(sb.addGlslc(.{ .source = source, .include_dirs = dirs, .defines = &.{"SKETCHBOOK_WEB"} }).spirv); + _ = web.addCopyFile(translate.addOutputFileArg(b.fmt("{s}.wgsl", .{name})), b.fmt("{s}.wgsl", .{name})); + web_source.print(b.graph.arena, "pub const {s}: []const u8 = @embedFile(\"{s}.wgsl\");\n", .{ name, name }) catch @panic("OOM"); + } + const shaders: Shaders = .{ + .module = b.createModule(.{ .root_source_file = native.add("spirv.zig", native_source.items) }), + .web = b.createModule(.{ .root_source_file = web.add("wgsl.zig", web_source.items) }), + }; + sb.shared.web_twins.put(b.graph.arena, shaders.module, shaders.web) catch @panic("OOM"); + return shaders; + } + + /// `glslc`, from shaderc or the Vulkan SDK. + fn glslc(sb: Sketchbook) Build.LazyPath { + const sdk = if (sb.b.graph.environ_map.get("VULKAN_SDK")) |dir| sb.b.pathJoin(&.{ dir, "bin", "glslc" }) else "glslc"; + return sb.program(&sb.shared.glslc, &.{ "glslc", sdk }); + } + + /// `naga`, from `cargo install naga-cli`: on PATH, or where cargo puts it. + fn naga(sb: Sketchbook) Build.LazyPath { + const env = &sb.b.graph.environ_map; + const cargo = if (env.get("CARGO_HOME")) |home| sb.b.pathJoin(&.{ home, "bin", "naga" }) else if (env.get("HOME")) |home| sb.b.pathJoin(&.{ home, ".cargo", "bin", "naga" }) else "naga"; + return sb.program(&sb.shared.naga, &.{ "naga", cargo }); + } + + /// A tool found on PATH when a step needs it, looked up once. + fn program(sb: Sketchbook, slot: *?Build.LazyPath, names: []const []const u8) Build.LazyPath { + if (slot.* == null) slot.* = sb.b.findProgramLazy(.{ .names = names }); + return slot.*.?; + } + + /// A self-signed certificate for serving over https, made once by + /// `openssl` and kept in the cache: browsers ask to trust it once. + fn certificate(sb: Sketchbook) Shared.Certificate { + if (sb.shared.certificate == null) { + const b = sb.b; + const run = Build.Step.Run.create(b, "self-signed certificate"); + run.addFileArg(sb.program(&sb.shared.openssl, &.{"openssl"})); + run.addArgs(&.{ "req", "-x509", "-newkey", "ec", "-pkeyopt", "ec_paramgen_curve:prime256v1", "-nodes", "-days", "3650", "-subj", "/CN=sketchbook" }); + run.addArg("-keyout"); + const key = run.addOutputFileArg("key.pem"); + run.addArg("-out"); + // Its progress on stderr is not a warning. + _ = run.captureStdErr(.{}); + sb.shared.certificate = .{ .cert = run.addOutputFileArg("cert.pem"), .key = key }; + } + return sb.shared.certificate.?; + } + + /// `module` as the web build compiles it: each shader module in its + /// imports, however deep, replaced by its WGSL twin, and each module on + /// the way copied to import the copy. Modules nothing swaps below are + /// shared as they are. + fn webModule(sb: Sketchbook, module: *Build.Module, copies: *std.AutoArrayHashMapUnmanaged(*Build.Module, *Build.Module)) *Build.Module { + const arena = sb.b.graph.arena; + if (sb.shared.web_twins.get(module)) |twin| return twin; + if (copies.get(module)) |done| return done; + // An import cycle comes back here as itself: unchanged, which is + // right unless a shader module hides inside the cycle. + copies.put(arena, module, module) catch @panic("OOM"); + var swapped = false; + const imports = arena.alloc(*Build.Module, module.import_table.count()) catch @panic("OOM"); + for (module.import_table.values(), imports) |imported, *web| { + web.* = sb.webModule(imported, copies); + swapped = swapped or web.* != imported; } - return b.createModule(.{ .root_source_file = files.add("spirv.zig", source.items) }); + if (!swapped) return module; + const copy = arena.create(Build.Module) catch @panic("OOM"); + copy.init(module.owner, .{ .existing = module }); + copy.import_table = .empty; + copy.cached_graph = .{ .modules = &.{}, .names = &.{} }; + for (module.import_table.keys(), imports) |name, web| copy.addImport(name, web); + copies.put(arena, module, copy) catch @panic("OOM"); + return copy; } pub fn app(sb: Sketchbook, options: App.Options) App { @@ -340,11 +532,35 @@ pub const App = struct { pub const WebOptions = struct { /// Directories copied next to the page, under their own `dest`. assets: []const Asset = &.{}, - /// Replaces the generic page; `{title}` is substituted either way. + /// Replaces the generic page; `{title}` is substituted either way, + /// and `{head}` with the links an installable app needs. index_html: ?Build.LazyPath = null, optimize: std.builtin.OptimizeMode = .ReleaseSafe, + /// Makes the page an installable app that opens offline. + pwa: ?Pwa = null, pub const Asset = struct { source: Build.LazyPath, dest: []const u8 }; + + /// A progressive web app: a manifest naming it, icons, and a + /// service worker (`web/sw.js`) that keeps a copy of everything the + /// page fetches, network first, so it opens offline and still shows + /// each new build at once. Browsers allow service workers only in + /// secure contexts (https, or localhost). + pub const Pwa = struct { + /// The app's name by default. + name: ?[]const u8 = null, + short_name: ?[]const u8 = null, + description: []const u8 = "", + /// CSS colours: behind the page while it starts, and the + /// browser's chrome around it. + background: []const u8 = "#000000", + theme: []const u8 = "#000000", + /// Square PNGs, served as `icon-<size>.png`. Browsers want 192 + /// and 512 pixels to offer installing. + icons: []const Icon = &.{}, + + pub const Icon = struct { source: Build.LazyPath, size: u32 }; + }; }; pub const Web = struct { @@ -356,9 +572,12 @@ pub const App = struct { /// The browser build, as one directory ready to serve. pub fn addWeb(app: App, options: WebOptions) Web { const b = app.sb.b; + var copies: std.AutoArrayHashMapUnmanaged(*Build.Module, *Build.Module) = .empty; + var web_app = app; + web_app.module = app.sb.webModule(app.module, &copies); const wasm = b.addExecutable(.{ .name = "sketch", - .root_module = app.wrap("comptime { @import(\"sketch\").web.exportWeb(@import(\"app\")); }", b.resolveTargetQuery(.{ .cpu_arch = .wasm32, .os_tag = .freestanding }), options.optimize), + .root_module = web_app.wrap("comptime { @import(\"sketch\").web.exportWeb(@import(\"app\")); }", b.resolveTargetQuery(.{ .cpu_arch = .wasm32, .os_tag = .freestanding }), options.optimize), }); wasm.entry = .disabled; wasm.rdynamic = true; @@ -373,11 +592,87 @@ pub const App = struct { })); page.addFileArg(options.index_html orelse app.sb.web.path(b, "index.html")); page.addArg(app.name); + page.addArg(if (options.pwa) |pwa| app.addPwa(files, pwa) else ""); _ = files.addCopyFile(page.addOutputFileArg("index.html"), "index.html"); for (options.assets) |asset| _ = files.addCopyDirectory(asset.source, asset.dest, .{}); return .{ .wasm = wasm, .dir = files.getDirectory() }; } + /// Writes the manifest, icons and service worker into `files`, and + /// returns what the page's `<head>` needs for them. + fn addPwa(app: App, files: *Build.Step.WriteFile, pwa: WebOptions.Pwa) []const u8 { + const b = app.sb.b; + const arena = b.graph.arena; + const Icon = struct { src: []const u8, sizes: []const u8, type: []const u8 = "image/png", purpose: []const u8 = "any" }; + const icons = arena.alloc(Icon, pwa.icons.len) catch @panic("OOM"); + var touch: ?[]const u8 = null; + for (pwa.icons, icons) |icon, *entry| { + const name = b.fmt("icon-{d}.png", .{icon.size}); + _ = files.addCopyFile(icon.source, name); + entry.* = .{ .src = name, .sizes = b.fmt("{d}x{d}", .{ icon.size, icon.size }) }; + // iOS takes its home-screen icon from the page, not the manifest. + if (icon.size >= 180 and (touch == null or icon.size < 512)) touch = name; + } + const name = pwa.name orelse app.name; + const manifest = std.json.Stringify.valueAlloc(arena, .{ + .name = name, + .short_name = pwa.short_name orelse name, + .description = pwa.description, + // Relative: the app works under any path it is served from. + .start_url = ".", + .scope = ".", + .display = "standalone", + .background_color = pwa.background, + .theme_color = pwa.theme, + .icons = icons, + }, .{ .whitespace = .indent_2 }) catch @panic("OOM"); + _ = files.add("manifest.webmanifest", manifest); + _ = files.addCopyFile(app.sb.web.path(b, "sw.js"), "sw.js"); + return b.fmt( + \\<link rel="manifest" href="manifest.webmanifest"> + \\<meta name="theme-color" content="{s}">{s} + , .{ pwa.theme, if (touch) |icon| b.fmt("\n<link rel=\"apple-touch-icon\" href=\"{s}\">", .{icon}) else "" }); + } + + /// Serves an installed web build (`install`, under the prefix) with + /// `miniserve`, over https on port 8042 of every interface, until + /// stopped. Browsers give WebGPU and the microphone only to secure + /// contexts, and https makes the page one from any device on the + /// network, once the self-signed certificate is accepted; miniserve + /// prints the addresses as QR codes, for phones. It serves the installed + /// files, so `zig build web --watch` alongside shows each rebuild on + /// reload. Its defaults are environment variables, so miniserve's + /// options after `--` replace them (`-p 9000`). + pub fn addServe(app: App, install: *Build.Step.InstallDir) *Build.Step.Run { + const sb = app.sb; + const b = sb.b; + const prefix_relative = switch (install.options.install_dir) { + .prefix => install.options.install_subdir, + .custom => |dir| b.pathJoin(&.{ dir, install.options.install_subdir }), + else => @panic("addServe: install the web build under the prefix"), + }; + const certificate = sb.certificate(); + const run = Build.Step.Run.create(b, "miniserve"); + run.addFileArg(sb.program(&sb.shared.miniserve, &.{"miniserve"})); + run.addPrefixedFileArg("--tls-cert=", certificate.cert); + run.addPrefixedFileArg("--tls-key=", certificate.key); + for ([_][2][]const u8{ + .{ "MINISERVE_PORT", "8042" }, + .{ "MINISERVE_INTERFACE", "0.0.0.0" }, + .{ "MINISERVE_INDEX", "index.html" }, + .{ "MINISERVE_DISABLE_INDEXING", "true" }, + .{ "MINISERVE_QRCODE", "true" }, + // A rebuilt sketch shows on reload. + .{ "MINISERVE_HEADER", "Cache-Control:no-cache" }, + }) |env| run.setEnvironmentVariable(env[0], env[1]); + run.addPassthruArgs(); + run.addDirectoryArg(b.graph.path(.install_prefix, prefix_relative)); + run.step.dependOn(&install.step); + run.stdio = .inherit; + run.has_side_effects = true; + return run; + } + pub const RunOptions = struct { /// State file, relative to the working directory. state: ?[]const u8 = null, @@ -393,10 +688,10 @@ pub const App = struct { const run = b.addRunArtifact(app.sb.host_exe); run.step.dependOn(&install.step); run.addArgs(&.{ "--state", options.state orelse b.fmt("{s}.state", .{app.name}) }); - run.addArg(b.getInstallPath(.lib, library.out_filename)); + run.addFileArg(b.graph.path(.install_lib, library.out_filename)); run.setCwd(options.cwd orelse b.path(".")); run.stdio = .inherit; - if (b.args) |args| run.addArgs(args); + run.addPassthruArgs(); return run; } @@ -417,7 +712,8 @@ pub const App = struct { run.addArgs(&.{ "--headless", "--fresh", "--script" }); run.addFileArg(script); run.addArg("--state"); - _ = run.addOutputFileArg(b.fmt("{s}.state", .{app.name})); + // Named per script: runs of one app may share an output directory. + _ = run.addOutputFileArg(b.fmt("{s}-{s}.state", .{ app.name, std.fs.path.stem(script.getDisplayName()) })); for (options.defines) |pair| run.addArgs(&.{ "--define", b.fmt("{s}={s}", .{ pair[0], pair[1] }) }); if (library) run.addArtifactArg(artifact); run.setCwd(options.cwd orelse b.path(".")); @@ -433,7 +729,7 @@ pub const App = struct { run.addArg(state orelse b.fmt("{s}.state", .{app.name})); run.setCwd(b.path(".")); run.stdio = .inherit; - if (b.args) |args| run.addArgs(args); + run.addPassthruArgs(); return run; } @@ -470,7 +766,8 @@ pub const App = struct { /// The conventional wiring. Installs the executable, and adds: /// `run` (host + hot reload), `lib` (rebuild the library; use `--watch`), - /// `web`, `probe`, `e2e` (scripts) and `test`. + /// `web` and `serve` (with `web` options), `probe`, `e2e` (scripts) and + /// `test`. pub fn addSteps(app: App, options: StepsOptions) Steps { const b = app.sb.b; const library = app.addLibrary(); @@ -484,7 +781,9 @@ pub const App = struct { const web = if (options.web) |web_options| blk: { const web = app.addWeb(web_options); - b.step("web", "Build the browser version into zig-out/web").dependOn(&b.addInstallDirectory(.{ .source_dir = web.dir, .install_dir = .prefix, .install_subdir = "web" }).step); + const install = b.addInstallDirectory(.{ .source_dir = web.dir, .install_dir = .prefix, .install_subdir = "web" }); + b.step("web", "Build the browser version into zig-out/web").dependOn(&install.step); + b.step("serve", "Serve zig-out/web over https on port 8042, to the local network too (-- -p PORT for another)").dependOn(&app.addServe(install).step); break :blk web; } else null; @@ -499,7 +798,7 @@ pub const App = struct { } }; -/// Fills `{title}` in the page template. +/// Fills `{title}` and `{head}` in the page template. const page_tool = \\const std = @import("std"); \\pub fn main(init: std.process.Init) !void { @@ -507,11 +806,13 @@ const page_tool = \\ _ = args.skip(); \\ const template = args.next().?; \\ const title = args.next().?; + \\ const head = args.next().?; \\ const out = args.next().?; \\ const cwd = std.Io.Dir.cwd(); \\ const arena = init.arena.allocator(); \\ const source = try cwd.readFileAlloc(init.io, template, arena, .limited(1 << 20)); - \\ const page = try std.mem.replaceOwned(u8, arena, source, "{title}", title); + \\ const titled = try std.mem.replaceOwned(u8, arena, source, "{title}", title); + \\ const page = try std.mem.replaceOwned(u8, arena, titled, "{head}", head); \\ try cwd.writeFile(init.io, .{ .sub_path = out, .data = page }); \\} ; diff --git a/build.zig.zon b/build.zig.zon index 7922d7f..4688729 100644 --- a/build.zig.zon +++ b/build.zig.zon @@ -3,15 +3,15 @@ .version = "0.1.0", .dependencies = .{ .sdl = .{ - .url = "git+https://github.com/castholm/SDL.git#fb2d799c4778832a34ccb3739e40dded700684bd", - .hash = "sdl-0.5.3+3.4.14-SDL--v4eqAGuIKFsspMVxBxZf1OIEmmH-yHDdEl9ZRdX", + .url = "git+https://github.com/castholm/SDL.git#88f45bb1621c4373dd3633280ca01446b3eb8b11", + .hash = "sdl-0.6.1-dev+3.4.18-SDL--pegqAGHW2CzWPJalwfjV_8Q5qdmSjk3uPRTsFUE", }, .translate_c = .{ - .url = "git+https://codeberg.org/ziglang/translate-c?ref=zig-0.16.x#6fe0ffc4549f15c5f2d9432c2b4460ba90ff85ac", - .hash = "translate_c-1.0.0-Q_BUWo_5BgD4flHdUhA31zOz0XvZk9k7lQv1ouzyNXj2", + .url = "git+https://codeberg.org/ziglang/translate-c?ref=zig-0.17.x#cfac21f3199b2b02ce3fd7db652aa456421fe4ab", + .hash = "translate_c-2.0.2-Q_BUWvBNBwDnyehSiuAJmdFcMYHYGA-Jc3KiYdrV2Dqx", }, }, - .minimum_zig_version = "0.16.0", - .paths = .{ "build.zig", "build.zig.zon", "src", "web" }, + .minimum_zig_version = "0.17.0", + .paths = .{ "build.zig", "build.zig.zon", "src", "vendor", "web" }, .fingerprint = 0x6d6d2e26aa60b45e, } diff --git a/src/abi.zig b/src/abi.zig index 4782599..ca1333e 100644 --- a/src/abi.zig +++ b/src/abi.zig @@ -13,6 +13,9 @@ const std = @import("std"); pub const layout = @import("layout.zig"); pub const Key = @import("keys.zig").Key; +/// Bytes read back from a GPU buffer (see `Command.download`). +pub const Download = struct { key: u32, bytes: []const u8 }; + pub const Rect = extern struct { x: f32 = 0, y: f32 = 0, @@ -71,11 +74,25 @@ pub const Event = extern struct { text, /// Focus loss, or the console took the pointer: drop armed gestures. cancel, + /// Horizontal wheel delta; appended to preserve recorded event values. + wheel_horizontal, _, }; /// `flags` bit for a held key's auto-repeat. pub const repeat: u8 = 1; + /// For mouse and wheel events: the modifier keys held as it happened, + /// as `Key.Mod` bits. Read this rather than `Input.held`, which is the + /// state after the frame's events, and misses modifiers that come with + /// no key events at all (a trackpad's pinch is ctrl+wheel). + pub fn modifiers(e: Event) u32 { + return e.flags >> 4; + } + /// `flags` for an event with `mods` (`Key.Mod` bits) held. + pub fn withModifiers(mods: u32) u8 { + return @intCast((mods & 0xf) << 4); + } + pub fn wheel(e: Event) f32 { return @bitCast(e.value); } @@ -112,25 +129,38 @@ pub const Input = struct { started: bool = false, /// The console is open and owns the keyboard. console: bool = false, + /// The canvas. width: f32 = 0, height: f32 = 0, + /// The room the host has for the canvas, in the canvas's units at 1:1: + /// the window's size (CSS pixels on the web). A sketch that sets its + /// canvas to it fills any window, a phone's included, at its own scale. + window_width: f32 = 0, + window_height: f32 = 0, mouse_x: f32 = 0, mouse_y: f32 = 0, /// Held mouse buttons, bit `button - 1`. buttons: u32 = 0, /// Held keys by USB HID scancode, maintained from the events fed so far. - keys: std.StaticBitSet(256) = .initEmpty(), + keys: std.StaticBitSet(256) = .empty, streams: []const Stream = &.{}, sample_rate: u32 = 48_000, + /// Mono samples at `sample_rate` recorded since the previous frame, while + /// `draw` declares `Command.microphone`. Only `draw` sees them: they come + /// from a device, not from events, so they do not replay. + microphone: []const f32 = &.{}, + /// Finished `Command.download`s, oldest first. Like the microphone, + /// only `draw` sees them, and they do not replay. + downloads: []const Download = &.{}, /// `key` is any enum whose values are HID scancodes, like `sketch.Key`. pub fn held(in: *const Input, key: anytype) bool { - return in.keys.isSet(@intFromEnum(key)); + return in.keys.isSet(@backingInt(key)); } /// Went down this frame; auto-repeats excluded. pub fn pressed(in: *const Input, key: anytype) bool { for (in.events) |e| { - if (e.kind == .key_down and e.code == @intFromEnum(key) and e.flags & Event.repeat == 0) return true; + if (e.kind == .key_down and e.code == @backingInt(key) and e.flags & Event.repeat == 0) return true; } return false; } @@ -141,7 +171,7 @@ pub const Input = struct { /// Follows mouse, button and key state through one fed event. pub fn track(in: *Input, e: Event) void { switch (e.kind) { - .mouse_move, .wheel => { + .mouse_move, .wheel, .wheel_horizontal => { in.mouse_x = e.x; in.mouse_y = e.y; }, @@ -155,7 +185,7 @@ pub const Input = struct { .key_up => in.keys.unset(@min(e.code, 255)), .cancel => { in.buttons = 0; - in.keys = .initEmpty(); + in.keys = .empty; }, else => {}, } @@ -167,6 +197,9 @@ pub const Input = struct { pub const Image = enum(u32) { none = 0, _ }; pub const Audio = enum(u32) { none = 0, _ }; +/// GPU memory made by `Host.newBuffer`, filled by `Command.upload` or by +/// compute shaders, and read by any shader. Contents stay on the GPU. +pub const Buffer = enum(u32) { none = 0, _ }; pub const Cursor = enum { hidden, arrow, pointer }; /// Voices are reconciled, not triggered: `draw` declares every voice that @@ -185,7 +218,10 @@ pub const Voice = struct { }; /// Slices must stay alive until the host consumes the frame, right after -/// `draw` returns: literals, the frame arena and session memory all qualify. +/// `draw` returns. `sketch.Frame` copies the small ones (text, GPU +/// descriptions, uniforms) into scratch memory itself; bulk data (vertices, +/// samples, pixels, uploads) must already live in scratch, session or store +/// memory, or be a literal. pub const Command = union(enum) { clear: Color, rect: struct { dst: Rect, color: Color }, @@ -205,6 +241,25 @@ pub const Command = union(enum) { /// Renders a GPU pass into an image made by `Host.newImage`, when /// `Host.gpu` says the host can. gpu: gpu.Pass, + /// Bytes into a buffer made by `Host.newBuffer`, at `offset`. + upload: struct { buffer: Buffer, offset: u32 = 0, bytes: []const u8 }, + /// Runs a compute shader over buffers (see `abi.gpu`). + compute: gpu.Dispatch, + /// Keeps the default recording device open while declared, like a + /// voice; what it hears arrives in `Input.microphone`. + microphone, + /// Asks the user for a file. The answer arrives later as the console + /// line `<command> <path>`, so it reaches the sketch's `command` (and + /// the console log) like anything typed. `pattern` filters by extension: + /// "wav;flac;ogg", or empty for any file. + open_file: struct { command: []const u8, pattern: []const u8 = "" }, + /// Copies `size` bytes of a buffer from `offset` back to the CPU. They + /// arrive in `Input.downloads` under `key` once the GPU has got there, + /// usually a frame or two later; the frame does not wait. + download: struct { buffer: Buffer, offset: u32 = 0, size: u32, key: u32 = 0 }, + /// Runs a console line after this frame, as if typed: the way `draw`, + /// which cannot write State, hands it something only it learns. + run: []const u8, quit, }; @@ -213,13 +268,44 @@ pub const Command = union(enum) { /// description (so an edited shader simply becomes a new pipeline after a /// reload), and runs the pass. Sketches never touch a graphics API. /// -/// Shaders are SPIR-V. Resources follow SDL_GPU's SPIR-V convention: -/// vertex uniforms in set 1, fragment uniforms in set 3, binding = slot. -/// Uniform bytes must follow std140. +/// Shaders are SPIR-V for native hosts and WGSL for the web host; the +/// build makes both from GLSL (`addShaders` in build.zig), and a sketch +/// passes whichever its build gave it. Resources follow SDL_GPU's SPIR-V +/// convention, which the WGSL keeps: vertex uniforms in set 1, fragment +/// uniforms in set 3, binding = slot. Uniform bytes must follow std140. pub const gpu = struct { pub const Shader = struct { - spirv: []const u8, + /// SPIR-V, or WGSL on the web. + code: []const u8, + uniform_buffers: u32 = 0, + /// Read-only storage buffers: set 0 in a vertex shader, set 2 in a + /// fragment shader. + storage_buffers: u32 = 0, + }; + + /// A compute shader. SDL_GPU's SPIR-V sets: read-only buffers in set 0, + /// read-write buffers in set 1, uniforms in set 2. + pub const Kernel = struct { + /// SPIR-V, or WGSL on the web. + code: []const u8, + readonly_buffers: u32 = 0, + readwrite_buffers: u32 = 0, uniform_buffers: u32 = 0, + /// The shader's `local_size`. + threads: [3]u32 = .{ 64, 1, 1 }, + }; + + /// One run of a kernel, in a compute pass of its own: everything it + /// writes is visible to the commands after it. + pub const Dispatch = struct { + /// Names it in GPU timings (`--profile-gpu`). + label: []const u8 = "", + kernel: Kernel, + readonly: []const Buffer = &.{}, + readwrite: []const Buffer = &.{}, + /// Uniform slot 0, std140. + uniforms: []const u8 = &.{}, + groups: [3]u32 = .{ 1, 1, 1 }, }; pub const Format = enum { float, float2, float3, float4, ubyte4_norm }; @@ -258,12 +344,17 @@ pub const gpu = struct { pub const Draw = struct { /// Index into `Pass.pipelines`. pipeline: u32, - vertices: []const u8, + /// Empty for a pipeline without attributes, whose vertex shader + /// makes its own vertices from `gl_VertexIndex`. + vertices: []const u8 = &.{}, vertex_count: u32, stencil_reference: u8 = 0, /// Uniform slot 0 of each stage. vertex_uniforms: []const u8 = &.{}, fragment_uniforms: []const u8 = &.{}, + /// Read-only storage buffers of each stage, binding = index. + vertex_storage: []const Buffer = &.{}, + fragment_storage: []const Buffer = &.{}, scissor: ?Rect = null, }; @@ -276,9 +367,9 @@ pub const gpu = struct { pub fn attributes(comptime V: type) []const Attribute { return comptime blk: { if (@typeInfo(V).@"struct".layout != .@"extern") @compileError(@typeName(V) ++ " must be extern to be a vertex"); - const fields = @typeInfo(V).@"struct".fields; - var out: [fields.len]Attribute = undefined; - for (fields, 0..) |f, i| out[i] = .{ .location = i, .offset = @offsetOf(V, f.name), .format = formatOf(f.type) }; + const s = @typeInfo(V).@"struct"; + var out: [s.field_names.len]Attribute = undefined; + for (s.field_names, s.field_types, 0..) |name, F, i| out[i] = .{ .location = i, .offset = @offsetOf(V, name), .format = formatOf(F) }; const frozen = out; break :blk &frozen; }; @@ -336,6 +427,8 @@ pub const gpu = struct { } pub const Pass = struct { + /// Names it in GPU timings (`--profile-gpu`). + label: []const u8 = "", /// The target; its size is the image's. image: Image, width: u32, @@ -366,14 +459,22 @@ pub const Host = struct { newImage: *const fn (*anyopaque, u32, u32) Image, /// Releases an image; its handle draws nothing afterwards. freeImage: *const fn (*anyopaque, Image) void, - /// Whether `Command.gpu` runs here: the host has a GPU device. + /// Whether `Command.gpu` and `Command.compute` run here: the host + /// has a GPU device. gpu: *const fn (*anyopaque) bool, + /// A GPU buffer of `size` bytes, zeroed; `.none` without a GPU. + newBuffer: *const fn (*anyopaque, u32) Buffer, + freeBuffer: *const fn (*anyopaque, Buffer) void, /// Zero while unknown. imageSize: *const fn (*anyopaque, Image) [2]u32, /// `.none` if the file does not load. loadSound: *const fn (*anyopaque, []const u8) Audio, - /// WAV bytes in memory. Not cached. `.none` if they do not decode. + /// WAV, FLAC or Ogg Vorbis bytes in memory. Not cached. `.none` if + /// they do not decode. decodeSound: *const fn (*anyopaque, []const u8) Audio, + /// A loaded sound's samples: mono, at `Input.sample_rate`, alive as + /// long as the host. Empty for `.none`. + samples: *const fn (*anyopaque, Audio) []const f32, /// The whole file, mapped for the life of the host; empty if missing. readFile: *const fn (*anyopaque, []const u8) []const u8, env: *const fn (*anyopaque, []const u8) ?[]const u8, @@ -387,7 +488,7 @@ pub const Host = struct { } pub fn print(h: Host, comptime fmt: []const u8, args: anytype) void { var buf: [512]u8 = undefined; - h.log(std.fmt.bufPrint(&buf, fmt, args) catch "(message too long)"); + h.log(std.mem.print(&buf, fmt, args) catch "(message too long)"); } pub fn loadImage(h: Host, path: []const u8) Image { return h.vtable.loadImage(h.ctx, path); @@ -404,6 +505,12 @@ pub const Host = struct { pub fn gpu(h: Host) bool { return h.vtable.gpu(h.ctx); } + pub fn newBuffer(h: Host, size: u32) Buffer { + return h.vtable.newBuffer(h.ctx, size); + } + pub fn freeBuffer(h: Host, buffer: Buffer) void { + h.vtable.freeBuffer(h.ctx, buffer); + } pub fn imageSize(h: Host, image: Image) [2]u32 { return h.vtable.imageSize(h.ctx, image); } @@ -413,6 +520,9 @@ pub const Host = struct { pub fn decodeSound(h: Host, wav: []const u8) Audio { return h.vtable.decodeSound(h.ctx, wav); } + pub fn samples(h: Host, audio: Audio) []const f32 { + return h.vtable.samples(h.ctx, audio); + } pub fn readFile(h: Host, path: []const u8) []const u8 { return h.vtable.readFile(h.ctx, path); } @@ -424,15 +534,24 @@ pub const Host = struct { } }; -/// What the host passes to every call. +/// What the host passes to every call: services, input, and memory by +/// lifetime, shortest first. pub const Context = struct { host: Host, input: *const Input, - /// Frame lifetime: reset after the host consumes the frame. - arena: std.mem.Allocator, - /// Session lifetime: survives reloads, released when the session resets. - gpa: std.mem.Allocator, - /// Filled by `draw`, using `arena`. + /// Frame lifetime: all of it is released after the host consumes the + /// frame. + scratch: std.mem.Allocator, + /// This process: survives reloads; frees one allocation at a time, and + /// releases the rest when Session's layout changes. + session: std.mem.Allocator, + /// On disk, next to the state file: survives reloads and restarts, at + /// the same addresses, so Store may point into it. + store: std.mem.Allocator, + /// The sketch's `Session` and `Store` values. + session_root: [*]u8, + store_root: [*]u8, + /// Filled by `draw`, using `scratch`. commands: *std.ArrayList(Command), }; @@ -470,12 +589,13 @@ pub const Info = struct { height: u32, state: Layout, session: Layout, + store: Layout, /// Console completions offered by the sketch. commands: []const []const u8 = &.{}, }; /// Which memory `sketch_defaults` fills. -pub const Memory = enum(u8) { state, session }; +pub const Memory = enum(u8) { state, session, store }; /// The entry points of a sketch, exported from libraries under these names. /// What became of a console line offered to a sketch. @@ -495,10 +615,11 @@ pub const Sketch = struct { init: *const fn (*const Context, [*]u8) callconv(.c) void, /// Deterministic: new state from state and input, nothing else. update: *const fn (*const Context, [*]u8) callconv(.c) void, - /// Presentation: reads state, may change session, appends commands. - draw: *const fn (*const Context, [*]const u8, [*]u8) callconv(.c) void, - /// A console line the host did not understand; true when handled. - command: *const fn (*const Context, [*]u8, [*]u8, *const []const u8) callconv(.c) Outcome, + /// Presentation: reads state, may change session and store, appends + /// commands. + draw: *const fn (*const Context, [*]const u8) callconv(.c) void, + /// A console line the host did not understand. + command: *const fn (*const Context, [*]u8, *const []const u8) callconv(.c) Outcome, pub const symbols = .{ .info = "sketch_info", @@ -514,7 +635,7 @@ pub const Sketch = struct { pub const hash: u64 = blk: { @setEvalBranchQuota(1_000_000); var h: u64 = 0; - for (.{ Rect, Color, Event, Mode, Stream, Input, Image, Audio, Cursor, Voice, Command, gpu.Pass, gpu.Pipeline, gpu.Draw, Host, Host.VTable, Context, Field, Layout, Info, Memory, Sketch }) |T| { + for (.{ Rect, Color, Event, Mode, Stream, Input, Image, Audio, Cursor, Voice, Command, gpu.Pass, gpu.Pipeline, gpu.Draw, gpu.Dispatch, Host, Host.VTable, Context, Field, Layout, Info, Memory, Sketch }) |T| { h = std.hash.Wyhash.hash(h, @typeName(T) ++ layout.signature(T)); } break :blk h; @@ -528,7 +649,7 @@ test "vertex attributes and std140 checks come from the types" { try std.testing.expectEqual(@as(u32, 12), attrs[2].offset); const Block = extern struct { color: [4]f32, size: [2]f32, pad: [2]f32, stops: [4][4]f32 }; comptime gpu.assertStd140(Block); - const p = gpu.pipeline(Vertex, .{ .vertex = .{ .spirv = "" }, .fragment = .{ .spirv = "" } }); + const p = gpu.pipeline(Vertex, .{ .vertex = .{ .code = "" }, .fragment = .{ .code = "" } }); try std.testing.expectEqual(@as(u32, 20), p.stride); } diff --git a/src/glsl.zig b/src/glsl.zig index 995985b..a1cd2d6 100644 --- a/src/glsl.zig +++ b/src/glsl.zig @@ -27,10 +27,10 @@ pub fn source(comptime T: type, comptime name: []const u8) []const u8 { comptime { @setEvalBranchQuota(100_000); const tag = T.glsl; - const fields = @typeInfo(T).@"struct".fields; + const s = @typeInfo(T).@"struct"; var out: []const u8 = "// Generated from the Zig struct " ++ @typeName(T) ++ "; do not edit.\n"; if (@TypeOf(tag) == @TypeOf(.enum_literal) and tag == .vertex) { - for (fields, 0..) |f, i| out = out ++ std.fmt.comptimePrint("layout(location = {d}) in {s} {s};\n", .{ i, typeName(f.type, .vertex), f.name }); + for (s.field_names, s.field_types, 0..) |name_, F, i| out = out ++ std.fmt.comptimePrint("layout(location = {d}) in {s} {s};\n", .{ i, typeName(F, .vertex), name_ }); } else if (@TypeOf(tag) == @TypeOf(.enum_literal) and tag == .plain) { out = out ++ "struct " ++ name ++ " {\n" ++ members(T) ++ "};\n"; } else if (@hasField(@TypeOf(tag), "uniform")) { @@ -46,15 +46,16 @@ pub fn source(comptime T: type, comptime name: []const u8) []const u8 { fn members(comptime T: type) []const u8 { comptime { var out: []const u8 = ""; - for (@typeInfo(T).@"struct".fields) |f| { - const array_of_composites = switch (@typeInfo(f.type)) { + const s = @typeInfo(T).@"struct"; + for (s.field_names, s.field_types) |name, F| { + const array_of_composites = switch (@typeInfo(F)) { .array => |a| @typeInfo(a.child) == .array or @typeInfo(a.child) == .@"struct", else => false, }; out = out ++ if (array_of_composites) - std.fmt.comptimePrint(" {s} {s}[{d}];\n", .{ typeName(@typeInfo(f.type).array.child, .uniform), f.name, @typeInfo(f.type).array.len }) + std.fmt.comptimePrint(" {s} {s}[{d}];\n", .{ typeName(@typeInfo(F).array.child, .uniform), name, @typeInfo(F).array.len }) else - std.fmt.comptimePrint(" {s} {s};\n", .{ typeName(f.type, .uniform), f.name }); + std.fmt.comptimePrint(" {s} {s};\n", .{ typeName(F, .uniform), name }); } return out; } @@ -103,9 +104,10 @@ pub fn generate(comptime root: type, comptime paths: []const []const u8, io: std pub fn assertStd140(comptime U: type) void { comptime { if (@typeInfo(U).@"struct".layout != .@"extern") @compileError(@typeName(U) ++ " must be extern to be a uniform block"); - for (@typeInfo(U).@"struct".fields) |f| { - const want = std140Align(f.type); - if (@offsetOf(U, f.name) % want != 0) @compileError(std.fmt.comptimePrint("{s}.{s} sits at {d}; std140 wants a multiple of {d}", .{ @typeName(U), f.name, @offsetOf(U, f.name), want })); + const s = @typeInfo(U).@"struct"; + for (s.field_names, s.field_types) |name, F| { + const want = std140Align(F); + if (@offsetOf(U, name) % want != 0) @compileError(std.fmt.comptimePrint("{s}.{s} sits at {d}; std140 wants a multiple of {d}", .{ @typeName(U), name, @offsetOf(U, name), want })); } if (@sizeOf(U) % 16 != 0) @compileError(@typeName(U) ++ " must be a multiple of 16 bytes"); } diff --git a/src/host/Console.zig b/src/host/Console.zig index e6be2f0..b22facb 100644 --- a/src/host/Console.zig +++ b/src/host/Console.zig @@ -95,7 +95,7 @@ fn complete(self: *Console, extra: []const []const u8) void { first = candidate; common = candidate.len; } - const match = std.fmt.bufPrint(matches[used..], "{s} ", .{candidate}) catch continue; + const match = std.mem.print(matches[used..], "{s} ", .{candidate}) catch continue; used += match.len; count += 1; }; @@ -120,7 +120,7 @@ pub fn log(self: *Console, message: []const u8) void { pub fn print(self: *Console, comptime fmt: []const u8, args: anytype) void { var buf: [512]u8 = undefined; - self.log(std.fmt.bufPrint(&buf, fmt, args) catch "(message too long)"); + self.log(std.mem.print(&buf, fmt, args) catch "(message too long)"); } /// Runs one line. Host commands first; anything else goes to the sketch. diff --git a/src/host/Host.zig b/src/host/Host.zig index d511099..7ebe54a 100644 --- a/src/host/Host.zig +++ b/src/host/Host.zig @@ -25,6 +25,8 @@ pub const Options = struct { headless: bool = false, /// Forget the state file's contents: scripts start from a known place. fresh: bool = false, + /// Time every GPU command alone and print the times at exit. + profile_gpu: bool = false, }; gpa: std.mem.Allocator, @@ -45,6 +47,10 @@ images: gfx.Images, audio: sound.Audio, state: memory.StateFile, session: memory.Session, +store: memory.StoreFile, +/// Images and buffers the sketch made (`newImage`, `decodeImage`, +/// `newBuffer`): Session holds their handles, so they go when it resets. +made: struct { images: std.ArrayList(abi.Image) = .empty, buffers: std.ArrayList(abi.Buffer) = .empty } = .{}, recordings: []align(memory.page) u8, timeline: Timeline, console: Console = .{}, @@ -71,6 +77,7 @@ started: bool = true, last_now: u64 = 0, /// Saved after the next frame is drawn (scripts). screenshot: ?[:0]const u8 = null, +dialog: Dialog = .{}, pub fn init(self: *Host, gpa: std.mem.Allocator, io: std.Io, environ: *const std.process.Environ.Map, options: Options, linked: ?abi.Sketch) !void { if (options.headless) { @@ -93,7 +100,7 @@ pub fn init(self: *Host, gpa: std.mem.Allocator, io: std.Io, environ: *const std const info = sketch.info(); var title_buf: [256]u8 = undefined; - const title = std.fmt.bufPrintZ(&title_buf, "{s}", .{info.name}) catch "sketch"; + const title = std.mem.printSentinel(&title_buf, "{s}", .{info.name}, 0) catch "sketch"; const window = c.SDL_CreateWindow(title.ptr, @intCast(info.width), @intCast(info.height), c.SDL_WINDOW_RESIZABLE) orelse return sdlError("SDL_CreateWindow"); errdefer c.SDL_DestroyWindow(window); // The 2D renderer runs on the GPU device that also runs sketches' GPU @@ -117,8 +124,14 @@ pub fn init(self: *Host, gpa: std.mem.Allocator, io: std.Io, environ: *const std var state = try memory.StateFile.open(gpa, options.state); errdefer state.close(); var inputs_buf: [std.fs.max_path_bytes]u8 = undefined; - const recordings = try mapRecordings(try std.fmt.bufPrint(&inputs_buf, "{s}.inputs", .{options.state})); + const recordings = try mapRecordings(try std.mem.print(&inputs_buf, "{s}.inputs", .{options.state})); errdefer std.posix.munmap(recordings); + var store_buf: [std.fs.max_path_bytes]u8 = undefined; + var store = memory.StoreFile.open(gpa, try std.mem.print(&store_buf, "{s}.store", .{options.state}), options.fresh) catch |err| { + if (err == error.StoreInUse) std.log.err("{s}.store is open in another process", .{options.state}); + return err; + }; + errdefer store.close(); self.* = .{ .gpa = gpa, @@ -132,6 +145,7 @@ pub fn init(self: *Host, gpa: std.mem.Allocator, io: std.Io, environ: *const std .audio = .{ .gpa = gpa }, .state = state, .session = .init(gpa), + .store = store, .recordings = recordings, .timeline = undefined, .sketch = sketch, @@ -141,6 +155,9 @@ pub fn init(self: *Host, gpa: std.mem.Allocator, io: std.Io, environ: *const std }; self.timeline = .{ .bank = .{ .state = &self.state, .inputs = recordings } }; if (gpu) |device| self.passes = .init(gpa, device); + if (options.profile_gpu) if (self.passes) |*g| { + g.profile = .init(gpa); + }; self.audio.open() catch |err| std.log.warn("audio unavailable: {t}", .{err}); if (options.library) |path| if (linked == null) { self.watch = Watch.init(path) catch null; @@ -156,6 +173,7 @@ pub fn init(self: *Host, gpa: std.mem.Allocator, io: std.Io, environ: *const std } pub fn deinit(self: *Host) void { + self.dialog.deinit(); if (self.watch) |w| w.deinit(); self.audio.deinit(); if (self.passes) |*g| g.deinit(&self.images); @@ -173,6 +191,9 @@ pub fn deinit(self: *Host) void { self.files.deinit(self.gpa); std.posix.munmap(self.recordings); self.session.deinit(self.gpa); + self.store.close(); + self.made.images.deinit(self.gpa); + self.made.buffers.deinit(self.gpa); self.state.close(); self.pending.deinit(self.gpa); self.frame_arena.deinit(); @@ -242,16 +263,30 @@ fn draw(self: *Host, now: u64) !void { self.input.console = self.console.visible; self.input.width = @floatFromInt(self.canvas[0]); self.input.height = @floatFromInt(self.canvas[1]); + // Headless runs have no window to fill: the canvas is all the room. + self.input.window_width = self.input.width; + self.input.window_height = self.input.height; + if (!self.options.headless) { + var w: c_int = 0; + var h: c_int = 0; + if (c.SDL_GetWindowSize(self.window, &w, &h) and w > 0 and h > 0) { + self.input.window_width = @floatFromInt(w); + self.input.window_height = @floatFromInt(h); + } + } self.input.streams = self.audio.streamStatus(); self.input.sample_rate = sound.sample_rate; + self.input.microphone = self.audio.record(arena); + self.input.downloads = if (self.passes) |*g| try g.collect(arena) else &.{}; self.input.frame += 1; const ctx = self.context(); - self.sketch.draw(&ctx, self.live().ptr, self.session.bytes.ptr); + self.sketch.draw(&ctx, self.live().ptr); _ = c.SDL_SetRenderClipRect(self.renderer, null); _ = c.SDL_SetRenderDrawColor(self.renderer, 0, 0, 0, 255); _ = c.SDL_RenderClear(self.renderer); self.audio.beginFrame(); + var lines: std.ArrayList([]const u8) = .empty; for (self.commands.items) |command| switch (command) { .canvas => |size| self.setCanvas(size.w, size.h), .cursor => |kind| self.cursor = kind, @@ -259,9 +294,16 @@ fn draw(self: *Host, now: u64) !void { .audio => |a| self.audio.push(a.key, a.samples, a.clear), .quit => self.quit = true, .gpu => |pass| if (self.passes) |*g| g.run(&self.images, pass) catch |err| self.print("gpu pass: {t} ({s})", .{ err, c.SDL_GetError() }), + .upload => |u| if (self.passes) |*g| g.write(u.buffer, u.offset, u.bytes) catch |err| self.print("gpu upload: {t} ({s})", .{ err, c.SDL_GetError() }), + .compute => |d| if (self.passes) |*g| g.compute(d) catch |err| self.print("gpu compute: {t} ({s})", .{ err, c.SDL_GetError() }), + .microphone => self.audio.listen(), + .download => |d| if (self.passes) |*g| g.download(d) catch |err| self.print("gpu download: {t} ({s})", .{ err, c.SDL_GetError() }), + .open_file => |o| self.openFile(o.command, o.pattern), + .run => |line| try lines.append(arena, line), else => gfx.draw(self.renderer, &self.images, command), }; self.audio.endFrame(); + for (lines.items) |line| self.runLine(line, now) catch {}; _ = c.SDL_SetRenderClipRect(self.renderer, null); var extra: std.ArrayList(abi.Command) = .empty; @@ -284,12 +326,72 @@ fn draw(self: *Host, now: u64) !void { if (self.passes) |*g| g.endFrame(); } +/// A file dialog: shown on request, its answer run as a console line from +/// the main loop. SDL may answer on another thread. +const Dialog = struct { + mutex: ?*c.SDL_Mutex = null, + /// Shown and not yet answered (main thread only). + showing: bool = false, + command: [64]u8 = undefined, + command_len: usize = 0, + /// The filter SDL reads until it answers. + pattern: [128:0]u8 = @splat(0), + filter: c.SDL_DialogFileFilter = undefined, + /// "<command> <path>" once answered; empty when cancelled. + answer: [std.fs.max_path_bytes + 64]u8 = undefined, + answer_len: ?usize = null, + + fn deinit(self: *Dialog) void { + if (self.mutex) |m| c.SDL_DestroyMutex(m); + self.mutex = null; + } + + fn callback(userdata: ?*anyopaque, files: [*c]const [*c]const u8, _: c_int) callconv(.c) void { + const self: *Dialog = @ptrCast(@alignCast(userdata orelse return)); + const m = self.mutex orelse return; + c.SDL_LockMutex(m); + defer c.SDL_UnlockMutex(m); + const first = if (files != null and files[0] != null) std.mem.span(files[0]) else ""; + const line = if (first.len == 0) "" else std.mem.print(&self.answer, "{s} {s}", .{ self.command[0..self.command_len], first }) catch ""; + self.answer_len = line.len; + } + + /// The answer's console line, once; empty when cancelled. + fn take(self: *Dialog, out: []u8) ?[]const u8 { + const m = self.mutex orelse return null; + c.SDL_LockMutex(m); + defer c.SDL_UnlockMutex(m); + const len = self.answer_len orelse return null; + self.answer_len = null; + self.showing = false; + @memcpy(out[0..len], self.answer[0..len]); + return out[0..len]; + } +}; + +fn openFile(self: *Host, command: []const u8, pattern: []const u8) void { + const d = &self.dialog; + if (self.options.headless or d.showing) return; + if (command.len > d.command.len or pattern.len > d.pattern.len) return self.print("open_file: command or pattern too long", .{}); + if (d.mutex == null) d.mutex = c.SDL_CreateMutex() orelse return self.print("open_file: {s}", .{c.SDL_GetError()}); + @memcpy(d.command[0..command.len], command); + d.command_len = command.len; + @memset(&d.pattern, 0); + @memcpy(d.pattern[0..pattern.len], pattern); + d.filter = .{ .name = "Files", .pattern = &d.pattern }; + d.showing = true; + c.SDL_ShowOpenFileDialog(Dialog.callback, d, self.window, if (pattern.len > 0) &d.filter else null, @intFromBool(pattern.len > 0), null, false); +} + fn context(self: *Host) abi.Context { return .{ .host = .{ .ctx = self, .vtable = &api }, .input = &self.input, - .arena = self.frame_arena.allocator(), - .gpa = self.session.heap.allocator(), + .scratch = self.frame_arena.allocator(), + .session = self.session.heap.allocator(), + .store = self.store.allocator(), + .session_root = self.session.bytes.ptr, + .store_root = self.store.root(), .commands = &self.commands, }; } @@ -298,7 +400,7 @@ fn context(self: *Host) abi.Context { /// console; while it is open it owns the keyboard, and pointer events over it /// become a cancel, so a half-made click never lands behind it. pub fn route(self: *Host, e: abi.Event) !void { - if (e.kind == .key_down and e.code == @intFromEnum(abi.Key.grave)) { + if (e.kind == .key_down and e.code == @backingInt(abi.Key.grave)) { if (e.flags & abi.Event.repeat != 0) return; self.console.visible = !self.console.visible; return self.pending.append(self.gpa, .{ .kind = .cancel, .time = e.time }); @@ -316,7 +418,7 @@ pub fn route(self: *Host, e: abi.Event) !void { return; }, .key_up => return, - .mouse_move, .mouse_down, .mouse_up, .wheel => if (e.y >= @as(f32, @floatFromInt(self.canvas[1])) - Console.panel_height) { + .mouse_move, .mouse_down, .mouse_up, .wheel, .wheel_horizontal => if (e.y >= @as(f32, @floatFromInt(self.canvas[1])) - Console.panel_height) { return self.pending.append(self.gpa, .{ .kind = .cancel, .time = e.time }); }, else => {}, @@ -339,7 +441,7 @@ pub fn runLine(self: *Host, line: []const u8, now: u64) !void { pub fn forward(self: *Host, _: *Console, line: []const u8, now: u64) abi.Outcome { self.input.now = now; const ctx = self.context(); - return self.sketch.command(&ctx, self.live().ptr, self.session.bytes.ptr, &line); + return self.sketch.command(&ctx, self.live().ptr, &line); } /// Fresh state: declared defaults, then the sketch's `init`. @@ -373,6 +475,8 @@ fn adopt(self: *Host, sketch: abi.Sketch) !void { if (info.abi != abi.hash) return error.AbiMismatch; const adoption = try self.state.adopt(self.gpa, info.state, sketch.defaults); const session_reset = try self.session.adopt(self.gpa, info.session, sketch.defaults); + if (session_reset) self.releaseMade(); + const stored = try self.store.adopt(self.gpa, info.store, sketch.defaults); _ = self.info_arena.reset(.retain_capacity); const arena = self.info_arena.allocator(); self.name = try arena.dupe(u8, info.name); @@ -388,6 +492,33 @@ fn adopt(self: *Host, sketch: abi.Sketch) !void { } if (!self.state.used(0)) self.resetState(); if (session_reset and self.reloads > 0) self.print("session layout changed: session reset", .{}); + switch (stored) { + .unchanged, .created => {}, + .migrated => |m| self.print("store layout changed: kept {d} fields, reset {d} to defaults", .{ m.kept, m.reset }), + } +} + +/// Frees what the old Session made: its handles are gone with it. +fn releaseMade(self: *Host) void { + for (self.made.images.items) |image| { + if (self.passes) |*g| g.forget(&self.images, image); + self.images.free(image); + } + if (self.passes) |*g| for (self.made.buffers.items) |buffer| g.freeBuffer(buffer); + self.made.images.clearRetainingCapacity(); + self.made.buffers.clearRetainingCapacity(); +} + +fn remember(self: *Host, comptime kind: enum { images, buffers }, handle: anytype) @TypeOf(handle) { + if (handle != .none) @field(self.made, @tagName(kind)).append(self.gpa, handle) catch {}; + return handle; +} + +fn forget(list: anytype, handle: anytype) void { + for (list.items, 0..) |h, i| if (h == handle) { + _ = list.swapRemove(i); + return; + }; } fn setCanvas(self: *Host, w: u32, h: u32) void { @@ -421,7 +552,7 @@ fn applyCursor(self: *Host) void { pub fn print(self: *Host, comptime fmt: []const u8, args: anytype) void { var buf: [512]u8 = undefined; - const message = std.fmt.bufPrint(&buf, fmt, args) catch "(message too long)"; + const message = std.mem.print(&buf, fmt, args) catch "(message too long)"; self.console.log(message); // Headless runs are tests: they report through their own results. if (!self.options.headless) std.debug.print("{s}\n", .{message}); @@ -436,18 +567,21 @@ pub fn poll(self: *Host) !void { switch (event.type) { c.SDL_EVENT_QUIT => self.quit = true, c.SDL_EVENT_WINDOW_FOCUS_LOST => try self.route(.{ .kind = .cancel, .time = time }), - c.SDL_EVENT_MOUSE_MOTION => try self.route(.{ .kind = .mouse_move, .time = time, .x = event.motion.x, .y = event.motion.y }), + c.SDL_EVENT_MOUSE_MOTION => try self.route(.{ .kind = .mouse_move, .time = time, .x = event.motion.x, .y = event.motion.y, .flags = held() }), c.SDL_EVENT_MOUSE_BUTTON_DOWN, c.SDL_EVENT_MOUSE_BUTTON_UP => try self.route(.{ .kind = if (event.type == c.SDL_EVENT_MOUSE_BUTTON_DOWN) .mouse_down else .mouse_up, .time = time, .x = event.button.x, .y = event.button.y, .code = event.button.button, + .flags = held(), }), c.SDL_EVENT_MOUSE_WHEEL => { - const delta = event.wheel.y * @as(f32, if (event.wheel.direction == c.SDL_MOUSEWHEEL_FLIPPED) -1 else 1); - if (!std.math.isFinite(delta) or delta == 0) continue; - try self.route(.{ .kind = .wheel, .time = time, .x = event.wheel.mouse_x, .y = event.wheel.mouse_y, .value = @bitCast(delta) }); + const direction: f32 = if (event.wheel.direction == c.SDL_MOUSEWHEEL_FLIPPED) -1 else 1; + const dx = event.wheel.x * direction; + const dy = event.wheel.y * direction; + if (std.math.isFinite(dy) and dy != 0) try self.route(.{ .kind = .wheel, .time = time, .x = event.wheel.mouse_x, .y = event.wheel.mouse_y, .value = @bitCast(dy), .flags = held() }); + if (std.math.isFinite(dx) and dx != 0) try self.route(.{ .kind = .wheel_horizontal, .time = time, .x = event.wheel.mouse_x, .y = event.wheel.mouse_y, .value = @bitCast(dx), .flags = held() }); }, c.SDL_EVENT_KEY_DOWN, c.SDL_EVENT_KEY_UP => try self.route(.{ .kind = if (event.type == c.SDL_EVENT_KEY_DOWN) .key_down else .key_up, @@ -463,6 +597,16 @@ pub fn poll(self: *Host) !void { else => {}, } } + var buf: [@sizeOf(@FieldType(Dialog, "answer"))]u8 = undefined; + if (self.dialog.take(&buf)) |line| if (line.len > 0) { + self.console.print("> {s}", .{line}); + self.runLine(line, c.SDL_GetTicksNS()) catch {}; + }; +} + +/// The modifiers held now, for a pointer event's `flags`. +fn held() u8 { + return abi.Event.withModifiers(modifiers(c.SDL_GetModState())); } fn modifiers(mod: c.SDL_Keymod) u32 { @@ -475,7 +619,7 @@ fn modifiers(mod: c.SDL_Keymod) u32 { } fn consoleKey(e: abi.Event) ?Console.Key { - const key: abi.Key = @enumFromInt(@as(u8, @truncate(e.code))); + const key: abi.Key = @fromBackingInt(@intCast(@as(u8, @truncate(e.code)))); if (key == .u and e.value & abi.Key.Mod.ctrl != 0) return .clear; if (e.flags & abi.Event.repeat != 0 and key.isEnter()) return null; if (key.isEnter()) return .submit; @@ -509,9 +653,12 @@ const api: abi.Host.VTable = .{ .newImage = apiNewImage, .freeImage = apiFreeImage, .gpu = apiGpu, + .newBuffer = apiNewBuffer, + .freeBuffer = apiFreeBuffer, .imageSize = apiImageSize, .loadSound = apiLoadSound, .decodeSound = apiDecodeSound, + .samples = apiSamples, .readFile = apiReadFile, .env = apiEnv, .clock = apiClock, @@ -527,19 +674,38 @@ fn apiLoadImage(ctx: *anyopaque, path: []const u8) abi.Image { return hostOf(ctx).images.load(path); } fn apiDecodeImage(ctx: *anyopaque, png: []const u8) abi.Image { - return hostOf(ctx).images.decode(png); + const self = hostOf(ctx); + return self.remember(.images, self.images.decode(png)); } fn apiNewImage(ctx: *anyopaque, w: u32, h: u32) abi.Image { - return hostOf(ctx).images.new(w, h); + const self = hostOf(ctx); + return self.remember(.images, self.images.new(w, h)); } fn apiFreeImage(ctx: *anyopaque, image: abi.Image) void { const self = hostOf(ctx); + forget(&self.made.images, image); if (self.passes) |*g| g.forget(&self.images, image); self.images.free(image); } fn apiGpu(ctx: *anyopaque) bool { return hostOf(ctx).gpu != null; } +fn apiNewBuffer(ctx: *anyopaque, size: u32) abi.Buffer { + const self = hostOf(ctx); + const g = &(self.passes orelse return .none); + return self.remember(.buffers, g.newBuffer(size) catch |err| blk: { + self.print("gpu buffer: {t} ({s})", .{ err, c.SDL_GetError() }); + break :blk .none; + }); +} +fn apiFreeBuffer(ctx: *anyopaque, buffer: abi.Buffer) void { + const self = hostOf(ctx); + forget(&self.made.buffers, buffer); + if (self.passes) |*g| g.freeBuffer(buffer); +} +fn apiSamples(ctx: *anyopaque, audio: abi.Audio) []const f32 { + return hostOf(ctx).audio.samplesOf(audio); +} fn apiImageSize(ctx: *anyopaque, image: abi.Image) [2]u32 { return hostOf(ctx).images.size(image); } diff --git a/src/host/audio.zig b/src/host/audio.zig index 87332b1..aebb909 100644 --- a/src/host/audio.zig +++ b/src/host/audio.zig @@ -1,9 +1,11 @@ -//! Sound output: decoded clips played by reconciled voices, plus raw streams -//! the sketch fills itself. The audio thread only ever sees SDL's own copies -//! of the samples; sketch code and memory never run on it. +//! Sound: decoded clips played by reconciled voices, raw streams the sketch +//! fills itself, and the microphone, open while declared. The audio thread +//! only ever sees SDL's own copies of the samples; sketch code and memory +//! never run on it. const std = @import("std"); const c = @import("c"); const abi = @import("abi"); +const memory = @import("memory.zig"); pub const sample_rate = 48_000; const spec: c.SDL_AudioSpec = .{ .format = c.SDL_AUDIO_F32, .channels = 1, .freq = sample_rate }; @@ -21,6 +23,11 @@ pub const Audio = struct { streams: std.ArrayList(Stream) = .empty, status: std.ArrayList(abi.Stream) = .empty, scope: Scope = .{}, + /// The default recording device, bound to this stream while declared. + microphone: ?*c.SDL_AudioStream = null, + listening: bool = false, + /// Set when opening failed, so it is tried once per declaration. + deaf: bool = false, const Playing = struct { voice: abi.Voice, cursor: usize = 0, done: bool = false, seen: bool = true }; const Stream = struct { key: u32, out: *c.SDL_AudioStream, pushed: u64 = 0 }; @@ -34,6 +41,7 @@ pub const Audio = struct { } pub fn deinit(self: *Audio) void { + if (self.microphone) |m| c.SDL_DestroyAudioStream(m); self.scope.deinit(); for (self.streams.items) |s| c.SDL_DestroyAudioStream(s.out); if (self.voices_out) |s| c.SDL_DestroyAudioStream(s); @@ -70,7 +78,7 @@ pub const Audio = struct { /// `.none` when `wav` does not decode. pub fn decode(self: *Audio, wav: []const u8) abi.Audio { - return self.add(decodeWav(self.gpa, wav) catch return .none); + return self.add(decodeBytes(self.gpa, wav) catch return .none); } fn add(self: *Audio, samples: []f32) abi.Audio { @@ -78,17 +86,52 @@ pub const Audio = struct { self.gpa.free(samples); return .none; }; - return @enumFromInt(self.clips.items.len); + return @fromBackingInt(@intCast(self.clips.items.len)); } - fn samplesOf(self: *Audio, audio: abi.Audio) []const f32 { - const index = @intFromEnum(audio); + pub fn samplesOf(self: *Audio, audio: abi.Audio) []const f32 { + const index = @backingInt(audio); if (index == 0 or index > self.clips.items.len) return &.{}; return self.clips.items[index - 1]; } pub fn beginFrame(self: *Audio) void { for (self.voices.items) |*v| v.seen = false; + self.listening = false; + } + + /// The microphone was declared this frame. + pub fn listen(self: *Audio) void { + self.listening = true; + } + + /// What the microphone heard since the last call, into `arena`. + pub fn record(self: *Audio, arena: std.mem.Allocator) []const f32 { + const m = self.microphone orelse return &.{}; + const available = c.SDL_GetAudioStreamAvailable(m); + if (available <= 0) return &.{}; + const samples = arena.alloc(f32, @as(usize, @intCast(available)) / @sizeOf(f32)) catch return &.{}; + const got = c.SDL_GetAudioStreamData(m, samples.ptr, @intCast(samples.len * @sizeOf(f32))); + if (got <= 0) return &.{}; + return samples[0 .. @as(usize, @intCast(got)) / @sizeOf(f32)]; + } + + /// Opens or closes the microphone to match this frame's declaration. + fn reconcileMicrophone(self: *Audio) void { + if (!self.listening) { + if (self.microphone) |m| c.SDL_DestroyAudioStream(m); + self.microphone = null; + self.deaf = false; + return; + } + if (self.microphone != null or self.deaf) return; + const m = c.SDL_OpenAudioDeviceStream(c.SDL_AUDIO_DEVICE_DEFAULT_RECORDING, &spec, null, null) orelse { + std.log.warn("microphone: {s}", .{c.SDL_GetError()}); + self.deaf = true; + return; + }; + if (!c.SDL_ResumeAudioStreamDevice(m)) std.log.warn("microphone: {s}", .{c.SDL_GetError()}); + self.microphone = m; } /// Reconciles one declared voice with what is playing. @@ -128,8 +171,10 @@ pub const Audio = struct { return &self.streams.items[self.streams.items.len - 1]; } - /// Drops undeclared voices and keeps the voice mix queued ahead. + /// Drops undeclared voices, keeps the voice mix queued ahead, and opens + /// or closes the microphone. pub fn endFrame(self: *Audio) void { + self.reconcileMicrophone(); var i: usize = 0; while (i < self.voices.items.len) { if (self.voices.items[i].seen) i += 1 else _ = self.voices.swapRemove(i); @@ -183,28 +228,23 @@ pub const Audio = struct { } }; -fn decodeWav(gpa: std.mem.Allocator, wav: []const u8) ![]f32 { - const io = c.SDL_IOFromConstMem(wav.ptr, wav.len) orelse return error.OpenWav; - var from: c.SDL_AudioSpec = undefined; - var data: [*c]u8 = null; - var len: u32 = 0; - if (!c.SDL_LoadWAV_IO(io, true, &from, &data, &len)) return error.DecodeWav; - defer c.SDL_free(data); - return convert(gpa, &from, data[0..len]); +/// WAV, FLAC or Ogg Vorbis, by content (see `vendor/codecs.h`). +fn decodeBytes(gpa: std.mem.Allocator, bytes: []const u8) ![]f32 { + var channels: u32 = 0; + var rate: u32 = 0; + var frames: u64 = 0; + const raw = c.sketchbook_decode_audio(bytes.ptr, bytes.len, &channels, &rate, &frames) orelse return error.DecodeAudio; + defer c.sketchbook_free_audio(raw); + if (rate == 0 or rate > 192000 or channels == 0 or channels > 8 or frames == 0 or frames > @as(u64, rate) * 600) return error.UnsupportedSoundFormat; + const from: c.SDL_AudioSpec = .{ .format = c.SDL_AUDIO_F32, .channels = @intCast(channels), .freq = @intCast(rate) }; + return convert(gpa, &from, std.mem.sliceAsBytes(raw[0..@intCast(frames * channels)])); } fn decodeFile(gpa: std.mem.Allocator, path: []const u8) ![]f32 { - var buf: [std.fs.max_path_bytes]u8 = undefined; - const z = try std.fmt.bufPrintZ(&buf, "{s}", .{path}); - var info: c.SF_INFO = std.mem.zeroes(c.SF_INFO); - const file = c.sf_open(z.ptr, c.SFM_READ, &info) orelse return error.OpenSoundFile; - defer _ = c.sf_close(file); - if (info.samplerate <= 0 or info.samplerate > 192000 or info.channels <= 0 or info.channels > 8 or info.frames <= 0 or info.frames > @as(i64, info.samplerate) * 600) return error.UnsupportedSoundFormat; - const raw = try gpa.alloc(f32, @intCast(info.frames * info.channels)); - defer gpa.free(raw); - if (c.sf_readf_float(file, raw.ptr, info.frames) != info.frames) return error.TruncatedAudio; - const from: c.SDL_AudioSpec = .{ .format = c.SDL_AUDIO_F32, .channels = info.channels, .freq = info.samplerate }; - return convert(gpa, &from, std.mem.sliceAsBytes(raw)); + const bytes = memory.mapReadOnly(path); + if (bytes.len == 0) return error.OpenSoundFile; + defer std.posix.munmap(bytes); + return decodeBytes(gpa, bytes); } /// To host format: f32 mono at `sample_rate`, all finite. diff --git a/src/host/fields.zig b/src/host/fields.zig index 215f176..ba8fc03 100644 --- a/src/host/fields.zig +++ b/src/host/fields.zig @@ -100,7 +100,7 @@ test "fields format, parse and compare by kind" { const S = extern struct { mood: enum(u32) { calm, angry = 5 } = .calm, x: i16 = -3, speed: f32 = 1.5, on: bool = false }; var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); defer arena.deinit(); - const layout = try @import("memory.zig").Layout.copy(arena.allocator(), abi.layout.of(S, .forbidden)); + const layout = try @import("memory.zig").Layout.copy(arena.allocator(), abi.layout.of(S, .state)); var s: S = .{}; const bytes = std.mem.asBytes(&s); try std.testing.expect(matches(layout.find("x").?, bytes, "-3")); diff --git a/src/host/gfx.zig b/src/host/gfx.zig index e00d325..16148a6 100644 --- a/src/host/gfx.zig +++ b/src/host/gfx.zig @@ -27,7 +27,7 @@ pub const Images = struct { pub fn load(self: *Images, path: []const u8) abi.Image { if (self.paths.get(path)) |image| return image; var buf: [std.fs.max_path_bytes]u8 = undefined; - const z = std.fmt.bufPrintZ(&buf, "{s}", .{path}) catch return .none; + const z = std.mem.printSentinel(&buf, "{s}", .{path}, 0) catch return .none; const surface = c.SDL_LoadPNG(z.ptr); if (surface == null) std.log.warn("image {s}: {s}", .{ path, c.SDL_GetError() }); const image = self.add(surface); @@ -83,7 +83,7 @@ pub const Images = struct { } pub fn get(self: *Images, image: abi.Image) ?*Entry { - const index = @intFromEnum(image); + const index = @backingInt(image); if (index == 0 or index > self.entries.items.len) return null; return &self.entries.items[index - 1]; } @@ -99,7 +99,7 @@ pub const Images = struct { fn append(self: *Images, entry: Entry) abi.Image { self.entries.append(self.gpa, entry) catch return .none; - return @enumFromInt(self.entries.items.len); + return @fromBackingInt(@intCast(self.entries.items.len)); } fn setup(texture: *c.SDL_Texture) void { @@ -112,6 +112,7 @@ pub const Images = struct { pub fn draw(renderer: *c.SDL_Renderer, images: *Images, command: abi.Command) void { switch (command) { .clear => |color| { + if (@import("builtin").os.tag == .linux) c.sk_text_begin_frame(renderer); setColor(renderer, color); _ = c.SDL_RenderClear(renderer); }, @@ -146,8 +147,10 @@ pub fn draw(renderer: *c.SDL_Renderer, images: *Images, command: abi.Command) vo } } -/// SDL's built-in 8x8 font, `scale` times larger. Stops at the first NUL. +/// Native shaped text on Linux, using a 13px monospace face at scale 1. +/// Other targets retain SDL's built-in 8x8 fallback. Coordinates are line-box tops. pub fn drawText(renderer: *c.SDL_Renderer, x: f32, y: f32, scale: f32, bytes: []const u8, color: abi.Color) void { + if (@import("builtin").os.tag == .linux and c.sk_text_draw(renderer, x, y, scale, bytes.ptr, bytes.len, color.r, color.g, color.b, color.a)) return; var buf: [1024]u8 = undefined; const len = @min(std.mem.indexOfScalar(u8, bytes, 0) orelse bytes.len, buf.len - 1); @memcpy(buf[0..len], bytes[0..len]); @@ -157,9 +160,22 @@ pub fn drawText(renderer: *c.SDL_Renderer, x: f32, y: f32, scale: f32, bytes: [] var sy: f32 = 1; _ = c.SDL_GetRenderScale(renderer, &sx, &sy); const s = @max(scale, 1); + // The clip rect scales with the renderer: shrink it to match, to the + // whole pixels around it (so it may let up to `s - 1` pixels more through). + var clip: c.SDL_Rect = undefined; + const clipped = c.SDL_RenderClipEnabled(renderer) and c.SDL_GetRenderClipRect(renderer, &clip); _ = c.SDL_SetRenderScale(renderer, sx * s, sy * s); + if (clipped and s != 1) { + const x0 = @floor(@as(f32, @floatFromInt(clip.x)) / s); + const y0 = @floor(@as(f32, @floatFromInt(clip.y)) / s); + const x1 = @ceil(@as(f32, @floatFromInt(clip.x + clip.w)) / s); + const y1 = @ceil(@as(f32, @floatFromInt(clip.y + clip.h)) / s); + const scaled: c.SDL_Rect = .{ .x = @intFromFloat(x0), .y = @intFromFloat(y0), .w = @intFromFloat(x1 - x0), .h = @intFromFloat(y1 - y0) }; + _ = c.SDL_SetRenderClipRect(renderer, &scaled); + } _ = c.SDL_RenderDebugText(renderer, x / s, y / s, &buf); _ = c.SDL_SetRenderScale(renderer, sx, sy); + if (clipped and s != 1) _ = c.SDL_SetRenderClipRect(renderer, &clip); } fn missing(renderer: *c.SDL_Renderer, dst: abi.Rect) void { diff --git a/src/host/gpu.zig b/src/host/gpu.zig index 34c817f..9c77205 100644 --- a/src/host/gpu.zig +++ b/src/host/gpu.zig @@ -1,7 +1,7 @@ -//! Runs `abi.gpu` passes with SDL_GPU: the translation from GPU work as data -//! to a real API. Shaders and pipelines are built on first use and cached by -//! the hash of their description; entries a sketch stops using (an edited -//! shader, after a reload) are released after a while. +//! Runs `abi.gpu` passes and dispatches with SDL_GPU: the translation from +//! GPU work as data to a real API. Shaders and pipelines are built on first +//! use and cached by the hash of their description; entries a sketch stops +//! using (an edited shader, after a reload) are released after a while. const Gpu = @This(); const std = @import("std"); const c = @import("c"); @@ -17,6 +17,11 @@ device: *c.SDL_GPUDevice, stencil_format: c.SDL_GPUTextureFormat, shaders: std.AutoHashMapUnmanaged(u64, Cached(c.SDL_GPUShader)) = .empty, pipelines: std.AutoHashMapUnmanaged(u64, Cached(c.SDL_GPUGraphicsPipeline)) = .empty, +kernels: std.AutoHashMapUnmanaged(u64, Cached(c.SDL_GPUComputePipeline)) = .empty, +/// Storage buffers by handle: `Buffer` n is `buffers[n - 1]`, null once freed. +buffers: std.ArrayList(?Storage) = .empty, +/// Uploads to storage buffers go through this, in chunks of its size. +staging: ?*c.SDL_GPUTransferBuffer = null, /// Render targets per image: the resolved texture the renderer shows, and /// multisampled and stencil attachments as passes ask for them. targets: std.AutoHashMapUnmanaged(abi.Image, Target) = .empty, @@ -25,11 +30,89 @@ vertex_capacity: u32 = 0, upload: ?*c.SDL_GPUTransferBuffer = null, upload_capacity: u32 = 0, frame: u64 = 0, +/// Set by `--profile-gpu`: every command is timed alone (see `Profile`). +profile: ?Profile = null, +/// Downloads in flight, oldest first, and transfer buffers to reuse. +downloads: std.ArrayList(Download) = .empty, +spare: std.ArrayList(Transfer) = .empty, + +const Transfer = struct { buffer: *c.SDL_GPUTransferBuffer, size: u32 }; +const Download = struct { key: u32, size: u32, transfer: Transfer, fence: *c.SDL_GPUFence }; + +/// GPU time per labelled command. Each command starts on an idle GPU and is +/// waited for, so its time is its own: the total of a frame is what the +/// GPU spends on it, though the waiting makes the frame itself slower. +pub const Profile = struct { + entries: std.StringArrayHashMapUnmanaged(Entry) = .empty, + arena: std.heap.ArenaAllocator, + + const Entry = struct { count: u64 = 0, total: u64 = 0, worst: u64 = 0 }; + + pub fn init(gpa: std.mem.Allocator) Profile { + return .{ .arena = .init(gpa) }; + } + + pub fn deinit(p: *Profile) void { + p.arena.deinit(); + } + + fn add(p: *Profile, label: []const u8, ns: u64) void { + const a = p.arena.allocator(); + const entry = p.entries.getOrPut(a, label) catch return; + if (!entry.found_existing) { + entry.key_ptr.* = a.dupe(u8, label) catch "?"; + entry.value_ptr.* = .{}; + } + entry.value_ptr.count += 1; + entry.value_ptr.total += ns; + entry.value_ptr.worst = @max(entry.value_ptr.worst, ns); + } + + pub fn report(p: *Profile, frames: u64) void { + std.debug.print("gpu time over {d} frames (ms)\n{s: <24} {s: >7} {s: >9} {s: >9} {s: >11}\n", .{ frames, "command", "count", "mean", "worst", "per frame" }); + var sum: u64 = 0; + var it = p.entries.iterator(); + while (it.next()) |e| { + const v = e.value_ptr.*; + sum += v.total; + std.debug.print("{s: <24} {d: >7} {d: >9.3} {d: >9.3} {d: >11.3}\n", .{ e.key_ptr.*, v.count, ms(v.total) / @as(f64, @floatFromInt(v.count)), ms(v.worst), ms(v.total) / @as(f64, @floatFromInt(@max(frames, 1))) }); + } + std.debug.print("{s: <24} {s: >7} {s: >9} {s: >9} {d: >11.3}\n", .{ "all", "", "", "", ms(sum) / @as(f64, @floatFromInt(@max(frames, 1))) }); + } + + fn ms(ns: u64) f64 { + return @as(f64, @floatFromInt(ns)) / std.time.ns_per_ms; + } +}; + +/// Submits `cmd`; when profiling, waits for it and books its time from +/// `start` (taken on an idle GPU, before recording) under `label`. +fn submit(g: *Gpu, cmd: *c.SDL_GPUCommandBuffer, label: []const u8, start: u64) !void { + const p = &(g.profile orelse { + if (!c.SDL_SubmitGPUCommandBuffer(cmd)) return error.GpuSubmit; + return; + }); + const fence = c.SDL_SubmitGPUCommandBufferAndAcquireFence(cmd) orelse return error.GpuSubmit; + defer c.SDL_ReleaseGPUFence(g.device, fence); + _ = c.SDL_WaitForGPUFences(g.device, true, &fence, 1); + p.add(label, c.SDL_GetTicksNS() - start); +} + +/// When profiling: waits for the GPU to be idle, then the time to start from. +fn startTiming(g: *Gpu) u64 { + if (g.profile != null) _ = c.SDL_WaitForGPUIdle(g.device); + return c.SDL_GetTicksNS(); +} + +/// The most bytes one buffer upload stages at a time. +const staging_size = 4 << 20; fn Cached(comptime T: type) type { return struct { object: *T, used: u64 }; } +const Storage = struct { buffer: *c.SDL_GPUBuffer, size: u32 }; + const Target = struct { width: u32, height: u32, @@ -50,27 +133,49 @@ pub fn init(gpa: std.mem.Allocator, device: *c.SDL_GPUDevice) Gpu { } pub fn deinit(g: *Gpu, images: *gfx.Images) void { + if (g.profile) |*p| { + p.report(g.frame); + p.deinit(); + } var shaders = g.shaders.valueIterator(); while (shaders.next()) |s| c.SDL_ReleaseGPUShader(g.device, s.object); var pipelines = g.pipelines.valueIterator(); while (pipelines.next()) |p| c.SDL_ReleaseGPUGraphicsPipeline(g.device, p.object); + var kernels = g.kernels.valueIterator(); + while (kernels.next()) |k| c.SDL_ReleaseGPUComputePipeline(g.device, k.object); + for (g.buffers.items) |entry| if (entry) |b| c.SDL_ReleaseGPUBuffer(g.device, b.buffer); + for (g.downloads.items) |d| { + _ = c.SDL_WaitForGPUFences(g.device, true, &d.fence, 1); + c.SDL_ReleaseGPUFence(g.device, d.fence); + c.SDL_ReleaseGPUTransferBuffer(g.device, d.transfer.buffer); + } + for (g.spare.items) |t| c.SDL_ReleaseGPUTransferBuffer(g.device, t.buffer); + g.downloads.deinit(g.gpa); + g.spare.deinit(g.gpa); + if (g.staging) |b| c.SDL_ReleaseGPUTransferBuffer(g.device, b); var targets = g.targets.iterator(); while (targets.next()) |entry| g.releaseTarget(images, entry.key_ptr.*, entry.value_ptr.*); if (g.vertices) |b| c.SDL_ReleaseGPUBuffer(g.device, b); if (g.upload) |b| c.SDL_ReleaseGPUTransferBuffer(g.device, b); g.shaders.deinit(g.gpa); g.pipelines.deinit(g.gpa); + g.kernels.deinit(g.gpa); + g.buffers.deinit(g.gpa); g.targets.deinit(g.gpa); } /// Releases what nothing has used for a while. pub fn endFrame(g: *Gpu) void { g.frame += 1; - inline for (.{ &g.pipelines, &g.shaders }) |map| { + inline for (.{ &g.pipelines, &g.shaders, &g.kernels }) |map| { var it = map.iterator(); while (it.next()) |entry| { if (g.frame - entry.value_ptr.used < keep_frames) continue; - if (@TypeOf(map) == @TypeOf(&g.pipelines)) c.SDL_ReleaseGPUGraphicsPipeline(g.device, entry.value_ptr.object) else c.SDL_ReleaseGPUShader(g.device, entry.value_ptr.object); + switch (@TypeOf(map)) { + @TypeOf(&g.pipelines) => c.SDL_ReleaseGPUGraphicsPipeline(g.device, entry.value_ptr.object), + @TypeOf(&g.kernels) => c.SDL_ReleaseGPUComputePipeline(g.device, entry.value_ptr.object), + else => c.SDL_ReleaseGPUShader(g.device, entry.value_ptr.object), + } map.removeByPtr(entry.key_ptr); // The iterator is invalid after a removal: the rest waits a frame. break; @@ -83,6 +188,7 @@ pub fn run(g: *Gpu, images: *gfx.Images, pass: abi.gpu.Pass) !void { const samples = g.supportedSamples(pass.samples); const t = try g.renderTarget(images, pass.image, pass.width, pass.height, samples, pass.stencil); + const start = g.startTiming(); // Every draw's vertices go up in one copy. var total: u32 = 0; for (pass.draws) |d| total += @intCast(std.mem.alignForward(usize, d.vertices.len, 4)); @@ -137,7 +243,10 @@ pub fn run(g: *Gpu, images: *gfx.Images, pass: abi.gpu.Pass) !void { c.SDL_BindGPUGraphicsPipeline(rp, built); bound = built; } - c.SDL_BindGPUVertexBuffers(rp, 0, &.{ .buffer = g.vertices, .offset = offset }, 1); + if (d.vertices.len > 0) c.SDL_BindGPUVertexBuffers(rp, 0, &.{ .buffer = g.vertices, .offset = offset }, 1); + var bound_storage: [8]*c.SDL_GPUBuffer = undefined; + if (d.vertex_storage.len > 0) c.SDL_BindGPUVertexStorageBuffers(rp, 0, &bound_storage, g.resolve(d.vertex_storage, &bound_storage) catch continue); + if (d.fragment_storage.len > 0) c.SDL_BindGPUFragmentStorageBuffers(rp, 0, &bound_storage, g.resolve(d.fragment_storage, &bound_storage) catch continue); if (d.vertex_uniforms.len > 0) c.SDL_PushGPUVertexUniformData(cmd, 0, d.vertex_uniforms.ptr, @intCast(d.vertex_uniforms.len)); if (d.fragment_uniforms.len > 0) c.SDL_PushGPUFragmentUniformData(cmd, 0, d.fragment_uniforms.ptr, @intCast(d.fragment_uniforms.len)); if (pass.stencil) c.SDL_SetGPUStencilReference(rp, d.stencil_reference); @@ -147,7 +256,171 @@ pub fn run(g: *Gpu, images: *gfx.Images, pass: abi.gpu.Pass) !void { c.SDL_DrawGPUPrimitives(rp, d.vertex_count, 1, 0, 0); } c.SDL_EndGPURenderPass(rp); + try g.submit(cmd, if (pass.label.len > 0) pass.label else "pass", start); +} + +/// Runs one dispatch in a compute pass of its own. +pub fn compute(g: *Gpu, d: abi.gpu.Dispatch) !void { + if (d.groups[0] == 0 or d.groups[1] == 0 or d.groups[2] == 0) return; + const built = try g.kernel(d.kernel); + const start = g.startTiming(); + var readonly: [8]*c.SDL_GPUBuffer = undefined; + const readonly_count = try g.resolve(d.readonly, &readonly); + var readwrite: [8]*c.SDL_GPUBuffer = undefined; + const readwrite_count = try g.resolve(d.readwrite, &readwrite); + var bindings: [8]c.SDL_GPUStorageBufferReadWriteBinding = undefined; + for (readwrite[0..readwrite_count], bindings[0..readwrite_count]) |b, *binding| binding.* = std.mem.zeroInit(c.SDL_GPUStorageBufferReadWriteBinding, .{ .buffer = b }); + const cmd = c.SDL_AcquireGPUCommandBuffer(g.device) orelse return error.GpuCommandBuffer; + const pass = c.SDL_BeginGPUComputePass(cmd, null, 0, &bindings, readwrite_count) orelse { + _ = c.SDL_CancelGPUCommandBuffer(cmd); + return error.GpuComputePass; + }; + c.SDL_BindGPUComputePipeline(pass, built); + if (readonly_count > 0) c.SDL_BindGPUComputeStorageBuffers(pass, 0, &readonly, readonly_count); + if (d.uniforms.len > 0) c.SDL_PushGPUComputeUniformData(cmd, 0, d.uniforms.ptr, @intCast(d.uniforms.len)); + c.SDL_DispatchGPUCompute(pass, d.groups[0], d.groups[1], d.groups[2]); + c.SDL_EndGPUComputePass(pass); + try g.submit(cmd, if (d.label.len > 0) d.label else "compute", start); +} + +/// A zeroed storage buffer, readable by every stage and writable by compute. +pub fn newBuffer(g: *Gpu, size: u32) !abi.Buffer { + if (size == 0) return error.EmptyBuffer; + const usage = c.SDL_GPU_BUFFERUSAGE_GRAPHICS_STORAGE_READ | c.SDL_GPU_BUFFERUSAGE_COMPUTE_STORAGE_READ | c.SDL_GPU_BUFFERUSAGE_COMPUTE_STORAGE_WRITE; + const buffer = c.SDL_CreateGPUBuffer(g.device, &.{ .usage = usage, .size = size, .props = 0 }) orelse return error.GpuBuffer; + errdefer c.SDL_ReleaseGPUBuffer(g.device, buffer); + try g.buffers.append(g.gpa, .{ .buffer = buffer, .size = size }); + const handle: abi.Buffer = @fromBackingInt(@intCast(g.buffers.items.len)); + // Vulkan leaves new memory undefined; sketches get zeroes, copied from + // one zeroed staging block as many times as it takes. + const staging = try g.stagingBuffer(); + const block: u32 = @min(size, staging_size); + const mapped: [*]u8 = @ptrCast(c.SDL_MapGPUTransferBuffer(g.device, staging, true) orelse return error.GpuMap); + @memset(mapped[0..block], 0); + c.SDL_UnmapGPUTransferBuffer(g.device, staging); + const cmd = c.SDL_AcquireGPUCommandBuffer(g.device) orelse return error.GpuCommandBuffer; + const copy = c.SDL_BeginGPUCopyPass(cmd); + var offset: u32 = 0; + while (offset < size) : (offset += block) { + c.SDL_UploadToGPUBuffer(copy, &.{ .transfer_buffer = staging, .offset = 0 }, &.{ .buffer = buffer, .offset = offset, .size = @min(block, size - offset) }, false); + } + c.SDL_EndGPUCopyPass(copy); if (!c.SDL_SubmitGPUCommandBuffer(cmd)) return error.GpuSubmit; + return handle; +} + +fn stagingBuffer(g: *Gpu) !*c.SDL_GPUTransferBuffer { + if (g.staging == null) g.staging = c.SDL_CreateGPUTransferBuffer(g.device, &.{ .usage = c.SDL_GPU_TRANSFERBUFFERUSAGE_UPLOAD, .size = staging_size, .props = 0 }) orelse return error.GpuBuffer; + return g.staging.?; +} + +pub fn freeBuffer(g: *Gpu, handle: abi.Buffer) void { + const index = @backingInt(handle); + if (index == 0 or index > g.buffers.items.len) return; + const entry = &g.buffers.items[index - 1]; + if (entry.*) |b| c.SDL_ReleaseGPUBuffer(g.device, b.buffer); + entry.* = null; +} + +/// Copies `bytes` into a buffer at `offset`, through the staging buffer. +pub fn write(g: *Gpu, handle: abi.Buffer, offset: u32, bytes: []const u8) !void { + const target = g.storage(handle) orelse return error.UnknownBuffer; + if (@as(u64, offset) + bytes.len > target.size) return error.UploadOutOfBounds; + const staging = try g.stagingBuffer(); + var done: usize = 0; + while (done < bytes.len) { + const chunk = bytes[done..][0..@min(staging_size, bytes.len - done)]; + const start = g.startTiming(); + // Cycling: a chunk still in flight keeps its own memory. + const mapped: [*]u8 = @ptrCast(c.SDL_MapGPUTransferBuffer(g.device, staging, true) orelse return error.GpuMap); + @memcpy(mapped[0..chunk.len], chunk); + c.SDL_UnmapGPUTransferBuffer(g.device, staging); + const cmd = c.SDL_AcquireGPUCommandBuffer(g.device) orelse return error.GpuCommandBuffer; + const copy = c.SDL_BeginGPUCopyPass(cmd); + c.SDL_UploadToGPUBuffer(copy, &.{ .transfer_buffer = staging, .offset = 0 }, &.{ .buffer = target.buffer, .offset = offset + @as(u32, @intCast(done)), .size = @intCast(chunk.len) }, false); + c.SDL_EndGPUCopyPass(copy); + try g.submit(cmd, "upload", start); + done += chunk.len; + } +} + +/// Starts copying `size` bytes of a buffer from `offset` back to the CPU; +/// `collect` hands them over once the GPU is done, a frame or more later. +pub fn download(g: *Gpu, d: @FieldType(abi.Command, "download")) !void { + const source = g.storage(d.buffer) orelse return error.UnknownBuffer; + if (d.size == 0 or @as(u64, d.offset) + d.size > source.size) return error.DownloadOutOfBounds; + const transfer = for (g.spare.items, 0..) |t, i| { + if (t.size >= d.size) break g.spare.swapRemove(i); + } else Transfer{ + .buffer = c.SDL_CreateGPUTransferBuffer(g.device, &.{ .usage = c.SDL_GPU_TRANSFERBUFFERUSAGE_DOWNLOAD, .size = d.size, .props = 0 }) orelse return error.GpuBuffer, + .size = d.size, + }; + errdefer g.spare.append(g.gpa, transfer) catch c.SDL_ReleaseGPUTransferBuffer(g.device, transfer.buffer); + const cmd = c.SDL_AcquireGPUCommandBuffer(g.device) orelse return error.GpuCommandBuffer; + const copy = c.SDL_BeginGPUCopyPass(cmd); + c.SDL_DownloadFromGPUBuffer(copy, &.{ .buffer = source.buffer, .offset = d.offset, .size = d.size }, &.{ .transfer_buffer = transfer.buffer, .offset = 0 }); + c.SDL_EndGPUCopyPass(copy); + const fence = c.SDL_SubmitGPUCommandBufferAndAcquireFence(cmd) orelse return error.GpuSubmit; + g.downloads.append(g.gpa, .{ .key = d.key, .size = d.size, .transfer = transfer, .fence = fence }) catch |err| { + _ = c.SDL_WaitForGPUFences(g.device, true, &fence, 1); + c.SDL_ReleaseGPUFence(g.device, fence); + return err; + }; +} + +/// The downloads the GPU has finished, copied into `arena`, in the order +/// they were asked for. +pub fn collect(g: *Gpu, arena: std.mem.Allocator) ![]const abi.Download { + var done: std.ArrayList(abi.Download) = .empty; + while (g.downloads.items.len > 0) { + const d = g.downloads.items[0]; + if (!c.SDL_QueryGPUFence(g.device, d.fence)) break; + _ = g.downloads.orderedRemove(0); + c.SDL_ReleaseGPUFence(g.device, d.fence); + defer g.spare.append(g.gpa, d.transfer) catch c.SDL_ReleaseGPUTransferBuffer(g.device, d.transfer.buffer); + const mapped: [*]const u8 = @ptrCast(c.SDL_MapGPUTransferBuffer(g.device, d.transfer.buffer, false) orelse return error.GpuMap); + defer c.SDL_UnmapGPUTransferBuffer(g.device, d.transfer.buffer); + try done.append(arena, .{ .key = d.key, .bytes = try arena.dupe(u8, mapped[0..d.size]) }); + } + return done.items; +} + +fn storage(g: *Gpu, handle: abi.Buffer) ?Storage { + const index = @backingInt(handle); + if (index == 0 or index > g.buffers.items.len) return null; + return g.buffers.items[index - 1]; +} + +/// The SDL buffers behind `handles`, into `out`; returns how many. +fn resolve(g: *Gpu, handles: []const abi.Buffer, out: []*c.SDL_GPUBuffer) !u32 { + if (handles.len > out.len) return error.TooManyBuffers; + for (handles, out[0..handles.len]) |handle, *b| b.* = (g.storage(handle) orelse return error.UnknownBuffer).buffer; + return @intCast(handles.len); +} + +fn kernel(g: *Gpu, k: abi.gpu.Kernel) !*c.SDL_GPUComputePipeline { + var h = std.hash.Wyhash.init(0); + h.update(k.code); + inline for (.{ k.readonly_buffers, k.readwrite_buffers, k.uniform_buffers, k.threads[0], k.threads[1], k.threads[2] }) |v| h.update(std.mem.asBytes(&@as(u32, v))); + const key = h.final(); + const entry = try g.kernels.getOrPut(g.gpa, key); + if (!entry.found_existing) { + errdefer _ = g.kernels.remove(key); + var info: c.SDL_GPUComputePipelineCreateInfo = std.mem.zeroes(c.SDL_GPUComputePipelineCreateInfo); + info.code_size = k.code.len; + info.code = k.code.ptr; + info.entrypoint = "main"; + info.format = c.SDL_GPU_SHADERFORMAT_SPIRV; + info.num_readonly_storage_buffers = k.readonly_buffers; + info.num_readwrite_storage_buffers = k.readwrite_buffers; + info.num_uniform_buffers = k.uniform_buffers; + info.threadcount_x = k.threads[0]; + info.threadcount_y = k.threads[1]; + info.threadcount_z = k.threads[2]; + entry.value_ptr.object = c.SDL_CreateGPUComputePipeline(g.device, &info) orelse return error.GpuComputePipeline; + } + entry.value_ptr.used = g.frame; + return entry.value_ptr.object; } fn supportedSamples(g: *Gpu, wanted: u8) u8 { @@ -236,21 +509,22 @@ fn reserve(g: *Gpu, bytes: u32) !void { fn shader(g: *Gpu, s: abi.gpu.Shader, stage: c.SDL_GPUShaderStage) !struct { *c.SDL_GPUShader, u64 } { var h = std.hash.Wyhash.init(stage); - h.update(s.spirv); + h.update(s.code); h.update(std.mem.asBytes(&s.uniform_buffers)); + h.update(std.mem.asBytes(&s.storage_buffers)); const key = h.final(); const entry = try g.shaders.getOrPut(g.gpa, key); if (!entry.found_existing) { errdefer _ = g.shaders.remove(key); entry.value_ptr.object = c.SDL_CreateGPUShader(g.device, &.{ - .code_size = s.spirv.len, - .code = s.spirv.ptr, + .code_size = s.code.len, + .code = s.code.ptr, .entrypoint = "main", .format = c.SDL_GPU_SHADERFORMAT_SPIRV, .stage = stage, .num_samplers = 0, .num_storage_textures = 0, - .num_storage_buffers = 0, + .num_storage_buffers = s.storage_buffers, .num_uniform_buffers = s.uniform_buffers, .props = 0, }) orelse return error.GpuShader; @@ -264,9 +538,9 @@ fn pipeline(g: *Gpu, p: abi.gpu.Pipeline, samples: u8, stencil: bool) !*c.SDL_GP const fragment, const fragment_key = try g.shader(p.fragment, c.SDL_GPU_SHADERSTAGE_FRAGMENT); // Field by field: no padding bytes in the key. var h = std.hash.Wyhash.init(vertex_key ^ (fragment_key *% 31)); - inline for (.{ p.stride, @intFromEnum(p.primitive), @intFromEnum(p.blend), @intFromBool(p.write_color), samples, @intFromBool(stencil) }) |v| h.update(std.mem.asBytes(&@as(u32, v))); - for (p.attributes) |a| inline for (.{ a.location, @intFromEnum(a.format), a.offset }) |v| h.update(std.mem.asBytes(&@as(u32, v))); - if (p.stencil) |st| inline for (.{ st.front, st.back }) |face| inline for (.{ @intFromEnum(face.compare), @intFromEnum(face.pass), @intFromEnum(face.fail) }) |v| h.update(std.mem.asBytes(&@as(u32, v))); + inline for (.{ p.stride, @backingInt(p.primitive), @backingInt(p.blend), @intFromBool(p.write_color), samples, @intFromBool(stencil) }) |v| h.update(std.mem.asBytes(&@as(u32, v))); + for (p.attributes) |a| inline for (.{ a.location, @backingInt(a.format), a.offset }) |v| h.update(std.mem.asBytes(&@as(u32, v))); + if (p.stencil) |st| inline for (.{ st.front, st.back }) |face| inline for (.{ @backingInt(face.compare), @backingInt(face.pass), @backingInt(face.fail) }) |v| h.update(std.mem.asBytes(&@as(u32, v))); const key = h.final(); const entry = try g.pipelines.getOrPut(g.gpa, key); if (!entry.found_existing) { @@ -310,7 +584,8 @@ fn pipeline(g: *Gpu, p: abi.gpu.Pipeline, samples: u8, stencil: bool) !*c.SDL_GP info.fragment_shader = fragment; info.vertex_input_state = .{ .vertex_buffer_descriptions = &.{ .slot = 0, .pitch = p.stride, .input_rate = c.SDL_GPU_VERTEXINPUTRATE_VERTEX, .instance_step_rate = 0 }, - .num_vertex_buffers = 1, + // No attributes: the vertex shader makes its own vertices. + .num_vertex_buffers = @intFromBool(count > 0), .vertex_attributes = &attributes, .num_vertex_attributes = @intCast(count), }; @@ -338,9 +613,9 @@ fn pipeline(g: *Gpu, p: abi.gpu.Pipeline, samples: u8, stencil: bool) !*c.SDL_GP fn stencilFace(face: abi.gpu.StencilFace) c.SDL_GPUStencilOpState { // Both enums follow SDL's order, after its INVALID. return .{ - .compare_op = @as(u32, @intFromEnum(face.compare)) + 1, - .pass_op = @as(u32, @intFromEnum(face.pass)) + 1, - .fail_op = @as(u32, @intFromEnum(face.fail)) + 1, + .compare_op = @as(u32, @backingInt(face.compare)) + 1, + .pass_op = @as(u32, @backingInt(face.pass)) + 1, + .fail_op = @as(u32, @backingInt(face.fail)) + 1, .depth_fail_op = c.SDL_GPU_STENCILOP_KEEP, }; } diff --git a/src/host/library.zig b/src/host/library.zig index 6c9338f..6958eec 100644 --- a/src/host/library.zig +++ b/src/host/library.zig @@ -17,10 +17,10 @@ pub const Library = struct { const object = c.SDL_LoadObject(path.ptr) orelse return error.LoadLibrary; errdefer c.SDL_UnloadObject(object); var sketch: abi.Sketch = undefined; - inline for (@typeInfo(abi.Sketch).@"struct".fields) |field| { - const symbol = @field(abi.Sketch.symbols, field.name); + inline for (comptime std.meta.fieldNames(abi.Sketch)) |name| { + const symbol = @field(abi.Sketch.symbols, name); const address = c.SDL_LoadFunction(object, symbol) orelse return error.MissingEntryPoint; - @field(sketch, field.name) = @ptrCast(address); + @field(sketch, name) = @ptrCast(address); } if (sketch.info().abi != abi.hash) return error.AbiMismatch; return .{ .object = object, .sketch = sketch }; @@ -29,7 +29,7 @@ pub const Library = struct { /// Loads `source` through a uniquely named copy in `scratch_dir`. pub fn openCopy(io: std.Io, source: []const u8, scratch_dir: []const u8, serial: usize) !Library { var buf: [std.fs.max_path_bytes]u8 = undefined; - const copy = try std.fmt.bufPrintZ(&buf, "{s}/sketchbook-{d}-{d}.so", .{ scratch_dir, linux.getpid(), serial }); + const copy = try std.mem.printSentinel(&buf, "{s}/sketchbook-{d}-{d}.so", .{ scratch_dir, linux.getpid(), serial }, 0); const cwd = std.Io.Dir.cwd(); try cwd.copyFile(source, cwd, copy, io, .{}); defer cwd.deleteFile(io, copy) catch {}; @@ -53,7 +53,7 @@ pub const Watch = struct { errdefer _ = linux.close(fd); var buf: [std.fs.max_path_bytes]u8 = undefined; const dir = std.fs.path.dirname(path) orelse "."; - const dir_z = try std.fmt.bufPrintZ(&buf, "{s}", .{dir}); + const dir_z = try std.mem.printSentinel(&buf, "{s}", .{dir}, 0); if (@as(isize, @bitCast(linux.inotify_add_watch(fd, dir_z, linux.IN.MOVED_TO | linux.IN.CLOSE_WRITE))) < 0) return error.WatchFailed; return .{ .fd = fd, .name = std.fs.path.basename(path) }; } diff --git a/src/host/memory.zig b/src/host/memory.zig index dd9f632..fc44e50 100644 --- a/src/host/memory.zig +++ b/src/host/memory.zig @@ -4,6 +4,12 @@ //! slots, plus a header describing the layout, so the state outlives restarts, //! a probe can read it from another process, and a reload with a changed //! layout can carry the old values over field by field. +//! +//! `StoreFile`: a heap in a file, mapped at the same address in every run, +//! so what it holds may point into it: the sketch's `Store` and everything +//! allocated for it outlive restarts. +//! +//! `Session`: this process's memory, kept across reloads. const std = @import("std"); const posix = std.posix; const linux = std.os.linux; @@ -197,7 +203,7 @@ pub const StateFile = struct { try self.resize(stride); const h = self.header(); h.* = .{ .size = new.size, .alignment = new.alignment, .hash = new.hash, .stride = stride, .used = old_used, .field_count = 0, .strings_len = 0 }; - try self.writeLayout(); + try writeLayout(self.map[0..header_len], self.layout); var total: Migration = .{}; for (old_slots, 0..) |maybe_old, i| { @@ -232,33 +238,34 @@ pub const StateFile = struct { } self.map = try mapFile(self.fd, len, .{ .READ = true, .WRITE = true }); } +}; - fn writeLayout(self: *StateFile) !void { - const h = self.header(); - const fields: [*]StoredField = @ptrCast(@alignCast(self.map.ptr + @sizeOf(Header))); - const strings_start = @sizeOf(Header) + self.layout.fields.len * @sizeOf(StoredField); - var strings: usize = 0; - for (self.layout.fields, 0..) |leaf, i| { - if (strings_start + strings + leaf.path.len + leaf.tags.len > header_len) return error.LayoutTooLarge; - const base = strings_start + strings; - @memcpy(self.map[base..][0..leaf.path.len], leaf.path); - @memcpy(self.map[base + leaf.path.len ..][0..leaf.tags.len], leaf.tags); - fields[i] = .{ - .offset = leaf.offset, - .size = leaf.size, - .hash = leaf.hash, - .kind = leaf.kind, - .path = @intCast(strings), - .path_len = @intCast(leaf.path.len), - .tags = @intCast(strings + leaf.path.len), - .tags_len = @intCast(leaf.tags.len), - }; - strings += leaf.path.len + leaf.tags.len; - } - h.field_count = @intCast(self.layout.fields.len); - h.strings_len = @intCast(strings); +/// Writes `layout`'s leaves after the `Header` at the start of `region`. +fn writeLayout(region: []u8, layout: Layout) !void { + const h: *Header = @ptrCast(@alignCast(region.ptr)); + const fields: [*]StoredField = @ptrCast(@alignCast(region.ptr + @sizeOf(Header))); + const strings_start = @sizeOf(Header) + layout.fields.len * @sizeOf(StoredField); + var strings: usize = 0; + for (layout.fields, 0..) |leaf, i| { + if (strings_start + strings + leaf.path.len + leaf.tags.len > region.len) return error.LayoutTooLarge; + const base = strings_start + strings; + @memcpy(region[base..][0..leaf.path.len], leaf.path); + @memcpy(region[base + leaf.path.len ..][0..leaf.tags.len], leaf.tags); + fields[i] = .{ + .offset = leaf.offset, + .size = leaf.size, + .hash = leaf.hash, + .kind = leaf.kind, + .path = @intCast(strings), + .path_len = @intCast(leaf.path.len), + .tags = @intCast(strings + leaf.path.len), + .tags_len = @intCast(leaf.tags.len), + }; + strings += leaf.path.len + leaf.tags.len; } -}; + h.field_count = @intCast(layout.fields.len); + h.strings_len = @intCast(strings); +} /// The layout stored in a state file's header, or null if it has none. pub fn readLayout(arena: std.mem.Allocator, map: []const u8) !?Layout { @@ -306,15 +313,211 @@ pub fn mapReadOnly(path: []const u8) []align(page) const u8 { return posix.mmap(null, @intCast(len), .{ .READ = true }, .{ .TYPE = .SHARED }, fd, 0) catch &.{}; } +/// Where every store is mapped, in every process: far from where Linux puts +/// executables (0x55...) and mappings (0x7f...). +pub const store_base: usize = 0x2000_0000_0000; +/// The most a store can grow to. Only reserved: pages exist as they are used. +pub const store_reserve: usize = 64 << 30; +pub const store_magic = "SKSTORE1".*; +/// The layout region is a state file header; the allocator's own record +/// follows, then the heap. +const meta_offset = header_len; +const heap_offset = 2 * header_len; + +/// A heap in a file, with the sketch's `Store` at its root. Mapped at +/// `store_base` in every run, so pointers it holds into itself stay valid +/// across restarts. Not crash-safe: a crash in the middle of an allocation +/// can leave it inconsistent (`--fresh` starts over). +/// +/// Blocks come in power-of-two sizes, each class with a free list threaded +/// through the free blocks, so growing a list by doubling reuses what it +/// frees. The file grows by mapping more of it in place: nothing moves. +pub const StoreFile = struct { + fd: posix.fd_t, + /// Bytes of the file mapped at `store_base`. + len: usize, + /// The root's layout, owned by `arena`. + layout: Layout, + arena: std.heap.ArenaAllocator, + + const Meta = extern struct { + magic: [8]u8 = store_magic, + base: u64 = store_base, + /// Offset of the first byte never handed out. + top: u64 = heap_offset, + /// Offset of the sketch's Store, or 0 before the first adopt. + root: u64 = 0, + /// Offsets of the first free block of each size class (2^i bytes), 0 + /// when none; a free block's first 8 bytes hold the next one's. + free: [64]u64 = @splat(0), + }; + + pub fn open(gpa: std.mem.Allocator, path: []const u8, fresh: bool) !StoreFile { + const fd = try posix.openat(posix.AT.FDCWD, path, .{ .ACCMODE = .RDWR, .CREAT = true }, 0o600); + errdefer _ = posix.system.close(fd); + // The same file mapped by two processes would be two heaps writing + // over each other. + if (linux.errno(linux.flock(fd, posix.LOCK.EX | posix.LOCK.NB)) != .SUCCESS) return error.StoreInUse; + const reserved = posix.mmap(@ptrFromInt(store_base), store_reserve, .{}, .{ .TYPE = .PRIVATE, .ANONYMOUS = true, .NORESERVE = true, .FIXED_NOREPLACE = true }, -1, 0) catch return error.StoreAddressTaken; + // Old kernels treat FIXED_NOREPLACE as a hint. + if (@intFromPtr(reserved.ptr) != store_base) { + posix.munmap(reserved); + return error.StoreAddressTaken; + } + errdefer posix.munmap(reserved); + var self: StoreFile = .{ .fd = fd, .len = 0, .layout = .{}, .arena = .init(gpa) }; + errdefer self.arena.deinit(); + const size: usize = @intCast(try fileSize(fd)); + if (!fresh and size >= heap_offset) { + try self.grow(size); + const m = self.meta(); + if (std.mem.eql(u8, &m.magic, &store_magic) and m.base == store_base and m.top <= size) { + self.layout = try readLayout(self.arena.allocator(), self.region()) orelse .{}; + return self; + } + } + try self.wipe(); + return self; + } + + pub fn close(self: *StoreFile) void { + posix.munmap(@as([*]align(page) u8, @ptrFromInt(store_base))[0..store_reserve]); + _ = posix.system.close(self.fd); + self.arena.deinit(); + } + + /// Empties the store: no root, nothing allocated. + fn wipe(self: *StoreFile) !void { + // Mapped pages past the new end would fault: map the reserve back. + if (self.len > heap_offset) { + _ = try posix.mmap(@ptrFromInt(store_base + heap_offset), self.len - heap_offset, .{}, .{ .TYPE = .PRIVATE, .ANONYMOUS = true, .NORESERVE = true, .FIXED = true }, -1, 0); + self.len = heap_offset; + } + try truncate(self.fd, heap_offset); + try self.grow(heap_offset); + @memset(self.region(), 0); + self.header().* = .{ .magic = store_magic, .size = 0, .alignment = 1, .hash = 0, .stride = 0, .used = 0, .field_count = 0, .strings_len = 0 }; + self.meta().* = .{}; + self.layout = .{}; + } + + /// Maps the file up to `need` bytes, growing it when it is shorter. + fn grow(self: *StoreFile, need: usize) !void { + if (need <= self.len) return; + const len = std.mem.alignForward(usize, @max(need, self.len * 2, 1 << 20), page); + if (len > store_reserve) return error.StoreFull; + if (try fileSize(self.fd) < len) try truncate(self.fd, len); + _ = try posix.mmap(@ptrFromInt(store_base + self.len), len - self.len, .{ .READ = true, .WRITE = true }, .{ .TYPE = .SHARED, .FIXED = true }, self.fd, self.len); + self.len = len; + } + + fn region(_: *StoreFile) []u8 { + return @as([*]u8, @ptrFromInt(store_base))[0..header_len]; + } + fn header(_: *StoreFile) *Header { + return @ptrFromInt(store_base); + } + fn meta(_: *StoreFile) *Meta { + return @ptrFromInt(store_base + meta_offset); + } + + /// The sketch's Store. + pub fn root(self: *StoreFile) [*]u8 { + const offset = self.meta().root; + return @ptrFromInt(store_base + if (offset == 0) heap_offset else offset); + } + + pub const Adoption = StateFile.Adoption; + + /// Makes the root hold `new`'s layout: the new defaults, then every field + /// whose path and type are unchanged, copied over. Pointer fields keep + /// pointing where they did, since the store never moves. + pub fn adopt(self: *StoreFile, gpa: std.mem.Allocator, new: abi.Layout, defaults: *const fn (abi.Memory, [*]u8) callconv(.c) void) !Adoption { + const m = self.meta(); + if (m.root != 0 and self.layout.hash == new.hash and self.layout.size == new.size) return .unchanged; + const a = self.allocator(); + const alignment: std.mem.Alignment = .fromByteUnits(@max(new.alignment, 1)); + const fresh = a.rawAlloc(@max(new.size, 1), alignment, @returnAddress()) orelse return error.StoreFull; + @memset(fresh[0..@max(new.size, 1)], 0); + defaults(.store, fresh); + + var kept = self.arena; + defer kept.deinit(); + self.arena = .init(gpa); + const old_layout = self.layout; + self.layout = try Layout.copy(self.arena.allocator(), new); + var result: Adoption = .created; + if (m.root != 0) { + const old: [*]u8 = @ptrFromInt(store_base + m.root); + result = .{ .migrated = migrate(old_layout, old[0..old_layout.size], self.layout, fresh[0..new.size]) }; + a.rawFree(old[0..@max(old_layout.size, 1)], .fromByteUnits(@max(old_layout.alignment, 1)), @returnAddress()); + } + self.header().* = .{ .magic = store_magic, .size = new.size, .alignment = new.alignment, .hash = new.hash, .stride = 0, .used = 1, .field_count = 0, .strings_len = 0 }; + try writeLayout(self.region(), self.layout); + m.root = @intFromPtr(fresh) - store_base; + return result; + } + + pub fn allocator(self: *StoreFile) std.mem.Allocator { + return .{ .ptr = self, .vtable = &.{ .alloc = alloc, .resize = resize, .remap = remap, .free = free } }; + } + + fn class(len: usize, alignment: std.mem.Alignment) u6 { + return @intCast(std.math.log2_int_ceil(usize, @max(len, alignment.toByteUnits(), 16))); + } + + fn alloc(ctx: *anyopaque, len: usize, alignment: std.mem.Alignment, _: usize) ?[*]u8 { + const self: *StoreFile = @ptrCast(@alignCast(ctx)); + const m = self.meta(); + const c = class(len, alignment); + const size = @as(usize, 1) << c; + if (m.free[c] != 0) { + const block: [*]u8 = @ptrFromInt(store_base + m.free[c]); + if (alignment.check(@intFromPtr(block))) { + m.free[c] = @as(*const u64, @ptrCast(@alignCast(block))).*; + return block; + } + } + const offset = std.mem.alignForward(usize, m.top, @max(@min(size, page), alignment.toByteUnits())); + self.grow(offset + size) catch return null; + m.top = offset + size; + return @ptrFromInt(store_base + offset); + } + + /// In place within the block's size class. + fn resize(_: *anyopaque, memory: []u8, alignment: std.mem.Alignment, new_len: usize, _: usize) bool { + return class(memory.len, alignment) == class(new_len, alignment); + } + + fn remap(ctx: *anyopaque, memory: []u8, alignment: std.mem.Alignment, new_len: usize, ret: usize) ?[*]u8 { + return if (resize(ctx, memory, alignment, new_len, ret)) memory.ptr else null; + } + + fn free(ctx: *anyopaque, memory: []u8, alignment: std.mem.Alignment, _: usize) void { + const self: *StoreFile = @ptrCast(@alignCast(ctx)); + const m = self.meta(); + const c = class(memory.len, alignment); + @as(*u64, @ptrCast(@alignCast(memory.ptr))).* = m.free[c]; + m.free[c] = @intFromPtr(memory.ptr) - store_base; + } +}; + +fn truncate(fd: posix.fd_t, len: usize) !void { + switch (posix.errno(posix.system.ftruncate(fd, @intCast(len)))) { + .SUCCESS => {}, + else => |err| return posix.unexpectedErrno(err), + } +} + /// Session memory: survives reloads while its layout hash is unchanged. -/// `heap` is released with it. +/// Everything allocated from `heap` is released with it. pub const Session = struct { bytes: []align(16) u8 = &.{}, hash: u64 = 0, - heap: std.heap.ArenaAllocator, + heap: Heap, pub fn init(gpa: std.mem.Allocator) Session { - return .{ .heap = .init(gpa) }; + return .{ .heap = .{ .gpa = gpa } }; } pub fn deinit(self: *Session, gpa: std.mem.Allocator) void { @@ -328,7 +531,7 @@ pub const Session = struct { if (layout.alignment > 16) return error.SessionOveraligned; gpa.free(self.bytes); self.bytes = &.{}; - _ = self.heap.reset(.free_all); + self.heap.reset(); self.bytes = try gpa.alignedAlloc(u8, .@"16", @max(layout.size, 1)); @memset(self.bytes, 0); defaults(.session, self.bytes.ptr); @@ -337,13 +540,88 @@ pub const Session = struct { } }; +/// The session's allocator: general purpose, so memory frees one allocation +/// at a time (a growing list does not leave its old copies behind), and it +/// remembers what is still live, so a session reset releases all of it. +pub const Heap = struct { + gpa: std.mem.Allocator, + live: std.AutoHashMapUnmanaged(usize, Block) = .empty, + + const Block = struct { len: usize, alignment: std.mem.Alignment }; + + pub fn allocator(self: *Heap) std.mem.Allocator { + return .{ .ptr = self, .vtable = &.{ .alloc = alloc, .resize = resize, .remap = remap, .free = free } }; + } + + /// Releases every live allocation. + pub fn reset(self: *Heap) void { + var it = self.live.iterator(); + while (it.next()) |entry| { + const memory: [*]u8 = @ptrFromInt(entry.key_ptr.*); + self.gpa.rawFree(memory[0..entry.value_ptr.len], entry.value_ptr.alignment, @returnAddress()); + } + self.live.clearRetainingCapacity(); + } + + pub fn deinit(self: *Heap) void { + self.reset(); + self.live.deinit(self.gpa); + } + + fn alloc(ctx: *anyopaque, len: usize, alignment: std.mem.Alignment, ret: usize) ?[*]u8 { + const self: *Heap = @ptrCast(@alignCast(ctx)); + self.live.ensureUnusedCapacity(self.gpa, 1) catch return null; + const memory = self.gpa.rawAlloc(len, alignment, ret) orelse return null; + self.live.putAssumeCapacity(@intFromPtr(memory), .{ .len = len, .alignment = alignment }); + return memory; + } + + fn resize(ctx: *anyopaque, memory: []u8, alignment: std.mem.Alignment, new_len: usize, ret: usize) bool { + const self: *Heap = @ptrCast(@alignCast(ctx)); + if (!self.gpa.rawResize(memory, alignment, new_len, ret)) return false; + self.live.getPtr(@intFromPtr(memory.ptr)).?.len = new_len; + return true; + } + + fn remap(ctx: *anyopaque, memory: []u8, alignment: std.mem.Alignment, new_len: usize, ret: usize) ?[*]u8 { + const self: *Heap = @ptrCast(@alignCast(ctx)); + self.live.ensureUnusedCapacity(self.gpa, 1) catch return null; + const moved = self.gpa.rawRemap(memory, alignment, new_len, ret) orelse return null; + _ = self.live.remove(@intFromPtr(memory.ptr)); + self.live.putAssumeCapacity(@intFromPtr(moved), .{ .len = new_len, .alignment = alignment }); + return moved; + } + + fn free(ctx: *anyopaque, memory: []u8, alignment: std.mem.Alignment, ret: usize) void { + const self: *Heap = @ptrCast(@alignCast(ctx)); + _ = self.live.remove(@intFromPtr(memory.ptr)); + self.gpa.rawFree(memory, alignment, ret); + } +}; + +test "the session heap frees one allocation at a time, and the rest on reset" { + var heap: Heap = .{ .gpa = std.testing.allocator }; + defer heap.deinit(); + const a = heap.allocator(); + var list: std.ArrayList(u32) = .empty; + for (0..1000) |i| try list.append(a, @intCast(i)); + try std.testing.expectEqual(@as(usize, 1), heap.live.count()); + const loose = try a.alloc(u8, 100); + a.free(loose); + try std.testing.expectEqual(@as(usize, 1), heap.live.count()); + _ = try a.create(u64); + heap.reset(); + try std.testing.expectEqual(@as(usize, 0), heap.live.count()); + // std.testing.allocator fails the test if anything leaked. +} + test "enums migrate by tag name" { const V1 = struct { mood: enum(u8) { calm, lost, angry } = .calm, gone: enum(u8) { a, b } = .b }; const V2 = struct { mood: enum(u16) { angry = 7, calm, sleepy } = .sleepy, gone: enum(u8) { c } = .c }; var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); defer arena.deinit(); - const l1 = try Layout.copy(arena.allocator(), abi.layout.of(V1, .forbidden)); - const l2 = try Layout.copy(arena.allocator(), abi.layout.of(V2, .forbidden)); + const l1 = try Layout.copy(arena.allocator(), abi.layout.of(V1, .state)); + const l2 = try Layout.copy(arena.allocator(), abi.layout.of(V2, .state)); for ([_]struct { @FieldType(V1, "mood"), @FieldType(V2, "mood") }{ .{ .calm, .calm }, .{ .angry, .angry }, .{ .lost, .sleepy } }) |case| { const old: V1 = .{ .mood = case[0] }; var new: V2 = .{}; @@ -358,8 +636,8 @@ test "migration keeps unchanged fields and defaults the rest" { const V2 = extern struct { b: f32 = 20, a: u32 = 10, b2: [2]u8 = .{ 7, 7 } }; var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); defer arena.deinit(); - const l1 = try Layout.copy(arena.allocator(), abi.layout.of(V1, .forbidden)); - const l2 = try Layout.copy(arena.allocator(), abi.layout.of(V2, .forbidden)); + const l1 = try Layout.copy(arena.allocator(), abi.layout.of(V1, .state)); + const l2 = try Layout.copy(arena.allocator(), abi.layout.of(V2, .state)); const old: V1 = .{ .a = 42, .b = 0.5 }; var new: V2 = .{}; const m = migrate(l1, std.mem.asBytes(&old), l2, std.mem.asBytes(&new)); @@ -382,12 +660,12 @@ test "state file migrates live state and snapshots, and a reopen reads the layou var tmp = std.testing.tmpDir(.{}); defer tmp.cleanup(); var buf: [std.fs.max_path_bytes]u8 = undefined; - const path = try std.fmt.bufPrint(&buf, ".zig-cache/tmp/{s}/test.state", .{&tmp.sub_path}); + const path = try std.mem.print(&buf, ".zig-cache/tmp/{s}/test.state", .{&tmp.sub_path}); const gpa = std.testing.allocator; { var file = try StateFile.open(gpa, path); defer file.close(); - try std.testing.expectEqual(StateFile.Adoption.created, try file.adopt(gpa, abi.layout.of(V1, .forbidden), D.v1)); + try std.testing.expectEqual(StateFile.Adoption.created, try file.adopt(gpa, abi.layout.of(V1, .state), D.v1)); file.markUsed(0, true); @as(*V1, @ptrCast(file.slot(0).ptr)).score = 99; @memcpy(file.slot(3), file.slot(0)); @@ -396,7 +674,7 @@ test "state file migrates live state and snapshots, and a reopen reads the layou var file = try StateFile.open(gpa, path); defer file.close(); try std.testing.expectEqual(abi.layout.hash(V1), file.layout.hash); - const adoption = try file.adopt(gpa, abi.layout.of(V2, .forbidden), D.v2); + const adoption = try file.adopt(gpa, abi.layout.of(V2, .state), D.v2); try std.testing.expectEqual(@as(usize, 1), adoption.migrated.reset); for ([_]usize{ 0, 3 }) |i| { const s: *V2 = @ptrCast(file.slot(i).ptr); @@ -404,3 +682,44 @@ test "state file migrates live state and snapshots, and a reopen reads the layou } try std.testing.expect(!file.used(1)); } + +test "a store keeps its root and what it points to across a reopen, and migrates by field" { + const V1 = struct { list: std.ArrayList(u32) = .empty, count: u32 = 0 }; + const V2 = struct { added: u8 = 7, list: std.ArrayList(u32) = .empty }; + const D = struct { + fn v1(_: abi.Memory, p: [*]u8) callconv(.c) void { + @as(*V1, @ptrCast(@alignCast(p))).* = .{}; + } + fn v2(_: abi.Memory, p: [*]u8) callconv(.c) void { + @as(*V2, @ptrCast(@alignCast(p))).* = .{}; + } + }; + var tmp = std.testing.tmpDir(.{}); + defer tmp.cleanup(); + var buf: [std.fs.max_path_bytes]u8 = undefined; + const path = try std.mem.print(&buf, ".zig-cache/tmp/{s}/test.store", .{&tmp.sub_path}); + const gpa = std.testing.allocator; + { + var store = try StoreFile.open(gpa, path, false); + defer store.close(); + try std.testing.expectEqual(StoreFile.Adoption.created, try store.adopt(gpa, abi.layout.of(V1, .store), D.v1)); + const v: *V1 = @ptrCast(@alignCast(store.root())); + for (0..10_000) |i| try v.list.append(store.allocator(), @intCast(i)); + v.count = 3; + try std.testing.expectError(error.StoreInUse, StoreFile.open(gpa, path, false)); + } + { + var store = try StoreFile.open(gpa, path, false); + defer store.close(); + try std.testing.expectEqual(StoreFile.Adoption.unchanged, try store.adopt(gpa, abi.layout.of(V1, .store), D.v1)); + const adoption = try store.adopt(gpa, abi.layout.of(V2, .store), D.v2); + try std.testing.expectEqual(@as(usize, 1), adoption.migrated.reset); + const v: *V2 = @ptrCast(@alignCast(store.root())); + try std.testing.expectEqual(@as(u8, 7), v.added); + try std.testing.expectEqual(@as(u32, 9999), v.list.items[9999]); + v.list.deinit(store.allocator()); + } + var store = try StoreFile.open(gpa, path, true); + defer store.close(); + try std.testing.expectEqual(StoreFile.Adoption.created, try store.adopt(gpa, abi.layout.of(V2, .store), D.v2)); +} diff --git a/src/host/root.zig b/src/host/root.zig index 637a325..2876c0f 100644 --- a/src/host/root.zig +++ b/src/host/root.zig @@ -6,6 +6,8 @@ //! --headless offscreen video, silent audio //! --script FILE play a script instead of reading devices //! --fresh start from default state: no snapshots or recordings +//! --profile-gpu time each GPU command alone, print the times at exit +//! --console LINE run a console line once the sketch is loaded (repeatable) //! --define NAME=VALUE substitute {NAME} in the script const std = @import("std"); const c = @import("c"); @@ -24,6 +26,8 @@ pub fn main(init: std.process.Init, linked: ?abi.Sketch) !void { var script_path: ?[]const u8 = null; var defines: std.ArrayList([2][]const u8) = .empty; defer defines.deinit(gpa); + var lines: std.ArrayList([]const u8) = .empty; + defer lines.deinit(gpa); var args = init.minimal.args.iterate(); _ = args.skip(); @@ -34,6 +38,10 @@ pub fn main(init: std.process.Init, linked: ?abi.Sketch) !void { options.fresh = true; } else if (std.mem.eql(u8, arg, "--headless")) { options.headless = true; + } else if (std.mem.eql(u8, arg, "--profile-gpu")) { + options.profile_gpu = true; + } else if (std.mem.eql(u8, arg, "--console")) { + try lines.append(gpa, args.next() orelse fatal("--console needs a line")); } else if (std.mem.eql(u8, arg, "--script")) { script_path = args.next() orelse fatal("--script needs a file"); } else if (std.mem.eql(u8, arg, "--define")) { @@ -56,6 +64,7 @@ pub fn main(init: std.process.Init, linked: ?abi.Sketch) !void { var host: Host = undefined; try host.init(gpa, init.io, init.environ_map, options, linked); defer host.deinit(); + for (lines.items) |line| host.runLine(line, c.SDL_GetTicksNS()) catch {}; if (script_path) |path| { const source = try std.Io.Dir.cwd().readFileAlloc(init.io, path, gpa, .limited(1 << 20)); defer gpa.free(source); @@ -71,6 +80,8 @@ const usage = \\ --headless offscreen video, silent audio, no vsync \\ --script FILE play a script instead of reading devices \\ --fresh start from default state: no snapshots or recordings + \\ --profile-gpu time each GPU command alone, print the times at exit + \\ --console LINE run a console line once the sketch is loaded (repeatable) \\ --define NAME=VALUE substitute {NAME} in the script \\ ; diff --git a/src/host/script.zig b/src/host/script.zig index 38954f3..cf551cd 100644 --- a/src/host/script.zig +++ b/src/host/script.zig @@ -4,7 +4,7 @@ //! # comment //! click 710 320 move there, press and release the left button //! move|down|up X Y one pointer event, then a frame -//! wheel X Y DELTA +//! wheel|wheel-x X Y DELTA //! cancel what losing focus sends //! key NAME press and release a key (`sketch.Key` names) //! type TEXT text input, as typed into a focused field @@ -56,23 +56,26 @@ fn step(host: *Host, arena: std.mem.Allocator, command: []const u8, rest: []cons const eq = std.mem.eql; if (eq(u8, command, "click")) { const x, const y = try point(words); - try host.route(.{ .kind = .mouse_move, .time = now.*, .x = x, .y = y }); - try host.route(.{ .kind = .mouse_down, .time = now.*, .x = x, .y = y, .code = 1 }); - try host.route(.{ .kind = .mouse_up, .time = now.*, .x = x, .y = y, .code = 1 }); + try host.route(.{ .kind = .mouse_move, .time = now.*, .x = x, .y = y, .flags = held(host) }); + try host.route(.{ .kind = .mouse_down, .time = now.*, .x = x, .y = y, .code = 1, .flags = held(host) }); + try host.route(.{ .kind = .mouse_up, .time = now.*, .x = x, .y = y, .code = 1, .flags = held(host) }); } else if (eq(u8, command, "move") or eq(u8, command, "down") or eq(u8, command, "up")) { const x, const y = try point(words); const kind: abi.Event.Kind = if (eq(u8, command, "move")) .mouse_move else if (eq(u8, command, "down")) .mouse_down else .mouse_up; - try host.route(.{ .kind = kind, .time = now.*, .x = x, .y = y, .code = if (kind == .mouse_move) 0 else 1 }); - } else if (eq(u8, command, "wheel")) { + try host.route(.{ .kind = kind, .time = now.*, .x = x, .y = y, .code = if (kind == .mouse_move) 0 else 1, .flags = held(host) }); + } else if (eq(u8, command, "wheel") or eq(u8, command, "wheel-x")) { const x, const y = try point(words); const delta = try std.fmt.parseFloat(f32, words.next() orelse return error.ExpectedDelta); - try host.route(.{ .kind = .wheel, .time = now.*, .x = x, .y = y, .value = @bitCast(delta) }); + try host.route(.{ .kind = if (eq(u8, command, "wheel-x")) .wheel_horizontal else .wheel, .time = now.*, .x = x, .y = y, .value = @bitCast(delta), .flags = held(host) }); } else if (eq(u8, command, "cancel")) { try host.route(.{ .kind = .cancel, .time = now.* }); } else if (eq(u8, command, "key")) { const key = std.meta.stringToEnum(abi.Key, rest) orelse return error.UnknownKey; - try host.route(.{ .kind = .key_down, .time = now.*, .code = @intFromEnum(key) }); - try host.route(.{ .kind = .key_up, .time = now.*, .code = @intFromEnum(key) }); + try host.route(.{ .kind = .key_down, .time = now.*, .code = @backingInt(key) }); + try host.route(.{ .kind = .key_up, .time = now.*, .code = @backingInt(key) }); + } else if (eq(u8, command, "hold") or eq(u8, command, "release")) { + const key = std.meta.stringToEnum(abi.Key, rest) orelse return error.UnknownKey; + try host.route(.{ .kind = if (eq(u8, command, "hold")) .key_down else .key_up, .time = now.*, .code = @backingInt(key) }); } else if (eq(u8, command, "type")) { for (rest) |byte| try host.route(.{ .kind = .text, .time = now.*, .value = byte }); } else if (eq(u8, command, "console")) { @@ -109,7 +112,7 @@ fn step(host: *Host, arena: std.mem.Allocator, command: []const u8, rest: []cons return; } else if (eq(u8, command, "screenshot")) { if (std.fs.path.dirname(rest)) |dir| try std.Io.Dir.cwd().createDirPath(host.io, dir); - host.screenshot = try arena.dupeZ(u8, rest); + host.screenshot = try arena.dupeSentinel(u8, rest, 0); } else if (eq(u8, command, "load")) { try host.load(rest); } else if (eq(u8, command, "reload")) { @@ -131,6 +134,16 @@ fn step(host: *Host, arena: std.mem.Allocator, command: []const u8, rest: []cons try advance(host, now); } +/// The modifiers `hold` has down, for a pointer event's `flags`. +fn held(host: *const Host) u8 { + var mods: u32 = 0; + if (host.input.held(abi.Key.left_shift) or host.input.held(abi.Key.right_shift)) mods |= abi.Key.Mod.shift; + if (host.input.held(abi.Key.left_ctrl) or host.input.held(abi.Key.right_ctrl)) mods |= abi.Key.Mod.ctrl; + if (host.input.held(abi.Key.left_alt) or host.input.held(abi.Key.right_alt)) mods |= abi.Key.Mod.alt; + if (host.input.held(abi.Key.left_gui) or host.input.held(abi.Key.right_gui)) mods |= abi.Key.Mod.gui; + return abi.Event.withModifiers(mods); +} + fn advance(host: *Host, now: *u64) !void { try host.frame(now.*); now.* += frame_ns; diff --git a/src/keys.zig b/src/keys.zig index b195d2a..7e13ba0 100644 --- a/src/keys.zig +++ b/src/keys.zig @@ -100,8 +100,8 @@ pub const Key = enum(u8) { test "HID values match the specification's anchors" { const std = @import("std"); - try std.testing.expectEqual(4, @intFromEnum(Key.a)); - try std.testing.expectEqual(39, @intFromEnum(Key.@"0")); - try std.testing.expectEqual(53, @intFromEnum(Key.grave)); - try std.testing.expectEqual(82, @intFromEnum(Key.up)); + try std.testing.expectEqual(4, @backingInt(Key.a)); + try std.testing.expectEqual(39, @backingInt(Key.@"0")); + try std.testing.expectEqual(53, @backingInt(Key.grave)); + try std.testing.expectEqual(82, @backingInt(Key.up)); } diff --git a/src/layout.zig b/src/layout.zig index e779260..5c97ca1 100644 --- a/src/layout.zig +++ b/src/layout.zig @@ -5,18 +5,26 @@ //! type, so a host can tell, across a reload, which bytes still mean the same //! thing. //! -//! State is saved to disk and snapshotted, so it may not hold pointers at all. -//! Session memory lives only in the host process and may point into memory -//! the host owns (its heap, mapped files), never into the sketch library. +//! What each kind of sketch memory may hold: +//! - State is saved to disk and snapshotted: no pointers, and no host +//! handles (images, sounds, buffers), which only mean something in the +//! process that made them. +//! - Session lives in this process: anything, as long as pointers go into +//! memory the host owns (its heaps, mapped files), never into the sketch +//! library, which a reload unmaps. +//! - Store is a file mapped at the same address in every run: pointers into +//! the store itself, but no host handles. const std = @import("std"); const abi = @import("abi.zig"); -pub const Pointers = enum { forbidden, host_only }; - -pub fn of(comptime T: type, comptime pointers: Pointers) abi.Layout { +pub fn of(comptime T: type, comptime memory: abi.Memory) abi.Layout { return comptime blk: { @setEvalBranchQuota(1_000_000); - if (pointers == .forbidden) noPointers(T, @typeName(T)); + switch (memory) { + .state => check(T, "State", @typeName(T), false), + .store => check(T, "Store", @typeName(T), true), + .session => {}, + } const Leaves = struct { const list = flatten(T, "", 0); const fields: [list.len]abi.Field = list[0..list.len].*; @@ -34,14 +42,15 @@ fn flatten(comptime T: type, comptime prefix: []const u8, comptime base: usize) comptime { if (@typeInfo(T) != .@"struct") return &.{leaf(T, if (prefix.len == 0) "value" else prefix, base)}; var leaves: []const abi.Field = &.{}; - for (@typeInfo(T).@"struct".fields) |field| { - if (field.is_comptime or @sizeOf(field.type) == 0) continue; - const path = if (prefix.len == 0) field.name else prefix ++ "." ++ field.name; - const offset = base + @offsetOf(T, field.name); - leaves = leaves ++ if (@typeInfo(field.type) == .@"struct" and @typeInfo(field.type).@"struct".backing_integer == null) - flatten(field.type, path, offset) + const s = @typeInfo(T).@"struct"; + for (s.field_names, s.field_types, s.field_attrs) |name, F, attrs| { + if (attrs.@"comptime" or @sizeOf(F) == 0) continue; + const path = if (prefix.len == 0) name else prefix ++ "." ++ name; + const offset = base + @offsetOf(T, name); + leaves = leaves ++ if (@typeInfo(F) == .@"struct" and @typeInfo(F).@"struct".backing_integer == null) + flatten(F, path, offset) else - &[_]abi.Field{leaf(field.type, path, offset)}; + &[_]abi.Field{leaf(F, path, offset)}; } return leaves; } @@ -68,23 +77,28 @@ fn leaf(comptime T: type, comptime path: []const u8, comptime offset: usize) abi } } -fn noPointers(comptime T: type, comptime path: []const u8) void { - comptime switch (@typeInfo(T)) { - .pointer => @compileError("State cannot hold pointers (" ++ path ++ "): it is saved to disk and outlives every address. Store handles, indices or offsets, or keep it in Session."), - .array => |a| noPointers(a.child, path ++ "[]"), - .vector => |v| noPointers(v.child, path ++ "[]"), - .optional => |o| noPointers(o.child, path), - .@"struct" => |s| for (s.fields) |f| noPointers(f.type, path ++ "." ++ f.name), - .@"union" => |u| for (u.fields) |f| noPointers(f.type, path ++ "." ++ f.name), - else => {}, - }; +fn check(comptime T: type, comptime memory: []const u8, comptime path: []const u8, comptime pointers: bool) void { + comptime { + if (T == abi.Image or T == abi.Audio or T == abi.Buffer) @compileError(memory ++ " cannot hold host handles (" ++ path ++ "): they mean nothing to the next process. Keep them in Session."); + switch (@typeInfo(T)) { + // Not followed: types can point to themselves. + .pointer => if (!pointers) @compileError(memory ++ " cannot hold pointers (" ++ path ++ "): it is saved to disk and outlives every address. Keep indices or offsets, or put it in Store or Session."), + .array => |a| check(a.child, memory, path ++ "[]", pointers), + .vector => |v| check(v.child, memory, path ++ "[]", pointers), + .optional => |o| check(o.child, memory, path, pointers), + .@"struct" => |s| for (s.field_names, s.field_types) |name, F| check(F, memory, path ++ "." ++ name, pointers), + .@"union" => |u| for (u.field_names, u.field_types) |name, F| check(F, memory, path ++ "." ++ name, pointers), + else => {}, + } + } } fn tags(comptime E: type) [:0]const u8 { comptime { var text: []const u8 = ""; - for (@typeInfo(E).@"enum".fields, 0..) |field, i| { - text = text ++ (if (i == 0) "" else ",") ++ std.fmt.comptimePrint("{s}={d}", .{ field.name, field.value }); + const e = @typeInfo(E).@"enum"; + for (e.field_names, e.field_values, 0..) |name, value, i| { + text = text ++ (if (i == 0) "" else ",") ++ std.fmt.comptimePrint("{s}={d}", .{ name, value }); } return std.fmt.comptimePrint("{s}", .{text}); } @@ -112,16 +126,16 @@ pub fn signature(comptime T: type) []const u8 { .optional => |o| "?" ++ signature(o.child), .@"struct" => |s| blk: { var text: []const u8 = std.fmt.comptimePrint("s{d}{{", .{@sizeOf(T)}); - for (s.fields) |field| { - if (field.is_comptime) continue; - text = text ++ std.fmt.comptimePrint("{s}@{d}:", .{ field.name, @offsetOf(T, field.name) }) ++ signature(field.type) ++ ";"; + for (s.field_names, s.field_types, s.field_attrs) |name, F, attrs| { + if (attrs.@"comptime") continue; + text = text ++ std.fmt.comptimePrint("{s}@{d}:", .{ name, @offsetOf(T, name) }) ++ signature(F) ++ ";"; } break :blk text ++ "}"; }, .@"union" => |u| blk: { var text: []const u8 = std.fmt.comptimePrint("u{d}{{", .{@sizeOf(T)}); if (u.tag_type) |tag| text = text ++ signature(tag) ++ "|"; - for (u.fields) |field| text = text ++ field.name ++ ":" ++ signature(field.type) ++ ";"; + for (u.field_names, u.field_types) |name, F| text = text ++ name ++ ":" ++ signature(F) ++ ";"; break :blk text ++ "}"; }, .pointer => |p| std.fmt.comptimePrint("*{s}:{s}", .{ @tagName(p.size), @typeName(p.child) }), @@ -135,7 +149,7 @@ pub fn signature(comptime T: type) []const u8 { test "leaves flatten structs, keep arrays whole and reject nothing pointer-free" { const Inner = extern struct { a: u8 = 0, b: enum(u8) { x, y = 5 } = .x }; const S = extern struct { inner: Inner = .{}, list: [4]f32 = @splat(0), flag: bool = false }; - const l = of(S, .forbidden); + const l = of(S, .state); const fields = l.fields; try std.testing.expectEqual(4, fields.len); try std.testing.expectEqualStrings("inner.a", std.mem.span(fields[0].path)); diff --git a/src/scope/layouts.zig b/src/scope/layouts.zig new file mode 100644 index 0000000..d8fc484 --- /dev/null +++ b/src/scope/layouts.zig @@ -0,0 +1,120 @@ +//! What the shaders read, as Zig structs: the only definition. The shaders +//! include GLSL generated from these (see sketchbook's `glsl.zig`). No +//! imports, so the shader build can compile this alone. + +/// FFT length: the analysis window, zero-padded to twice its size so the +/// spectrum is sampled twice as finely as the window resolves. +pub const fft_size = 8192; +pub const window = 4096; +/// The most columns a view can have; the column buffers hold this many. +pub const max_columns = 2048; +/// The tallest spectrogram; its buffer holds this many rows of +/// `max_columns`. +pub const max_rows = 1024; +// All three analyses read samples `offset + i` of their sample buffer for +// `0 <= i < samples`, so one buffer can hold several sounds. +/// The spectrogram's window, zero-padded to `spectrogram_fft` points. +pub const spectrogram_window = 2048; +pub const spectrogram_fft = 4096; + +/// spectrum.comp: the window ending at `end`, and the columns to fill. +pub const SpectrumParams = extern struct { + offset: u32, + /// Sample index one past the window's last sample. + end: u32, + /// Samples on the tape. + samples: u32, + columns: u32, + /// Nonzero: forget the smoothed levels and peaks. + reset: u32, + rate: f32, + min_freq: f32, + max_freq: f32, + /// Seconds since the previous analysis, for smoothing. + dt: f32, + pad0: u32 = 0, + pad1: u32 = 0, + pad2: u32 = 0, + + pub const glsl = .{ .uniform = .{ .set = 2, .instance = "u" } }; +}; + +/// waveform.comp: column `c` covers samples `first + c * spp` up to the next +/// column's start, and is written to slot `(slot + c) % ring`: the live +/// view writes its newest columns into a ring, the others from slot 0. +pub const WaveParams = extern struct { + offset: u32, + first: u32, + samples: u32, + columns: u32, + spp: f32, + slot: u32, + ring: u32, + pad: u32 = 0, + + pub const glsl = .{ .uniform = .{ .set = 2, .instance = "u" } }; +}; + +pub const Tint = enum(u32) { take, recording, live }; + +/// waveform.frag. Pixel column `x` shows slot `(x + scroll) % width`; only +/// the last `filled` columns have anything in them. +pub const WaveView = extern struct { + size: [2]f32, + /// Pixels; negative hides it. + playhead: f32, + gain: f32, + /// A `Tint`. + tint: u32, + scroll: u32, + filled: u32, + pad: u32 = 0, + + pub const glsl = .{ .uniform = .{ .set = 3, .instance = "u" } }; +}; + +/// spectrum.frag. +pub const SpectrumView = extern struct { + size: [2]f32, + floor_db: f32, + ceil_db: f32, + min_freq: f32, + max_freq: f32, + pad: [2]f32 = .{ 0, 0 }, + + pub const glsl = .{ .uniform = .{ .set = 3, .instance = "u" } }; +}; + +/// spectrogram.comp: the same columns as `WaveParams`, each analysed into +/// `rows` log-spaced frequency rows, row 0 the lowest. +/// Columns land in slots as `WaveParams` says. +pub const SpectrogramParams = extern struct { + offset: u32, + first: u32, + samples: u32, + columns: u32, + rows: u32, + spp: f32, + rate: f32, + min_freq: f32, + max_freq: f32, + slot: u32, + ring: u32, + pad: u32 = 0, + + pub const glsl = .{ .uniform = .{ .set = 2, .instance = "u" } }; +}; + +/// spectrogram.frag; `scroll` and `filled` as in `WaveView`. +pub const SpectrogramView = extern struct { + size: [2]f32, + floor_db: f32, + ceil_db: f32, + /// Pixels; negative hides it. + playhead: f32, + scroll: u32, + filled: u32, + pad: f32 = 0, + + pub const glsl = .{ .uniform = .{ .set = 3, .instance = "u" } }; +}; diff --git a/src/scope/root.zig b/src/scope/root.zig new file mode 100644 index 0000000..a71fa53 --- /dev/null +++ b/src/scope/root.zig @@ -0,0 +1,394 @@ +//! Live views of a stream of sound, analysed and drawn on the GPU: the last +//! few seconds as a scrolling waveform over its spectrogram, the smoothed +//! spectrum of the newest samples, and an oscilloscope. A sketch feeds a +//! `Scope` (kept in its Session) with `push` and draws it once a frame with +//! `draw`, giving the rectangles of the views it wants. +//! +//! The analyses (see the shaders for the details): +//! - spectrogram: a 2048-point Blackman-Harris window per column, and below +//! 500 Hz a second analysis decimated 4× (6 Hz resolution) averaged over +//! three windows; log-frequency rows; +//! - spectrum: a 4096-point window zero-padded to 8192, log-frequency +//! columns, fast attack and slow release, falling peaks; +//! - waveform: min, max and RMS per column, joined into a line when zoomed +//! past one sample per column. +//! +//! Columns of the scrolling views are analysed once, when everything their +//! windows reach has arrived, into rings the views read from the oldest; so +//! the spectrogram runs `reach` samples (about 130 ms) behind the newest. +const std = @import("std"); +const sk = @import("sketch"); +const gpu = sk.abi.gpu; +pub const layouts = @import("scope_layouts"); +const shaders = @import("scope_shaders"); + +pub const min_freq = 20.0; +pub const floor_db = -96.0; +/// The samples kept: enough for a frame's new columns and all they read. +pub const history = 1 << 16; +/// How far the spectrogram reads past a column's centre (its long +/// analysis: 2048 points decimated 4×, three windows a quarter apart). +pub const reach = 6160; + +pub const Tint = layouts.Tint; + +/// Where and how to draw; views left null are not analysed. +pub const Views = struct { + /// Seconds the waveform and spectrogram span, ending now. + seconds: f32 = 6, + /// Both scroll together, column for column: give them the same width. + wave: ?sk.Rect = null, + spectrogram: ?sk.Rect = null, + spectrum: ?sk.Rect = null, + oscilloscope: ?Oscilloscope = null, + gain: f32 = 1, + tint: Tint = .live, + + pub const Oscilloscope = struct { + rect: sk.Rect, + /// The samples across it. + samples: u32 = 1024, + /// How many samples before the newest it ends: a sketch steadies a + /// periodic wave by ending it at the same phase each frame. + end: u32 = 0, + }; +}; + +pub const Scope = struct { + samples: [history]f32 = @splat(0), + /// Samples pushed since the start. + heard: u64 = 0, + /// Pushed since the last `draw`, so the GPU copy is stale. + fresh: bool = false, + /// The next column to analyse, counted from the first sample; how many + /// have been since the view was (re)made; their count and width. + next: u64 = 0, + filled: u32 = 0, + columns: u32 = 0, + spp: u32 = 0, + gpu: Gpu = .{}, + + pub fn push(s: *Scope, heard: []const f32) void { + if (heard.len == 0) return; + const h = &s.samples; + if (heard.len >= h.len) { + @memcpy(h, heard[heard.len - h.len ..]); + } else { + std.mem.copyForwards(f32, h[0 .. h.len - heard.len], h[heard.len..]); + @memcpy(h[h.len - heard.len ..], heard); + } + s.heard += heard.len; + s.fresh = true; + } + + /// Analyses what is new and draws the views; needs `f.host.gpu()`. + pub fn draw(s: *Scope, f: *sk.Frame, v: Views) void { + const g = &s.gpu; + const rate: f64 = @floatFromInt(f.in.sample_rate); + g.setup(f); + if (f.in.reloaded) g.forget(); + if (s.fresh) f.upload(g.tape, 0, std.mem.sliceAsBytes(&s.samples)); + defer s.fresh = false; + + if (v.wave orelse v.spectrogram) |r| s.scroll(f, v, r, rate); + if (v.spectrum) |r| s.spectrum(f, r, rate); + if (v.oscilloscope) |o| s.oscilloscope(f, v, o); + } + + /// The waveform and spectrogram, scrolling. + fn scroll(s: *Scope, f: *sk.Frame, v: Views, r: sk.Rect, rate: f64) void { + const g = &s.gpu; + const columns: u32 = @intFromFloat(std.math.clamp(r.w, 1, layouts.max_columns)); + const spp: u32 = @max(1, @as(u32, @intFromFloat(@round(v.seconds * rate / @as(f64, @floatFromInt(columns)))))); + if (columns != s.columns or spp != s.spp or f.in.reloaded) { + s.columns = columns; + s.spp = spp; + s.filled = 0; + s.next = 0; + } + const rows: u32 = if (v.spectrogram) |sr| @intFromFloat(std.math.clamp(sr.h, 1, layouts.max_rows)) else 1; + g.levelsFor(f, rows); + + // History index of sample `i`: `i - base`. A column is ready once + // all it reads has arrived, and analysable while all it reads is kept. + const base: i64 = @as(i64, @intCast(s.heard)) - history; + const ready = (s.heard -| reach) / spp; + const kept: u64 = @intCast(@max(0, @divFloor(base + reach + spp - 1, spp))); + s.next = @max(s.next, kept, ready -| columns); + if (ready > s.next) { + const count: u32 = @intCast(ready - s.next); + const first: u32 = @intCast(@as(i64, @intCast(s.next * spp)) - base); + const slot: u32 = @intCast(s.next % columns); + if (v.wave != null) f.compute(.{ + .label = "scope waveform.comp", + .kernel = waveform_kernel, + .readonly = &.{g.tape}, + .readwrite = &.{g.wave.columns}, + .uniforms = gpu.uniform(f.scratch, layouts.WaveParams{ .offset = 0, .first = first, .samples = history, .columns = count, .spp = @floatFromInt(spp), .slot = slot, .ring = columns }), + .groups = .{ count, 1, 1 }, + }); + if (v.spectrogram != null) f.compute(.{ + .label = "scope spectrogram.comp", + .kernel = spectrogram_kernel, + .readonly = &.{g.tape}, + .readwrite = &.{g.levels}, + .uniforms = gpu.uniform(f.scratch, layouts.SpectrogramParams{ + .offset = 0, + .first = first, + .samples = history, + .columns = count, + .rows = rows, + .spp = @floatFromInt(spp), + .rate = @floatCast(rate), + .min_freq = min_freq, + .max_freq = maxFreq(rate), + .slot = slot, + .ring = columns, + }), + .groups = .{ count, 1, 1 }, + }); + s.filled = @intCast(@min(columns, s.filled + count)); + s.next = ready; + } + + const scroll_at: u32 = @intCast(s.next % columns); + if (v.wave) |wr| { + const size = target(f, &g.wave, wr, layouts.max_rows); + const view: layouts.WaveView = .{ + .size = .{ @floatFromInt(size[0]), @floatFromInt(size[1]) }, + .playhead = -1, + .gain = v.gain, + .tint = @intFromEnum(v.tint), + .scroll = scroll_at, + .filled = s.filled, + }; + if (g.wave.redraw(std.hash.Wyhash.hash(0, std.mem.asBytes(&view)))) wavePass(f, &g.wave, size, view); + f.image(g.wave.image, wr, .{}); + } + if (v.spectrogram) |sr| { + const size = target(f, &g.spectrogram, sr, rows); + const view: layouts.SpectrogramView = .{ + .size = .{ @floatFromInt(size[0]), @floatFromInt(size[1]) }, + .floor_db = floor_db, + .ceil_db = -6, + .playhead = -1, + .scroll = scroll_at, + .filled = s.filled, + }; + if (g.spectrogram.redraw(std.hash.Wyhash.hash(0, std.mem.asBytes(&view)))) f.gpu(.{ + .label = "scope spectrogram.frag", + .image = g.spectrogram.image, + .width = size[0], + .height = size[1], + .clear = null, + .pipelines = &spectrogram_pipelines, + .draws = &.{.{ .pipeline = 0, .vertex_count = 3, .fragment_uniforms = gpu.uniform(f.scratch, view), .fragment_storage = &.{g.levels} }}, + }); + f.image(g.spectrogram.image, sr, .{}); + } + } + + /// The spectrum of the newest samples, smoothed frame to frame. + fn spectrum(s: *Scope, f: *sk.Frame, r: sk.Rect, rate: f64) void { + const g = &s.gpu; + const size = target(f, &g.spectrum, r, layouts.max_rows); + const dt: f64 = @min(@as(f64, @floatFromInt(f.in.dt)) / std.time.ns_per_s, 0.1); + f.compute(.{ + .label = "scope spectrum.comp", + .kernel = spectrum_kernel, + .readonly = &.{g.tape}, + .readwrite = &.{ g.scratch, g.spectrum.columns }, + .uniforms = gpu.uniform(f.scratch, layouts.SpectrumParams{ + .offset = 0, + .end = history, + .samples = history, + .columns = size[0], + .reset = @intFromBool(!g.spectrum_started), + .rate = @floatCast(rate), + .min_freq = min_freq, + .max_freq = maxFreq(rate), + .dt = @floatCast(dt), + }), + }); + g.spectrum_started = true; + f.gpu(.{ + .label = "scope spectrum.frag", + .image = g.spectrum.image, + .width = size[0], + .height = size[1], + .clear = null, + .pipelines = &spectrum_pipelines, + .draws = &.{.{ + .pipeline = 0, + .vertex_count = 3, + .fragment_uniforms = gpu.uniform(f.scratch, layouts.SpectrumView{ + .size = .{ @floatFromInt(size[0]), @floatFromInt(size[1]) }, + .floor_db = floor_db, + .ceil_db = 0, + .min_freq = min_freq, + .max_freq = maxFreq(rate), + }), + .fragment_storage = &.{g.spectrum.columns}, + }}, + }); + f.image(g.spectrum.image, r, .{}); + } + + /// The newest `samples` (less `end`), as a line. + fn oscilloscope(s: *Scope, f: *sk.Frame, v: Views, o: Views.Oscilloscope) void { + const g = &s.gpu; + const size = target(f, &g.oscilloscope, o.rect, layouts.max_rows); + const n: u32 = @min(o.samples, history / 2); + const end: u32 = @min(o.end, history / 2); + const first: u32 = history - n - end; + f.compute(.{ + .label = "scope oscilloscope", + .kernel = waveform_kernel, + .readonly = &.{g.tape}, + .readwrite = &.{g.oscilloscope.columns}, + .uniforms = gpu.uniform(f.scratch, layouts.WaveParams{ + .offset = 0, + .first = first, + .samples = history, + .columns = size[0], + .spp = @as(f32, @floatFromInt(n)) / @as(f32, @floatFromInt(size[0])), + .slot = 0, + .ring = layouts.max_columns, + }), + .groups = .{ size[0], 1, 1 }, + }); + wavePass(f, &g.oscilloscope, size, .{ + .size = .{ @floatFromInt(size[0]), @floatFromInt(size[1]) }, + .playhead = -1, + .gain = v.gain, + .tint = @intFromEnum(v.tint), + .scroll = 0, + .filled = size[0], + }); + f.image(g.oscilloscope.image, o.rect, .{}); + } +}; + +/// The highest frequency the views show. +pub fn maxFreq(rate: f64) f32 { + return @floatCast(@min(20_000, rate / 2)); +} + +/// Where `hz` falls across a spectrum view (or up a spectrogram, from its +/// bottom): 0 to 1, log frequency. +pub fn place(hz: f64, rate: f64) f32 { + return @floatCast(@log(hz / min_freq) / @log(maxFreq(rate) / min_freq)); +} + +// --------------------------------------------------------------------------- +// GPU objects. + +/// One view's image and columns, and the inputs it was last drawn from. +const View = struct { + image: sk.Image = .none, + size: [2]u32 = .{ 0, 0 }, + columns: sk.Buffer = .none, + drawn: u64 = 0, + + fn redraw(v: *View, key: u64) bool { + if (key == v.drawn) return false; + v.drawn = key; + return true; + } +}; + +const Gpu = struct { + ready: bool = false, + tape: sk.Buffer = .none, + scratch: sk.Buffer = .none, + levels: sk.Buffer = .none, + levels_rows: u32 = 0, + wave: View = .{}, + spectrogram: View = .{}, + spectrum: View = .{}, + oscilloscope: View = .{}, + spectrum_started: bool = false, + + fn setup(g: *Gpu, f: *sk.Frame) void { + if (g.ready) return; + g.ready = true; + g.tape = f.host.newBuffer(history * @sizeOf(f32)); + g.scratch = f.host.newBuffer(layouts.fft_size * 2 * @sizeOf([2]f32)); + g.wave.columns = f.host.newBuffer(layouts.max_columns * @sizeOf([4]f32)); + g.oscilloscope.columns = f.host.newBuffer(layouts.max_columns * @sizeOf([4]f32)); + g.spectrum.columns = f.host.newBuffer(layouts.max_columns * @sizeOf([2]f32)); + } + + /// The spectrogram's levels, `rows` of `max_columns`. + fn levelsFor(g: *Gpu, f: *sk.Frame, rows: u32) void { + if (rows <= g.levels_rows) return; + if (g.levels != .none) f.host.freeBuffer(g.levels); + g.levels = f.host.newBuffer(layouts.max_columns * rows * @sizeOf(f32)); + g.levels_rows = rows; + g.forget(); + } + + /// New shaders, or new buffers: draw everything again. + fn forget(g: *Gpu) void { + g.wave.drawn = 0; + g.spectrogram.drawn = 0; + g.spectrum_started = false; + } +}; + +/// The view's image at `r`'s size, (re)made when that changes. +fn target(f: *sk.Frame, v: *View, r: sk.Rect, max_h: u32) [2]u32 { + const want: [2]u32 = .{ + @intFromFloat(std.math.clamp(r.w, 1, layouts.max_columns)), + @intFromFloat(std.math.clamp(r.h, 1, @as(f32, @floatFromInt(max_h)))), + }; + if (v.image == .none or v.size[0] != want[0] or v.size[1] != want[1]) { + if (v.image != .none) f.host.freeImage(v.image); + v.image = f.host.newImage(want[0], want[1]); + v.size = want; + v.drawn = 0; + } + return want; +} + +fn wavePass(f: *sk.Frame, v: *View, size: [2]u32, view: layouts.WaveView) void { + f.gpu(.{ + .label = "scope waveform.frag", + .image = v.image, + .width = size[0], + .height = size[1], + .clear = null, + .pipelines = &wave_pipelines, + .draws = &.{.{ .pipeline = 0, .vertex_count = 3, .fragment_uniforms = gpu.uniform(f.scratch, view), .fragment_storage = &.{v.columns} }}, + }); +} + +fn viewPipeline(fragment: []const u8) gpu.Pipeline { + return .{ + .vertex = .{ .code = shaders.view_vert }, + .fragment = .{ .code = fragment, .uniform_buffers = 1, .storage_buffers = 1 }, + .stride = 0, + .attributes = &.{}, + .blend = .none, + }; +} +const wave_pipelines = [_]gpu.Pipeline{viewPipeline(shaders.waveform_frag)}; +const spectrum_pipelines = [_]gpu.Pipeline{viewPipeline(shaders.spectrum_frag)}; +const spectrogram_pipelines = [_]gpu.Pipeline{viewPipeline(shaders.spectrogram_frag)}; + +const spectrum_kernel: gpu.Kernel = .{ .code = shaders.spectrum_comp, .readonly_buffers = 1, .readwrite_buffers = 2, .uniform_buffers = 1, .threads = .{ 256, 1, 1 } }; +const spectrogram_kernel: gpu.Kernel = .{ .code = shaders.spectrogram_comp, .readonly_buffers = 1, .readwrite_buffers = 1, .uniform_buffers = 1, .threads = .{ 256, 1, 1 } }; +const waveform_kernel: gpu.Kernel = .{ .code = shaders.waveform_comp, .readonly_buffers = 1, .readwrite_buffers = 1, .uniform_buffers = 1, .threads = .{ 64, 1, 1 } }; + +test "the history keeps the newest samples in order" { + var s: Scope = .{}; + s.push(&.{ 1, 2, 3 }); + s.push(&.{4}); + try std.testing.expectEqualSlices(f32, &.{ 1, 2, 3, 4 }, s.samples[history - 4 ..]); + try std.testing.expectEqual(@as(u64, 4), s.heard); +} + +test "frequencies place on a log axis" { + try std.testing.expectEqual(@as(f32, 0), place(min_freq, 48_000)); + try std.testing.expectApproxEqAbs(@as(f32, 1), place(20_000, 48_000), 1e-6); +} diff --git a/src/scope/shaders/spectrogram.comp b/src/scope/shaders/spectrogram.comp new file mode 100644 index 0000000..b32f4c5 --- /dev/null +++ b/src/scope/shaders/spectrogram.comp @@ -0,0 +1,208 @@ +#version 450 +#extension GL_GOOGLE_include_directive : require +// The spectrogram, one workgroup per column, the same columns as the +// waveform (waveform.comp), so the two line up in time. +// +// Two resolutions per column, because one window cannot serve both ends of +// a log axis: 2048 samples resolve 23 Hz, which is sharp in time and fine +// in the kilohertz, but leaves 20 to 100 Hz (a quarter of the height) to +// four bins. So the low end is analysed again from the signal low-passed +// and decimated to a quarter of the rate: the same 2048 points then span +// 170 ms and resolve 6 Hz. Rows take the long analysis below 250 Hz, the +// short one above 500 Hz, and blend in between. +// +// A single window's spectrum of noise scatters as much as its level (each +// bin is chi-squared with 2 degrees of freedom), which in dB shows as +// speckle full of black holes, and finer bins only make more of them. So +// the long analysis averages three windows a quarter window apart in time +// (Welch): steady lines stay as sharp, noise settles. Multitaper averaging +// was tried and measured against a reference (see test/README.md): sine +// tapers and DPSS both widen narrow lines into slabs and leak, which a +// 96 dB scale shows. +// +// Each analysis is a Blackman-Harris window zero-padded to 4096 points, +// transformed as 2048 complex points (even and odd samples as real and +// imaginary parts) by an in-place radix-2 FFT in shared memory, then split +// into the real signal's bins. Zoomed out, a column averages the power of +// up to four windows spread across its samples, so it neither skips sound +// nor flickers as the view moves; zoomed in, neighbouring windows overlap. +// Rows are log-spaced: a Catmull-Rom spline through the bins' levels where +// rows are finer than bins, the loudest bin where they are coarser. +layout(local_size_x = 256) in; + +#include "SpectrogramParams.glsl" + +layout(set = 0, binding = 0) readonly buffer Tape { float tape[]; }; +// MAX_ROWS rows of MAX_COLUMNS levels, dBFS; row 0 is the lowest frequency. +layout(set = 1, binding = 0) buffer Levels { float levels[]; }; + +// Match layouts.zig. +const uint MAX_COLUMNS = 2048; +const uint W = 2048; +const uint N = 4096; +const uint HALF = N / 2; +const uint LOG2_HALF = 11; +const uint THREADS = 256; +// The long analysis: every DECIMATE-th sample of the low-passed signal. +const uint DECIMATE = 4; +// Its bins needed below the blend: 600 Hz at 12 kHz / 4096 is bin 205. +const uint LONG_BINS = 256; +// Windows the long analysis averages per column, a quarter window apart. +const uint WELCH = 3; +const float BLEND_LO = 250.0; +const float BLEND_HI = 500.0; +const float PI = 3.14159265358979; +const float SILENCE = -160.0; +// Columns past the end of the tape. +const float NO_TAPE = 1000.0; +const float A0 = 0.35875; +const float A1 = 0.48829; +const float A2 = 0.14128; +const float A3 = 0.01168; + +shared vec2 z[HALF]; +// Power per bin, summed over windows; then dBFS. +shared float short_bins[HALF + 1]; +shared float long_bins[LONG_BINS]; + +float sample_at(int i) { + return i >= 0 && i < int(u.samples) ? tape[u.offset + uint(i)] : 0.0; +} + +// The signal low-passed for decimation: a triangle over 7 samples (two +// boxes of 4), whose zeros sit on 12 and 24 kHz, exactly where anything +// that would fold onto the low bins comes from. +float low_passed(int i) { + return (sample_at(i - 3) + 2.0 * sample_at(i - 2) + 3.0 * sample_at(i - 1) + 4.0 * sample_at(i) + 3.0 * sample_at(i + 1) + 2.0 * sample_at(i + 2) + sample_at(i + 3)) / 16.0; +} + +float blackman_harris(uint i) { + float a = 2.0 * PI * float(i) / float(W - 1); + return A0 - A1 * cos(a) + A2 * cos(2.0 * a) - A3 * cos(3.0 * a); +} + +vec2 mul(vec2 a, vec2 b) { + return vec2(a.x * b.x - a.y * b.y, a.x * b.y + a.y * b.x); +} + +float windowed(int centre, uint i, bool long_window) { + if (long_window) return low_passed(centre + (int(i) - int(W / 2)) * int(DECIMATE)) * blackman_harris(i); + return sample_at(centre + int(i) - int(W / 2)) * blackman_harris(i); +} + +// Adds the power spectrum of the window centred on sample `centre`. +void analyse(int centre, bool long_window) { + uint t = gl_LocalInvocationID.x; + // Even and odd samples packed as one complex sequence, in bit-reversed + // order for the in-place transform. Past W it is all padding. + for (uint i = t; i < HALF; i += THREADS) { + vec2 v = vec2(0.0); + if (2u * i < W) v = vec2(windowed(centre, 2u * i, long_window), windowed(centre, 2u * i + 1u, long_window)); + z[bitfieldReverse(i) >> (32u - LOG2_HALF)] = v; + } + barrier(); + for (uint s = 1u; s <= LOG2_HALF; s++) { + uint half_size = 1u << (s - 1u); + for (uint j = t; j < HALF / 2u; j += THREADS) { + uint k = j & (half_size - 1u); + uint i0 = ((j >> (s - 1u)) << s) + k; + uint i1 = i0 + half_size; + float angle = -PI * float(k) / float(half_size); + vec2 b = mul(vec2(cos(angle), sin(angle)), z[i1]); + vec2 a = z[i0]; + z[i0] = a + b; + z[i1] = a - b; + } + barrier(); + } + // The real signal's bins from the packed transform: + // X[k] = E[k] + e^(-2 pi i k / N) O[k], with E and O the even and odd + // samples' spectra, recovered from Z[k] and conj(Z[HALF - k]). + uint count = long_window ? LONG_BINS : HALF + 1u; + for (uint k = t; k < count; k += THREADS) { + vec2 zk = z[k % HALF]; + vec2 zc = z[(HALF - k) % HALF]; + zc.y = -zc.y; + vec2 even = 0.5 * (zk + zc); + vec2 d = 0.5 * (zk - zc); + vec2 odd = vec2(d.y, -d.x); + float angle = -2.0 * PI * float(k) / float(N); + vec2 x = even + mul(vec2(cos(angle), sin(angle)), odd); + if (long_window) long_bins[k] += dot(x, x); + else short_bins[k] += dot(x, x); + } + barrier(); +} + +float bin_level(uint k, bool long_window) { + return long_window ? long_bins[min(k, LONG_BINS - 1u)] : short_bins[min(k, HALF)]; +} + +// The level of the row spanning bins `lo` to `hi` of one analysis. +float row_level(float lo, float hi, bool long_window) { + if (hi - lo < 1.0) { + float mid = sqrt(lo * hi); + uint i = uint(floor(mid)); + float f = mid - floor(mid); + float p0 = bin_level(max(i, 1u) - 1u, long_window); + float p1 = bin_level(i, long_window); + float p2 = bin_level(i + 1u, long_window); + float p3 = bin_level(i + 2u, long_window); + float v = p1 + 0.5 * f * (p2 - p0 + f * (2.0 * p0 - 5.0 * p1 + 4.0 * p2 - p3 + f * (3.0 * (p1 - p2) + p3 - p0))); + return clamp(v, SILENCE, max(max(p1, p2), max(p0, p3))); + } + float v = SILENCE; + for (uint i = uint(floor(lo)); i <= uint(ceil(hi)); i++) v = max(v, bin_level(i, long_window)); + return v; +} + +// Average power over `count` windows to dBFS: a full-scale sine at a bin's +// centre gives (W * A0 / 2) squared. +float to_db(float power, uint count) { + float ratio = power / float(count) / (0.25 * float(W) * A0 * float(W) * A0); + return ratio > 0.0 ? max(10.0 * log(ratio) / log(10.0), SILENCE) : SILENCE; +} + +void main() { + uint c = gl_WorkGroupID.x; + uint t = gl_LocalInvocationID.x; + for (uint k = t; k <= HALF; k += THREADS) short_bins[k] = 0.0; + for (uint k = t; k < LONG_BINS; k += THREADS) long_bins[k] = 0.0; + barrier(); + + float start = float(c) * u.spp; + uint short_windows = uint(clamp(ceil(u.spp / float(W / 2u)), 1.0, 4.0)); + for (uint w = 0u; w < short_windows; w++) { + analyse(int(float(u.first) + start + u.spp * (float(w) + 0.5) / float(short_windows)), false); + } + // Zoomed out, the windows spread across the column average already; + // zoomed in, WELCH windows a quarter window apart around its centre. + uint long_windows = uint(clamp(ceil(u.spp / float(W * DECIMATE / 2u)), 1.0, 4.0)); + if (long_windows == 1u) { + int centre = int(float(u.first) + start + 0.5 * u.spp); + for (uint w = 0u; w < WELCH; w++) analyse(centre + (int(w) - int(WELCH / 2u)) * int(W * DECIMATE / 4u), true); + long_windows = WELCH; + } else { + for (uint w = 0u; w < long_windows; w++) { + analyse(int(float(u.first) + start + u.spp * (float(w) + 0.5) / float(long_windows)), true); + } + } + + for (uint k = t; k <= HALF; k += THREADS) short_bins[k] = to_db(short_bins[k], short_windows); + for (uint k = t; k < LONG_BINS; k += THREADS) long_bins[k] = to_db(long_bins[k], long_windows); + barrier(); + + bool inside = float(u.first) + start < float(u.samples); + float short_hz = u.rate / float(N); + float long_hz = short_hz / float(DECIMATE); + float ratio = log(u.max_freq / u.min_freq); + for (uint r = t; r < u.rows; r += THREADS) { + float lo = u.min_freq * exp(ratio * float(r) / float(u.rows)); + float hi = u.min_freq * exp(ratio * float(r + 1u) / float(u.rows)); + float blend = smoothstep(BLEND_LO, BLEND_HI, sqrt(lo * hi)); + float v = 0.0; + if (blend < 1.0) v += (1.0 - blend) * row_level(lo / long_hz, hi / long_hz, true); + if (blend > 0.0) v += blend * row_level(lo / short_hz, hi / short_hz, false); + levels[r * MAX_COLUMNS + (u.slot + c) % u.ring] = inside ? v : NO_TAPE; + } +} diff --git a/src/scope/shaders/spectrogram.frag b/src/scope/shaders/spectrogram.frag new file mode 100644 index 0000000..9cb5e3d --- /dev/null +++ b/src/scope/shaders/spectrogram.frag @@ -0,0 +1,45 @@ +#version 450 +#extension GL_GOOGLE_include_directive : require +// The spectrogram (spectrogram.comp): time across, as the waveform above +// it, log frequency up, each level through a heat gradient (inferno: +// black, violet, red, orange, pale yellow), with the playhead. +layout(location = 0) out vec4 out_color; + +#include "SpectrogramView.glsl" + +layout(set = 2, binding = 0) readonly buffer Levels { float levels[]; }; + +// Match layouts.zig and spectrogram.comp. +const uint MAX_COLUMNS = 2048; +const float NO_TAPE = 1000.0; + +// Matplotlib's inferno, as a polynomial fit (Matt Zucker's, CC0). +vec3 inferno(float t) { + const vec3 c0 = vec3(0.0002189403691192265, 0.001651004631001012, -0.01948089843709184); + const vec3 c1 = vec3(0.1065134194856116, 0.5639564367884091, 3.932712388889277); + const vec3 c2 = vec3(11.60249308247187, -3.972853965665698, -15.9423941062914); + const vec3 c3 = vec3(-41.70399613139459, 17.43639888205313, 44.35414519872813); + const vec3 c4 = vec3(77.162935699427, -33.40235894210092, -81.80730925738993); + const vec3 c5 = vec3(-71.31942824499214, 32.62606426397723, 73.20951985803202); + const vec3 c6 = vec3(25.13112622477341, -12.24266895238567, -23.07032500287172); + return clamp(c0 + t * (c1 + t * (c2 + t * (c3 + t * (c4 + t * (c5 + t * c6))))), 0.0, 1.0); +} + +void main() { + vec2 p = gl_FragCoord.xy; + uint row = uint(u.size.y - p.y); + uint x = uint(p.x); + float db = levels[row * MAX_COLUMNS + (x + u.scroll) % uint(u.size.x)]; + if (x + u.filled < uint(u.size.x)) db = NO_TAPE; + vec3 color = vec3(0.035, 0.03, 0.05); + if (db < NO_TAPE) { + float t = clamp((db - u.floor_db) / (u.ceil_db - u.floor_db), 0.0, 1.0); + // A touch of contrast: quiet stays dark, loud blooms. + color = inferno(pow(t, 1.3)); + } + if (u.playhead >= 0.0) { + float d = abs(p.x - u.playhead); + color = mix(color, vec3(1.0, 0.85, 0.3), clamp(1.5 - d, 0.0, 1.0) * 0.8); + } + out_color = vec4(color, 1.0); +} diff --git a/src/scope/shaders/spectrum.comp b/src/scope/shaders/spectrum.comp new file mode 100644 index 0000000..e576e30 --- /dev/null +++ b/src/scope/shaders/spectrum.comp @@ -0,0 +1,128 @@ +#version 450 +#extension GL_GOOGLE_include_directive : require +// The spectrum at the playhead, entirely on the GPU, in one workgroup: +// +// 1. a Blackman-Harris window of the tape ending at `u.end`, zero-padded to +// N. Its sidelobes sit below -92 dB, so the finely sampled spectrum shows +// the signal rather than the window's ripples; +// 2. a radix-2 Stockham FFT, ping-ponging between the scratch halves; +// 3. levels in dBFS per bin; +// 4. one value per log-spaced column. Where columns are narrower than a bin +// (the low end), a Catmull-Rom spline through the bins' levels; where a +// column spans several bins (the high end), their loudest. The zero +// padding already interpolates between the window's own bins, so the +// curve is continuous at every zoom; +// 5. smoothing in time (fast attack, slow release) and falling peaks, kept +// in the column buffer between frames. +layout(local_size_x = 256) in; + +#include "SpectrumParams.glsl" + +layout(set = 0, binding = 0) readonly buffer Tape { float tape[]; }; +// Two halves of N complex values, which the workgroup's threads read from +// each other: coherent, for Vulkan; WGSL's storageBarrier covers it. +#ifdef SKETCHBOOK_WEB +layout(set = 1, binding = 0) buffer Scratch { vec2 scratch[]; }; +#else +layout(set = 1, binding = 0) coherent buffer Scratch { vec2 scratch[]; }; +#endif +// x: smoothed level, y: peak, both dBFS. +layout(set = 1, binding = 1) buffer Columns { vec2 columns[]; }; + +// Match layouts.zig. +const uint N = 8192; +const uint LOG2_N = 13; +const uint W = 4096; +const uint THREADS = 256; +const float PI = 3.14159265358979; +const float SILENCE = -160.0; +// Blackman-Harris, 4 terms. A0 is also its coherent gain. +const float A0 = 0.35875; +const float A1 = 0.48829; +const float A2 = 0.14128; +const float A3 = 0.01168; + +void sync() { + memoryBarrierBuffer(); + barrier(); +} + +void main() { + uint t = gl_LocalInvocationID.x; + + // 1. Window. + int first = int(u.end) - int(W); + for (uint i = t; i < N; i += THREADS) { + float x = 0.0; + int at = first + int(i); + if (i < W && at >= 0 && at < int(u.samples)) { + float a = 2.0 * PI * float(i) / float(W - 1); + x = tape[u.offset + uint(at)] * (A0 - A1 * cos(a) + A2 * cos(2.0 * a) - A3 * cos(3.0 * a)); + } + scratch[i] = vec2(x, 0.0); + } + sync(); + + // 2. FFT: after stage s, sub-transforms of 2p = 2^(s+1) points, in order. + for (uint s = 0; s < LOG2_N; s++) { + uint p = 1u << s; + uint from = (s & 1u) * N; + uint to = N - from; + for (uint i = t; i < N / 2; i += THREADS) { + uint k = i & (p - 1u); + vec2 a = scratch[from + i]; + vec2 b = scratch[from + i + N / 2]; + float angle = -PI * float(k) / float(p); + vec2 w = vec2(cos(angle), sin(angle)); + b = vec2(b.x * w.x - b.y * w.y, b.x * w.y + b.y * w.x); + uint j = (i << 1u) - k; + scratch[to + j] = a + b; + scratch[to + j + p] = a - b; + } + sync(); + } + uint result = (LOG2_N & 1u) * N; + + // 3. Levels into the other half. A full-scale sine peaks at W * A0 / 2: + // the window sums to W * A0, and a real sine splits between ±f. + uint levels = N - result; + for (uint i = t; i <= N / 2; i += THREADS) { + float amplitude = length(scratch[result + i]) * 2.0 / (float(W) * A0); + scratch[levels + i].x = amplitude > 0.0 ? max(20.0 * log(amplitude) / log(10.0), SILENCE) : SILENCE; + } + sync(); + + // 4 and 5. Columns. + float bin_hz = u.rate / float(N); + float ratio = log(u.max_freq / u.min_freq); + for (uint c = t; c < u.columns; c += THREADS) { + float lo = u.min_freq * exp(ratio * float(c) / float(u.columns)) / bin_hz; + float hi = u.min_freq * exp(ratio * float(c + 1u) / float(u.columns)) / bin_hz; + float mid = sqrt(lo * hi); + float v; + if (hi - lo < 1.0) { + uint i = uint(floor(mid)); + float f = mid - floor(mid); + float p0 = scratch[levels + min(max(i, 1u) - 1u, N / 2)].x; + float p1 = scratch[levels + min(i, N / 2)].x; + float p2 = scratch[levels + min(i + 1u, N / 2)].x; + float p3 = scratch[levels + min(i + 2u, N / 2)].x; + v = p1 + 0.5 * f * (p2 - p0 + f * (2.0 * p0 - 5.0 * p1 + 4.0 * p2 - p3 + f * (3.0 * (p1 - p2) + p3 - p0))); + v = clamp(v, SILENCE, max(max(p1, p2), max(p0, p3))); + } else { + v = SILENCE; + uint end = min(uint(ceil(hi)), N / 2); + for (uint i = uint(floor(lo)); i <= end; i++) v = max(v, scratch[levels + i].x); + } + vec2 old = columns[c]; + vec2 now; + if (u.reset != 0u) { + now = vec2(v, v); + } else { + float rate = v > old.x ? 40.0 : 6.0; + now.x = mix(old.x, v, 1.0 - exp(-rate * u.dt)); + now.y = max(now.x, old.y - 24.0 * u.dt); + } + columns[c] = now; + } +} diff --git a/src/scope/shaders/spectrum.frag b/src/scope/shaders/spectrum.frag new file mode 100644 index 0000000..e88b091 --- /dev/null +++ b/src/scope/shaders/spectrum.frag @@ -0,0 +1,51 @@ +#version 450 +#extension GL_GOOGLE_include_directive : require +// The spectrum from its columns (spectrum.comp): one column per pixel on a +// log frequency axis. The curve is drawn by its distance along the normal, +// from the neighbours' slope, so steep parts stay as thick as flat ones. +layout(location = 0) out vec4 out_color; + +#include "SpectrumView.glsl" + +layout(set = 2, binding = 0) readonly buffer Columns { vec2 columns[]; }; + +float y_of(float db) { + return u.size.y * (1.0 - clamp((db - u.floor_db) / (u.ceil_db - u.floor_db), 0.0, 1.0)); +} + +float x_of(float hz) { + return u.size.x * log(hz / u.min_freq) / log(u.max_freq / u.min_freq); +} + +// How close `p` is to a grid line, 0..1. +float grid(vec2 p) { + const float lines[10] = float[](20.0, 50.0, 100.0, 200.0, 500.0, 1000.0, 2000.0, 5000.0, 10000.0, 20000.0); + float near = 1e9; + for (int i = 0; i < 10; i++) near = min(near, abs(p.x - x_of(lines[i]))); + float step = u.size.y * 12.0 / (u.ceil_db - u.floor_db); + float db = mod(p.y, step); + near = min(near, min(db, step - db)); + return clamp(1.0 - near, 0.0, 1.0); +} + +void main() { + vec2 p = gl_FragCoord.xy; + int n = int(u.size.x); + int x = int(p.x); + float y = y_of(columns[x].x); + float slope = (y_of(columns[min(x + 1, n - 1)].x) - y_of(columns[max(x - 1, 0)].x)) * 0.5; + float distance = abs(p.y - y) / sqrt(1.0 + slope * slope); + + vec3 color = vec3(0.055, 0.06, 0.07); + color = mix(color, vec3(0.16, 0.17, 0.2), grid(p)); + + // Fill under the curve, by frequency, fading downwards. + float along = p.x / u.size.x; + vec3 hue = mix(vec3(0.2, 0.75, 0.9), vec3(0.85, 0.35, 0.9), along); + if (p.y > y) color = mix(color, hue, 0.35 * (1.0 - (p.y - y) / max(u.size.y - y, 1.0)) + 0.08); + // Peaks, then the curve. + float peak = abs(p.y - y_of(columns[x].y)); + color = mix(color, vec3(0.9, 0.9, 0.95), clamp(1.0 - peak, 0.0, 1.0) * 0.45); + color = mix(color, mix(hue, vec3(1.0), 0.5), clamp(1.6 - distance, 0.0, 1.0)); + out_color = vec4(color, 1.0); +} diff --git a/src/scope/shaders/view.vert b/src/scope/shaders/view.vert new file mode 100644 index 0000000..d9461e9 --- /dev/null +++ b/src/scope/shaders/view.vert @@ -0,0 +1,7 @@ +#version 450 +// One triangle over the whole target, from the vertex index alone: the +// fragment shaders work in pixels (gl_FragCoord). +void main() { + vec2 p = vec2((gl_VertexIndex << 1) & 2, gl_VertexIndex & 2); + gl_Position = vec4(p * 2.0 - 1.0, 0.0, 1.0); +} diff --git a/src/scope/shaders/waveform.comp b/src/scope/shaders/waveform.comp new file mode 100644 index 0000000..d1d64e5 --- /dev/null +++ b/src/scope/shaders/waveform.comp @@ -0,0 +1,72 @@ +#version 450 +#extension GL_GOOGLE_include_directive : require +// One workgroup per waveform column: the lowest and highest the signal goes +// in it, and its RMS. Each column also includes the linearly interpolated +// signal at its own edges, which it shares with its neighbours, so that +// zoomed in past one sample per column the columns still join into one +// continuous line. +layout(local_size_x = 64) in; + +#include "WaveParams.glsl" + +layout(set = 0, binding = 0) readonly buffer Tape { float tape[]; }; +// x: min, y: max, z: RMS, w: 1 where the column has tape under it. +layout(set = 1, binding = 0) buffer Columns { vec4 columns[]; }; + +const uint THREADS = 64; +shared float lows[THREADS]; +shared float highs[THREADS]; +shared float squares[THREADS]; + +float sample_at(int i) { + return i >= 0 && i < int(u.samples) ? tape[u.offset + uint(i)] : 0.0; +} + +// The signal at `t` samples after `u.first`, between samples linearly. +float signal(float t) { + float whole = floor(t); + int i = int(u.first) + int(whole); + return mix(sample_at(i), sample_at(i + 1), t - whole); +} + +void main() { + uint c = gl_WorkGroupID.x; + uint t = gl_LocalInvocationID.x; + float start = float(c) * u.spp; + float end = start + u.spp; + int first = int(ceil(start)); + int last = min(int(floor(end)), int(u.samples) - int(u.first) - 1); + + float lo = 1e9; + float hi = -1e9; + float sq = 0.0; + for (int i = first + int(t); i <= last; i += int(THREADS)) { + float v = tape[u.offset + u.first + uint(i)]; + lo = min(lo, v); + hi = max(hi, v); + sq += v * v; + } + if (t == 0u) { + float a = signal(start); + float b = signal(min(end, float(u.samples - u.first - 1u))); + lo = min(lo, min(a, b)); + hi = max(hi, max(a, b)); + } + lows[t] = lo; + highs[t] = hi; + squares[t] = sq; + barrier(); + for (uint stride = THREADS / 2u; stride > 0u; stride >>= 1u) { + if (t < stride) { + lows[t] = min(lows[t], lows[t + stride]); + highs[t] = max(highs[t], highs[t + stride]); + squares[t] += squares[t + stride]; + } + barrier(); + } + if (t == 0u) { + float count = float(max(last - first + 1, 1)); + float inside = float(u.first) + start < float(u.samples) ? 1.0 : 0.0; + columns[(u.slot + c) % u.ring] = vec4(lows[0], highs[0], sqrt(squares[0] / count), inside); + } +} diff --git a/src/scope/shaders/waveform.frag b/src/scope/shaders/waveform.frag new file mode 100644 index 0000000..37be7ce --- /dev/null +++ b/src/scope/shaders/waveform.frag @@ -0,0 +1,56 @@ +#version 450 +#extension GL_GOOGLE_include_directive : require +// The waveform from its columns (waveform.comp): one column per pixel, the +// band between its low and high filled, anti-aliased, at least a pixel +// thick; the RMS inside it brighter; played and unplayed in two colors, +// red while recording, green for the live microphone. +layout(location = 0) out vec4 out_color; + +#include "WaveView.glsl" + +layout(set = 2, binding = 0) readonly buffer Columns { vec4 columns[]; }; + +// Match layouts.Tint. +const uint TINT_RECORDING = 1u; +const uint TINT_LIVE = 2u; + +void main() { + vec2 p = gl_FragCoord.xy; + uint x = uint(p.x); + vec4 col = columns[(x + u.scroll) % uint(u.size.x)]; + if (x + u.filled < uint(u.size.x)) col.w = 0.0; + float mid = u.size.y * 0.5; + float scale = u.size.y * 0.46 * u.gain; + + vec3 color = vec3(0.055, 0.06, 0.07); + if (col.w > 0.0) color = vec3(0.075, 0.08, 0.095); + // Center line. + color = mix(color, vec3(0.2, 0.22, 0.26), clamp(1.0 - abs(p.y - mid), 0.0, 1.0) * 0.6); + + if (col.w > 0.0) { + float top = mid - col.y * scale - 0.5; + float bottom = mid - col.x * scale + 0.5; + float outside = max(top - p.y, p.y - bottom); + float cover = clamp(0.5 - outside, 0.0, 1.0); + float rms = col.z * scale; + float inner = clamp(rms - abs(p.y - mid) + 0.5, 0.0, 1.0); + bool played = u.playhead >= 0.0 && p.x < u.playhead; + vec3 band = played ? vec3(0.35, 0.8, 0.95) : vec3(0.45, 0.5, 0.58); + vec3 core = played ? vec3(0.7, 0.95, 1.0) : vec3(0.65, 0.7, 0.78); + if (u.tint == TINT_RECORDING) { + band = vec3(0.95, 0.35, 0.35); + core = vec3(1.0, 0.6, 0.55); + } else if (u.tint == TINT_LIVE) { + band = vec3(0.35, 0.75, 0.5); + core = vec3(0.65, 1.0, 0.75); + } + color = mix(color, mix(band, core, inner), cover); + } + + // Playhead. + if (u.playhead >= 0.0) { + float d = abs(p.x - u.playhead); + color = mix(color, vec3(1.0, 0.85, 0.3), clamp(1.5 - d, 0.0, 1.0)); + } + out_color = vec4(color, 1.0); +} diff --git a/src/sketch/request.zig b/src/sketch/request.zig new file mode 100644 index 0000000..0b8b400 --- /dev/null +++ b/src/sketch/request.zig @@ -0,0 +1,59 @@ +//! Requests from `update` that `draw` carries out once. `update` sees only +//! State and input, so when a key should do something to Session or Store +//! (start the microphone, open a dialog, seek a stream), State holds the +//! request and Session remembers which ones it has handled: +//! +//! // State +//! seek: sk.Request(f32) = .{}, +//! // Session +//! seek: sk.Handled = .{}, +//! +//! // update +//! if (in.pressed(.home)) s.seek.send(0); +//! // draw +//! if (ss.seek.take(s.seek)) |to| cursor = to * len; +const std = @import("std"); + +/// A request carrying `T`, starting at `T`'s zero value. +pub fn Request(comptime T: type) type { + return struct { + value: T = std.mem.zeroes(T), + /// Bumped by every `send`, so the same value can be asked for twice. + serial: u32 = 0, + + pub const Value = T; + + pub fn send(r: *@This(), value: T) void { + r.value = value; + r.serial +%= 1; + } + }; +} + +/// Which request a Session has carried out. +pub const Handled = struct { + serial: ?u32 = null, + + /// The request's value if it is new to this Session. A fresh Session (a + /// restart, or Session's layout changed) counts earlier requests as + /// handled: nothing re-fires from saved state, as `Input.started` asks + /// of `update`. So take every request on every frame, even when its + /// value will be ignored: the first take only catches up. + pub fn take(h: *Handled, request: anytype) ?@TypeOf(request).Value { + defer h.serial = request.serial; + const last = h.serial orelse return null; + return if (last == request.serial) null else request.value; + } +}; + +test "a request is taken once, and none from before the session" { + var r: Request(f32) = .{}; + r.send(0.5); + var h: Handled = .{}; + try std.testing.expectEqual(@as(?f32, null), h.take(r)); + r.send(0.25); + try std.testing.expectEqual(@as(?f32, 0.25), h.take(r)); + try std.testing.expectEqual(@as(?f32, null), h.take(r)); + r.send(0.25); + try std.testing.expectEqual(@as(?f32, 0.25), h.take(r)); +} diff --git a/src/sketch/root.zig b/src/sketch/root.zig index 0567839..94ce617 100644 --- a/src/sketch/root.zig +++ b/src/sketch/root.zig @@ -5,17 +5,21 @@ //! pub const name = "my sketch"; //! pub const width = 960; //! pub const height = 600; -//! pub const State = struct { ... }; // persisted and snapshotted: no pointers -//! pub const Session = struct { ... }; // survives reloads, never saved +//! pub const State = struct { ... }; // saved, snapshotted, replayed: no pointers +//! pub const Session = struct { ... }; // this process: survives reloads +//! pub const Store = struct { ... }; // on disk: survives restarts too //! pub const commands = .{"hello"}; // console completions //! pub fn init(state: *State, host: sk.Host) void //! pub fn update(state: *State, in: *const sk.Input) void -//! pub fn draw(state: *const State, session: *Session, f: *sk.Frame) void -//! pub fn command(state: *State, session: *Session, c: *sk.Console, line: []const u8) !bool +//! pub fn draw(state: *const State, session: *Session, store: *Store, f: *sk.Frame) void +//! pub fn command(state: *State, store: *Store, c: *sk.Console, line: []const u8) !bool //! -//! `update` must be deterministic: the host replays recorded input through it. -//! `draw` presents state as pictures and sound, and may only change `Session`. -//! Geometry and colors are duck-typed: see `shape`. +//! Functions take the parameters they name, by type, in any order: `draw` +//! may ask for `*Session`, `*Store`, both or neither. `update` must be +//! deterministic, so it gets `State` and `Input` only: the host replays +//! recorded input through it. `draw` presents state as pictures and sound, +//! and changes only Session and Store. Geometry and colors are duck-typed: +//! see `shape`. const std = @import("std"); pub const abi = @import("abi"); pub const shape = @import("shape.zig"); @@ -23,6 +27,9 @@ pub const Key = abi.Key; pub const Camera = @import("camera.zig").Camera; pub const tween = @import("tween.zig"); pub const web = @import("web.zig"); +const request = @import("request.zig"); +pub const Request = request.Request; +pub const Handled = request.Handled; pub const Rect = abi.Rect; pub const Color = abi.Color; @@ -32,25 +39,36 @@ pub const Mode = abi.Mode; pub const Host = abi.Host; pub const Image = abi.Image; pub const Audio = abi.Audio; +pub const Buffer = abi.Buffer; pub const Cursor = abi.Cursor; pub const Voice = abi.Voice; -/// What `draw` receives: input, host services, allocators and the command list. +/// What `draw` receives: input, host services, memory by lifetime and the +/// command list. Commands may be built from anything: the methods copy what +/// they keep (text, GPU descriptions) into `scratch`, except bulk data +/// (vertices, samples, pixels, uploads), which must outlive the frame. pub const Frame = struct { ctx: *const abi.Context, in: *const Input, host: Host, - /// Frame lifetime. - arena: std.mem.Allocator, - /// Session lifetime; survives reloads. - gpa: std.mem.Allocator, + /// Released once the frame is drawn. + scratch: std.mem.Allocator, + /// For Session: this process. + session: std.mem.Allocator, + /// For Store: on disk, across restarts. + store: std.mem.Allocator, pub fn init(ctx: *const abi.Context) Frame { - return .{ .ctx = ctx, .in = ctx.input, .host = ctx.host, .arena = ctx.arena, .gpa = ctx.gpa }; + return .{ .ctx = ctx, .in = ctx.input, .host = ctx.host, .scratch = ctx.scratch, .session = ctx.session, .store = ctx.store }; } pub fn emit(f: *Frame, command: abi.Command) void { - f.ctx.commands.append(f.arena, command) catch f.host.log("frame arena full: command dropped"); + f.ctx.commands.append(f.scratch, command) catch f.host.log("out of scratch memory: command dropped"); + } + + /// A scratch copy of `items`. + pub fn keep(f: *Frame, comptime T: type, items: []const T) []const T { + return f.scratch.dupe(T, items) catch &.{}; } pub fn clear(f: *Frame, color: anytype) void { @@ -69,17 +87,15 @@ pub const Frame = struct { .tint = if (@hasField(O, "tint")) shape.color(options.tint) else .white, } }); } - /// `bytes` must live until the frame is consumed: literals, `arena` and - /// session memory all do. pub fn text(f: *Frame, x: f32, y: f32, scale: f32, bytes: []const u8, color: anytype) void { - f.emit(.{ .text = .{ .x = x, .y = y, .scale = scale, .color = shape.color(color), .bytes = bytes } }); + f.emit(.{ .text = .{ .x = x, .y = y, .scale = scale, .color = shape.color(color), .bytes = f.keep(u8, bytes) } }); } pub fn print(f: *Frame, x: f32, y: f32, scale: f32, color: anytype, comptime fmt: []const u8, args: anytype) void { - f.text(x, y, scale, f.format(fmt, args), color); + f.emit(.{ .text = .{ .x = x, .y = y, .scale = scale, .color = shape.color(color), .bytes = f.format(fmt, args) } }); } - /// Formats into the frame arena. + /// Formats into scratch memory. pub fn format(f: *Frame, comptime fmt: []const u8, args: anytype) []const u8 { - return std.fmt.allocPrint(f.arena, fmt, args) catch "?"; + return f.scratch.print(fmt, args) catch "?"; } /// A rect, or `null` to stop clipping. pub fn clip(f: *Frame, r: anytype) void { @@ -101,18 +117,67 @@ pub const Frame = struct { f.emit(.{ .pixels = .{ .image = img, .w = w, .h = h, .rgba = rgba, .premultiplied = options.premultiplied } }); } /// Runs GPU work described as data (see `abi.gpu`) into `pass.image`. + /// Vertices must outlive the frame; the rest is copied. pub fn gpu(f: *Frame, pass: abi.gpu.Pass) void { - f.emit(.{ .gpu = pass }); + var kept = pass; + kept.label = f.keep(u8, pass.label); + kept.pipelines = f.keep(abi.gpu.Pipeline, pass.pipelines); + const draws = f.scratch.dupe(abi.gpu.Draw, pass.draws) catch return; + for (draws) |*d| { + d.vertex_uniforms = f.keep(u8, d.vertex_uniforms); + d.fragment_uniforms = f.keep(u8, d.fragment_uniforms); + d.vertex_storage = f.keep(abi.Buffer, d.vertex_storage); + d.fragment_storage = f.keep(abi.Buffer, d.fragment_storage); + } + kept.draws = draws; + f.emit(.{ .gpu = kept }); + } + /// `bytes` into a buffer made by `Host.newBuffer`. + pub fn upload(f: *Frame, buffer: abi.Buffer, offset: u32, bytes: []const u8) void { + f.emit(.{ .upload = .{ .buffer = buffer, .offset = offset, .bytes = bytes } }); + } + /// Runs a compute shader; later commands see what it wrote. + pub fn compute(f: *Frame, dispatch: abi.gpu.Dispatch) void { + var kept = dispatch; + kept.label = f.keep(u8, dispatch.label); + kept.readonly = f.keep(abi.Buffer, dispatch.readonly); + kept.readwrite = f.keep(abi.Buffer, dispatch.readwrite); + kept.uniforms = f.keep(u8, dispatch.uniforms); + f.emit(.{ .compute = kept }); + } + /// Keeps the microphone open this frame: next frame's `in.microphone` + /// holds what it heard. + pub fn microphone(f: *Frame) void { + f.emit(.microphone); + } + /// Reads `size` bytes of a GPU buffer back: they arrive in a later + /// frame's `in.downloads` under `key`. Results lag a frame or two, so + /// it suits measurements (a detected pitch, a level), not every pixel. + pub fn download(f: *Frame, buffer: abi.Buffer, options: struct { offset: u32 = 0, size: u32, key: u32 = 0 }) void { + f.emit(.{ .download = .{ .buffer = buffer, .offset = options.offset, .size = options.size, .key = options.key } }); + } + /// Asks the user for a file; the answer comes back as the console line + /// `<command> <path>`. Emit it once per request, not every frame. + pub fn openFile(f: *Frame, command: []const u8, options: struct { pattern: []const u8 = "" }) void { + f.emit(.{ .open_file = .{ .command = f.keep(u8, command), .pattern = f.keep(u8, options.pattern) } }); + } + /// Runs `line` through the console (and so the sketch's `command`) + /// after this frame, before the next update. + pub fn run(f: *Frame, line: []const u8) void { + f.emit(.{ .run = f.keep(u8, line) }); } pub fn quit(f: *Frame) void { f.emit(.quit); } }; -/// What `command` receives: somewhere to answer, and the current input. +/// What `command` receives: somewhere to answer, the current input, and the +/// allocators for Session and Store. pub const Console = struct { host: Host, in: *const Input, + session: std.mem.Allocator, + store: std.mem.Allocator, pub fn log(c: *Console, message: []const u8) void { c.host.log(message); @@ -144,25 +209,64 @@ pub fn Glue(comptime App: type) type { return struct { pub const State = if (@hasDecl(App, "State")) App.State else struct {}; pub const Session = if (@hasDecl(App, "Session")) App.Session else struct {}; + pub const Store = if (@hasDecl(App, "Store")) App.Store else struct {}; pub const description: abi.Info = .{ .abi = abi.hash, .name = if (@hasDecl(App, "name")) App.name else "sketch", .width = if (@hasDecl(App, "width")) App.width else 960, .height = if (@hasDecl(App, "height")) App.height else 600, - .state = abi.layout.of(State, .forbidden), - .session = abi.layout.of(Session, .host_only), + .state = abi.layout.of(State, .state), + .session = abi.layout.of(Session, .session), + .store = abi.layout.of(Store, .store), .commands = if (@hasDecl(App, "commands")) &App.commands else &.{}, }; - fn state(ptr: [*]u8) *State { - return @ptrCast(@alignCast(ptr)); - } - fn session(ptr: [*]u8) *Session { - return @ptrCast(@alignCast(ptr)); - } - fn arity(comptime func: anytype) usize { - return @typeInfo(@TypeOf(func)).@"fn".params.len; + /// What a call can offer, by type. + const Offer = struct { + ctx: *const abi.Context, + state: [*]u8, + frame: ?*Frame = null, + console: ?*Console = null, + line: []const u8 = "", + }; + + /// Calls `func` with the arguments its parameter types ask for. + fn call(comptime func: anytype, comptime name: []const u8, offer: Offer) @typeInfo(@TypeOf(func)).@"fn".return_type.? { + const F = @TypeOf(func); + var args: std.meta.ArgsTuple(F) = undefined; + inline for (@typeInfo(F).@"fn".param_types, 0..) |P, i| { + const T = P orelse @compileError(name ++ ": generic parameters are not supported"); + const is = struct { + fn named(comptime n: []const u8) bool { + return std.mem.eql(u8, name, n); + } + }; + if (comptime is.named("update") and (T == *Session or T == *Store)) + @compileError("update must be deterministic: it gets State and Input only, so a replay reproduces it"); + if (comptime is.named("draw") and T == *State) @compileError("draw only reads State: take *const State"); + if (comptime T == *Frame and !is.named("draw")) @compileError(name ++ " gets no Frame: only draw does"); + if (comptime T == *Console and !is.named("command")) @compileError(name ++ " gets no Console: only command does"); + args[i] = if (T == *State or T == *const State) + @ptrCast(@alignCast(offer.state)) + else if (T == *Session or T == *const Session) + @ptrCast(@alignCast(offer.ctx.session_root)) + else if (T == *Store or T == *const Store) + @ptrCast(@alignCast(offer.ctx.store_root)) + else if (T == *const Input) + offer.ctx.input + else if (T == Host) + offer.ctx.host + else if (T == *Frame) + offer.frame.? + else if (T == *Console) + offer.console.? + else if (T == []const u8) + offer.line + else + @compileError(name ++ ": no argument of type " ++ @typeName(T)); + } + return @call(.auto, func, args); } pub fn info() callconv(.c) *const abi.Info { @@ -170,36 +274,26 @@ pub fn Glue(comptime App: type) type { } pub fn defaults(which: abi.Memory, ptr: [*]u8) callconv(.c) void { switch (which) { - .state => state(ptr).* = .{}, - .session => session(ptr).* = .{}, + .state => @as(*State, @ptrCast(@alignCast(ptr))).* = .{}, + .session => @as(*Session, @ptrCast(@alignCast(ptr))).* = .{}, + .store => @as(*Store, @ptrCast(@alignCast(ptr))).* = .{}, } } pub fn init(ctx: *const abi.Context, s: [*]u8) callconv(.c) void { - if (!@hasDecl(App, "init")) return; - switch (comptime arity(App.init)) { - 1 => App.init(state(s)), - else => App.init(state(s), ctx.host), - } + if (@hasDecl(App, "init")) call(App.init, "init", .{ .ctx = ctx, .state = s }); } pub fn update(ctx: *const abi.Context, s: [*]u8) callconv(.c) void { - if (@hasDecl(App, "update")) App.update(state(s), ctx.input); + if (@hasDecl(App, "update")) call(App.update, "update", .{ .ctx = ctx, .state = s }); } - pub fn draw(ctx: *const abi.Context, s: [*]const u8, ss: [*]u8) callconv(.c) void { + pub fn draw(ctx: *const abi.Context, s: [*]const u8) callconv(.c) void { if (!@hasDecl(App, "draw")) return; var f: Frame = .init(ctx); - const current: *const State = @ptrCast(@alignCast(s)); - switch (comptime arity(App.draw)) { - 2 => App.draw(current, &f), - else => App.draw(current, session(ss), &f), - } + call(App.draw, "draw", .{ .ctx = ctx, .state = @constCast(s), .frame = &f }); } - pub fn command(ctx: *const abi.Context, s: [*]u8, ss: [*]u8, line: *const []const u8) callconv(.c) abi.Outcome { + pub fn command(ctx: *const abi.Context, s: [*]u8, line: *const []const u8) callconv(.c) abi.Outcome { if (!@hasDecl(App, "command")) return .unknown; - var c: Console = .{ .host = ctx.host, .in = ctx.input }; - const taken = switch (comptime arity(App.command)) { - 3 => App.command(state(s), &c, line.*), - else => App.command(state(s), session(ss), &c, line.*), - } catch |err| { + var c: Console = .{ .host = ctx.host, .in = ctx.input, .session = ctx.session, .store = ctx.store }; + const taken = call(App.command, "command", .{ .ctx = ctx, .state = s, .console = &c, .line = line.* }) catch |err| { c.print("{t}. Type help for usage.", .{err}); return .failed; }; @@ -209,6 +303,7 @@ pub fn Glue(comptime App: type) type { } test { + _ = request; _ = abi.layout; _ = shape; _ = @import("camera.zig"); diff --git a/src/sketch/shape.zig b/src/sketch/shape.zig index 9bd4cee..e7575e1 100644 --- a/src/sketch/shape.zig +++ b/src/sketch/shape.zig @@ -52,7 +52,7 @@ pub const Point = struct { x: f32 = 0, y: f32 = 0 }; /// `abi.Rect`. pub fn Out(comptime T: type) type { if (@typeInfo(T) != .@"struct") return T; - for (@typeInfo(T).@"struct".fields) |field| switch (@typeInfo(field.type)) { + for (@typeInfo(T).@"struct".field_types) |F| switch (@typeInfo(F)) { .comptime_int, .comptime_float => return if (isRect(T)) abi.Rect else Point, else => {}, }; @@ -84,19 +84,20 @@ pub fn make(comptime T: type, x: f32, y: f32, w: f32, h: f32) T { .array => |a| return if (a.len == 2) .{ cast(a.child, x), cast(a.child, y) } else .{ cast(a.child, x), cast(a.child, y), cast(a.child, w), cast(a.child, h) }, .@"struct" => { var result: T = undefined; - inline for (@typeInfo(T).@"struct".fields) |field| { - @field(result, field.name) = if (comptime eql(field.name, "x")) - cast(field.type, x) - else if (comptime eql(field.name, "y")) - cast(field.type, y) - else if (comptime eql(field.name, "w") or eql(field.name, "width")) - cast(field.type, w) - else if (comptime eql(field.name, "h") or eql(field.name, "height")) - cast(field.type, h) - else if (field.defaultValue()) |default| + const s = @typeInfo(T).@"struct"; + inline for (s.field_names, s.field_types, s.field_attrs) |name, F, attrs| { + @field(result, name) = if (comptime eql(name, "x")) + cast(F, x) + else if (comptime eql(name, "y")) + cast(F, y) + else if (comptime eql(name, "w") or eql(name, "width")) + cast(F, w) + else if (comptime eql(name, "h") or eql(name, "height")) + cast(F, h) + else if (comptime attrs.defaultValue(F)) |default| default else - @compileError(@typeName(T) ++ "." ++ field.name ++ " has no default"); + @compileError(@typeName(T) ++ "." ++ name ++ " has no default"); } return result; }, diff --git a/src/sketch/tween.zig b/src/sketch/tween.zig index d3b53a1..6d1cef1 100644 --- a/src/sketch/tween.zig +++ b/src/sketch/tween.zig @@ -25,7 +25,7 @@ pub fn lerp(a: anytype, b: @TypeOf(a), t: f32) @TypeOf(a) { }, .@"struct" => |s| blk: { var out: T = undefined; - inline for (s.fields) |field| @field(out, field.name) = lerp(@field(a, field.name), @field(b, field.name), t); + inline for (s.field_names) |name| @field(out, name) = lerp(@field(a, name), @field(b, name), t); break :blk out; }, else => if (t >= 1) b else a, @@ -49,14 +49,14 @@ pub fn lerpFields(a: anytype, b: @TypeOf(a), t: f32, comptime names: anytype) @T pub fn toward(a: anytype, target: anytype, t: f32) @TypeOf(a) { const T = @TypeOf(a); var out = a; - inline for (@typeInfo(@TypeOf(target)).@"struct".fields) |field| { - if (!@hasField(T, field.name)) @compileError(@typeName(T) ++ " has no field '" ++ field.name ++ "' to animate"); - const F = @FieldType(T, field.name); + inline for (@typeInfo(@TypeOf(target)).@"struct".field_names) |name| { + if (!@hasField(T, name)) @compileError(@typeName(T) ++ " has no field '" ++ name ++ "' to animate"); + const F = @FieldType(T, name); const goal: F = switch (@typeInfo(F)) { - .@"struct" => toward(@field(a, field.name), @field(target, field.name), 1), - else => @field(target, field.name), + .@"struct" => toward(@field(a, name), @field(target, name), 1), + else => @field(target, name), }; - @field(out, field.name) = lerp(@field(a, field.name), goal, t); + @field(out, name) = lerp(@field(a, name), goal, t); } return out; } diff --git a/src/sketch/web.zig b/src/sketch/web.zig index e91a0ad..bc7fead 100644 --- a/src/sketch/web.zig +++ b/src/sketch/web.zig @@ -1,6 +1,8 @@ -//! Runs a sketch in a browser. `web/host.js` forwards input and implements a -//! handful of drawing imports; state, `update` and `draw` are the same Zig as -//! on native. There is no console, timeline or persistence on the web. +//! Runs a sketch in a browser. `web/host.js` forwards input and implements +//! the imports below: drawing on a 2D canvas, GPU work on WebGPU, sound and +//! the microphone on Web Audio, files through the page. State, `update` and +//! `draw` are the same Zig as on native. There is no timeline, and nothing +//! persists: Store lasts as long as the page. const std = @import("std"); const abi = @import("abi"); const root = @import("root.zig"); @@ -11,7 +13,9 @@ const js = struct { extern "host" fn imageWidth(image: u32) u32; extern "host" fn imageHeight(image: u32) u32; extern "host" fn newImage(w: u32, h: u32) u32; + extern "host" fn freeImage(image: u32) void; extern "host" fn clock() f64; + extern "host" fn openFile(command: [*]const u8, command_len: usize, pattern: [*]const u8, pattern_len: usize) void; extern "canvas" fn clear(r: u32, g: u32, b: u32, a: u32) void; extern "canvas" fn fill(x: f32, y: f32, w: f32, h: f32, r: u32, g: u32, b: u32, a: u32) void; extern "canvas" fn image(id: u32, x: f32, y: f32, w: f32, h: f32, sx: f32, sy: f32, sw: f32, sh: f32, alpha: u32) void; @@ -20,16 +24,40 @@ const js = struct { extern "canvas" fn canvas(w: u32, h: u32) void; extern "canvas" fn cursor(kind: u32) void; extern "canvas" fn pixels(id: u32, w: u32, h: u32, ptr: [*]const u8, len: usize, premultiplied: u32) void; + /// Nonzero once the page has a WebGPU device. + extern "gpu" fn available() u32; + extern "gpu" fn newBuffer(size: u32) u32; + extern "gpu" fn freeBuffer(id: u32) void; + extern "gpu" fn write(id: u32, offset: u32, ptr: [*]const u8, len: usize) void; + /// A `Dispatch` or a `Pass`, described in JSON (see `describe`). + extern "gpu" fn compute(ptr: [*]const u8, len: usize) void; + extern "gpu" fn pass(ptr: [*]const u8, len: usize) void; + /// Answers through `sk_download`. + extern "gpu" fn download(id: u32, offset: u32, size: u32, key: u32) void; + extern "audio" fn rate() u32; + extern "audio" fn push(key: u32, ptr: [*]const f32, len: usize, clear: u32) void; + extern "audio" fn queued(key: u32) u32; + /// Keeps the microphone open; it answers through `sk_microphone`. + extern "audio" fn listen() void; }; -/// Exports `App` for `web/host.js`: `sk_init`, `sk_event`, `sk_frame`, and -/// for tests to read state by path, `sk_field` (a number) and `sk_tag` (an -/// enum's tag name). +/// Exports `App` for `web/host.js`: `sk_init`, `sk_event`, `sk_frame`, +/// `sk_window` (the room for the canvas); +/// `sk_alloc` and the deliveries into what it returns (`sk_microphone`, +/// `sk_download`, `sk_sound`, `sk_command`); and for tests to read state by +/// path, `sk_buffer`, `sk_field` (a number) and `sk_tag` (an enum's tag +/// name). pub fn exportWeb(comptime App: type) void { const D = Driver(App); @export(&D.init, .{ .name = "sk_init" }); @export(&D.event, .{ .name = "sk_event" }); @export(&D.frame, .{ .name = "sk_frame" }); + @export(&D.window, .{ .name = "sk_window" }); + @export(&D.alloc, .{ .name = "sk_alloc" }); + @export(&D.microphone, .{ .name = "sk_microphone" }); + @export(&D.download, .{ .name = "sk_download" }); + @export(&D.sound, .{ .name = "sk_sound" }); + @export(&D.command, .{ .name = "sk_command" }); @export(&D.buffer, .{ .name = "sk_buffer" }); @export(&D.field, .{ .name = "sk_field" }); @export(&D.tag, .{ .name = "sk_tag" }); @@ -37,24 +65,36 @@ pub fn exportWeb(comptime App: type) void { fn Driver(comptime App: type) type { const G = root.Glue(App); + const gpa = std.heap.wasm_allocator; return struct { var state: G.State = undefined; var session: G.Session = undefined; + // No disk on the web: the store lasts as long as the page. + var store: G.Store = undefined; var events: [1024]abi.Event = undefined; var event_count: usize = 0; var input: abi.Input = .{}; var commands: std.ArrayList(abi.Command) = .empty; - var arena: std.heap.ArenaAllocator = .init(std.heap.wasm_allocator); - var session_heap: std.heap.ArenaAllocator = .init(std.heap.wasm_allocator); + var scratch: std.heap.ArenaAllocator = .init(gpa); var last: u64 = 0; var path: [256]u8 = undefined; + /// What arrived since the last frame, handed to the next one. + var heard: std.ArrayList(f32) = .empty; + var downloads: std.ArrayList(abi.Download) = .empty; + /// Streams the sketch has pushed to, by key. + var streams: std.ArrayList(abi.Stream) = .empty; + /// Sounds the page decoded, by name: `Audio` n is `sounds[n - 1]`. + var sounds: std.ArrayList(struct { name: []const u8, samples: []const f32 }) = .empty; fn context() abi.Context { return .{ .host = .{ .ctx = @ptrCast(&input), .vtable = &vtable }, .input = &input, - .arena = arena.allocator(), - .gpa = session_heap.allocator(), + .scratch = scratch.allocator(), + .session = gpa, + .store = gpa, + .session_root = std.mem.asBytes(&session), + .store_root = std.mem.asBytes(&store), .commands = &commands, }; } @@ -62,21 +102,33 @@ fn Driver(comptime App: type) type { fn init() callconv(.c) void { G.defaults(.state, std.mem.asBytes(&state)); G.defaults(.session, std.mem.asBytes(&session)); + G.defaults(.store, std.mem.asBytes(&store)); input.width = G.description.width; input.height = G.description.height; + if (input.window_width == 0) { + input.window_width = input.width; + input.window_height = input.height; + } + input.sample_rate = js.rate(); js.canvas(G.description.width, G.description.height); const ctx = context(); G.init(&ctx, std.mem.asBytes(&state)); } - fn event(kind: u32, code: u32, x: f32, y: f32, value: u32) callconv(.c) void { + /// The page's room for the canvas, in CSS pixels. + fn window(w: f32, h: f32) callconv(.c) void { + input.window_width = w; + input.window_height = h; + } + + fn event(kind: u32, code: u32, x: f32, y: f32, value: u32, flags: u32) callconv(.c) void { const k = std.enums.fromInt(abi.Event.Kind, kind) orelse return; if (event_count == events.len) { // Never leave a gesture armed when the queue overflows. events[0] = .{ .kind = .cancel }; event_count = 1; } - events[event_count] = .{ .kind = k, .code = @intCast(code), .x = x, .y = y, .value = value }; + events[event_count] = .{ .kind = k, .code = @intCast(code), .x = x, .y = y, .value = value, .flags = @truncate(flags) }; event_count += 1; } @@ -92,27 +144,39 @@ fn Driver(comptime App: type) type { e.time = now; input.track(e.*); } - _ = arena.reset(.retain_capacity); + _ = scratch.reset(.retain_capacity); commands = .empty; input.now = now; input.dt = dt; input.frame += 1; input.events = events[0..event_count]; input.started = input.frame == 1; + for (streams.items) |*s| s.queued = js.queued(s.key); + input.streams = streams.items; + input.microphone = heard.items; + input.downloads = downloads.items; const ctx = context(); G.update(&ctx, std.mem.asBytes(&state)); input.events = &.{}; event_count = 0; - G.draw(&ctx, std.mem.asBytes(&state), std.mem.asBytes(&session)); + G.draw(&ctx, std.mem.asBytes(&state)); js.clip(0, 0, 0, 0, 0); - for (commands.items) |command| submit(command); + for (commands.items) |c| submit(c); + // After the frame, as on native: lines `draw` asked to run. + for (commands.items) |c| if (c == .run) run(c.run); + + heard.clearRetainingCapacity(); + for (downloads.items) |d| gpa.free(d.bytes); + downloads.clearRetainingCapacity(); + input.microphone = &.{}; + input.downloads = &.{}; } - fn submit(command: abi.Command) void { - switch (command) { + fn submit(c: abi.Command) void { + switch (c) { .clear => |v| js.clear(v.r, v.g, v.b, v.a), .rect => |v| js.fill(v.dst.x, v.dst.y, v.dst.w, v.dst.h, v.color.r, v.color.g, v.color.b, v.color.a), - .image => |v| js.image(@intFromEnum(v.image), v.dst.x, v.dst.y, v.dst.w, v.dst.h, v.src.x, v.src.y, v.src.w, v.src.h, v.tint.a), + .image => |v| js.image(@backingInt(v.image), v.dst.x, v.dst.y, v.dst.w, v.dst.h, v.src.x, v.src.y, v.src.w, v.src.h, v.tint.a), .text => |v| js.text(v.x, v.y, v.scale, v.bytes.ptr, v.bytes.len, v.color.r, v.color.g, v.color.b, v.color.a), .clip => |v| if (v) |r| js.clip(1, r.x, r.y, r.w, r.h) else js.clip(0, 0, 0, 0, 0), .canvas => |v| { @@ -120,20 +184,138 @@ fn Driver(comptime App: type) type { input.height = @floatFromInt(v.h); js.canvas(v.w, v.h); }, - .cursor => |v| js.cursor(@intFromEnum(v)), - .pixels => |v| js.pixels(@intFromEnum(v.image), v.w, v.h, v.rgba.ptr, v.rgba.len, @intFromBool(v.premultiplied)), - // No sound or GPU passes on the web yet (Host.gpu says so); - // quitting a page means nothing. - .voice, .audio, .gpu, .quit => {}, + .cursor => |v| js.cursor(@backingInt(v)), + .pixels => |v| js.pixels(@backingInt(v.image), v.w, v.h, v.rgba.ptr, v.rgba.len, @intFromBool(v.premultiplied)), + .audio => |v| { + const s = stream(v.key); + if (v.clear) s.pushed = 0; + s.pushed += v.samples.len; + js.push(v.key, v.samples.ptr, v.samples.len, @intFromBool(v.clear)); + }, + .microphone => js.listen(), + .upload => |v| js.write(@backingInt(v.buffer), v.offset, v.bytes.ptr, v.bytes.len), + .compute => |v| { + const json = describe(v); + js.compute(json.ptr, json.len); + }, + .gpu => |v| { + const json = describe(v); + js.pass(json.ptr, json.len); + }, + .download => |v| js.download(@backingInt(v.buffer), v.offset, v.size, v.key), + .open_file => |v| js.openFile(v.command.ptr, v.command.len, v.pattern.ptr, v.pattern.len), + // Synthesised voices are not on the web yet; quitting a page + // means nothing; `run` lines go after the frame. + .voice, .quit, .run => {}, } } + fn stream(key: u32) *abi.Stream { + for (streams.items) |*s| if (s.key == key) return s; + streams.append(gpa, .{ .key = key }) catch return &streams.items[0]; + return &streams.items[streams.items.len - 1]; + } + + fn run(line: []const u8) void { + const ctx = context(); + if (G.command(&ctx, std.mem.asBytes(&state), &line) == .unknown) { + var buf: [160]u8 = undefined; + const message = std.fmt.bufPrint(&buf, "unknown command: {s}", .{line[0..@min(line.len, 120)]}) catch return; + js.log(message.ptr, message.len); + } + } + + /// GPU work as JSON for host.js: buffers by handle, bytes (code, + /// uniforms, vertices) as `[pointer, length]` into wasm memory, and + /// each shader with a hash of its code to cache it by. + fn describe(value: anytype) []const u8 { + var out: std.Io.Writer.Allocating = .init(scratch.allocator()); + var json: std.json.Stringify = .{ .writer = &out.writer }; + write(&json, value) catch return "{}"; + return out.written(); + } + + fn write(json: *std.json.Stringify, value: anytype) !void { + const T = @TypeOf(value); + if (T == []const u8) { + try json.beginArray(); + try json.write(@intFromPtr(value.ptr)); + try json.write(value.len); + return json.endArray(); + } + if (T == abi.gpu.Shader or T == abi.gpu.Kernel) { + // The code once more, by hash: compiled modules are cached. + try json.beginObject(); + try json.objectField("hash"); + try json.print("\"{x}\"", .{std.hash.Wyhash.hash(0, value.code)}); + inline for (@typeInfo(T).@"struct".field_names) |name| { + try json.objectField(name); + try write(json, @field(value, name)); + } + return json.endObject(); + } + switch (@typeInfo(T)) { + .@"struct" => |s| { + try json.beginObject(); + inline for (s.field_names) |name| { + try json.objectField(name); + try write(json, @field(value, name)); + } + try json.endObject(); + }, + .pointer => { + try json.beginArray(); + for (value) |item| try write(json, item); + try json.endArray(); + }, + .array => { + try json.beginArray(); + for (value) |item| try write(json, item); + try json.endArray(); + }, + .optional => if (value) |v| try write(json, v) else try json.write(null), + .@"enum" => |e| if (e.mode == .exhaustive) try json.write(@tagName(value)) else try json.write(@backingInt(value)), + else => try json.write(value), + } + } + + /// Memory for host.js to write a delivery into; the delivery takes + /// it over. + fn alloc(len: usize) callconv(.c) ?[*]u8 { + const bytes = gpa.alloc(u8, len) catch return null; + return bytes.ptr; + } + + fn microphone(ptr: [*]f32, len: usize) callconv(.c) void { + heard.appendSlice(gpa, ptr[0..len]) catch {}; + gpa.free(std.mem.sliceAsBytes(ptr[0..len])); + } + + fn download(key: u32, ptr: [*]u8, len: usize) callconv(.c) void { + downloads.append(gpa, .{ .key = key, .bytes = ptr[0..len] }) catch gpa.free(ptr[0..len]); + } + + /// A decoded sound for `loadSound(name)`; a name seen before is + /// replaced. + fn sound(name: [*]u8, name_len: usize, decoded: [*]f32, len: usize) callconv(.c) void { + for (sounds.items) |*s| if (std.mem.eql(u8, s.name, name[0..name_len])) { + gpa.free(name[0..name_len]); + s.samples = decoded[0..len]; + return; + }; + sounds.append(gpa, .{ .name = name[0..name_len], .samples = decoded[0..len] }) catch {}; + } + + /// Runs a console line through the sketch's `command`. + fn command(ptr: [*]u8, len: usize) callconv(.c) void { + run(ptr[0..len]); + gpa.free(ptr[0..len]); + } + fn buffer() callconv(.c) [*]u8 { return &path; } - /// The state field at the path written into `sk_buffer`, as a number - /// (enums by value); NaN when there is no such scalar field. /// Writes the tag name of the enum field at the path in `buffer` over /// it, and returns its length; -1 when there is no such enum field. fn tag(len: usize) callconv(.c) i32 { @@ -152,6 +334,8 @@ fn Driver(comptime App: type) type { return -1; } + /// The state field at the path written into `sk_buffer`, as a number + /// (enums by value); NaN when there is no such scalar field. fn field(len: usize) callconv(.c) f64 { const name = path[0..@min(len, path.len)]; const bytes = std.mem.asBytes(&state); @@ -180,9 +364,12 @@ fn Driver(comptime App: type) type { .newImage = newImage, .freeImage = freeImage, .gpu = gpu, + .newBuffer = newBuffer, + .freeBuffer = freeBuffer, .imageSize = imageSize, - .loadSound = unsupportedSound, + .loadSound = loadSound, .decodeSound = unsupportedSound, + .samples = samples, .readFile = readFile, .env = env, .clock = clock, @@ -191,24 +378,42 @@ fn Driver(comptime App: type) type { js.log(message.ptr, message.len); } fn loadImage(_: *anyopaque, p: []const u8) abi.Image { - return @enumFromInt(js.loadImage(p.ptr, p.len)); + return @fromBackingInt(@intCast(js.loadImage(p.ptr, p.len))); } fn unsupportedImage(_: *anyopaque, _: []const u8) abi.Image { return .none; } fn newImage(_: *anyopaque, w: u32, h: u32) abi.Image { - return @enumFromInt(js.newImage(w, h)); + return @fromBackingInt(@intCast(js.newImage(w, h))); + } + fn freeImage(_: *anyopaque, img: abi.Image) void { + js.freeImage(@backingInt(img)); } - fn freeImage(_: *anyopaque, _: abi.Image) void {} fn gpu(_: *anyopaque) bool { - return false; + return js.available() != 0; + } + fn newBuffer(_: *anyopaque, size: u32) abi.Buffer { + return @fromBackingInt(js.newBuffer(size)); + } + fn freeBuffer(_: *anyopaque, b: abi.Buffer) void { + js.freeBuffer(@backingInt(b)); } fn imageSize(_: *anyopaque, img: abi.Image) [2]u32 { - return .{ js.imageWidth(@intFromEnum(img)), js.imageHeight(@intFromEnum(img)) }; + return .{ js.imageWidth(@backingInt(img)), js.imageHeight(@backingInt(img)) }; + } + /// Only sounds the page has decoded (opened or dropped files): the + /// browser decodes asynchronously, so there is no reading on demand. + fn loadSound(_: *anyopaque, name: []const u8) abi.Audio { + for (sounds.items, 1..) |s, i| if (std.mem.eql(u8, s.name, name)) return @fromBackingInt(@intCast(i)); + return .none; } fn unsupportedSound(_: *anyopaque, _: []const u8) abi.Audio { return .none; } + fn samples(_: *anyopaque, a: abi.Audio) []const f32 { + const i = @backingInt(a); + return if (i == 0 or i > sounds.items.len) &.{} else sounds.items[i - 1].samples; + } fn readFile(_: *anyopaque, _: []const u8) []const u8 { return &.{}; } diff --git a/src/vector/Gpu.zig b/src/vector/Gpu.zig index a557537..5853145 100644 --- a/src/vector/Gpu.zig +++ b/src/vector/Gpu.zig @@ -19,8 +19,8 @@ const shaders = @import("vector_shaders"); pub const samples = 8; -const vertex: abi.gpu.Shader = .{ .spirv = shaders.path_vert, .uniform_buffers = 1 }; -const stencil_only: abi.gpu.Shader = .{ .spirv = shaders.stencil_frag }; +const vertex: abi.gpu.Shader = .{ .code = shaders.path_vert, .uniform_buffers = 1 }; +const stencil_only: abi.gpu.Shader = .{ .code = shaders.stencil_frag }; const pipelines = [_]abi.gpu.Pipeline{ // 0: nonzero winding. @@ -42,7 +42,7 @@ const pipelines = [_]abi.gpu.Pipeline{ // 2: cover. abi.gpu.pipeline(layouts.Vertex, .{ .vertex = vertex, - .fragment = .{ .spirv = shaders.cover_frag, .uniform_buffers = 1 }, + .fragment = .{ .code = shaders.cover_frag, .uniform_buffers = 1 }, .stencil = .{ .front = .{ .compare = .not_equal, .pass = .zero }, .back = .{ .compare = .not_equal, .pass = .zero }, @@ -116,8 +116,8 @@ fn block(paint: *const Paint, alpha: f32) layouts.Paint { const m = paint.to_gradient; var out: layouts.Paint = .{ .color = paint.color, - .row0 = .{ m.a, m.c, m.e, @floatFromInt(@intFromEnum(paint.kind)) }, - .row1 = .{ m.b, m.d, m.f, @floatFromInt(@intFromEnum(paint.spread)) }, + .row0 = .{ m.a, m.c, m.e, @floatFromInt(@backingInt(paint.kind)) }, + .row1 = .{ m.b, m.d, m.f, @floatFromInt(@backingInt(paint.spread)) }, .points = .{ paint.p0[0], paint.p0[1], paint.p1[0], paint.p1[1] }, .extra = .{ paint.radius, alpha, 0, 0 }, .offsets = @splat(@splat(0)), @@ -157,7 +157,7 @@ pub fn pass(g: *const Gpu) abi.gpu.Pass { test "a fill is a stencil fan and a covering quad with its paint" { var arena: std.heap.ArenaAllocator = .init(std.testing.allocator); defer arena.deinit(); - var g: Gpu = .{ .width = 100, .height = 50, .image = @enumFromInt(1) }; + var g: Gpu = .{ .width = 100, .height = 50, .image = @fromBackingInt(@intCast(1)) }; g.begin(arena.allocator(), .{ 0, 0, 0, 1 }); var p: Polygons = .init(arena.allocator()); try p.points.appendSlice(arena.allocator(), &.{ .{ 10, 10 }, .{ 40, 10 }, .{ 40, 30 }, .{ 10, 30 } }); diff --git a/src/vector/root.zig b/src/vector/root.zig index 7556948..0414286 100644 --- a/src/vector/root.zig +++ b/src/vector/root.zig @@ -69,12 +69,12 @@ pub fn Renderer(comptime backend: Backend) type { return r.target.height; } - /// Starts a picture. `f.arena` holds what the GPU pass needs until + /// Starts a picture. `f.scratch` holds what the GPU pass needs until /// the frame is consumed. pub fn begin(r: *Self, f: anytype, clear: Drawing.Color) void { switch (backend) { .cpu => r.target.clear(clear), - .gpu => r.target.begin(f.arena, clear), + .gpu => r.target.begin(f.scratch, clear), } } diff --git a/test/fonts.c b/test/fonts.c new file mode 100644 index 0000000..87a47bb --- /dev/null +++ b/test/fonts.c @@ -0,0 +1,78 @@ +/* Standalone Linux font regression/specimen. See vendor/fonts/README.md. */ +#include "../vendor/text.c" +#include <assert.h> +#include <stdio.h> + +static unsigned glyphs(PangoLayout *layout) { + unsigned count = 0; + PangoLayoutIter *it = pango_layout_get_iter(layout); + do { + PangoLayoutRun *run = pango_layout_iter_get_run_readonly(it); + if (run) count += run->glyphs->num_glyphs; + } while (pango_layout_iter_next_run(it)); + pango_layout_iter_free(it); + return count; +} +int main(int argc, char **argv) { + assert(SDL_Init(0)); + SDL_Surface *surface = SDL_CreateSurface(1440, 820, SDL_PIXELFORMAT_ARGB8888); + SDL_Renderer *r = SDL_CreateSoftwareRenderer(surface); + assert(r); + SDL_SetRenderDrawColor(r, 17, 21, 27, 255); SDL_RenderClear(r); + sk_text_begin_frame(r); + const char *samples[] = { + "Go Mono / native shaped text", + "Pardes / 46.8M accesses / 3.97M calls", + "0123456789 0x7ffed849acf0 () [] {} <> -> != ==", + "Latin: cafe\xcc\x81 / café / naïve / fi fl ffi", + "Greek / Cyrillic: λ → Ω Привет мир", + "Japanese fallback: 日本語 ひらがな カタカナ", + "Arabic shaping: العربية لا السلام عليكم", + "Hebrew bidi: שלום עולם | English 123", + "Symbols / emoji: ✓ ⌘ ⌥ ⌫ ⌜ ⌟ 😀 ★ ♥", + "draw.render → body_layer.renderBody → TerminalImages.draw" + }; + for (unsigned i=0; i<sizeof(samples)/sizeof(*samples); ++i) { + assert(sk_text_draw(r, 30, 25 + 53*i, i==0?2.2f:1.6f, samples[i], strlen(samples[i]), + i==0?228:165,i==0?234:178,i==0?242:195,255)); + TextCache *c=cache_for(r); + assert(pango_layout_get_unknown_glyphs_count(c->layout)==0); + } + TextCache *c=cache_for(r); + assert(sk_text_draw(r,30,575,1.0f,"e\xcc\x81",3,228,234,242,255)); + PangoRectangle combining, plain; + pango_layout_get_pixel_extents(c->layout,NULL,&combining); + assert(sk_text_draw(r,50,575,1.0f,"e",1,228,234,242,255)); + pango_layout_get_pixel_extents(c->layout,NULL,&plain); + assert(combining.width==plain.width); + assert(sk_text_draw(r,70,575,1.0f,"لا",strlen("لا"),228,234,242,255)); + assert(glyphs(c->layout)==1); + assert(sk_text_draw(r,120,575,1.0f,"0123456789",10,228,234,242,255)); + PangoRectangle bounds; pango_layout_get_pixel_extents(c->layout,NULL,&bounds); + assert(bounds.width>=75 && bounds.width<=85 && bounds.height>=13); + printf("scale 1: 10-digit advance %dpx, line box %dpx\n",bounds.width,bounds.height); + SDL_Rect clip={30,620,400,30}; SDL_SetRenderClipRect(r,&clip); + assert(sk_text_draw(r,30,617,1.6f,"Clipping preserves the baseline and glyph edges outside this box",strlen("Clipping preserves the baseline and glyph edges outside this box"),112,183,255,255)); + SDL_SetRenderClipRect(r,NULL); + assert(SDL_RenderPresent(r)); + if(argc>1) assert(SDL_SavePNG(surface,argv[1])); + assert(SDL_SetRenderLogicalPresentation(r,720,410,SDL_LOGICAL_PRESENTATION_LETTERBOX)); + assert(fabsf(density(r)-2)<0.01f); + assert(sk_text_draw(r,10,10,1.0f,"0123456789",10,228,234,242,255)); + pango_layout_get_pixel_extents(c->layout,NULL,&bounds); + assert(bounds.width>=150 && bounds.width<=170); + /* Eviction across frames and exhaustion inside one frame both stay bounded. */ + for(unsigned i=0;i<800;i++) { + if(i%100==0) sk_text_begin_frame(r); + char text[32]; int n=snprintf(text,sizeof(text),"cache %u",i); + assert(sk_text_draw(r,0,0,1,text,(size_t)n,255,255,255,255)); + assert(c->bytes<=CACHE_BYTES); + } + for(unsigned i=0;i<600;i++) { + char text[32]; int n=snprintf(text,sizeof(text),"transient %u",i); + assert(sk_text_draw(r,0,0,1,text,(size_t)n,255,255,255,255)); + assert(c->bytes<=CACHE_BYTES); + } + SDL_DestroyRenderer(r); SDL_DestroySurface(surface); SDL_Quit(); + puts("Go Mono, fallback, combining marks, Arabic shaping, 2x DPI, clipping and bounded cache passed"); +} diff --git a/vendor/README.md b/vendor/README.md new file mode 100644 index 0000000..97c311d --- /dev/null +++ b/vendor/README.md @@ -0,0 +1,7 @@ +Single-file codecs, vendored unchanged; `codecs.c` compiles them for the host. + +| File | Upstream | Version | +| --- | --- | --- | +| `dr_wav.h` | https://github.com/mackron/dr_libs | 0.14.6 | +| `dr_flac.h` | https://github.com/mackron/dr_libs | 0.13.4 | +| `stb_vorbis.c` | https://github.com/nothings/stb | v1.22 | diff --git a/vendor/codecs.c b/vendor/codecs.c new file mode 100644 index 0000000..7d54f34 --- /dev/null +++ b/vendor/codecs.c @@ -0,0 +1,57 @@ +#include <stdlib.h> +#include <string.h> +#include "codecs.h" + +#define DR_WAV_IMPLEMENTATION +#define DR_WAV_NO_STDIO +#include "dr_wav.h" +#define DR_FLAC_IMPLEMENTATION +#define DR_FLAC_NO_STDIO +#include "dr_flac.h" +#define STB_VORBIS_NO_STDIO +#define STB_VORBIS_NO_PUSHDATA_API +#include "stb_vorbis.c" + +static float *decode_vorbis(const void *data, size_t size, uint32_t *channels, uint32_t *rate, uint64_t *frames) { + if (size > INT32_MAX) return NULL; + int error = 0; + stb_vorbis *v = stb_vorbis_open_memory(data, (int)size, &error, NULL); + if (!v) return NULL; + stb_vorbis_info info = stb_vorbis_get_info(v); + unsigned total = stb_vorbis_stream_length_in_samples(v); + float *out = NULL; + if (info.channels > 0 && total > 0) out = malloc((size_t)total * info.channels * sizeof(float)); + if (!out) { + stb_vorbis_close(v); + return NULL; + } + size_t got = 0; + while (got < total) { + int n = stb_vorbis_get_samples_float_interleaved(v, info.channels, out + got * info.channels, (int)((total - got) * info.channels)); + if (n <= 0) break; + got += (size_t)n; + } + stb_vorbis_close(v); + *channels = (uint32_t)info.channels; + *rate = info.sample_rate; + *frames = got; + return out; +} + +float *sketchbook_decode_audio(const void *data, size_t size, uint32_t *channels, uint32_t *rate, uint64_t *frames) { + unsigned int ch = 0, sr = 0; + drwav_uint64 n = 0; + float *out = NULL; + if (size < 4) return NULL; + if (memcmp(data, "OggS", 4) == 0) return decode_vorbis(data, size, channels, rate, frames); + if (memcmp(data, "fLaC", 4) == 0) out = drflac_open_memory_and_read_pcm_frames_f32(data, size, &ch, &sr, &n, NULL); + else out = drwav_open_memory_and_read_pcm_frames_f32(data, size, &ch, &sr, &n, NULL); + *channels = ch; + *rate = sr; + *frames = n; + return out; +} + +void sketchbook_free_audio(float *samples) { + free(samples); +} diff --git a/vendor/codecs.h b/vendor/codecs.h new file mode 100644 index 0000000..a4cb120 --- /dev/null +++ b/vendor/codecs.h @@ -0,0 +1,9 @@ +/* Sound decoding for the host, from single-file libraries: dr_wav, dr_flac + * and stb_vorbis, compiled in codecs.c. */ +#include <stddef.h> +#include <stdint.h> + +/* WAV, FLAC or Ogg Vorbis bytes to interleaved f32, chosen by content. Null + * when they do not decode; otherwise release with sketchbook_free_audio. */ +float *sketchbook_decode_audio(const void *data, size_t size, uint32_t *channels, uint32_t *rate, uint64_t *frames); +void sketchbook_free_audio(float *samples); diff --git a/vendor/dr_flac.h b/vendor/dr_flac.h new file mode 100644 index 0000000..4ca68e9 --- /dev/null +++ b/vendor/dr_flac.h @@ -0,0 +1,12475 @@ +/* +FLAC audio decoder. Choice of public domain or MIT-0. See license statements at the end of this file. +dr_flac - v0.13.4 - TBD + +David Reid - [email protected] + +GitHub: https://github.com/mackron/dr_libs +*/ + +/* +Introduction +============ +dr_flac is a single file library. To use it, do something like the following in one .c file. + + ```c + #define DR_FLAC_IMPLEMENTATION + #include "dr_flac.h" + ``` + +You can then #include this file in other parts of the program as you would with any other header file. To decode audio data, do something like the following: + + ```c + drflac* pFlac = drflac_open_file("MySong.flac", NULL); + if (pFlac == NULL) { + // Failed to open FLAC file + } + + drflac_int32* pSamples = malloc(pFlac->totalPCMFrameCount * pFlac->channels * sizeof(drflac_int32)); + drflac_uint64 numberOfInterleavedSamplesActuallyRead = drflac_read_pcm_frames_s32(pFlac, pFlac->totalPCMFrameCount, pSamples); + ``` + +The drflac object represents the decoder. It is a transparent type so all the information you need, such as the number of channels and the bits per sample, +should be directly accessible - just make sure you don't change their values. Samples are always output as interleaved signed 32-bit PCM. In the example above +a native FLAC stream was opened, however dr_flac has seamless support for Ogg encapsulated FLAC streams as well. + +You do not need to decode the entire stream in one go - you just specify how many samples you'd like at any given time and the decoder will give you as many +samples as it can, up to the amount requested. Later on when you need the next batch of samples, just call it again. Example: + + ```c + while (drflac_read_pcm_frames_s32(pFlac, chunkSizeInPCMFrames, pChunkSamples) > 0) { + do_something(); + } + ``` + +You can seek to a specific PCM frame with `drflac_seek_to_pcm_frame()`. + +If you just want to quickly decode an entire FLAC file in one go you can do something like this: + + ```c + unsigned int channels; + unsigned int sampleRate; + drflac_uint64 totalPCMFrameCount; + drflac_int32* pSampleData = drflac_open_file_and_read_pcm_frames_s32("MySong.flac", &channels, &sampleRate, &totalPCMFrameCount, NULL); + if (pSampleData == NULL) { + // Failed to open and decode FLAC file. + } + + ... + + drflac_free(pSampleData, NULL); + ``` + +You can read samples as signed 16-bit integer and 32-bit floating-point PCM with the *_s16() and *_f32() family of APIs respectively, but note that these +should be considered lossy. + + +If you need access to metadata (album art, etc.), use `drflac_open_with_metadata()`, `drflac_open_file_with_metdata()` or `drflac_open_memory_with_metadata()`. +The rationale for keeping these APIs separate is that they're slightly slower than the normal versions and also just a little bit harder to use. dr_flac +reports metadata to the application through the use of a callback, and every metadata block is reported before `drflac_open_with_metdata()` returns. + +The main opening APIs (`drflac_open()`, etc.) will fail if the header is not present. The presents a problem in certain scenarios such as broadcast style +streams or internet radio where the header may not be present because the user has started playback mid-stream. To handle this, use the relaxed APIs: + + `drflac_open_relaxed()` + `drflac_open_with_metadata_relaxed()` + +It is not recommended to use these APIs for file based streams because a missing header would usually indicate a corrupt or perverse file. In addition, these +APIs can take a long time to initialize because they may need to spend a lot of time finding the first frame. + + + +Build Options +============= +#define these options before including this file. + +#define DR_FLAC_NO_STDIO + Disable `drflac_open_file()` and family. + +#define DR_FLAC_NO_OGG + Disables support for Ogg/FLAC streams. + +#define DR_FLAC_BUFFER_SIZE <number> + Defines the size of the internal buffer to store data from onRead(). This buffer is used to reduce the number of calls back to the client for more data. + Larger values means more memory, but better performance. My tests show diminishing returns after about 4KB (which is the default). Consider reducing this if + you have a very efficient implementation of onRead(), or increase it if it's very inefficient. Must be a multiple of 8. + +#define DR_FLAC_NO_CRC + Disables CRC checks. This will offer a performance boost when CRC is unnecessary. This will disable binary search seeking. When seeking, the seek table will + be used if available. Otherwise the seek will be performed using brute force. + +#define DR_FLAC_NO_SIMD + Disables SIMD optimizations (SSE on x86/x64 architectures, NEON on ARM architectures). Use this if you are having compatibility issues with your compiler. + +#define DR_FLAC_NO_WCHAR + Disables all functions ending with `_w`. Use this if your compiler does not provide wchar.h. Not required if DR_FLAC_NO_STDIO is also defined. + + + +Notes +===== +- dr_flac does not support changing the sample rate nor channel count mid stream. +- dr_flac is not thread-safe, but its APIs can be called from any thread so long as you do your own synchronization. +- When using Ogg encapsulation, a corrupted metadata block will result in `drflac_open_with_metadata()` and `drflac_open()` returning inconsistent samples due + to differences in corrupted stream recorvery logic between the two APIs. +*/ + +#ifndef dr_flac_h +#define dr_flac_h + +#ifdef __cplusplus +extern "C" { +#endif + +#define DRFLAC_STRINGIFY(x) #x +#define DRFLAC_XSTRINGIFY(x) DRFLAC_STRINGIFY(x) + +#define DRFLAC_VERSION_MAJOR 0 +#define DRFLAC_VERSION_MINOR 13 +#define DRFLAC_VERSION_REVISION 4 +#define DRFLAC_VERSION_STRING DRFLAC_XSTRINGIFY(DRFLAC_VERSION_MAJOR) "." DRFLAC_XSTRINGIFY(DRFLAC_VERSION_MINOR) "." DRFLAC_XSTRINGIFY(DRFLAC_VERSION_REVISION) + +#include <stddef.h> /* For size_t. */ + +/* Sized Types */ +typedef signed char drflac_int8; +typedef unsigned char drflac_uint8; +typedef signed short drflac_int16; +typedef unsigned short drflac_uint16; +typedef signed int drflac_int32; +typedef unsigned int drflac_uint32; +#if defined(_MSC_VER) && !defined(__clang__) + typedef signed __int64 drflac_int64; + typedef unsigned __int64 drflac_uint64; +#else + #if defined(__clang__) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))) + #pragma GCC diagnostic push + #pragma GCC diagnostic ignored "-Wlong-long" + #if defined(__clang__) + #pragma GCC diagnostic ignored "-Wc++11-long-long" + #endif + #endif + typedef signed long long drflac_int64; + typedef unsigned long long drflac_uint64; + #if defined(__clang__) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))) + #pragma GCC diagnostic pop + #endif +#endif +#if defined(__LP64__) || defined(_WIN64) || (defined(__x86_64__) && !defined(__ILP32__)) || defined(_M_X64) || defined(__ia64) || defined(_M_IA64) || defined(__aarch64__) || defined(_M_ARM64) || defined(__powerpc64__) + typedef drflac_uint64 drflac_uintptr; +#else + typedef drflac_uint32 drflac_uintptr; +#endif +typedef drflac_uint8 drflac_bool8; +typedef drflac_uint32 drflac_bool32; +#define DRFLAC_TRUE 1 +#define DRFLAC_FALSE 0 +/* End Sized Types */ + +/* Decorations */ +#if !defined(DRFLAC_API) + #if defined(DRFLAC_DLL) + #if defined(_WIN32) + #define DRFLAC_DLL_IMPORT __declspec(dllimport) + #define DRFLAC_DLL_EXPORT __declspec(dllexport) + #define DRFLAC_DLL_PRIVATE static + #else + #if defined(__GNUC__) && __GNUC__ >= 4 + #define DRFLAC_DLL_IMPORT __attribute__((visibility("default"))) + #define DRFLAC_DLL_EXPORT __attribute__((visibility("default"))) + #define DRFLAC_DLL_PRIVATE __attribute__((visibility("hidden"))) + #else + #define DRFLAC_DLL_IMPORT + #define DRFLAC_DLL_EXPORT + #define DRFLAC_DLL_PRIVATE static + #endif + #endif + + #if defined(DR_FLAC_IMPLEMENTATION) || defined(DRFLAC_IMPLEMENTATION) + #define DRFLAC_API DRFLAC_DLL_EXPORT + #else + #define DRFLAC_API DRFLAC_DLL_IMPORT + #endif + #define DRFLAC_PRIVATE DRFLAC_DLL_PRIVATE + #else + #define DRFLAC_API extern + #define DRFLAC_PRIVATE static + #endif +#endif +/* End Decorations */ + +#if defined(_MSC_VER) && _MSC_VER >= 1700 /* Visual Studio 2012 */ + #define DRFLAC_DEPRECATED __declspec(deprecated) +#elif (defined(__GNUC__) && __GNUC__ >= 4) /* GCC 4 */ + #define DRFLAC_DEPRECATED __attribute__((deprecated)) +#elif defined(__has_feature) /* Clang */ + #if __has_feature(attribute_deprecated) + #define DRFLAC_DEPRECATED __attribute__((deprecated)) + #else + #define DRFLAC_DEPRECATED + #endif +#else + #define DRFLAC_DEPRECATED +#endif + +DRFLAC_API void drflac_version(drflac_uint32* pMajor, drflac_uint32* pMinor, drflac_uint32* pRevision); +DRFLAC_API const char* drflac_version_string(void); + +/* Allocation Callbacks */ +typedef struct +{ + void* pUserData; + void* (* onMalloc)(size_t sz, void* pUserData); + void* (* onRealloc)(void* p, size_t sz, void* pUserData); + void (* onFree)(void* p, void* pUserData); +} drflac_allocation_callbacks; +/* End Allocation Callbacks */ + +/* +As data is read from the client it is placed into an internal buffer for fast access. This controls the size of that buffer. Larger values means more speed, +but also more memory. In my testing there is diminishing returns after about 4KB, but you can fiddle with this to suit your own needs. Must be a multiple of 8. +*/ +#ifndef DR_FLAC_BUFFER_SIZE +#define DR_FLAC_BUFFER_SIZE 4096 +#endif + + +/* Architecture Detection */ +#if defined(_WIN64) || defined(_LP64) || defined(__LP64__) +#define DRFLAC_64BIT +#endif + +#if defined(__x86_64__) || (defined(_M_X64) && !defined(_M_ARM64EC)) + #define DRFLAC_X64 +#elif defined(__i386) || defined(_M_IX86) + #define DRFLAC_X86 +#elif defined(__arm__) || defined(_M_ARM) || defined(__arm64) || defined(__arm64__) || defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC) + #define DRFLAC_ARM +#endif +/* End Architecture Detection */ + + +#ifdef DRFLAC_64BIT +typedef drflac_uint64 drflac_cache_t; +#else +typedef drflac_uint32 drflac_cache_t; +#endif + +/* The various metadata block types. */ +#define DRFLAC_METADATA_BLOCK_TYPE_STREAMINFO 0 +#define DRFLAC_METADATA_BLOCK_TYPE_PADDING 1 +#define DRFLAC_METADATA_BLOCK_TYPE_APPLICATION 2 +#define DRFLAC_METADATA_BLOCK_TYPE_SEEKTABLE 3 +#define DRFLAC_METADATA_BLOCK_TYPE_VORBIS_COMMENT 4 +#define DRFLAC_METADATA_BLOCK_TYPE_CUESHEET 5 +#define DRFLAC_METADATA_BLOCK_TYPE_PICTURE 6 +#define DRFLAC_METADATA_BLOCK_TYPE_INVALID 127 + +/* The various picture types specified in the PICTURE block. */ +#define DRFLAC_PICTURE_TYPE_OTHER 0 +#define DRFLAC_PICTURE_TYPE_FILE_ICON 1 +#define DRFLAC_PICTURE_TYPE_OTHER_FILE_ICON 2 +#define DRFLAC_PICTURE_TYPE_COVER_FRONT 3 +#define DRFLAC_PICTURE_TYPE_COVER_BACK 4 +#define DRFLAC_PICTURE_TYPE_LEAFLET_PAGE 5 +#define DRFLAC_PICTURE_TYPE_MEDIA 6 +#define DRFLAC_PICTURE_TYPE_LEAD_ARTIST 7 +#define DRFLAC_PICTURE_TYPE_ARTIST 8 +#define DRFLAC_PICTURE_TYPE_CONDUCTOR 9 +#define DRFLAC_PICTURE_TYPE_BAND 10 +#define DRFLAC_PICTURE_TYPE_COMPOSER 11 +#define DRFLAC_PICTURE_TYPE_LYRICIST 12 +#define DRFLAC_PICTURE_TYPE_RECORDING_LOCATION 13 +#define DRFLAC_PICTURE_TYPE_DURING_RECORDING 14 +#define DRFLAC_PICTURE_TYPE_DURING_PERFORMANCE 15 +#define DRFLAC_PICTURE_TYPE_SCREEN_CAPTURE 16 +#define DRFLAC_PICTURE_TYPE_BRIGHT_COLORED_FISH 17 +#define DRFLAC_PICTURE_TYPE_ILLUSTRATION 18 +#define DRFLAC_PICTURE_TYPE_BAND_LOGOTYPE 19 +#define DRFLAC_PICTURE_TYPE_PUBLISHER_LOGOTYPE 20 + +typedef enum +{ + drflac_container_native, + drflac_container_ogg, + drflac_container_unknown +} drflac_container; + +typedef enum +{ + DRFLAC_SEEK_SET, + DRFLAC_SEEK_CUR, + DRFLAC_SEEK_END +} drflac_seek_origin; + +/* The order of members in this structure is important because we map this directly to the raw data within the SEEKTABLE metadata block. */ +typedef struct +{ + drflac_uint64 firstPCMFrame; + drflac_uint64 flacFrameOffset; /* The offset from the first byte of the header of the first frame. */ + drflac_uint16 pcmFrameCount; +} drflac_seekpoint; + +typedef struct +{ + drflac_uint16 minBlockSizeInPCMFrames; + drflac_uint16 maxBlockSizeInPCMFrames; + drflac_uint32 minFrameSizeInPCMFrames; + drflac_uint32 maxFrameSizeInPCMFrames; + drflac_uint32 sampleRate; + drflac_uint8 channels; + drflac_uint8 bitsPerSample; + drflac_uint64 totalPCMFrameCount; + drflac_uint8 md5[16]; +} drflac_streaminfo; + +typedef struct +{ + /* + The metadata type. Use this to know how to interpret the data below. Will be set to one of the + DRFLAC_METADATA_BLOCK_TYPE_* tokens. + */ + drflac_uint32 type; + + /* The size in bytes of the block and the buffer pointed to by pRawData if it's non-NULL. */ + drflac_uint32 rawDataSize; + + /* The offset in the stream of the raw data. */ + drflac_uint64 rawDataOffset; + + /* + A pointer to the raw data. This points to a temporary buffer so don't hold on to it. It's best to + not modify the contents of this buffer. Use the structures below for more meaningful and structured + information about the metadata. It's possible for this to be null. + */ + const void* pRawData; + + union + { + drflac_streaminfo streaminfo; + + struct + { + int unused; + } padding; + + struct + { + drflac_uint32 id; + const void* pData; + drflac_uint32 dataSize; + } application; + + struct + { + drflac_uint32 seekpointCount; + const drflac_seekpoint* pSeekpoints; + } seektable; + + struct + { + drflac_uint32 vendorLength; + const char* vendor; + drflac_uint32 commentCount; + const void* pComments; + } vorbis_comment; + + struct + { + char catalog[128]; + drflac_uint64 leadInSampleCount; + drflac_bool32 isCD; + drflac_uint8 trackCount; + const void* pTrackData; + } cuesheet; + + struct + { + drflac_uint32 type; + drflac_uint32 mimeLength; + const char* mime; + drflac_uint32 descriptionLength; + const char* description; + drflac_uint32 width; + drflac_uint32 height; + drflac_uint32 colorDepth; + drflac_uint32 indexColorCount; + drflac_uint32 pictureDataSize; + drflac_uint64 pictureDataOffset; /* Offset from the start of the stream. */ + const drflac_uint8* pPictureData; + } picture; + } data; +} drflac_metadata; + + +/* +Callback for when data needs to be read from the client. + + +Parameters +---------- +pUserData (in) + The user data that was passed to drflac_open() and family. + +pBufferOut (out) + The output buffer. + +bytesToRead (in) + The number of bytes to read. + + +Return Value +------------ +The number of bytes actually read. + + +Remarks +------- +A return value of less than bytesToRead indicates the end of the stream. Do _not_ return from this callback until either the entire bytesToRead is filled or +you have reached the end of the stream. +*/ +typedef size_t (* drflac_read_proc)(void* pUserData, void* pBufferOut, size_t bytesToRead); + +/* +Callback for when data needs to be seeked. + + +Parameters +---------- +pUserData (in) + The user data that was passed to drflac_open() and family. + +offset (in) + The number of bytes to move, relative to the origin. Will never be negative. + +origin (in) + The origin of the seek - the current position, the start of the stream, or the end of the stream. + + +Return Value +------------ +Whether or not the seek was successful. + + +Remarks +------- +Seeking relative to the start and the current position must always be supported. If seeking from the end of the stream is not supported, return DRFLAC_FALSE. + +When seeking to a PCM frame using drflac_seek_to_pcm_frame(), dr_flac may call this with an offset beyond the end of the FLAC stream. This needs to be detected +and handled by returning DRFLAC_FALSE. +*/ +typedef drflac_bool32 (* drflac_seek_proc)(void* pUserData, int offset, drflac_seek_origin origin); + +/* +Callback for when the current position in the stream needs to be retrieved. + + +Parameters +---------- +pUserData (in) + The user data that was passed to drflac_open() and family. + +pCursor (out) + A pointer to a variable to receive the current position in the stream. + + +Return Value +------------ +Whether or not the operation was successful. +*/ +typedef drflac_bool32 (* drflac_tell_proc)(void* pUserData, drflac_int64* pCursor); + +/* +Callback for when a metadata block is read. + + +Parameters +---------- +pUserData (in) + The user data that was passed to drflac_open() and family. + +pMetadata (in) + A pointer to a structure containing the data of the metadata block. + + +Remarks +------- +Use pMetadata->type to determine which metadata block is being handled and how to read the data. This +will be set to one of the DRFLAC_METADATA_BLOCK_TYPE_* tokens. +*/ +typedef void (* drflac_meta_proc)(void* pUserData, drflac_metadata* pMetadata); + + +/* Structure for internal use. Only used for decoders opened with drflac_open_memory. */ +typedef struct +{ + const drflac_uint8* data; + size_t dataSize; + size_t currentReadPos; +} drflac__memory_stream; + +/* Structure for internal use. Used for bit streaming. */ +typedef struct +{ + /* The function to call when more data needs to be read. */ + drflac_read_proc onRead; + + /* The function to call when the current read position needs to be moved. */ + drflac_seek_proc onSeek; + + /* The function to call when the current read position needs to be retrieved. */ + drflac_tell_proc onTell; + + /* The user data to pass around to onRead and onSeek. */ + void* pUserData; + + + /* + The number of unaligned bytes in the L2 cache. This will always be 0 until the end of the stream is hit. At the end of the + stream there will be a number of bytes that don't cleanly fit in an L1 cache line, so we use this variable to know whether + or not the bistreamer needs to run on a slower path to read those last bytes. This will never be more than sizeof(drflac_cache_t). + */ + size_t unalignedByteCount; + + /* The content of the unaligned bytes. */ + drflac_cache_t unalignedCache; + + /* The index of the next valid cache line in the "L2" cache. */ + drflac_uint32 nextL2Line; + + /* The number of bits that have been consumed by the cache. This is used to determine how many valid bits are remaining. */ + drflac_uint32 consumedBits; + + /* + The cached data which was most recently read from the client. There are two levels of cache. Data flows as such: + Client -> L2 -> L1. The L2 -> L1 movement is aligned and runs on a fast path in just a few instructions. + */ + drflac_cache_t cacheL2[DR_FLAC_BUFFER_SIZE/sizeof(drflac_cache_t)]; + drflac_cache_t cache; + + /* + CRC-16. This is updated whenever bits are read from the bit stream. Manually set this to 0 to reset the CRC. For FLAC, this + is reset to 0 at the beginning of each frame. + */ + drflac_uint16 crc16; + drflac_cache_t crc16Cache; /* A cache for optimizing CRC calculations. This is filled when when the L1 cache is reloaded. */ + drflac_uint32 crc16CacheIgnoredBytes; /* The number of bytes to ignore when updating the CRC-16 from the CRC-16 cache. */ +} drflac_bs; + +typedef struct +{ + /* The type of the subframe: SUBFRAME_CONSTANT, SUBFRAME_VERBATIM, SUBFRAME_FIXED or SUBFRAME_LPC. */ + drflac_uint8 subframeType; + + /* The number of wasted bits per sample as specified by the sub-frame header. */ + drflac_uint8 wastedBitsPerSample; + + /* The order to use for the prediction stage for SUBFRAME_FIXED and SUBFRAME_LPC. */ + drflac_uint8 lpcOrder; + + /* A pointer to the buffer containing the decoded samples in the subframe. This pointer is an offset from drflac::pExtraData. */ + drflac_int32* pSamplesS32; +} drflac_subframe; + +typedef struct +{ + /* + If the stream uses variable block sizes, this will be set to the index of the first PCM frame. If fixed block sizes are used, this will + always be set to 0. This is 64-bit because the decoded PCM frame number will be 36 bits. + */ + drflac_uint64 pcmFrameNumber; + + /* + If the stream uses fixed block sizes, this will be set to the frame number. If variable block sizes are used, this will always be 0. This + is 32-bit because in fixed block sizes, the maximum frame number will be 31 bits. + */ + drflac_uint32 flacFrameNumber; + + /* The sample rate of this frame. */ + drflac_uint32 sampleRate; + + /* The number of PCM frames in each sub-frame within this frame. */ + drflac_uint16 blockSizeInPCMFrames; + + /* + The channel assignment of this frame. This is not always set to the channel count. If interchannel decorrelation is being used this + will be set to DRFLAC_CHANNEL_ASSIGNMENT_LEFT_SIDE, DRFLAC_CHANNEL_ASSIGNMENT_RIGHT_SIDE or DRFLAC_CHANNEL_ASSIGNMENT_MID_SIDE. + */ + drflac_uint8 channelAssignment; + + /* The number of bits per sample within this frame. */ + drflac_uint8 bitsPerSample; + + /* The frame's CRC. */ + drflac_uint8 crc8; +} drflac_frame_header; + +typedef struct +{ + /* The header. */ + drflac_frame_header header; + + /* + The number of PCM frames left to be read in this FLAC frame. This is initially set to the block size. As PCM frames are read, + this will be decremented. When it reaches 0, the decoder will see this frame as fully consumed and load the next frame. + */ + drflac_uint32 pcmFramesRemaining; + + /* The list of sub-frames within the frame. There is one sub-frame for each channel, and there's a maximum of 8 channels. */ + drflac_subframe subframes[8]; +} drflac_frame; + +typedef struct +{ + /* The function to call when a metadata block is read. */ + drflac_meta_proc onMeta; + + /* The user data posted to the metadata callback function. */ + void* pUserDataMD; + + /* Memory allocation callbacks. */ + drflac_allocation_callbacks allocationCallbacks; + + + /* The sample rate. Will be set to something like 44100. */ + drflac_uint32 sampleRate; + + /* + The number of channels. This will be set to 1 for monaural streams, 2 for stereo, etc. Maximum 8. This is set based on the + value specified in the STREAMINFO block. + */ + drflac_uint8 channels; + + /* The bits per sample. Will be set to something like 16, 24, etc. */ + drflac_uint8 bitsPerSample; + + /* The maximum block size, in samples. This number represents the number of samples in each channel (not combined). */ + drflac_uint16 maxBlockSizeInPCMFrames; + + /* + The total number of PCM Frames making up the stream. Can be 0 in which case it's still a valid stream, but just means + the total PCM frame count is unknown. Likely the case with streams like internet radio. + */ + drflac_uint64 totalPCMFrameCount; + + + /* The container type. This is set based on whether or not the decoder was opened from a native or Ogg stream. */ + drflac_container container; + + /* The number of seekpoints in the seektable. */ + drflac_uint32 seekpointCount; + + + /* Information about the frame the decoder is currently sitting on. */ + drflac_frame currentFLACFrame; + + + /* The index of the PCM frame the decoder is currently sitting on. This is only used for seeking. */ + drflac_uint64 currentPCMFrame; + + /* The position of the first FLAC frame in the stream. This is only ever used for seeking. */ + drflac_uint64 firstFLACFramePosInBytes; + + + /* A hack to avoid a malloc() when opening a decoder with drflac_open_memory(). */ + drflac__memory_stream memoryStream; + + + /* A pointer to the decoded sample data. This is an offset of pExtraData. */ + drflac_int32* pDecodedSamples; + + /* A pointer to the seek table. This is an offset of pExtraData, or NULL if there is no seek table. */ + drflac_seekpoint* pSeekpoints; + + /* Internal use only. Only used with Ogg containers. Points to a drflac_oggbs object. This is an offset of pExtraData. */ + void* _oggbs; + + /* Internal use only. Used for profiling and testing different seeking modes. */ + drflac_bool32 _noSeekTableSeek : 1; + drflac_bool32 _noBinarySearchSeek : 1; + drflac_bool32 _noBruteForceSeek : 1; + + /* The bit streamer. The raw FLAC data is fed through this object. */ + drflac_bs bs; + + /* Variable length extra data. We attach this to the end of the object so we can avoid unnecessary mallocs. */ + drflac_uint8 pExtraData[1]; +} drflac; + + +/* +Opens a FLAC decoder. + + +Parameters +---------- +onRead (in) + The function to call when data needs to be read from the client. + +onSeek (in) + The function to call when the read position of the client data needs to move. + +onTell (in) + The function to call when the read position of the client needs to be queried. + +pUserData (in, optional) + A pointer to application defined data that will be passed to onRead and onSeek. + +pAllocationCallbacks (in, optional) + A pointer to application defined callbacks for managing memory allocations. + + +Return Value +------------ +Returns a pointer to an object representing the decoder. + + +Remarks +------- +Close the decoder with `drflac_close()`. + +`pAllocationCallbacks` can be NULL in which case it will use `DRFLAC_MALLOC`, `DRFLAC_REALLOC` and `DRFLAC_FREE`. + +This function will automatically detect whether or not you are attempting to open a native or Ogg encapsulated FLAC, both of which should work seamlessly +without any manual intervention. Ogg encapsulation also works with multiplexed streams which basically means it can play FLAC encoded audio tracks in videos. + +This is the lowest level function for opening a FLAC stream. You can also use `drflac_open_file()` and `drflac_open_memory()` to open the stream from a file or +from a block of memory respectively. + +The STREAMINFO block must be present for this to succeed. Use `drflac_open_relaxed()` to open a FLAC stream where the header may not be present. + +Use `drflac_open_with_metadata()` if you need access to metadata. + + +Seek Also +--------- +drflac_open_file() +drflac_open_memory() +drflac_open_with_metadata() +drflac_close() +*/ +DRFLAC_API drflac* drflac_open(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* +Opens a FLAC stream with relaxed validation of the header block. + + +Parameters +---------- +onRead (in) + The function to call when data needs to be read from the client. + +onSeek (in) + The function to call when the read position of the client data needs to move. + +onTell (in) + The function to call when the read position of the client needs to be queried. + +container (in) + Whether or not the FLAC stream is encapsulated using standard FLAC encapsulation or Ogg encapsulation. + +pUserData (in, optional) + A pointer to application defined data that will be passed to onRead and onSeek. + +pAllocationCallbacks (in, optional) + A pointer to application defined callbacks for managing memory allocations. + + +Return Value +------------ +A pointer to an object representing the decoder. + + +Remarks +------- +The same as drflac_open(), except attempts to open the stream even when a header block is not present. + +Because the header is not necessarily available, the caller must explicitly define the container (Native or Ogg). Do not set this to `drflac_container_unknown` +as that is for internal use only. + +Opening in relaxed mode will continue reading data from onRead until it finds a valid frame. If a frame is never found it will continue forever. To abort, +force your `onRead` callback to return 0, which dr_flac will use as an indicator that the end of the stream was found. + +Use `drflac_open_with_metadata_relaxed()` if you need access to metadata. +*/ +DRFLAC_API drflac* drflac_open_relaxed(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_container container, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* +Opens a FLAC decoder and notifies the caller of the metadata chunks (album art, etc.). + + +Parameters +---------- +onRead (in) + The function to call when data needs to be read from the client. + +onSeek (in) + The function to call when the read position of the client data needs to move. + +onTell (in) + The function to call when the read position of the client needs to be queried. + +onMeta (in) + The function to call for every metadata block. + +pUserData (in, optional) + A pointer to application defined data that will be passed to onRead, onSeek and onMeta. + +pAllocationCallbacks (in, optional) + A pointer to application defined callbacks for managing memory allocations. + + +Return Value +------------ +A pointer to an object representing the decoder. + + +Remarks +------- +Close the decoder with `drflac_close()`. + +`pAllocationCallbacks` can be NULL in which case it will use `DRFLAC_MALLOC`, `DRFLAC_REALLOC` and `DRFLAC_FREE`. + +This is slower than `drflac_open()`, so avoid this one if you don't need metadata. Internally, this will allocate and free memory on the heap for every +metadata block except for STREAMINFO and PADDING blocks. + +The caller is notified of the metadata via the `onMeta` callback. All metadata blocks will be handled before the function returns. This callback takes a +pointer to a `drflac_metadata` object which is a union containing the data of all relevant metadata blocks. Use the `type` member to discriminate against +the different metadata types. + +The STREAMINFO block must be present for this to succeed. Use `drflac_open_with_metadata_relaxed()` to open a FLAC stream where the header may not be present. + +Note that this will behave inconsistently with `drflac_open()` if the stream is an Ogg encapsulated stream and a metadata block is corrupted. This is due to +the way the Ogg stream recovers from corrupted pages. When `drflac_open_with_metadata()` is being used, the open routine will try to read the contents of the +metadata block, whereas `drflac_open()` will simply seek past it (for the sake of efficiency). This inconsistency can result in different samples being +returned depending on whether or not the stream is being opened with metadata. + + +Seek Also +--------- +drflac_open_file_with_metadata() +drflac_open_memory_with_metadata() +drflac_open() +drflac_close() +*/ +DRFLAC_API drflac* drflac_open_with_metadata(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_meta_proc onMeta, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* +The same as drflac_open_with_metadata(), except attempts to open the stream even when a header block is not present. + +See Also +-------- +drflac_open_with_metadata() +drflac_open_relaxed() +*/ +DRFLAC_API drflac* drflac_open_with_metadata_relaxed(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_meta_proc onMeta, drflac_container container, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* +Closes the given FLAC decoder. + + +Parameters +---------- +pFlac (in) + The decoder to close. + + +Remarks +------- +This will destroy the decoder object. + + +See Also +-------- +drflac_open() +drflac_open_with_metadata() +drflac_open_file() +drflac_open_file_w() +drflac_open_file_with_metadata() +drflac_open_file_with_metadata_w() +drflac_open_memory() +drflac_open_memory_with_metadata() +*/ +DRFLAC_API void drflac_close(drflac* pFlac); + + +/* +Reads sample data from the given FLAC decoder, output as interleaved signed 32-bit PCM. + + +Parameters +---------- +pFlac (in) + The decoder. + +framesToRead (in) + The number of PCM frames to read. + +pBufferOut (out, optional) + A pointer to the buffer that will receive the decoded samples. + + +Return Value +------------ +Returns the number of PCM frames actually read. If the return value is less than `framesToRead` it has reached the end. + + +Remarks +------- +pBufferOut can be null, in which case the call will act as a seek, and the return value will be the number of frames seeked. +*/ +DRFLAC_API drflac_uint64 drflac_read_pcm_frames_s32(drflac* pFlac, drflac_uint64 framesToRead, drflac_int32* pBufferOut); + + +/* +Reads sample data from the given FLAC decoder, output as interleaved signed 16-bit PCM. + + +Parameters +---------- +pFlac (in) + The decoder. + +framesToRead (in) + The number of PCM frames to read. + +pBufferOut (out, optional) + A pointer to the buffer that will receive the decoded samples. + + +Return Value +------------ +Returns the number of PCM frames actually read. If the return value is less than `framesToRead` it has reached the end. + + +Remarks +------- +pBufferOut can be null, in which case the call will act as a seek, and the return value will be the number of frames seeked. + +Note that this is lossy for streams where the bits per sample is larger than 16. +*/ +DRFLAC_API drflac_uint64 drflac_read_pcm_frames_s16(drflac* pFlac, drflac_uint64 framesToRead, drflac_int16* pBufferOut); + +/* +Reads sample data from the given FLAC decoder, output as interleaved 32-bit floating point PCM. + + +Parameters +---------- +pFlac (in) + The decoder. + +framesToRead (in) + The number of PCM frames to read. + +pBufferOut (out, optional) + A pointer to the buffer that will receive the decoded samples. + + +Return Value +------------ +Returns the number of PCM frames actually read. If the return value is less than `framesToRead` it has reached the end. + + +Remarks +------- +pBufferOut can be null, in which case the call will act as a seek, and the return value will be the number of frames seeked. + +Note that this should be considered lossy due to the nature of floating point numbers not being able to exactly represent every possible number. +*/ +DRFLAC_API drflac_uint64 drflac_read_pcm_frames_f32(drflac* pFlac, drflac_uint64 framesToRead, float* pBufferOut); + +/* +Seeks to the PCM frame at the given index. + + +Parameters +---------- +pFlac (in) + The decoder. + +pcmFrameIndex (in) + The index of the PCM frame to seek to. See notes below. + + +Return Value +------------- +`DRFLAC_TRUE` if successful; `DRFLAC_FALSE` otherwise. +*/ +DRFLAC_API drflac_bool32 drflac_seek_to_pcm_frame(drflac* pFlac, drflac_uint64 pcmFrameIndex); + + + +#ifndef DR_FLAC_NO_STDIO +/* +Opens a FLAC decoder from the file at the given path. + + +Parameters +---------- +pFileName (in) + The path of the file to open, either absolute or relative to the current directory. + +pAllocationCallbacks (in, optional) + A pointer to application defined callbacks for managing memory allocations. + + +Return Value +------------ +A pointer to an object representing the decoder. + + +Remarks +------- +Close the decoder with drflac_close(). + + +Remarks +------- +This will hold a handle to the file until the decoder is closed with drflac_close(). Some platforms will restrict the number of files a process can have open +at any given time, so keep this mind if you have many decoders open at the same time. + + +See Also +-------- +drflac_open_file_with_metadata() +drflac_open() +drflac_close() +*/ +DRFLAC_API drflac* drflac_open_file(const char* pFileName, const drflac_allocation_callbacks* pAllocationCallbacks); +DRFLAC_API drflac* drflac_open_file_w(const wchar_t* pFileName, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* +Opens a FLAC decoder from the file at the given path and notifies the caller of the metadata chunks (album art, etc.) + + +Parameters +---------- +pFileName (in) + The path of the file to open, either absolute or relative to the current directory. + +pAllocationCallbacks (in, optional) + A pointer to application defined callbacks for managing memory allocations. + +onMeta (in) + The callback to fire for each metadata block. + +pUserData (in) + A pointer to the user data to pass to the metadata callback. + +pAllocationCallbacks (in) + A pointer to application defined callbacks for managing memory allocations. + + +Remarks +------- +Look at the documentation for drflac_open_with_metadata() for more information on how metadata is handled. + + +See Also +-------- +drflac_open_with_metadata() +drflac_open() +drflac_close() +*/ +DRFLAC_API drflac* drflac_open_file_with_metadata(const char* pFileName, drflac_meta_proc onMeta, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks); +DRFLAC_API drflac* drflac_open_file_with_metadata_w(const wchar_t* pFileName, drflac_meta_proc onMeta, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks); +#endif + +/* +Opens a FLAC decoder from a pre-allocated block of memory + + +Parameters +---------- +pData (in) + A pointer to the raw encoded FLAC data. + +dataSize (in) + The size in bytes of `data`. + +pAllocationCallbacks (in) + A pointer to application defined callbacks for managing memory allocations. + + +Return Value +------------ +A pointer to an object representing the decoder. + + +Remarks +------- +This does not create a copy of the data. It is up to the application to ensure the buffer remains valid for the lifetime of the decoder. + + +See Also +-------- +drflac_open() +drflac_close() +*/ +DRFLAC_API drflac* drflac_open_memory(const void* pData, size_t dataSize, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* +Opens a FLAC decoder from a pre-allocated block of memory and notifies the caller of the metadata chunks (album art, etc.) + + +Parameters +---------- +pData (in) + A pointer to the raw encoded FLAC data. + +dataSize (in) + The size in bytes of `data`. + +onMeta (in) + The callback to fire for each metadata block. + +pUserData (in) + A pointer to the user data to pass to the metadata callback. + +pAllocationCallbacks (in) + A pointer to application defined callbacks for managing memory allocations. + + +Remarks +------- +Look at the documentation for drflac_open_with_metadata() for more information on how metadata is handled. + + +See Also +------- +drflac_open_with_metadata() +drflac_open() +drflac_close() +*/ +DRFLAC_API drflac* drflac_open_memory_with_metadata(const void* pData, size_t dataSize, drflac_meta_proc onMeta, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks); + + + +/* High Level APIs */ + +/* +Opens a FLAC stream from the given callbacks and fully decodes it in a single operation. The return value is a +pointer to the sample data as interleaved signed 32-bit PCM. The returned data must be freed with drflac_free(). + +You can pass in custom memory allocation callbacks via the pAllocationCallbacks parameter. This can be NULL in which +case it will use DRFLAC_MALLOC, DRFLAC_REALLOC and DRFLAC_FREE. + +Sometimes a FLAC file won't keep track of the total sample count. In this situation the function will continuously +read samples into a dynamically sized buffer on the heap until no samples are left. + +Do not call this function on a broadcast type of stream (like internet radio streams and whatnot). +*/ +DRFLAC_API drflac_int32* drflac_open_and_read_pcm_frames_s32(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, void* pUserData, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* Same as drflac_open_and_read_pcm_frames_s32(), except returns signed 16-bit integer samples. */ +DRFLAC_API drflac_int16* drflac_open_and_read_pcm_frames_s16(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, void* pUserData, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* Same as drflac_open_and_read_pcm_frames_s32(), except returns 32-bit floating-point samples. */ +DRFLAC_API float* drflac_open_and_read_pcm_frames_f32(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, void* pUserData, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); + +#ifndef DR_FLAC_NO_STDIO +/* Same as drflac_open_and_read_pcm_frames_s32() except opens the decoder from a file. */ +DRFLAC_API drflac_int32* drflac_open_file_and_read_pcm_frames_s32(const char* filename, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* Same as drflac_open_file_and_read_pcm_frames_s32(), except returns signed 16-bit integer samples. */ +DRFLAC_API drflac_int16* drflac_open_file_and_read_pcm_frames_s16(const char* filename, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* Same as drflac_open_file_and_read_pcm_frames_s32(), except returns 32-bit floating-point samples. */ +DRFLAC_API float* drflac_open_file_and_read_pcm_frames_f32(const char* filename, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); +#endif + +/* Same as drflac_open_and_read_pcm_frames_s32() except opens the decoder from a block of memory. */ +DRFLAC_API drflac_int32* drflac_open_memory_and_read_pcm_frames_s32(const void* data, size_t dataSize, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* Same as drflac_open_memory_and_read_pcm_frames_s32(), except returns signed 16-bit integer samples. */ +DRFLAC_API drflac_int16* drflac_open_memory_and_read_pcm_frames_s16(const void* data, size_t dataSize, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* Same as drflac_open_memory_and_read_pcm_frames_s32(), except returns 32-bit floating-point samples. */ +DRFLAC_API float* drflac_open_memory_and_read_pcm_frames_f32(const void* data, size_t dataSize, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks); + +/* +Frees memory that was allocated internally by dr_flac. + +Set pAllocationCallbacks to the same object that was passed to drflac_open_*_and_read_pcm_frames_*(). If you originally passed in NULL, pass in NULL for this. +*/ +DRFLAC_API void drflac_free(void* p, const drflac_allocation_callbacks* pAllocationCallbacks); + + +/* Structure representing an iterator for vorbis comments in a VORBIS_COMMENT metadata block. */ +typedef struct +{ + drflac_uint32 countRemaining; + const char* pRunningData; +} drflac_vorbis_comment_iterator; + +/* +Initializes a vorbis comment iterator. This can be used for iterating over the vorbis comments in a VORBIS_COMMENT +metadata block. +*/ +DRFLAC_API void drflac_init_vorbis_comment_iterator(drflac_vorbis_comment_iterator* pIter, drflac_uint32 commentCount, const void* pComments); + +/* +Goes to the next vorbis comment in the given iterator. If null is returned it means there are no more comments. The +returned string is NOT null terminated. +*/ +DRFLAC_API const char* drflac_next_vorbis_comment(drflac_vorbis_comment_iterator* pIter, drflac_uint32* pCommentLengthOut); + + +/* Structure representing an iterator for cuesheet tracks in a CUESHEET metadata block. */ +typedef struct +{ + drflac_uint32 countRemaining; + const char* pRunningData; +} drflac_cuesheet_track_iterator; + +/* The order of members here is important because we map this directly to the raw data within the CUESHEET metadata block. */ +typedef struct +{ + drflac_uint64 offset; + drflac_uint8 index; + drflac_uint8 reserved[3]; +} drflac_cuesheet_track_index; + +typedef struct +{ + drflac_uint64 offset; + drflac_uint8 trackNumber; + char ISRC[12]; + drflac_bool8 isAudio; + drflac_bool8 preEmphasis; + drflac_uint8 indexCount; + const drflac_cuesheet_track_index* pIndexPoints; +} drflac_cuesheet_track; + +/* +Initializes a cuesheet track iterator. This can be used for iterating over the cuesheet tracks in a CUESHEET metadata +block. +*/ +DRFLAC_API void drflac_init_cuesheet_track_iterator(drflac_cuesheet_track_iterator* pIter, drflac_uint32 trackCount, const void* pTrackData); + +/* Goes to the next cuesheet track in the given iterator. If DRFLAC_FALSE is returned it means there are no more comments. */ +DRFLAC_API drflac_bool32 drflac_next_cuesheet_track(drflac_cuesheet_track_iterator* pIter, drflac_cuesheet_track* pCuesheetTrack); + + +#ifdef __cplusplus +} +#endif +#endif /* dr_flac_h */ + + +/************************************************************************************************************************************************************ + ************************************************************************************************************************************************************ + + IMPLEMENTATION + + ************************************************************************************************************************************************************ + ************************************************************************************************************************************************************/ +#if defined(DR_FLAC_IMPLEMENTATION) || defined(DRFLAC_IMPLEMENTATION) +#ifndef dr_flac_c +#define dr_flac_c + +/* Disable some annoying warnings. */ +#if defined(__clang__) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))) + #pragma GCC diagnostic push + #if __GNUC__ >= 7 + #pragma GCC diagnostic ignored "-Wimplicit-fallthrough" + #endif +#endif + +#ifdef __linux__ + #ifndef _BSD_SOURCE + #define _BSD_SOURCE + #endif + #ifndef _DEFAULT_SOURCE + #define _DEFAULT_SOURCE + #endif + #ifndef __USE_BSD + #define __USE_BSD + #endif + #include <endian.h> +#endif + +#include <stdlib.h> +#include <string.h> + +/* Inline */ +#ifdef _MSC_VER + #define DRFLAC_INLINE __forceinline +#elif defined(__GNUC__) + /* + I've had a bug report where GCC is emitting warnings about functions possibly not being inlineable. This warning happens when + the __attribute__((always_inline)) attribute is defined without an "inline" statement. I think therefore there must be some + case where "__inline__" is not always defined, thus the compiler emitting these warnings. When using -std=c89 or -ansi on the + command line, we cannot use the "inline" keyword and instead need to use "__inline__". In an attempt to work around this issue + I am using "__inline__" only when we're compiling in strict ANSI mode. + */ + #if defined(__STRICT_ANSI__) + #define DRFLAC_GNUC_INLINE_HINT __inline__ + #else + #define DRFLAC_GNUC_INLINE_HINT inline + #endif + + #if (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 2)) || defined(__clang__) + #define DRFLAC_INLINE DRFLAC_GNUC_INLINE_HINT __attribute__((always_inline)) + #else + #define DRFLAC_INLINE DRFLAC_GNUC_INLINE_HINT + #endif +#elif defined(__WATCOMC__) + #define DRFLAC_INLINE __inline +#else + #define DRFLAC_INLINE +#endif +/* End Inline */ + +/* +Intrinsics Support + +There's a bug in GCC 4.2.x which results in an incorrect compilation error when using _mm_slli_epi32() where it complains with + + "error: shift must be an immediate" + +Unfortuantely dr_flac depends on this for a few things so we're just going to disable SSE on GCC 4.2 and below. +*/ +#if !defined(DR_FLAC_NO_SIMD) + #if defined(DRFLAC_X64) || defined(DRFLAC_X86) + #if defined(_MSC_VER) && !defined(__clang__) + /* MSVC. */ + #if _MSC_VER >= 1400 && !defined(DRFLAC_NO_SSE2) /* 2005 */ + #define DRFLAC_SUPPORT_SSE2 + #endif + #if _MSC_VER >= 1600 && !defined(DRFLAC_NO_SSE41) /* 2010 */ + #define DRFLAC_SUPPORT_SSE41 + #endif + #elif defined(__clang__) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))) + /* Assume GNUC-style. */ + #if defined(__SSE2__) && !defined(DRFLAC_NO_SSE2) + #define DRFLAC_SUPPORT_SSE2 + #endif + #if defined(__SSE4_1__) && !defined(DRFLAC_NO_SSE41) + #define DRFLAC_SUPPORT_SSE41 + #endif + #endif + + /* If at this point we still haven't determined compiler support for the intrinsics just fall back to __has_include. */ + #if !defined(__GNUC__) && !defined(__clang__) && defined(__has_include) + #if !defined(DRFLAC_SUPPORT_SSE2) && !defined(DRFLAC_NO_SSE2) && __has_include(<emmintrin.h>) + #define DRFLAC_SUPPORT_SSE2 + #endif + #if !defined(DRFLAC_SUPPORT_SSE41) && !defined(DRFLAC_NO_SSE41) && __has_include(<smmintrin.h>) + #define DRFLAC_SUPPORT_SSE41 + #endif + #endif + + #if defined(DRFLAC_SUPPORT_SSE41) + #include <smmintrin.h> + #elif defined(DRFLAC_SUPPORT_SSE2) + #include <emmintrin.h> + #endif + #endif + + #if defined(DRFLAC_ARM) + #if !defined(DRFLAC_NO_NEON) && (defined(__ARM_NEON) || defined(__aarch64__) || defined(_M_ARM64)) + #define DRFLAC_SUPPORT_NEON + #include <arm_neon.h> + #endif + #endif +#endif + +/* Compile-time CPU feature support. */ +#if !defined(DR_FLAC_NO_SIMD) && (defined(DRFLAC_X86) || defined(DRFLAC_X64)) + #if defined(_MSC_VER) && !defined(__clang__) + #if _MSC_VER >= 1400 + #include <intrin.h> + static void drflac__cpuid(int info[4], int fid) + { + __cpuid(info, fid); + } + #else + #define DRFLAC_NO_CPUID + #endif + #else + #if defined(__GNUC__) || defined(__clang__) + static void drflac__cpuid(int info[4], int fid) + { + /* + It looks like the -fPIC option uses the ebx register which GCC complains about. We can work around this by just using a different register, the + specific register of which I'm letting the compiler decide on. The "k" prefix is used to specify a 32-bit register. The {...} syntax is for + supporting different assembly dialects. + + What's basically happening is that we're saving and restoring the ebx register manually. + */ + #if defined(DRFLAC_X86) && defined(__PIC__) + __asm__ __volatile__ ( + "xchg{l} {%%}ebx, %k1;" + "cpuid;" + "xchg{l} {%%}ebx, %k1;" + : "=a"(info[0]), "=&r"(info[1]), "=c"(info[2]), "=d"(info[3]) : "a"(fid), "c"(0) + ); + #else + __asm__ __volatile__ ( + "cpuid" : "=a"(info[0]), "=b"(info[1]), "=c"(info[2]), "=d"(info[3]) : "a"(fid), "c"(0) + ); + #endif + } + #else + #define DRFLAC_NO_CPUID + #endif + #endif +#else + #define DRFLAC_NO_CPUID +#endif + +static DRFLAC_INLINE drflac_bool32 drflac_has_sse2(void) +{ +#if defined(DRFLAC_SUPPORT_SSE2) + #if (defined(DRFLAC_X64) || defined(DRFLAC_X86)) && !defined(DRFLAC_NO_SSE2) + #if defined(DRFLAC_X64) + return DRFLAC_TRUE; /* 64-bit targets always support SSE2. */ + #elif (defined(_M_IX86_FP) && _M_IX86_FP == 2) || defined(__SSE2__) + return DRFLAC_TRUE; /* If the compiler is allowed to freely generate SSE2 code we can assume support. */ + #else + #if defined(DRFLAC_NO_CPUID) + return DRFLAC_FALSE; + #else + int info[4]; + drflac__cpuid(info, 1); + return (info[3] & (1 << 26)) != 0; + #endif + #endif + #else + return DRFLAC_FALSE; /* SSE2 is only supported on x86 and x64 architectures. */ + #endif +#else + return DRFLAC_FALSE; /* No compiler support. */ +#endif +} + +static DRFLAC_INLINE drflac_bool32 drflac_has_sse41(void) +{ +#if defined(DRFLAC_SUPPORT_SSE41) + #if (defined(DRFLAC_X64) || defined(DRFLAC_X86)) && !defined(DRFLAC_NO_SSE41) + #if defined(__SSE4_1__) || defined(__AVX__) + return DRFLAC_TRUE; /* If the compiler is allowed to freely generate SSE41 code we can assume support. */ + #else + #if defined(DRFLAC_NO_CPUID) + return DRFLAC_FALSE; + #else + int info[4]; + drflac__cpuid(info, 1); + return (info[2] & (1 << 19)) != 0; + #endif + #endif + #else + return DRFLAC_FALSE; /* SSE41 is only supported on x86 and x64 architectures. */ + #endif +#else + return DRFLAC_FALSE; /* No compiler support. */ +#endif +} + + +#if defined(_MSC_VER) && _MSC_VER >= 1500 && (defined(DRFLAC_X86) || defined(DRFLAC_X64)) && !defined(__clang__) + #define DRFLAC_HAS_LZCNT_INTRINSIC +#elif (defined(__GNUC__) && ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 7))) + #define DRFLAC_HAS_LZCNT_INTRINSIC +#elif defined(__clang__) + #if defined(__has_builtin) + #if __has_builtin(__builtin_clzll) || __has_builtin(__builtin_clzl) + #define DRFLAC_HAS_LZCNT_INTRINSIC + #endif + #endif +#endif + +#if defined(_MSC_VER) && _MSC_VER >= 1400 && !defined(__clang__) + #define DRFLAC_HAS_BYTESWAP16_INTRINSIC + #define DRFLAC_HAS_BYTESWAP32_INTRINSIC + #define DRFLAC_HAS_BYTESWAP64_INTRINSIC +#elif defined(__clang__) + #if defined(__has_builtin) + #if __has_builtin(__builtin_bswap16) + #define DRFLAC_HAS_BYTESWAP16_INTRINSIC + #endif + #if __has_builtin(__builtin_bswap32) + #define DRFLAC_HAS_BYTESWAP32_INTRINSIC + #endif + #if __has_builtin(__builtin_bswap64) + #define DRFLAC_HAS_BYTESWAP64_INTRINSIC + #endif + #endif +#elif defined(__GNUC__) + #if ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)) + #define DRFLAC_HAS_BYTESWAP32_INTRINSIC + #define DRFLAC_HAS_BYTESWAP64_INTRINSIC + #endif + #if ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) + #define DRFLAC_HAS_BYTESWAP16_INTRINSIC + #endif +#elif defined(__WATCOMC__) && defined(__386__) + #define DRFLAC_HAS_BYTESWAP16_INTRINSIC + #define DRFLAC_HAS_BYTESWAP32_INTRINSIC + #define DRFLAC_HAS_BYTESWAP64_INTRINSIC + extern __inline drflac_uint16 _watcom_bswap16(drflac_uint16); + extern __inline drflac_uint32 _watcom_bswap32(drflac_uint32); + extern __inline drflac_uint64 _watcom_bswap64(drflac_uint64); +#pragma aux _watcom_bswap16 = \ + "xchg al, ah" \ + parm [ax] \ + value [ax] \ + modify nomemory; +#pragma aux _watcom_bswap32 = \ + "bswap eax" \ + parm [eax] \ + value [eax] \ + modify nomemory; +#pragma aux _watcom_bswap64 = \ + "bswap eax" \ + "bswap edx" \ + "xchg eax,edx" \ + parm [eax edx] \ + value [eax edx] \ + modify nomemory; +#endif + + +/* Standard library stuff. */ +#ifndef DRFLAC_ASSERT +#include <assert.h> +#define DRFLAC_ASSERT(expression) assert(expression) +#endif +#ifndef DRFLAC_MALLOC +#define DRFLAC_MALLOC(sz) malloc((sz)) +#endif +#ifndef DRFLAC_REALLOC +#define DRFLAC_REALLOC(p, sz) realloc((p), (sz)) +#endif +#ifndef DRFLAC_FREE +#define DRFLAC_FREE(p) free((p)) +#endif +#ifndef DRFLAC_COPY_MEMORY +#define DRFLAC_COPY_MEMORY(dst, src, sz) memcpy((dst), (src), (sz)) +#endif +#ifndef DRFLAC_ZERO_MEMORY +#define DRFLAC_ZERO_MEMORY(p, sz) memset((p), 0, (sz)) +#endif +#ifndef DRFLAC_ZERO_OBJECT +#define DRFLAC_ZERO_OBJECT(p) DRFLAC_ZERO_MEMORY((p), sizeof(*(p))) +#endif + +#define DRFLAC_MIN(a, b) (((a) < (b)) ? (a) : (b)) + +#define DRFLAC_MAX_SIMD_VECTOR_SIZE 64 /* 64 for AVX-512 in the future. */ + +/* Result Codes */ +typedef drflac_int32 drflac_result; +#define DRFLAC_SUCCESS 0 +#define DRFLAC_ERROR -1 /* A generic error. */ +#define DRFLAC_INVALID_ARGS -2 +#define DRFLAC_INVALID_OPERATION -3 +#define DRFLAC_OUT_OF_MEMORY -4 +#define DRFLAC_OUT_OF_RANGE -5 +#define DRFLAC_ACCESS_DENIED -6 +#define DRFLAC_DOES_NOT_EXIST -7 +#define DRFLAC_ALREADY_EXISTS -8 +#define DRFLAC_TOO_MANY_OPEN_FILES -9 +#define DRFLAC_INVALID_FILE -10 +#define DRFLAC_TOO_BIG -11 +#define DRFLAC_PATH_TOO_LONG -12 +#define DRFLAC_NAME_TOO_LONG -13 +#define DRFLAC_NOT_DIRECTORY -14 +#define DRFLAC_IS_DIRECTORY -15 +#define DRFLAC_DIRECTORY_NOT_EMPTY -16 +#define DRFLAC_END_OF_FILE -17 +#define DRFLAC_NO_SPACE -18 +#define DRFLAC_BUSY -19 +#define DRFLAC_IO_ERROR -20 +#define DRFLAC_INTERRUPT -21 +#define DRFLAC_UNAVAILABLE -22 +#define DRFLAC_ALREADY_IN_USE -23 +#define DRFLAC_BAD_ADDRESS -24 +#define DRFLAC_BAD_SEEK -25 +#define DRFLAC_BAD_PIPE -26 +#define DRFLAC_DEADLOCK -27 +#define DRFLAC_TOO_MANY_LINKS -28 +#define DRFLAC_NOT_IMPLEMENTED -29 +#define DRFLAC_NO_MESSAGE -30 +#define DRFLAC_BAD_MESSAGE -31 +#define DRFLAC_NO_DATA_AVAILABLE -32 +#define DRFLAC_INVALID_DATA -33 +#define DRFLAC_TIMEOUT -34 +#define DRFLAC_NO_NETWORK -35 +#define DRFLAC_NOT_UNIQUE -36 +#define DRFLAC_NOT_SOCKET -37 +#define DRFLAC_NO_ADDRESS -38 +#define DRFLAC_BAD_PROTOCOL -39 +#define DRFLAC_PROTOCOL_UNAVAILABLE -40 +#define DRFLAC_PROTOCOL_NOT_SUPPORTED -41 +#define DRFLAC_PROTOCOL_FAMILY_NOT_SUPPORTED -42 +#define DRFLAC_ADDRESS_FAMILY_NOT_SUPPORTED -43 +#define DRFLAC_SOCKET_NOT_SUPPORTED -44 +#define DRFLAC_CONNECTION_RESET -45 +#define DRFLAC_ALREADY_CONNECTED -46 +#define DRFLAC_NOT_CONNECTED -47 +#define DRFLAC_CONNECTION_REFUSED -48 +#define DRFLAC_NO_HOST -49 +#define DRFLAC_IN_PROGRESS -50 +#define DRFLAC_CANCELLED -51 +#define DRFLAC_MEMORY_ALREADY_MAPPED -52 +#define DRFLAC_AT_END -53 + +#define DRFLAC_CRC_MISMATCH -100 +/* End Result Codes */ + + +#define DRFLAC_SUBFRAME_CONSTANT 0 +#define DRFLAC_SUBFRAME_VERBATIM 1 +#define DRFLAC_SUBFRAME_FIXED 8 +#define DRFLAC_SUBFRAME_LPC 32 +#define DRFLAC_SUBFRAME_RESERVED 255 + +#define DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE 0 +#define DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE2 1 + +#define DRFLAC_CHANNEL_ASSIGNMENT_INDEPENDENT 0 +#define DRFLAC_CHANNEL_ASSIGNMENT_LEFT_SIDE 8 +#define DRFLAC_CHANNEL_ASSIGNMENT_RIGHT_SIDE 9 +#define DRFLAC_CHANNEL_ASSIGNMENT_MID_SIDE 10 + +#define drflac_align(x, a) ((((x) + (a) - 1) / (a)) * (a)) + +#define DRFLAC_SEEKPOINT_SIZE_IN_BYTES 18 +#define DRFLAC_CUESHEET_TRACK_SIZE_IN_BYTES 36 +#define DRFLAC_CUESHEET_TRACK_INDEX_SIZE_IN_BYTES 12 +#define DRFLAC_CUESHEET_TRACK_PADDED_SIZE_IN_BYTES drflac_align(DRFLAC_CUESHEET_TRACK_SIZE_IN_BYTES, sizeof(drflac_uint64)) + + +DRFLAC_API void drflac_version(drflac_uint32* pMajor, drflac_uint32* pMinor, drflac_uint32* pRevision) +{ + if (pMajor) { + *pMajor = DRFLAC_VERSION_MAJOR; + } + + if (pMinor) { + *pMinor = DRFLAC_VERSION_MINOR; + } + + if (pRevision) { + *pRevision = DRFLAC_VERSION_REVISION; + } +} + +DRFLAC_API const char* drflac_version_string(void) +{ + return DRFLAC_VERSION_STRING; +} + + +/* CPU caps. */ +#if defined(__has_feature) + #if __has_feature(thread_sanitizer) + #define DRFLAC_NO_THREAD_SANITIZE __attribute__((no_sanitize("thread"))) + #else + #define DRFLAC_NO_THREAD_SANITIZE + #endif +#else + #define DRFLAC_NO_THREAD_SANITIZE +#endif + +#if defined(DRFLAC_HAS_LZCNT_INTRINSIC) +static drflac_bool32 drflac__gIsLZCNTSupported = DRFLAC_FALSE; +#endif + +#ifndef DRFLAC_NO_CPUID +static drflac_bool32 drflac__gIsSSE2Supported = DRFLAC_FALSE; +static drflac_bool32 drflac__gIsSSE41Supported = DRFLAC_FALSE; + +/* +I've had a bug report that Clang's ThreadSanitizer presents a warning in this function. Having reviewed this, this does +actually make sense. However, since CPU caps should never differ for a running process, I don't think the trade off of +complicating internal API's by passing around CPU caps versus just disabling the warnings is worthwhile. I'm therefore +just going to disable these warnings. This is disabled via the DRFLAC_NO_THREAD_SANITIZE attribute. +*/ +DRFLAC_NO_THREAD_SANITIZE static void drflac__init_cpu_caps(void) +{ + static drflac_bool32 isCPUCapsInitialized = DRFLAC_FALSE; + + if (!isCPUCapsInitialized) { + /* LZCNT */ +#if defined(DRFLAC_HAS_LZCNT_INTRINSIC) + int info[4] = {0}; + drflac__cpuid(info, 0x80000001); + drflac__gIsLZCNTSupported = (info[2] & (1 << 5)) != 0; +#endif + + /* SSE2 */ + drflac__gIsSSE2Supported = drflac_has_sse2(); + + /* SSE4.1 */ + drflac__gIsSSE41Supported = drflac_has_sse41(); + + /* Initialized. */ + isCPUCapsInitialized = DRFLAC_TRUE; + } +} +#else +static drflac_bool32 drflac__gIsNEONSupported = DRFLAC_FALSE; + +static DRFLAC_INLINE drflac_bool32 drflac__has_neon(void) +{ +#if defined(DRFLAC_SUPPORT_NEON) + #if defined(DRFLAC_ARM) && !defined(DRFLAC_NO_NEON) + #if (defined(__ARM_NEON) || defined(__aarch64__) || defined(_M_ARM64)) + return DRFLAC_TRUE; /* If the compiler is allowed to freely generate NEON code we can assume support. */ + #else + /* TODO: Runtime check. */ + return DRFLAC_FALSE; + #endif + #else + return DRFLAC_FALSE; /* NEON is only supported on ARM architectures. */ + #endif +#else + return DRFLAC_FALSE; /* No compiler support. */ +#endif +} + +DRFLAC_NO_THREAD_SANITIZE static void drflac__init_cpu_caps(void) +{ + drflac__gIsNEONSupported = drflac__has_neon(); + +#if defined(DRFLAC_HAS_LZCNT_INTRINSIC) && defined(DRFLAC_ARM) && (defined(__ARM_ARCH) && __ARM_ARCH >= 5) + drflac__gIsLZCNTSupported = DRFLAC_TRUE; +#endif +} +#endif + + +/* Endian Management */ +static DRFLAC_INLINE drflac_bool32 drflac__is_little_endian(void) +{ +#if defined(DRFLAC_X86) || defined(DRFLAC_X64) + return DRFLAC_TRUE; +#elif defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && __BYTE_ORDER == __LITTLE_ENDIAN + return DRFLAC_TRUE; +#else + int n = 1; + return (*(char*)&n) == 1; +#endif +} + +static DRFLAC_INLINE drflac_uint16 drflac__swap_endian_uint16(drflac_uint16 n) +{ +#ifdef DRFLAC_HAS_BYTESWAP16_INTRINSIC + #if defined(_MSC_VER) && !defined(__clang__) + return _byteswap_ushort(n); + #elif defined(__GNUC__) || defined(__clang__) + return __builtin_bswap16(n); + #elif defined(__WATCOMC__) && defined(__386__) + return _watcom_bswap16(n); + #else + #error "This compiler does not support the byte swap intrinsic." + #endif +#else + return ((n & 0xFF00) >> 8) | + ((n & 0x00FF) << 8); +#endif +} + +static DRFLAC_INLINE drflac_uint32 drflac__swap_endian_uint32(drflac_uint32 n) +{ +#ifdef DRFLAC_HAS_BYTESWAP32_INTRINSIC + #if defined(_MSC_VER) && !defined(__clang__) + return _byteswap_ulong(n); + #elif defined(__GNUC__) || defined(__clang__) + #if defined(DRFLAC_ARM) && (defined(__ARM_ARCH) && __ARM_ARCH >= 6) && !defined(__ARM_ARCH_6M__) && !defined(DRFLAC_64BIT) /* <-- 64-bit inline assembly has not been tested, so disabling for now. */ + /* Inline assembly optimized implementation for ARM. In my testing, GCC does not generate optimized code with __builtin_bswap32(). */ + drflac_uint32 r; + __asm__ __volatile__ ( + #if defined(DRFLAC_64BIT) + "rev %w[out], %w[in]" : [out]"=r"(r) : [in]"r"(n) /* <-- This is untested. If someone in the community could test this, that would be appreciated! */ + #else + "rev %[out], %[in]" : [out]"=r"(r) : [in]"r"(n) + #endif + ); + return r; + #else + return __builtin_bswap32(n); + #endif + #elif defined(__WATCOMC__) && defined(__386__) + return _watcom_bswap32(n); + #else + #error "This compiler does not support the byte swap intrinsic." + #endif +#else + return ((n & 0xFF000000) >> 24) | + ((n & 0x00FF0000) >> 8) | + ((n & 0x0000FF00) << 8) | + ((n & 0x000000FF) << 24); +#endif +} + +static DRFLAC_INLINE drflac_uint64 drflac__swap_endian_uint64(drflac_uint64 n) +{ +#ifdef DRFLAC_HAS_BYTESWAP64_INTRINSIC + #if defined(_MSC_VER) && !defined(__clang__) + return _byteswap_uint64(n); + #elif defined(__GNUC__) || defined(__clang__) + return __builtin_bswap64(n); + #elif defined(__WATCOMC__) && defined(__386__) + return _watcom_bswap64(n); + #else + #error "This compiler does not support the byte swap intrinsic." + #endif +#else + /* Weird "<< 32" bitshift is required for C89 because it doesn't support 64-bit constants. Should be optimized out by a good compiler. */ + return ((n & ((drflac_uint64)0xFF000000 << 32)) >> 56) | + ((n & ((drflac_uint64)0x00FF0000 << 32)) >> 40) | + ((n & ((drflac_uint64)0x0000FF00 << 32)) >> 24) | + ((n & ((drflac_uint64)0x000000FF << 32)) >> 8) | + ((n & ((drflac_uint64)0xFF000000 )) << 8) | + ((n & ((drflac_uint64)0x00FF0000 )) << 24) | + ((n & ((drflac_uint64)0x0000FF00 )) << 40) | + ((n & ((drflac_uint64)0x000000FF )) << 56); +#endif +} + + +static DRFLAC_INLINE drflac_uint16 drflac__be2host_16(drflac_uint16 n) +{ + if (drflac__is_little_endian()) { + return drflac__swap_endian_uint16(n); + } + + return n; +} + +static DRFLAC_INLINE drflac_uint32 drflac__be2host_32(drflac_uint32 n) +{ + if (drflac__is_little_endian()) { + return drflac__swap_endian_uint32(n); + } + + return n; +} + +static DRFLAC_INLINE drflac_uint32 drflac__be2host_32_ptr_unaligned(const void* pData) +{ + const drflac_uint8* pNum = (drflac_uint8*)pData; + return *(pNum) << 24 | *(pNum+1) << 16 | *(pNum+2) << 8 | *(pNum+3); +} + +static DRFLAC_INLINE drflac_uint64 drflac__be2host_64(drflac_uint64 n) +{ + if (drflac__is_little_endian()) { + return drflac__swap_endian_uint64(n); + } + + return n; +} + + +static DRFLAC_INLINE drflac_uint32 drflac__le2host_32(drflac_uint32 n) +{ + if (!drflac__is_little_endian()) { + return drflac__swap_endian_uint32(n); + } + + return n; +} + +static DRFLAC_INLINE drflac_uint32 drflac__le2host_32_ptr_unaligned(const void* pData) +{ + const drflac_uint8* pNum = (drflac_uint8*)pData; + return *pNum | *(pNum+1) << 8 | *(pNum+2) << 16 | *(pNum+3) << 24; +} + + +static DRFLAC_INLINE drflac_uint32 drflac__unsynchsafe_32(drflac_uint32 n) +{ + drflac_uint32 result = 0; + result |= (n & 0x7F000000) >> 3; + result |= (n & 0x007F0000) >> 2; + result |= (n & 0x00007F00) >> 1; + result |= (n & 0x0000007F) >> 0; + + return result; +} + + + +/* The CRC code below is based on this document: http://zlib.net/crc_v3.txt */ +static drflac_uint8 drflac__crc8_table[] = { + 0x00, 0x07, 0x0E, 0x09, 0x1C, 0x1B, 0x12, 0x15, 0x38, 0x3F, 0x36, 0x31, 0x24, 0x23, 0x2A, 0x2D, + 0x70, 0x77, 0x7E, 0x79, 0x6C, 0x6B, 0x62, 0x65, 0x48, 0x4F, 0x46, 0x41, 0x54, 0x53, 0x5A, 0x5D, + 0xE0, 0xE7, 0xEE, 0xE9, 0xFC, 0xFB, 0xF2, 0xF5, 0xD8, 0xDF, 0xD6, 0xD1, 0xC4, 0xC3, 0xCA, 0xCD, + 0x90, 0x97, 0x9E, 0x99, 0x8C, 0x8B, 0x82, 0x85, 0xA8, 0xAF, 0xA6, 0xA1, 0xB4, 0xB3, 0xBA, 0xBD, + 0xC7, 0xC0, 0xC9, 0xCE, 0xDB, 0xDC, 0xD5, 0xD2, 0xFF, 0xF8, 0xF1, 0xF6, 0xE3, 0xE4, 0xED, 0xEA, + 0xB7, 0xB0, 0xB9, 0xBE, 0xAB, 0xAC, 0xA5, 0xA2, 0x8F, 0x88, 0x81, 0x86, 0x93, 0x94, 0x9D, 0x9A, + 0x27, 0x20, 0x29, 0x2E, 0x3B, 0x3C, 0x35, 0x32, 0x1F, 0x18, 0x11, 0x16, 0x03, 0x04, 0x0D, 0x0A, + 0x57, 0x50, 0x59, 0x5E, 0x4B, 0x4C, 0x45, 0x42, 0x6F, 0x68, 0x61, 0x66, 0x73, 0x74, 0x7D, 0x7A, + 0x89, 0x8E, 0x87, 0x80, 0x95, 0x92, 0x9B, 0x9C, 0xB1, 0xB6, 0xBF, 0xB8, 0xAD, 0xAA, 0xA3, 0xA4, + 0xF9, 0xFE, 0xF7, 0xF0, 0xE5, 0xE2, 0xEB, 0xEC, 0xC1, 0xC6, 0xCF, 0xC8, 0xDD, 0xDA, 0xD3, 0xD4, + 0x69, 0x6E, 0x67, 0x60, 0x75, 0x72, 0x7B, 0x7C, 0x51, 0x56, 0x5F, 0x58, 0x4D, 0x4A, 0x43, 0x44, + 0x19, 0x1E, 0x17, 0x10, 0x05, 0x02, 0x0B, 0x0C, 0x21, 0x26, 0x2F, 0x28, 0x3D, 0x3A, 0x33, 0x34, + 0x4E, 0x49, 0x40, 0x47, 0x52, 0x55, 0x5C, 0x5B, 0x76, 0x71, 0x78, 0x7F, 0x6A, 0x6D, 0x64, 0x63, + 0x3E, 0x39, 0x30, 0x37, 0x22, 0x25, 0x2C, 0x2B, 0x06, 0x01, 0x08, 0x0F, 0x1A, 0x1D, 0x14, 0x13, + 0xAE, 0xA9, 0xA0, 0xA7, 0xB2, 0xB5, 0xBC, 0xBB, 0x96, 0x91, 0x98, 0x9F, 0x8A, 0x8D, 0x84, 0x83, + 0xDE, 0xD9, 0xD0, 0xD7, 0xC2, 0xC5, 0xCC, 0xCB, 0xE6, 0xE1, 0xE8, 0xEF, 0xFA, 0xFD, 0xF4, 0xF3 +}; + +static drflac_uint16 drflac__crc16_table[] = { + 0x0000, 0x8005, 0x800F, 0x000A, 0x801B, 0x001E, 0x0014, 0x8011, + 0x8033, 0x0036, 0x003C, 0x8039, 0x0028, 0x802D, 0x8027, 0x0022, + 0x8063, 0x0066, 0x006C, 0x8069, 0x0078, 0x807D, 0x8077, 0x0072, + 0x0050, 0x8055, 0x805F, 0x005A, 0x804B, 0x004E, 0x0044, 0x8041, + 0x80C3, 0x00C6, 0x00CC, 0x80C9, 0x00D8, 0x80DD, 0x80D7, 0x00D2, + 0x00F0, 0x80F5, 0x80FF, 0x00FA, 0x80EB, 0x00EE, 0x00E4, 0x80E1, + 0x00A0, 0x80A5, 0x80AF, 0x00AA, 0x80BB, 0x00BE, 0x00B4, 0x80B1, + 0x8093, 0x0096, 0x009C, 0x8099, 0x0088, 0x808D, 0x8087, 0x0082, + 0x8183, 0x0186, 0x018C, 0x8189, 0x0198, 0x819D, 0x8197, 0x0192, + 0x01B0, 0x81B5, 0x81BF, 0x01BA, 0x81AB, 0x01AE, 0x01A4, 0x81A1, + 0x01E0, 0x81E5, 0x81EF, 0x01EA, 0x81FB, 0x01FE, 0x01F4, 0x81F1, + 0x81D3, 0x01D6, 0x01DC, 0x81D9, 0x01C8, 0x81CD, 0x81C7, 0x01C2, + 0x0140, 0x8145, 0x814F, 0x014A, 0x815B, 0x015E, 0x0154, 0x8151, + 0x8173, 0x0176, 0x017C, 0x8179, 0x0168, 0x816D, 0x8167, 0x0162, + 0x8123, 0x0126, 0x012C, 0x8129, 0x0138, 0x813D, 0x8137, 0x0132, + 0x0110, 0x8115, 0x811F, 0x011A, 0x810B, 0x010E, 0x0104, 0x8101, + 0x8303, 0x0306, 0x030C, 0x8309, 0x0318, 0x831D, 0x8317, 0x0312, + 0x0330, 0x8335, 0x833F, 0x033A, 0x832B, 0x032E, 0x0324, 0x8321, + 0x0360, 0x8365, 0x836F, 0x036A, 0x837B, 0x037E, 0x0374, 0x8371, + 0x8353, 0x0356, 0x035C, 0x8359, 0x0348, 0x834D, 0x8347, 0x0342, + 0x03C0, 0x83C5, 0x83CF, 0x03CA, 0x83DB, 0x03DE, 0x03D4, 0x83D1, + 0x83F3, 0x03F6, 0x03FC, 0x83F9, 0x03E8, 0x83ED, 0x83E7, 0x03E2, + 0x83A3, 0x03A6, 0x03AC, 0x83A9, 0x03B8, 0x83BD, 0x83B7, 0x03B2, + 0x0390, 0x8395, 0x839F, 0x039A, 0x838B, 0x038E, 0x0384, 0x8381, + 0x0280, 0x8285, 0x828F, 0x028A, 0x829B, 0x029E, 0x0294, 0x8291, + 0x82B3, 0x02B6, 0x02BC, 0x82B9, 0x02A8, 0x82AD, 0x82A7, 0x02A2, + 0x82E3, 0x02E6, 0x02EC, 0x82E9, 0x02F8, 0x82FD, 0x82F7, 0x02F2, + 0x02D0, 0x82D5, 0x82DF, 0x02DA, 0x82CB, 0x02CE, 0x02C4, 0x82C1, + 0x8243, 0x0246, 0x024C, 0x8249, 0x0258, 0x825D, 0x8257, 0x0252, + 0x0270, 0x8275, 0x827F, 0x027A, 0x826B, 0x026E, 0x0264, 0x8261, + 0x0220, 0x8225, 0x822F, 0x022A, 0x823B, 0x023E, 0x0234, 0x8231, + 0x8213, 0x0216, 0x021C, 0x8219, 0x0208, 0x820D, 0x8207, 0x0202 +}; + +static DRFLAC_INLINE drflac_uint8 drflac_crc8_byte(drflac_uint8 crc, drflac_uint8 data) +{ + return drflac__crc8_table[crc ^ data]; +} + +static DRFLAC_INLINE drflac_uint8 drflac_crc8(drflac_uint8 crc, drflac_uint32 data, drflac_uint32 count) +{ +#ifdef DR_FLAC_NO_CRC + (void)crc; + (void)data; + (void)count; + return 0; +#else +#if 0 + /* REFERENCE (use of this implementation requires an explicit flush by doing "drflac_crc8(crc, 0, 8);") */ + drflac_uint8 p = 0x07; + for (int i = count-1; i >= 0; --i) { + drflac_uint8 bit = (data & (1 << i)) >> i; + if (crc & 0x80) { + crc = ((crc << 1) | bit) ^ p; + } else { + crc = ((crc << 1) | bit); + } + } + return crc; +#else + drflac_uint32 wholeBytes; + drflac_uint32 leftoverBits; + drflac_uint64 leftoverDataMask; + + static drflac_uint64 leftoverDataMaskTable[8] = { + 0x00, 0x01, 0x03, 0x07, 0x0F, 0x1F, 0x3F, 0x7F + }; + + DRFLAC_ASSERT(count <= 32); + + wholeBytes = count >> 3; + leftoverBits = count - (wholeBytes*8); + leftoverDataMask = leftoverDataMaskTable[leftoverBits]; + + switch (wholeBytes) { + case 4: crc = drflac_crc8_byte(crc, (drflac_uint8)((data & (0xFF000000UL << leftoverBits)) >> (24 + leftoverBits))); + case 3: crc = drflac_crc8_byte(crc, (drflac_uint8)((data & (0x00FF0000UL << leftoverBits)) >> (16 + leftoverBits))); + case 2: crc = drflac_crc8_byte(crc, (drflac_uint8)((data & (0x0000FF00UL << leftoverBits)) >> ( 8 + leftoverBits))); + case 1: crc = drflac_crc8_byte(crc, (drflac_uint8)((data & (0x000000FFUL << leftoverBits)) >> ( 0 + leftoverBits))); + case 0: if (leftoverBits > 0) crc = (drflac_uint8)((crc << leftoverBits) ^ drflac__crc8_table[(crc >> (8 - leftoverBits)) ^ (data & leftoverDataMask)]); + } + return crc; +#endif +#endif +} + +static DRFLAC_INLINE drflac_uint16 drflac_crc16_byte(drflac_uint16 crc, drflac_uint8 data) +{ + return (crc << 8) ^ drflac__crc16_table[(drflac_uint8)(crc >> 8) ^ data]; +} + +static DRFLAC_INLINE drflac_uint16 drflac_crc16_cache(drflac_uint16 crc, drflac_cache_t data) +{ +#ifdef DRFLAC_64BIT + crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 56) & 0xFF)); + crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 48) & 0xFF)); + crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 40) & 0xFF)); + crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 32) & 0xFF)); +#endif + crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 24) & 0xFF)); + crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 16) & 0xFF)); + crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 8) & 0xFF)); + crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 0) & 0xFF)); + + return crc; +} + +static DRFLAC_INLINE drflac_uint16 drflac_crc16_bytes(drflac_uint16 crc, drflac_cache_t data, drflac_uint32 byteCount) +{ + switch (byteCount) + { +#ifdef DRFLAC_64BIT + case 8: crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 56) & 0xFF)); + case 7: crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 48) & 0xFF)); + case 6: crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 40) & 0xFF)); + case 5: crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 32) & 0xFF)); +#endif + case 4: crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 24) & 0xFF)); + case 3: crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 16) & 0xFF)); + case 2: crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 8) & 0xFF)); + case 1: crc = drflac_crc16_byte(crc, (drflac_uint8)((data >> 0) & 0xFF)); + } + + return crc; +} + +#if 0 +static DRFLAC_INLINE drflac_uint16 drflac_crc16__32bit(drflac_uint16 crc, drflac_uint32 data, drflac_uint32 count) +{ +#ifdef DR_FLAC_NO_CRC + (void)crc; + (void)data; + (void)count; + return 0; +#else +#if 0 + /* REFERENCE (use of this implementation requires an explicit flush by doing "drflac_crc16(crc, 0, 16);") */ + drflac_uint16 p = 0x8005; + for (int i = count-1; i >= 0; --i) { + drflac_uint16 bit = (data & (1ULL << i)) >> i; + if (r & 0x8000) { + r = ((r << 1) | bit) ^ p; + } else { + r = ((r << 1) | bit); + } + } + + return crc; +#else + drflac_uint32 wholeBytes; + drflac_uint32 leftoverBits; + drflac_uint64 leftoverDataMask; + + static drflac_uint64 leftoverDataMaskTable[8] = { + 0x00, 0x01, 0x03, 0x07, 0x0F, 0x1F, 0x3F, 0x7F + }; + + DRFLAC_ASSERT(count <= 64); + + wholeBytes = count >> 3; + leftoverBits = count & 7; + leftoverDataMask = leftoverDataMaskTable[leftoverBits]; + + switch (wholeBytes) { + default: + case 4: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (0xFF000000UL << leftoverBits)) >> (24 + leftoverBits))); + case 3: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (0x00FF0000UL << leftoverBits)) >> (16 + leftoverBits))); + case 2: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (0x0000FF00UL << leftoverBits)) >> ( 8 + leftoverBits))); + case 1: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (0x000000FFUL << leftoverBits)) >> ( 0 + leftoverBits))); + case 0: if (leftoverBits > 0) crc = (crc << leftoverBits) ^ drflac__crc16_table[(crc >> (16 - leftoverBits)) ^ (data & leftoverDataMask)]; + } + return crc; +#endif +#endif +} + +static DRFLAC_INLINE drflac_uint16 drflac_crc16__64bit(drflac_uint16 crc, drflac_uint64 data, drflac_uint32 count) +{ +#ifdef DR_FLAC_NO_CRC + (void)crc; + (void)data; + (void)count; + return 0; +#else + drflac_uint32 wholeBytes; + drflac_uint32 leftoverBits; + drflac_uint64 leftoverDataMask; + + static drflac_uint64 leftoverDataMaskTable[8] = { + 0x00, 0x01, 0x03, 0x07, 0x0F, 0x1F, 0x3F, 0x7F + }; + + DRFLAC_ASSERT(count <= 64); + + wholeBytes = count >> 3; + leftoverBits = count & 7; + leftoverDataMask = leftoverDataMaskTable[leftoverBits]; + + switch (wholeBytes) { + default: + case 8: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (((drflac_uint64)0xFF000000 << 32) << leftoverBits)) >> (56 + leftoverBits))); /* Weird "<< 32" bitshift is required for C89 because it doesn't support 64-bit constants. Should be optimized out by a good compiler. */ + case 7: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (((drflac_uint64)0x00FF0000 << 32) << leftoverBits)) >> (48 + leftoverBits))); + case 6: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (((drflac_uint64)0x0000FF00 << 32) << leftoverBits)) >> (40 + leftoverBits))); + case 5: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (((drflac_uint64)0x000000FF << 32) << leftoverBits)) >> (32 + leftoverBits))); + case 4: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (((drflac_uint64)0xFF000000 ) << leftoverBits)) >> (24 + leftoverBits))); + case 3: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (((drflac_uint64)0x00FF0000 ) << leftoverBits)) >> (16 + leftoverBits))); + case 2: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (((drflac_uint64)0x0000FF00 ) << leftoverBits)) >> ( 8 + leftoverBits))); + case 1: crc = drflac_crc16_byte(crc, (drflac_uint8)((data & (((drflac_uint64)0x000000FF ) << leftoverBits)) >> ( 0 + leftoverBits))); + case 0: if (leftoverBits > 0) crc = (crc << leftoverBits) ^ drflac__crc16_table[(crc >> (16 - leftoverBits)) ^ (data & leftoverDataMask)]; + } + return crc; +#endif +} + + +static DRFLAC_INLINE drflac_uint16 drflac_crc16(drflac_uint16 crc, drflac_cache_t data, drflac_uint32 count) +{ +#ifdef DRFLAC_64BIT + return drflac_crc16__64bit(crc, data, count); +#else + return drflac_crc16__32bit(crc, data, count); +#endif +} +#endif + + +#ifdef DRFLAC_64BIT +#define drflac__be2host__cache_line drflac__be2host_64 +#else +#define drflac__be2host__cache_line drflac__be2host_32 +#endif + +/* +BIT READING ATTEMPT #2 + +This uses a 32- or 64-bit bit-shifted cache - as bits are read, the cache is shifted such that the first valid bit is sitting +on the most significant bit. It uses the notion of an L1 and L2 cache (borrowed from CPU architecture), where the L1 cache +is a 32- or 64-bit unsigned integer (depending on whether or not a 32- or 64-bit build is being compiled) and the L2 is an +array of "cache lines", with each cache line being the same size as the L1. The L2 is a buffer of about 4KB and is where data +from onRead() is read into. +*/ +#define DRFLAC_CACHE_L1_SIZE_BYTES(bs) (sizeof((bs)->cache)) +#define DRFLAC_CACHE_L1_SIZE_BITS(bs) (sizeof((bs)->cache)*8) +#define DRFLAC_CACHE_L1_BITS_REMAINING(bs) (DRFLAC_CACHE_L1_SIZE_BITS(bs) - (bs)->consumedBits) +#define DRFLAC_CACHE_L1_SELECTION_MASK(_bitCount) (~((~(drflac_cache_t)0) >> (_bitCount))) +#define DRFLAC_CACHE_L1_SELECTION_SHIFT(bs, _bitCount) (DRFLAC_CACHE_L1_SIZE_BITS(bs) - (_bitCount)) +#define DRFLAC_CACHE_L1_SELECT(bs, _bitCount) (((bs)->cache) & DRFLAC_CACHE_L1_SELECTION_MASK(_bitCount)) +#define DRFLAC_CACHE_L1_SELECT_AND_SHIFT(bs, _bitCount) (DRFLAC_CACHE_L1_SELECT((bs), (_bitCount)) >> DRFLAC_CACHE_L1_SELECTION_SHIFT((bs), (_bitCount))) +#define DRFLAC_CACHE_L1_SELECT_AND_SHIFT_SAFE(bs, _bitCount)(DRFLAC_CACHE_L1_SELECT((bs), (_bitCount)) >> (DRFLAC_CACHE_L1_SELECTION_SHIFT((bs), (_bitCount)) & (DRFLAC_CACHE_L1_SIZE_BITS(bs)-1))) +#define DRFLAC_CACHE_L2_SIZE_BYTES(bs) (sizeof((bs)->cacheL2)) +#define DRFLAC_CACHE_L2_LINE_COUNT(bs) (DRFLAC_CACHE_L2_SIZE_BYTES(bs) / sizeof((bs)->cacheL2[0])) +#define DRFLAC_CACHE_L2_LINES_REMAINING(bs) (DRFLAC_CACHE_L2_LINE_COUNT(bs) - (bs)->nextL2Line) + + +#ifndef DR_FLAC_NO_CRC +static DRFLAC_INLINE void drflac__reset_crc16(drflac_bs* bs) +{ + bs->crc16 = 0; + bs->crc16CacheIgnoredBytes = bs->consumedBits >> 3; +} + +static DRFLAC_INLINE void drflac__update_crc16(drflac_bs* bs) +{ + if (bs->crc16CacheIgnoredBytes == 0) { + bs->crc16 = drflac_crc16_cache(bs->crc16, bs->crc16Cache); + } else { + bs->crc16 = drflac_crc16_bytes(bs->crc16, bs->crc16Cache, DRFLAC_CACHE_L1_SIZE_BYTES(bs) - bs->crc16CacheIgnoredBytes); + bs->crc16CacheIgnoredBytes = 0; + } +} + +static DRFLAC_INLINE drflac_uint16 drflac__flush_crc16(drflac_bs* bs) +{ + /* We should never be flushing in a situation where we are not aligned on a byte boundary. */ + DRFLAC_ASSERT((DRFLAC_CACHE_L1_BITS_REMAINING(bs) & 7) == 0); + + /* + The bits that were read from the L1 cache need to be accumulated. The number of bytes needing to be accumulated is determined + by the number of bits that have been consumed. + */ + if (DRFLAC_CACHE_L1_BITS_REMAINING(bs) == 0) { + drflac__update_crc16(bs); + } else { + /* We only accumulate the consumed bits. */ + bs->crc16 = drflac_crc16_bytes(bs->crc16, bs->crc16Cache >> DRFLAC_CACHE_L1_BITS_REMAINING(bs), (bs->consumedBits >> 3) - bs->crc16CacheIgnoredBytes); + + /* + The bits that we just accumulated should never be accumulated again. We need to keep track of how many bytes were accumulated + so we can handle that later. + */ + bs->crc16CacheIgnoredBytes = bs->consumedBits >> 3; + } + + return bs->crc16; +} +#endif + +static DRFLAC_INLINE drflac_bool32 drflac__reload_l1_cache_from_l2(drflac_bs* bs) +{ + size_t bytesRead; + size_t alignedL1LineCount; + + /* Fast path. Try loading straight from L2. */ + if (bs->nextL2Line < DRFLAC_CACHE_L2_LINE_COUNT(bs)) { + bs->cache = bs->cacheL2[bs->nextL2Line++]; + return DRFLAC_TRUE; + } + + /* + If we get here it means we've run out of data in the L2 cache. We'll need to fetch more from the client, if there's + any left. + */ + if (bs->unalignedByteCount > 0) { + return DRFLAC_FALSE; /* If we have any unaligned bytes it means there's no more aligned bytes left in the client. */ + } + + bytesRead = bs->onRead(bs->pUserData, bs->cacheL2, DRFLAC_CACHE_L2_SIZE_BYTES(bs)); + + bs->nextL2Line = 0; + if (bytesRead == DRFLAC_CACHE_L2_SIZE_BYTES(bs)) { + bs->cache = bs->cacheL2[bs->nextL2Line++]; + return DRFLAC_TRUE; + } + + + /* + If we get here it means we were unable to retrieve enough data to fill the entire L2 cache. It probably + means we've just reached the end of the file. We need to move the valid data down to the end of the buffer + and adjust the index of the next line accordingly. Also keep in mind that the L2 cache must be aligned to + the size of the L1 so we'll need to seek backwards by any misaligned bytes. + */ + alignedL1LineCount = bytesRead / DRFLAC_CACHE_L1_SIZE_BYTES(bs); + + /* We need to keep track of any unaligned bytes for later use. */ + bs->unalignedByteCount = bytesRead - (alignedL1LineCount * DRFLAC_CACHE_L1_SIZE_BYTES(bs)); + if (bs->unalignedByteCount > 0) { + bs->unalignedCache = bs->cacheL2[alignedL1LineCount]; + } + + if (alignedL1LineCount > 0) { + size_t offset = DRFLAC_CACHE_L2_LINE_COUNT(bs) - alignedL1LineCount; + size_t i; + for (i = alignedL1LineCount; i > 0; --i) { + bs->cacheL2[i-1 + offset] = bs->cacheL2[i-1]; + } + + bs->nextL2Line = (drflac_uint32)offset; + bs->cache = bs->cacheL2[bs->nextL2Line++]; + return DRFLAC_TRUE; + } else { + /* If we get into this branch it means we weren't able to load any L1-aligned data. */ + bs->nextL2Line = DRFLAC_CACHE_L2_LINE_COUNT(bs); + return DRFLAC_FALSE; + } +} + +static drflac_bool32 drflac__reload_cache(drflac_bs* bs) +{ + size_t bytesRead; + +#ifndef DR_FLAC_NO_CRC + drflac__update_crc16(bs); +#endif + + /* Fast path. Try just moving the next value in the L2 cache to the L1 cache. */ + if (drflac__reload_l1_cache_from_l2(bs)) { + bs->cache = drflac__be2host__cache_line(bs->cache); + bs->consumedBits = 0; +#ifndef DR_FLAC_NO_CRC + bs->crc16Cache = bs->cache; +#endif + return DRFLAC_TRUE; + } + + /* Slow path. */ + + /* + If we get here it means we have failed to load the L1 cache from the L2. Likely we've just reached the end of the stream and the last + few bytes did not meet the alignment requirements for the L2 cache. In this case we need to fall back to a slower path and read the + data from the unaligned cache. + */ + bytesRead = bs->unalignedByteCount; + if (bytesRead == 0) { + bs->consumedBits = DRFLAC_CACHE_L1_SIZE_BITS(bs); /* <-- The stream has been exhausted, so marked the bits as consumed. */ + return DRFLAC_FALSE; + } + + DRFLAC_ASSERT(bytesRead < DRFLAC_CACHE_L1_SIZE_BYTES(bs)); + bs->consumedBits = (drflac_uint32)(DRFLAC_CACHE_L1_SIZE_BYTES(bs) - bytesRead) * 8; + + bs->cache = drflac__be2host__cache_line(bs->unalignedCache); + bs->cache &= DRFLAC_CACHE_L1_SELECTION_MASK(DRFLAC_CACHE_L1_BITS_REMAINING(bs)); /* <-- Make sure the consumed bits are always set to zero. Other parts of the library depend on this property. */ + bs->unalignedByteCount = 0; /* <-- At this point the unaligned bytes have been moved into the cache and we thus have no more unaligned bytes. */ + +#ifndef DR_FLAC_NO_CRC + bs->crc16Cache = bs->cache >> bs->consumedBits; + bs->crc16CacheIgnoredBytes = bs->consumedBits >> 3; +#endif + return DRFLAC_TRUE; +} + +static void drflac__reset_cache(drflac_bs* bs) +{ + bs->nextL2Line = DRFLAC_CACHE_L2_LINE_COUNT(bs); /* <-- This clears the L2 cache. */ + bs->consumedBits = DRFLAC_CACHE_L1_SIZE_BITS(bs); /* <-- This clears the L1 cache. */ + bs->cache = 0; + bs->unalignedByteCount = 0; /* <-- This clears the trailing unaligned bytes. */ + bs->unalignedCache = 0; + +#ifndef DR_FLAC_NO_CRC + bs->crc16Cache = 0; + bs->crc16CacheIgnoredBytes = 0; +#endif +} + + +static DRFLAC_INLINE drflac_bool32 drflac__read_uint32(drflac_bs* bs, unsigned int bitCount, drflac_uint32* pResultOut) +{ + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pResultOut != NULL); + DRFLAC_ASSERT(bitCount > 0); + DRFLAC_ASSERT(bitCount <= 32); + + if (bs->consumedBits == DRFLAC_CACHE_L1_SIZE_BITS(bs)) { + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + } + + if (bitCount <= DRFLAC_CACHE_L1_BITS_REMAINING(bs)) { + /* + If we want to load all 32-bits from a 32-bit cache we need to do it slightly differently because we can't do + a 32-bit shift on a 32-bit integer. This will never be the case on 64-bit caches, so we can have a slightly + more optimal solution for this. + */ +#ifdef DRFLAC_64BIT + *pResultOut = (drflac_uint32)DRFLAC_CACHE_L1_SELECT_AND_SHIFT(bs, bitCount); + bs->consumedBits += bitCount; + bs->cache <<= bitCount; +#else + if (bitCount < DRFLAC_CACHE_L1_SIZE_BITS(bs)) { + *pResultOut = (drflac_uint32)DRFLAC_CACHE_L1_SELECT_AND_SHIFT(bs, bitCount); + bs->consumedBits += bitCount; + bs->cache <<= bitCount; + } else { + /* Cannot shift by 32-bits, so need to do it differently. */ + *pResultOut = (drflac_uint32)bs->cache; + bs->consumedBits = DRFLAC_CACHE_L1_SIZE_BITS(bs); + bs->cache = 0; + } +#endif + + return DRFLAC_TRUE; + } else { + /* It straddles the cached data. It will never cover more than the next chunk. We just read the number in two parts and combine them. */ + drflac_uint32 bitCountHi = DRFLAC_CACHE_L1_BITS_REMAINING(bs); + drflac_uint32 bitCountLo = bitCount - bitCountHi; + drflac_uint32 resultHi; + + DRFLAC_ASSERT(bitCountHi > 0); + DRFLAC_ASSERT(bitCountHi < 32); + resultHi = (drflac_uint32)DRFLAC_CACHE_L1_SELECT_AND_SHIFT(bs, bitCountHi); + + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + if (bitCountLo > DRFLAC_CACHE_L1_BITS_REMAINING(bs)) { + /* This happens when we get to end of stream */ + return DRFLAC_FALSE; + } + + *pResultOut = (resultHi << bitCountLo) | (drflac_uint32)DRFLAC_CACHE_L1_SELECT_AND_SHIFT(bs, bitCountLo); + bs->consumedBits += bitCountLo; + bs->cache <<= bitCountLo; + return DRFLAC_TRUE; + } +} + +static drflac_bool32 drflac__read_int32(drflac_bs* bs, unsigned int bitCount, drflac_int32* pResult) +{ + drflac_uint32 result; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pResult != NULL); + DRFLAC_ASSERT(bitCount > 0); + DRFLAC_ASSERT(bitCount <= 32); + + if (!drflac__read_uint32(bs, bitCount, &result)) { + return DRFLAC_FALSE; + } + + /* Do not attempt to shift by 32 as it's undefined. */ + if (bitCount < 32) { + drflac_uint32 signbit; + signbit = ((result >> (bitCount-1)) & 0x01); + result |= (~signbit + 1) << bitCount; + } + + *pResult = (drflac_int32)result; + return DRFLAC_TRUE; +} + +#ifdef DRFLAC_64BIT +static drflac_bool32 drflac__read_uint64(drflac_bs* bs, unsigned int bitCount, drflac_uint64* pResultOut) +{ + drflac_uint32 resultHi; + drflac_uint32 resultLo; + + DRFLAC_ASSERT(bitCount <= 64); + DRFLAC_ASSERT(bitCount > 32); + + if (!drflac__read_uint32(bs, bitCount - 32, &resultHi)) { + return DRFLAC_FALSE; + } + + if (!drflac__read_uint32(bs, 32, &resultLo)) { + return DRFLAC_FALSE; + } + + *pResultOut = (((drflac_uint64)resultHi) << 32) | ((drflac_uint64)resultLo); + return DRFLAC_TRUE; +} +#endif + +/* Function below is unused, but leaving it here in case I need to quickly add it again. */ +#if 0 +static drflac_bool32 drflac__read_int64(drflac_bs* bs, unsigned int bitCount, drflac_int64* pResultOut) +{ + drflac_uint64 result; + drflac_uint64 signbit; + + DRFLAC_ASSERT(bitCount <= 64); + + if (!drflac__read_uint64(bs, bitCount, &result)) { + return DRFLAC_FALSE; + } + + signbit = ((result >> (bitCount-1)) & 0x01); + result |= (~signbit + 1) << bitCount; + + *pResultOut = (drflac_int64)result; + return DRFLAC_TRUE; +} +#endif + +static drflac_bool32 drflac__read_uint16(drflac_bs* bs, unsigned int bitCount, drflac_uint16* pResult) +{ + drflac_uint32 result; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pResult != NULL); + DRFLAC_ASSERT(bitCount > 0); + DRFLAC_ASSERT(bitCount <= 16); + + if (!drflac__read_uint32(bs, bitCount, &result)) { + return DRFLAC_FALSE; + } + + *pResult = (drflac_uint16)result; + return DRFLAC_TRUE; +} + +#if 0 +static drflac_bool32 drflac__read_int16(drflac_bs* bs, unsigned int bitCount, drflac_int16* pResult) +{ + drflac_int32 result; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pResult != NULL); + DRFLAC_ASSERT(bitCount > 0); + DRFLAC_ASSERT(bitCount <= 16); + + if (!drflac__read_int32(bs, bitCount, &result)) { + return DRFLAC_FALSE; + } + + *pResult = (drflac_int16)result; + return DRFLAC_TRUE; +} +#endif + +static drflac_bool32 drflac__read_uint8(drflac_bs* bs, unsigned int bitCount, drflac_uint8* pResult) +{ + drflac_uint32 result; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pResult != NULL); + DRFLAC_ASSERT(bitCount > 0); + DRFLAC_ASSERT(bitCount <= 8); + + if (!drflac__read_uint32(bs, bitCount, &result)) { + return DRFLAC_FALSE; + } + + *pResult = (drflac_uint8)result; + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__read_int8(drflac_bs* bs, unsigned int bitCount, drflac_int8* pResult) +{ + drflac_int32 result; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pResult != NULL); + DRFLAC_ASSERT(bitCount > 0); + DRFLAC_ASSERT(bitCount <= 8); + + if (!drflac__read_int32(bs, bitCount, &result)) { + return DRFLAC_FALSE; + } + + *pResult = (drflac_int8)result; + return DRFLAC_TRUE; +} + + +static drflac_bool32 drflac__seek_bits(drflac_bs* bs, size_t bitsToSeek) +{ + if (bitsToSeek <= DRFLAC_CACHE_L1_BITS_REMAINING(bs)) { + bs->consumedBits += (drflac_uint32)bitsToSeek; + bs->cache <<= bitsToSeek; + return DRFLAC_TRUE; + } else { + /* It straddles the cached data. This function isn't called too frequently so I'm favouring simplicity here. */ + bitsToSeek -= DRFLAC_CACHE_L1_BITS_REMAINING(bs); + bs->consumedBits += DRFLAC_CACHE_L1_BITS_REMAINING(bs); + bs->cache = 0; + + /* Simple case. Seek in groups of the same number as bits that fit within a cache line. */ +#ifdef DRFLAC_64BIT + while (bitsToSeek >= DRFLAC_CACHE_L1_SIZE_BITS(bs)) { + drflac_uint64 bin; + if (!drflac__read_uint64(bs, DRFLAC_CACHE_L1_SIZE_BITS(bs), &bin)) { + return DRFLAC_FALSE; + } + bitsToSeek -= DRFLAC_CACHE_L1_SIZE_BITS(bs); + } +#else + while (bitsToSeek >= DRFLAC_CACHE_L1_SIZE_BITS(bs)) { + drflac_uint32 bin; + if (!drflac__read_uint32(bs, DRFLAC_CACHE_L1_SIZE_BITS(bs), &bin)) { + return DRFLAC_FALSE; + } + bitsToSeek -= DRFLAC_CACHE_L1_SIZE_BITS(bs); + } +#endif + + /* Whole leftover bytes. */ + while (bitsToSeek >= 8) { + drflac_uint8 bin; + if (!drflac__read_uint8(bs, 8, &bin)) { + return DRFLAC_FALSE; + } + bitsToSeek -= 8; + } + + /* Leftover bits. */ + if (bitsToSeek > 0) { + drflac_uint8 bin; + if (!drflac__read_uint8(bs, (drflac_uint32)bitsToSeek, &bin)) { + return DRFLAC_FALSE; + } + bitsToSeek = 0; /* <-- Necessary for the assert below. */ + } + + DRFLAC_ASSERT(bitsToSeek == 0); + return DRFLAC_TRUE; + } +} + + +/* This function moves the bit streamer to the first bit after the sync code (bit 15 of the of the frame header). It will also update the CRC-16. */ +static drflac_bool32 drflac__find_and_seek_to_next_sync_code(drflac_bs* bs) +{ + DRFLAC_ASSERT(bs != NULL); + + /* + The sync code is always aligned to 8 bits. This is convenient for us because it means we can do byte-aligned movements. The first + thing to do is align to the next byte. + */ + if (!drflac__seek_bits(bs, DRFLAC_CACHE_L1_BITS_REMAINING(bs) & 7)) { + return DRFLAC_FALSE; + } + + for (;;) { + drflac_uint8 hi; + +#ifndef DR_FLAC_NO_CRC + drflac__reset_crc16(bs); +#endif + + if (!drflac__read_uint8(bs, 8, &hi)) { + return DRFLAC_FALSE; + } + + if (hi == 0xFF) { + drflac_uint8 lo; + if (!drflac__read_uint8(bs, 6, &lo)) { + return DRFLAC_FALSE; + } + + if (lo == 0x3E) { + return DRFLAC_TRUE; + } else { + if (!drflac__seek_bits(bs, DRFLAC_CACHE_L1_BITS_REMAINING(bs) & 7)) { + return DRFLAC_FALSE; + } + } + } + } + + /* Should never get here. */ + /*return DRFLAC_FALSE;*/ +} + + +#if defined(DRFLAC_HAS_LZCNT_INTRINSIC) +#define DRFLAC_IMPLEMENT_CLZ_LZCNT +#endif +#if defined(_MSC_VER) && _MSC_VER >= 1400 && (defined(DRFLAC_X64) || defined(DRFLAC_X86)) && !defined(__clang__) +#define DRFLAC_IMPLEMENT_CLZ_MSVC +#endif +#if defined(__WATCOMC__) && defined(__386__) +#define DRFLAC_IMPLEMENT_CLZ_WATCOM +#endif +#ifdef __MRC__ +#include <intrinsics.h> +#define DRFLAC_IMPLEMENT_CLZ_MRC +#endif + +static DRFLAC_INLINE drflac_uint32 drflac__clz_software(drflac_cache_t x) +{ + drflac_uint32 n; + static drflac_uint32 clz_table_4[] = { + 0, + 4, + 3, 3, + 2, 2, 2, 2, + 1, 1, 1, 1, 1, 1, 1, 1 + }; + + if (x == 0) { + return sizeof(x)*8; + } + + n = clz_table_4[x >> (sizeof(x)*8 - 4)]; + if (n == 0) { +#ifdef DRFLAC_64BIT + if ((x & ((drflac_uint64)0xFFFFFFFF << 32)) == 0) { n = 32; x <<= 32; } + if ((x & ((drflac_uint64)0xFFFF0000 << 32)) == 0) { n += 16; x <<= 16; } + if ((x & ((drflac_uint64)0xFF000000 << 32)) == 0) { n += 8; x <<= 8; } + if ((x & ((drflac_uint64)0xF0000000 << 32)) == 0) { n += 4; x <<= 4; } +#else + if ((x & 0xFFFF0000) == 0) { n = 16; x <<= 16; } + if ((x & 0xFF000000) == 0) { n += 8; x <<= 8; } + if ((x & 0xF0000000) == 0) { n += 4; x <<= 4; } +#endif + n += clz_table_4[x >> (sizeof(x)*8 - 4)]; + } + + return n - 1; +} + +#ifdef DRFLAC_IMPLEMENT_CLZ_LZCNT +static DRFLAC_INLINE drflac_bool32 drflac__is_lzcnt_supported(void) +{ + /* Fast compile time check for ARM. */ +#if defined(DRFLAC_HAS_LZCNT_INTRINSIC) && defined(DRFLAC_ARM) && (defined(__ARM_ARCH) && __ARM_ARCH >= 5) + return DRFLAC_TRUE; +#elif defined(__MRC__) + return DRFLAC_TRUE; +#else + /* If the compiler itself does not support the intrinsic then we'll need to return false. */ + #ifdef DRFLAC_HAS_LZCNT_INTRINSIC + return drflac__gIsLZCNTSupported; + #else + return DRFLAC_FALSE; + #endif +#endif +} + +static DRFLAC_INLINE drflac_uint32 drflac__clz_lzcnt(drflac_cache_t x) +{ + /* + It's critical for competitive decoding performance that this function be highly optimal. With MSVC we can use the __lzcnt64() and __lzcnt() intrinsics + to achieve good performance, however on GCC and Clang it's a little bit more annoying. The __builtin_clzl() and __builtin_clzll() intrinsics leave + it undefined as to the return value when `x` is 0. We need this to be well defined as returning 32 or 64, depending on whether or not it's a 32- or + 64-bit build. To work around this we would need to add a conditional to check for the x = 0 case, but this creates unnecessary inefficiency. To work + around this problem I have written some inline assembly to emit the LZCNT (x86) or CLZ (ARM) instruction directly which removes the need to include + the conditional. This has worked well in the past, but for some reason Clang's MSVC compatible driver, clang-cl, does not seem to be handling this + in the same way as the normal Clang driver. It seems that `clang-cl` is just outputting the wrong results sometimes, maybe due to some register + getting clobbered? + + I'm not sure if this is a bug with dr_flac's inlined assembly (most likely), a bug in `clang-cl` or just a misunderstanding on my part with inline + assembly rules for `clang-cl`. If somebody can identify an error in dr_flac's inlined assembly I'm happy to get that fixed. + + Fortunately there is an easy workaround for this. Clang implements MSVC-specific intrinsics for compatibility. It also defines _MSC_VER for extra + compatibility. We can therefore just check for _MSC_VER and use the MSVC intrinsic which, fortunately for us, Clang supports. It would still be nice + to know how to fix the inlined assembly for correctness sake, however. + */ + +#if defined(_MSC_VER) /*&& !defined(__clang__)*/ /* <-- Intentionally wanting Clang to use the MSVC __lzcnt64/__lzcnt intrinsics due to above ^. */ + #ifdef DRFLAC_64BIT + return (drflac_uint32)__lzcnt64(x); + #else + return (drflac_uint32)__lzcnt(x); + #endif +#else + #if defined(__GNUC__) || defined(__clang__) + #if defined(DRFLAC_X64) + { + /* + A note on lzcnt. + + We check for the presence of the lzcnt instruction at runtime before calling this function, but we still generate this code. I have had + a report where the assembler does not recognize the lzcnt instruction. To work around this we are going to use `rep; bsr` instead which + has an identical byte encoding as lzcnt, and should hopefully improve compatibility with older assemblers. + */ + drflac_uint64 r; + __asm__ __volatile__ ( + "rep; bsr{q %1, %0| %0, %1}" : "=r"(r) : "r"(x) : "cc" + /*"lzcnt{ %1, %0| %0, %1}" : "=r"(r) : "r"(x) : "cc"*/ + ); + + return (drflac_uint32)r; + } + #elif defined(DRFLAC_X86) + { + drflac_uint32 r; + __asm__ __volatile__ ( + "rep; bsr{l %1, %0| %0, %1}" : "=r"(r) : "r"(x) : "cc" + /*"lzcnt{l %1, %0| %0, %1}" : "=r"(r) : "r"(x) : "cc"*/ + ); + + return r; + } + #elif defined(DRFLAC_ARM) && (defined(__ARM_ARCH) && __ARM_ARCH >= 5) && !defined(__ARM_ARCH_6M__) && !(defined(__thumb__) && !defined(__thumb2__)) && !defined(DRFLAC_64BIT) /* <-- I haven't tested 64-bit inline assembly, so only enabling this for the 32-bit build for now. */ + { + unsigned int r; + __asm__ __volatile__ ( + #if defined(DRFLAC_64BIT) + "clz %w[out], %w[in]" : [out]"=r"(r) : [in]"r"(x) /* <-- This is untested. If someone in the community could test this, that would be appreciated! */ + #else + "clz %[out], %[in]" : [out]"=r"(r) : [in]"r"(x) + #endif + ); + + return r; + } + #else + if (x == 0) { + return sizeof(x)*8; + } + #ifdef DRFLAC_64BIT + return (drflac_uint32)__builtin_clzll((drflac_uint64)x); + #else + return (drflac_uint32)__builtin_clzl((drflac_uint32)x); + #endif + #endif + #else + /* Unsupported compiler. */ + #error "This compiler does not support the lzcnt intrinsic." + #endif +#endif +} +#endif + +#ifdef DRFLAC_IMPLEMENT_CLZ_MSVC +#include <intrin.h> /* For BitScanReverse(). */ + +static DRFLAC_INLINE drflac_uint32 drflac__clz_msvc(drflac_cache_t x) +{ + drflac_uint32 n; + + if (x == 0) { + return sizeof(x)*8; + } + +#ifdef DRFLAC_64BIT + _BitScanReverse64((unsigned long*)&n, x); +#else + _BitScanReverse((unsigned long*)&n, x); +#endif + return sizeof(x)*8 - n - 1; +} +#endif + +#ifdef DRFLAC_IMPLEMENT_CLZ_WATCOM +static __inline drflac_uint32 drflac__clz_watcom (drflac_uint32); +#ifdef DRFLAC_IMPLEMENT_CLZ_WATCOM_LZCNT +/* Use the LZCNT instruction (only available on some processors since the 2010s). */ +#pragma aux drflac__clz_watcom_lzcnt = \ + "db 0F3h, 0Fh, 0BDh, 0C0h" /* lzcnt eax, eax */ \ + parm [eax] \ + value [eax] \ + modify nomemory; +#else +/* Use the 386+-compatible implementation. */ +#pragma aux drflac__clz_watcom = \ + "bsr eax, eax" \ + "xor eax, 31" \ + parm [eax] nomemory \ + value [eax] \ + modify exact [eax] nomemory; +#endif +#endif + +static DRFLAC_INLINE drflac_uint32 drflac__clz(drflac_cache_t x) +{ +#ifdef DRFLAC_IMPLEMENT_CLZ_LZCNT + if (drflac__is_lzcnt_supported()) { + return drflac__clz_lzcnt(x); + } else +#endif + { +#ifdef DRFLAC_IMPLEMENT_CLZ_MSVC + return drflac__clz_msvc(x); +#elif defined(DRFLAC_IMPLEMENT_CLZ_WATCOM_LZCNT) + return drflac__clz_watcom_lzcnt(x); +#elif defined(DRFLAC_IMPLEMENT_CLZ_WATCOM) + return (x == 0) ? sizeof(x)*8 : drflac__clz_watcom(x); +#elif defined(__MRC__) + return __cntlzw(x); +#else + return drflac__clz_software(x); +#endif + } +} + + +static DRFLAC_INLINE drflac_bool32 drflac__seek_past_next_set_bit(drflac_bs* bs, unsigned int* pOffsetOut) +{ + drflac_uint32 zeroCounter = 0; + drflac_uint32 setBitOffsetPlus1; + + while (bs->cache == 0) { + zeroCounter += (drflac_uint32)DRFLAC_CACHE_L1_BITS_REMAINING(bs); + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + } + + if (bs->cache == 1) { + /* Not catching this would lead to undefined behaviour: a shift of a 32-bit number by 32 or more is undefined */ + *pOffsetOut = zeroCounter + (drflac_uint32)DRFLAC_CACHE_L1_BITS_REMAINING(bs) - 1; + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + + return DRFLAC_TRUE; + } + + setBitOffsetPlus1 = drflac__clz(bs->cache); + setBitOffsetPlus1 += 1; + + if (setBitOffsetPlus1 > DRFLAC_CACHE_L1_BITS_REMAINING(bs)) { + /* This happens when we get to end of stream */ + return DRFLAC_FALSE; + } + + bs->consumedBits += setBitOffsetPlus1; + bs->cache <<= setBitOffsetPlus1; + + *pOffsetOut = zeroCounter + setBitOffsetPlus1 - 1; + return DRFLAC_TRUE; +} + + + +static drflac_bool32 drflac__seek_to_byte(drflac_bs* bs, drflac_uint64 offsetFromStart) +{ + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(offsetFromStart > 0); + + /* + Seeking from the start is not quite as trivial as it sounds because the onSeek callback takes a signed 32-bit integer (which + is intentional because it simplifies the implementation of the onSeek callbacks), however offsetFromStart is unsigned 64-bit. + To resolve we just need to do an initial seek from the start, and then a series of offset seeks to make up the remainder. + */ + if (offsetFromStart > 0x7FFFFFFF) { + drflac_uint64 bytesRemaining = offsetFromStart; + if (!bs->onSeek(bs->pUserData, 0x7FFFFFFF, DRFLAC_SEEK_SET)) { + return DRFLAC_FALSE; + } + bytesRemaining -= 0x7FFFFFFF; + + while (bytesRemaining > 0x7FFFFFFF) { + if (!bs->onSeek(bs->pUserData, 0x7FFFFFFF, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + bytesRemaining -= 0x7FFFFFFF; + } + + if (bytesRemaining > 0) { + if (!bs->onSeek(bs->pUserData, (int)bytesRemaining, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + } + } else { + if (!bs->onSeek(bs->pUserData, (int)offsetFromStart, DRFLAC_SEEK_SET)) { + return DRFLAC_FALSE; + } + } + + /* The cache should be reset to force a reload of fresh data from the client. */ + drflac__reset_cache(bs); + return DRFLAC_TRUE; +} + + +static drflac_result drflac__read_utf8_coded_number(drflac_bs* bs, drflac_uint64* pNumberOut, drflac_uint8* pCRCOut) +{ + drflac_uint8 crc; + drflac_uint64 result; + drflac_uint8 utf8[7] = {0}; + int byteCount; + int i; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pNumberOut != NULL); + DRFLAC_ASSERT(pCRCOut != NULL); + + crc = *pCRCOut; + + if (!drflac__read_uint8(bs, 8, utf8)) { + *pNumberOut = 0; + return DRFLAC_AT_END; + } + crc = drflac_crc8(crc, utf8[0], 8); + + if ((utf8[0] & 0x80) == 0) { + *pNumberOut = utf8[0]; + *pCRCOut = crc; + return DRFLAC_SUCCESS; + } + + /*byteCount = 1;*/ + if ((utf8[0] & 0xE0) == 0xC0) { + byteCount = 2; + } else if ((utf8[0] & 0xF0) == 0xE0) { + byteCount = 3; + } else if ((utf8[0] & 0xF8) == 0xF0) { + byteCount = 4; + } else if ((utf8[0] & 0xFC) == 0xF8) { + byteCount = 5; + } else if ((utf8[0] & 0xFE) == 0xFC) { + byteCount = 6; + } else if ((utf8[0] & 0xFF) == 0xFE) { + byteCount = 7; + } else { + *pNumberOut = 0; + return DRFLAC_CRC_MISMATCH; /* Bad UTF-8 encoding. */ + } + + /* Read extra bytes. */ + DRFLAC_ASSERT(byteCount > 1); + + result = (drflac_uint64)(utf8[0] & (0xFF >> (byteCount + 1))); + for (i = 1; i < byteCount; ++i) { + if (!drflac__read_uint8(bs, 8, utf8 + i)) { + *pNumberOut = 0; + return DRFLAC_AT_END; + } + crc = drflac_crc8(crc, utf8[i], 8); + + result = (result << 6) | (utf8[i] & 0x3F); + } + + *pNumberOut = result; + *pCRCOut = crc; + return DRFLAC_SUCCESS; +} + + +static DRFLAC_INLINE drflac_uint32 drflac__ilog2_u32(drflac_uint32 x) +{ +#if 1 /* Needs optimizing. */ + drflac_uint32 result = 0; + while (x > 0) { + result += 1; + x >>= 1; + } + + return result; +#endif +} + +static DRFLAC_INLINE drflac_bool32 drflac__use_64_bit_prediction(drflac_uint32 bitsPerSample, drflac_uint32 order, drflac_uint32 precision) +{ + /* https://web.archive.org/web/20220205005724/https://github.com/ietf-wg-cellar/flac-specification/blob/37a49aa48ba4ba12e8757badfc59c0df35435fec/rfc_backmatter.md */ + return bitsPerSample + precision + drflac__ilog2_u32(order) > 32; +} + + +/* +The next two functions are responsible for calculating the prediction. + +When the bits per sample is >16 we need to use 64-bit integer arithmetic because otherwise we'll run out of precision. It's +safe to assume this will be slower on 32-bit platforms so we use a more optimal solution when the bits per sample is <=16. +*/ +#if defined(__clang__) +__attribute__((no_sanitize("signed-integer-overflow"))) +#endif +static DRFLAC_INLINE drflac_int32 drflac__calculate_prediction_32(drflac_uint32 order, drflac_int32 shift, const drflac_int32* coefficients, drflac_int32* pDecodedSamples) +{ + drflac_int32 prediction = 0; + + DRFLAC_ASSERT(order <= 32); + + /* 32-bit version. */ + + /* VC++ optimizes this to a single jmp. I've not yet verified this for other compilers. */ + switch (order) + { + case 32: prediction += coefficients[31] * pDecodedSamples[-32]; + case 31: prediction += coefficients[30] * pDecodedSamples[-31]; + case 30: prediction += coefficients[29] * pDecodedSamples[-30]; + case 29: prediction += coefficients[28] * pDecodedSamples[-29]; + case 28: prediction += coefficients[27] * pDecodedSamples[-28]; + case 27: prediction += coefficients[26] * pDecodedSamples[-27]; + case 26: prediction += coefficients[25] * pDecodedSamples[-26]; + case 25: prediction += coefficients[24] * pDecodedSamples[-25]; + case 24: prediction += coefficients[23] * pDecodedSamples[-24]; + case 23: prediction += coefficients[22] * pDecodedSamples[-23]; + case 22: prediction += coefficients[21] * pDecodedSamples[-22]; + case 21: prediction += coefficients[20] * pDecodedSamples[-21]; + case 20: prediction += coefficients[19] * pDecodedSamples[-20]; + case 19: prediction += coefficients[18] * pDecodedSamples[-19]; + case 18: prediction += coefficients[17] * pDecodedSamples[-18]; + case 17: prediction += coefficients[16] * pDecodedSamples[-17]; + case 16: prediction += coefficients[15] * pDecodedSamples[-16]; + case 15: prediction += coefficients[14] * pDecodedSamples[-15]; + case 14: prediction += coefficients[13] * pDecodedSamples[-14]; + case 13: prediction += coefficients[12] * pDecodedSamples[-13]; + case 12: prediction += coefficients[11] * pDecodedSamples[-12]; + case 11: prediction += coefficients[10] * pDecodedSamples[-11]; + case 10: prediction += coefficients[ 9] * pDecodedSamples[-10]; + case 9: prediction += coefficients[ 8] * pDecodedSamples[- 9]; + case 8: prediction += coefficients[ 7] * pDecodedSamples[- 8]; + case 7: prediction += coefficients[ 6] * pDecodedSamples[- 7]; + case 6: prediction += coefficients[ 5] * pDecodedSamples[- 6]; + case 5: prediction += coefficients[ 4] * pDecodedSamples[- 5]; + case 4: prediction += coefficients[ 3] * pDecodedSamples[- 4]; + case 3: prediction += coefficients[ 2] * pDecodedSamples[- 3]; + case 2: prediction += coefficients[ 1] * pDecodedSamples[- 2]; + case 1: prediction += coefficients[ 0] * pDecodedSamples[- 1]; + } + + return (drflac_int32)(prediction >> shift); +} + +static DRFLAC_INLINE drflac_int32 drflac__calculate_prediction_64(drflac_uint32 order, drflac_int32 shift, const drflac_int32* coefficients, drflac_int32* pDecodedSamples) +{ + drflac_int64 prediction; + + DRFLAC_ASSERT(order <= 32); + + /* 64-bit version. */ + + /* This method is faster on the 32-bit build when compiling with VC++. See note below. */ +#ifndef DRFLAC_64BIT + if (order == 8) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + prediction += coefficients[4] * (drflac_int64)pDecodedSamples[-5]; + prediction += coefficients[5] * (drflac_int64)pDecodedSamples[-6]; + prediction += coefficients[6] * (drflac_int64)pDecodedSamples[-7]; + prediction += coefficients[7] * (drflac_int64)pDecodedSamples[-8]; + } + else if (order == 7) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + prediction += coefficients[4] * (drflac_int64)pDecodedSamples[-5]; + prediction += coefficients[5] * (drflac_int64)pDecodedSamples[-6]; + prediction += coefficients[6] * (drflac_int64)pDecodedSamples[-7]; + } + else if (order == 3) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + } + else if (order == 6) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + prediction += coefficients[4] * (drflac_int64)pDecodedSamples[-5]; + prediction += coefficients[5] * (drflac_int64)pDecodedSamples[-6]; + } + else if (order == 5) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + prediction += coefficients[4] * (drflac_int64)pDecodedSamples[-5]; + } + else if (order == 4) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + } + else if (order == 12) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + prediction += coefficients[4] * (drflac_int64)pDecodedSamples[-5]; + prediction += coefficients[5] * (drflac_int64)pDecodedSamples[-6]; + prediction += coefficients[6] * (drflac_int64)pDecodedSamples[-7]; + prediction += coefficients[7] * (drflac_int64)pDecodedSamples[-8]; + prediction += coefficients[8] * (drflac_int64)pDecodedSamples[-9]; + prediction += coefficients[9] * (drflac_int64)pDecodedSamples[-10]; + prediction += coefficients[10] * (drflac_int64)pDecodedSamples[-11]; + prediction += coefficients[11] * (drflac_int64)pDecodedSamples[-12]; + } + else if (order == 2) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + } + else if (order == 1) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + } + else if (order == 10) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + prediction += coefficients[4] * (drflac_int64)pDecodedSamples[-5]; + prediction += coefficients[5] * (drflac_int64)pDecodedSamples[-6]; + prediction += coefficients[6] * (drflac_int64)pDecodedSamples[-7]; + prediction += coefficients[7] * (drflac_int64)pDecodedSamples[-8]; + prediction += coefficients[8] * (drflac_int64)pDecodedSamples[-9]; + prediction += coefficients[9] * (drflac_int64)pDecodedSamples[-10]; + } + else if (order == 9) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + prediction += coefficients[4] * (drflac_int64)pDecodedSamples[-5]; + prediction += coefficients[5] * (drflac_int64)pDecodedSamples[-6]; + prediction += coefficients[6] * (drflac_int64)pDecodedSamples[-7]; + prediction += coefficients[7] * (drflac_int64)pDecodedSamples[-8]; + prediction += coefficients[8] * (drflac_int64)pDecodedSamples[-9]; + } + else if (order == 11) + { + prediction = coefficients[0] * (drflac_int64)pDecodedSamples[-1]; + prediction += coefficients[1] * (drflac_int64)pDecodedSamples[-2]; + prediction += coefficients[2] * (drflac_int64)pDecodedSamples[-3]; + prediction += coefficients[3] * (drflac_int64)pDecodedSamples[-4]; + prediction += coefficients[4] * (drflac_int64)pDecodedSamples[-5]; + prediction += coefficients[5] * (drflac_int64)pDecodedSamples[-6]; + prediction += coefficients[6] * (drflac_int64)pDecodedSamples[-7]; + prediction += coefficients[7] * (drflac_int64)pDecodedSamples[-8]; + prediction += coefficients[8] * (drflac_int64)pDecodedSamples[-9]; + prediction += coefficients[9] * (drflac_int64)pDecodedSamples[-10]; + prediction += coefficients[10] * (drflac_int64)pDecodedSamples[-11]; + } + else + { + int j; + + prediction = 0; + for (j = 0; j < (int)order; ++j) { + prediction += coefficients[j] * (drflac_int64)pDecodedSamples[-j-1]; + } + } +#endif + + /* + VC++ optimizes this to a single jmp instruction, but only the 64-bit build. The 32-bit build generates less efficient code for some + reason. The ugly version above is faster so we'll just switch between the two depending on the target platform. + */ +#ifdef DRFLAC_64BIT + prediction = 0; + switch (order) + { + case 32: prediction += coefficients[31] * (drflac_int64)pDecodedSamples[-32]; + case 31: prediction += coefficients[30] * (drflac_int64)pDecodedSamples[-31]; + case 30: prediction += coefficients[29] * (drflac_int64)pDecodedSamples[-30]; + case 29: prediction += coefficients[28] * (drflac_int64)pDecodedSamples[-29]; + case 28: prediction += coefficients[27] * (drflac_int64)pDecodedSamples[-28]; + case 27: prediction += coefficients[26] * (drflac_int64)pDecodedSamples[-27]; + case 26: prediction += coefficients[25] * (drflac_int64)pDecodedSamples[-26]; + case 25: prediction += coefficients[24] * (drflac_int64)pDecodedSamples[-25]; + case 24: prediction += coefficients[23] * (drflac_int64)pDecodedSamples[-24]; + case 23: prediction += coefficients[22] * (drflac_int64)pDecodedSamples[-23]; + case 22: prediction += coefficients[21] * (drflac_int64)pDecodedSamples[-22]; + case 21: prediction += coefficients[20] * (drflac_int64)pDecodedSamples[-21]; + case 20: prediction += coefficients[19] * (drflac_int64)pDecodedSamples[-20]; + case 19: prediction += coefficients[18] * (drflac_int64)pDecodedSamples[-19]; + case 18: prediction += coefficients[17] * (drflac_int64)pDecodedSamples[-18]; + case 17: prediction += coefficients[16] * (drflac_int64)pDecodedSamples[-17]; + case 16: prediction += coefficients[15] * (drflac_int64)pDecodedSamples[-16]; + case 15: prediction += coefficients[14] * (drflac_int64)pDecodedSamples[-15]; + case 14: prediction += coefficients[13] * (drflac_int64)pDecodedSamples[-14]; + case 13: prediction += coefficients[12] * (drflac_int64)pDecodedSamples[-13]; + case 12: prediction += coefficients[11] * (drflac_int64)pDecodedSamples[-12]; + case 11: prediction += coefficients[10] * (drflac_int64)pDecodedSamples[-11]; + case 10: prediction += coefficients[ 9] * (drflac_int64)pDecodedSamples[-10]; + case 9: prediction += coefficients[ 8] * (drflac_int64)pDecodedSamples[- 9]; + case 8: prediction += coefficients[ 7] * (drflac_int64)pDecodedSamples[- 8]; + case 7: prediction += coefficients[ 6] * (drflac_int64)pDecodedSamples[- 7]; + case 6: prediction += coefficients[ 5] * (drflac_int64)pDecodedSamples[- 6]; + case 5: prediction += coefficients[ 4] * (drflac_int64)pDecodedSamples[- 5]; + case 4: prediction += coefficients[ 3] * (drflac_int64)pDecodedSamples[- 4]; + case 3: prediction += coefficients[ 2] * (drflac_int64)pDecodedSamples[- 3]; + case 2: prediction += coefficients[ 1] * (drflac_int64)pDecodedSamples[- 2]; + case 1: prediction += coefficients[ 0] * (drflac_int64)pDecodedSamples[- 1]; + } +#endif + + return (drflac_int32)(prediction >> shift); +} + + +#if 0 +/* +Reference implementation for reading and decoding samples with residual. This is intentionally left unoptimized for the +sake of readability and should only be used as a reference. +*/ +static drflac_bool32 drflac__decode_samples_with_residual__rice__reference(drflac_bs* bs, drflac_uint32 bitsPerSample, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 lpcOrder, drflac_int32 lpcShift, drflac_uint32 lpcPrecision, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + drflac_uint32 i; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pSamplesOut != NULL); + + for (i = 0; i < count; ++i) { + drflac_uint32 zeroCounter = 0; + for (;;) { + drflac_uint8 bit; + if (!drflac__read_uint8(bs, 1, &bit)) { + return DRFLAC_FALSE; + } + + if (bit == 0) { + zeroCounter += 1; + } else { + break; + } + } + + drflac_uint32 decodedRice; + if (riceParam > 0) { + if (!drflac__read_uint32(bs, riceParam, &decodedRice)) { + return DRFLAC_FALSE; + } + } else { + decodedRice = 0; + } + + decodedRice |= (zeroCounter << riceParam); + if ((decodedRice & 0x01)) { + decodedRice = ~(decodedRice >> 1); + } else { + decodedRice = (decodedRice >> 1); + } + + + if (drflac__use_64_bit_prediction(bitsPerSample, lpcOrder, lpcPrecision)) { + pSamplesOut[i] = decodedRice + drflac__calculate_prediction_64(lpcOrder, lpcShift, coefficients, pSamplesOut + i); + } else { + pSamplesOut[i] = decodedRice + drflac__calculate_prediction_32(lpcOrder, lpcShift, coefficients, pSamplesOut + i); + } + } + + return DRFLAC_TRUE; +} +#endif + +#if 0 +static drflac_bool32 drflac__read_rice_parts__reference(drflac_bs* bs, drflac_uint8 riceParam, drflac_uint32* pZeroCounterOut, drflac_uint32* pRiceParamPartOut) +{ + drflac_uint32 zeroCounter = 0; + drflac_uint32 decodedRice; + + for (;;) { + drflac_uint8 bit; + if (!drflac__read_uint8(bs, 1, &bit)) { + return DRFLAC_FALSE; + } + + if (bit == 0) { + zeroCounter += 1; + } else { + break; + } + } + + if (riceParam > 0) { + if (!drflac__read_uint32(bs, riceParam, &decodedRice)) { + return DRFLAC_FALSE; + } + } else { + decodedRice = 0; + } + + *pZeroCounterOut = zeroCounter; + *pRiceParamPartOut = decodedRice; + return DRFLAC_TRUE; +} +#endif + +#if 0 +static DRFLAC_INLINE drflac_bool32 drflac__read_rice_parts(drflac_bs* bs, drflac_uint8 riceParam, drflac_uint32* pZeroCounterOut, drflac_uint32* pRiceParamPartOut) +{ + drflac_cache_t riceParamMask; + drflac_uint32 zeroCounter; + drflac_uint32 setBitOffsetPlus1; + drflac_uint32 riceParamPart; + drflac_uint32 riceLength; + + DRFLAC_ASSERT(riceParam > 0); /* <-- riceParam should never be 0. drflac__read_rice_parts__param_equals_zero() should be used instead for this case. */ + + riceParamMask = DRFLAC_CACHE_L1_SELECTION_MASK(riceParam); + + zeroCounter = 0; + while (bs->cache == 0) { + zeroCounter += (drflac_uint32)DRFLAC_CACHE_L1_BITS_REMAINING(bs); + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + } + + setBitOffsetPlus1 = drflac__clz(bs->cache); + zeroCounter += setBitOffsetPlus1; + setBitOffsetPlus1 += 1; + + riceLength = setBitOffsetPlus1 + riceParam; + if (riceLength < DRFLAC_CACHE_L1_BITS_REMAINING(bs)) { + riceParamPart = (drflac_uint32)((bs->cache & (riceParamMask >> setBitOffsetPlus1)) >> DRFLAC_CACHE_L1_SELECTION_SHIFT(bs, riceLength)); + + bs->consumedBits += riceLength; + bs->cache <<= riceLength; + } else { + drflac_uint32 bitCountLo; + drflac_cache_t resultHi; + + bs->consumedBits += riceLength; + bs->cache <<= setBitOffsetPlus1 & (DRFLAC_CACHE_L1_SIZE_BITS(bs)-1); /* <-- Equivalent to "if (setBitOffsetPlus1 < DRFLAC_CACHE_L1_SIZE_BITS(bs)) { bs->cache <<= setBitOffsetPlus1; }" */ + + /* It straddles the cached data. It will never cover more than the next chunk. We just read the number in two parts and combine them. */ + bitCountLo = bs->consumedBits - DRFLAC_CACHE_L1_SIZE_BITS(bs); + resultHi = DRFLAC_CACHE_L1_SELECT_AND_SHIFT(bs, riceParam); /* <-- Use DRFLAC_CACHE_L1_SELECT_AND_SHIFT_SAFE() if ever this function allows riceParam=0. */ + + if (bs->nextL2Line < DRFLAC_CACHE_L2_LINE_COUNT(bs)) { +#ifndef DR_FLAC_NO_CRC + drflac__update_crc16(bs); +#endif + bs->cache = drflac__be2host__cache_line(bs->cacheL2[bs->nextL2Line++]); + bs->consumedBits = 0; +#ifndef DR_FLAC_NO_CRC + bs->crc16Cache = bs->cache; +#endif + } else { + /* Slow path. We need to fetch more data from the client. */ + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + if (bitCountLo > DRFLAC_CACHE_L1_BITS_REMAINING(bs)) { + /* This happens when we get to end of stream */ + return DRFLAC_FALSE; + } + } + + riceParamPart = (drflac_uint32)(resultHi | DRFLAC_CACHE_L1_SELECT_AND_SHIFT_SAFE(bs, bitCountLo)); + + bs->consumedBits += bitCountLo; + bs->cache <<= bitCountLo; + } + + pZeroCounterOut[0] = zeroCounter; + pRiceParamPartOut[0] = riceParamPart; + + return DRFLAC_TRUE; +} +#endif + +static DRFLAC_INLINE drflac_bool32 drflac__read_rice_parts_x1(drflac_bs* bs, drflac_uint8 riceParam, drflac_uint32* pZeroCounterOut, drflac_uint32* pRiceParamPartOut) +{ + drflac_uint32 riceParamPlus1 = riceParam + 1; + /*drflac_cache_t riceParamPlus1Mask = DRFLAC_CACHE_L1_SELECTION_MASK(riceParamPlus1);*/ + drflac_uint32 riceParamPlus1Shift = DRFLAC_CACHE_L1_SELECTION_SHIFT(bs, riceParamPlus1); + drflac_uint32 riceParamPlus1MaxConsumedBits = DRFLAC_CACHE_L1_SIZE_BITS(bs) - riceParamPlus1; + + /* + The idea here is to use local variables for the cache in an attempt to encourage the compiler to store them in registers. I have + no idea how this will work in practice... + */ + drflac_cache_t bs_cache = bs->cache; + drflac_uint32 bs_consumedBits = bs->consumedBits; + + /* The first thing to do is find the first unset bit. Most likely a bit will be set in the current cache line. */ + drflac_uint32 lzcount = drflac__clz(bs_cache); + if (lzcount < sizeof(bs_cache)*8) { + pZeroCounterOut[0] = lzcount; + + /* + It is most likely that the riceParam part (which comes after the zero counter) is also on this cache line. When extracting + this, we include the set bit from the unary coded part because it simplifies cache management. This bit will be handled + outside of this function at a higher level. + */ + extract_rice_param_part: + bs_cache <<= lzcount; + bs_consumedBits += lzcount; + + if (bs_consumedBits <= riceParamPlus1MaxConsumedBits) { + /* Getting here means the rice parameter part is wholly contained within the current cache line. */ + pRiceParamPartOut[0] = (drflac_uint32)(bs_cache >> riceParamPlus1Shift); + bs_cache <<= riceParamPlus1; + bs_consumedBits += riceParamPlus1; + } else { + drflac_uint32 riceParamPartHi; + drflac_uint32 riceParamPartLo; + drflac_uint32 riceParamPartLoBitCount; + + /* + Getting here means the rice parameter part straddles the cache line. We need to read from the tail of the current cache + line, reload the cache, and then combine it with the head of the next cache line. + */ + + /* Grab the high part of the rice parameter part. */ + riceParamPartHi = (drflac_uint32)(bs_cache >> riceParamPlus1Shift); + + /* Before reloading the cache we need to grab the size in bits of the low part. */ + riceParamPartLoBitCount = bs_consumedBits - riceParamPlus1MaxConsumedBits; + DRFLAC_ASSERT(riceParamPartLoBitCount > 0 && riceParamPartLoBitCount < 32); + + /* Now reload the cache. */ + if (bs->nextL2Line < DRFLAC_CACHE_L2_LINE_COUNT(bs)) { + #ifndef DR_FLAC_NO_CRC + drflac__update_crc16(bs); + #endif + bs_cache = drflac__be2host__cache_line(bs->cacheL2[bs->nextL2Line++]); + bs_consumedBits = riceParamPartLoBitCount; + #ifndef DR_FLAC_NO_CRC + bs->crc16Cache = bs_cache; + #endif + } else { + /* Slow path. We need to fetch more data from the client. */ + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + if (riceParamPartLoBitCount > DRFLAC_CACHE_L1_BITS_REMAINING(bs)) { + /* This happens when we get to end of stream */ + return DRFLAC_FALSE; + } + + bs_cache = bs->cache; + bs_consumedBits = bs->consumedBits + riceParamPartLoBitCount; + } + + /* We should now have enough information to construct the rice parameter part. */ + riceParamPartLo = (drflac_uint32)(bs_cache >> (DRFLAC_CACHE_L1_SELECTION_SHIFT(bs, riceParamPartLoBitCount))); + pRiceParamPartOut[0] = riceParamPartHi | riceParamPartLo; + + bs_cache <<= riceParamPartLoBitCount; + } + } else { + /* + Getting here means there are no bits set on the cache line. This is a less optimal case because we just wasted a call + to drflac__clz() and we need to reload the cache. + */ + drflac_uint32 zeroCounter = (drflac_uint32)(DRFLAC_CACHE_L1_SIZE_BITS(bs) - bs_consumedBits); + for (;;) { + if (bs->nextL2Line < DRFLAC_CACHE_L2_LINE_COUNT(bs)) { + #ifndef DR_FLAC_NO_CRC + drflac__update_crc16(bs); + #endif + bs_cache = drflac__be2host__cache_line(bs->cacheL2[bs->nextL2Line++]); + bs_consumedBits = 0; + #ifndef DR_FLAC_NO_CRC + bs->crc16Cache = bs_cache; + #endif + } else { + /* Slow path. We need to fetch more data from the client. */ + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + + bs_cache = bs->cache; + bs_consumedBits = bs->consumedBits; + } + + lzcount = drflac__clz(bs_cache); + zeroCounter += lzcount; + + if (lzcount < sizeof(bs_cache)*8) { + break; + } + } + + pZeroCounterOut[0] = zeroCounter; + goto extract_rice_param_part; + } + + /* Make sure the cache is restored at the end of it all. */ + bs->cache = bs_cache; + bs->consumedBits = bs_consumedBits; + + return DRFLAC_TRUE; +} + +static DRFLAC_INLINE drflac_bool32 drflac__seek_rice_parts(drflac_bs* bs, drflac_uint8 riceParam) +{ + drflac_uint32 riceParamPlus1 = riceParam + 1; + drflac_uint32 riceParamPlus1MaxConsumedBits = DRFLAC_CACHE_L1_SIZE_BITS(bs) - riceParamPlus1; + + /* + The idea here is to use local variables for the cache in an attempt to encourage the compiler to store them in registers. I have + no idea how this will work in practice... + */ + drflac_cache_t bs_cache = bs->cache; + drflac_uint32 bs_consumedBits = bs->consumedBits; + + /* The first thing to do is find the first unset bit. Most likely a bit will be set in the current cache line. */ + drflac_uint32 lzcount = drflac__clz(bs_cache); + if (lzcount < sizeof(bs_cache)*8) { + /* + It is most likely that the riceParam part (which comes after the zero counter) is also on this cache line. When extracting + this, we include the set bit from the unary coded part because it simplifies cache management. This bit will be handled + outside of this function at a higher level. + */ + extract_rice_param_part: + bs_cache <<= lzcount; + bs_consumedBits += lzcount; + + if (bs_consumedBits <= riceParamPlus1MaxConsumedBits) { + /* Getting here means the rice parameter part is wholly contained within the current cache line. */ + bs_cache <<= riceParamPlus1; + bs_consumedBits += riceParamPlus1; + } else { + /* + Getting here means the rice parameter part straddles the cache line. We need to read from the tail of the current cache + line, reload the cache, and then combine it with the head of the next cache line. + */ + + /* Before reloading the cache we need to grab the size in bits of the low part. */ + drflac_uint32 riceParamPartLoBitCount = bs_consumedBits - riceParamPlus1MaxConsumedBits; + DRFLAC_ASSERT(riceParamPartLoBitCount > 0 && riceParamPartLoBitCount < 32); + + /* Now reload the cache. */ + if (bs->nextL2Line < DRFLAC_CACHE_L2_LINE_COUNT(bs)) { + #ifndef DR_FLAC_NO_CRC + drflac__update_crc16(bs); + #endif + bs_cache = drflac__be2host__cache_line(bs->cacheL2[bs->nextL2Line++]); + bs_consumedBits = riceParamPartLoBitCount; + #ifndef DR_FLAC_NO_CRC + bs->crc16Cache = bs_cache; + #endif + } else { + /* Slow path. We need to fetch more data from the client. */ + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + + if (riceParamPartLoBitCount > DRFLAC_CACHE_L1_BITS_REMAINING(bs)) { + /* This happens when we get to end of stream */ + return DRFLAC_FALSE; + } + + bs_cache = bs->cache; + bs_consumedBits = bs->consumedBits + riceParamPartLoBitCount; + } + + bs_cache <<= riceParamPartLoBitCount; + } + } else { + /* + Getting here means there are no bits set on the cache line. This is a less optimal case because we just wasted a call + to drflac__clz() and we need to reload the cache. + */ + for (;;) { + if (bs->nextL2Line < DRFLAC_CACHE_L2_LINE_COUNT(bs)) { + #ifndef DR_FLAC_NO_CRC + drflac__update_crc16(bs); + #endif + bs_cache = drflac__be2host__cache_line(bs->cacheL2[bs->nextL2Line++]); + bs_consumedBits = 0; + #ifndef DR_FLAC_NO_CRC + bs->crc16Cache = bs_cache; + #endif + } else { + /* Slow path. We need to fetch more data from the client. */ + if (!drflac__reload_cache(bs)) { + return DRFLAC_FALSE; + } + + bs_cache = bs->cache; + bs_consumedBits = bs->consumedBits; + } + + lzcount = drflac__clz(bs_cache); + if (lzcount < sizeof(bs_cache)*8) { + break; + } + } + + goto extract_rice_param_part; + } + + /* Make sure the cache is restored at the end of it all. */ + bs->cache = bs_cache; + bs->consumedBits = bs_consumedBits; + + return DRFLAC_TRUE; +} + + +static drflac_bool32 drflac__decode_samples_with_residual__rice__scalar_zeroorder(drflac_bs* bs, drflac_uint32 bitsPerSample, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 order, drflac_int32 shift, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + drflac_uint32 t[2] = {0x00000000, 0xFFFFFFFF}; + drflac_uint32 zeroCountPart0; + drflac_uint32 riceParamPart0; + drflac_uint32 riceParamMask; + drflac_uint32 i; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pSamplesOut != NULL); + + (void)bitsPerSample; + (void)order; + (void)shift; + (void)coefficients; + + riceParamMask = (drflac_uint32)~((~0UL) << riceParam); + + i = 0; + while (i < count) { + /* Rice extraction. */ + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart0, &riceParamPart0)) { + return DRFLAC_FALSE; + } + + /* Rice reconstruction. */ + riceParamPart0 &= riceParamMask; + riceParamPart0 |= (zeroCountPart0 << riceParam); + riceParamPart0 = (riceParamPart0 >> 1) ^ t[riceParamPart0 & 0x01]; + + pSamplesOut[i] = riceParamPart0; + + i += 1; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_samples_with_residual__rice__scalar(drflac_bs* bs, drflac_uint32 bitsPerSample, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 lpcOrder, drflac_int32 lpcShift, drflac_uint32 lpcPrecision, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + drflac_uint32 t[2] = {0x00000000, 0xFFFFFFFF}; + drflac_uint32 zeroCountPart0 = 0; + drflac_uint32 zeroCountPart1 = 0; + drflac_uint32 zeroCountPart2 = 0; + drflac_uint32 zeroCountPart3 = 0; + drflac_uint32 riceParamPart0 = 0; + drflac_uint32 riceParamPart1 = 0; + drflac_uint32 riceParamPart2 = 0; + drflac_uint32 riceParamPart3 = 0; + drflac_uint32 riceParamMask; + const drflac_int32* pSamplesOutEnd; + drflac_uint32 i; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pSamplesOut != NULL); + + if (lpcOrder == 0) { + return drflac__decode_samples_with_residual__rice__scalar_zeroorder(bs, bitsPerSample, count, riceParam, lpcOrder, lpcShift, coefficients, pSamplesOut); + } + + riceParamMask = (drflac_uint32)~((~0UL) << riceParam); + pSamplesOutEnd = pSamplesOut + (count & ~3); + + if (drflac__use_64_bit_prediction(bitsPerSample, lpcOrder, lpcPrecision)) { + while (pSamplesOut < pSamplesOutEnd) { + /* + Rice extraction. It's faster to do this one at a time against local variables than it is to use the x4 version + against an array. Not sure why, but perhaps it's making more efficient use of registers? + */ + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart0, &riceParamPart0) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart1, &riceParamPart1) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart2, &riceParamPart2) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart3, &riceParamPart3)) { + return DRFLAC_FALSE; + } + + riceParamPart0 &= riceParamMask; + riceParamPart1 &= riceParamMask; + riceParamPart2 &= riceParamMask; + riceParamPart3 &= riceParamMask; + + riceParamPart0 |= (zeroCountPart0 << riceParam); + riceParamPart1 |= (zeroCountPart1 << riceParam); + riceParamPart2 |= (zeroCountPart2 << riceParam); + riceParamPart3 |= (zeroCountPart3 << riceParam); + + riceParamPart0 = (riceParamPart0 >> 1) ^ t[riceParamPart0 & 0x01]; + riceParamPart1 = (riceParamPart1 >> 1) ^ t[riceParamPart1 & 0x01]; + riceParamPart2 = (riceParamPart2 >> 1) ^ t[riceParamPart2 & 0x01]; + riceParamPart3 = (riceParamPart3 >> 1) ^ t[riceParamPart3 & 0x01]; + + pSamplesOut[0] = riceParamPart0 + drflac__calculate_prediction_64(lpcOrder, lpcShift, coefficients, pSamplesOut + 0); + pSamplesOut[1] = riceParamPart1 + drflac__calculate_prediction_64(lpcOrder, lpcShift, coefficients, pSamplesOut + 1); + pSamplesOut[2] = riceParamPart2 + drflac__calculate_prediction_64(lpcOrder, lpcShift, coefficients, pSamplesOut + 2); + pSamplesOut[3] = riceParamPart3 + drflac__calculate_prediction_64(lpcOrder, lpcShift, coefficients, pSamplesOut + 3); + + pSamplesOut += 4; + } + } else { + while (pSamplesOut < pSamplesOutEnd) { + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart0, &riceParamPart0) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart1, &riceParamPart1) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart2, &riceParamPart2) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart3, &riceParamPart3)) { + return DRFLAC_FALSE; + } + + riceParamPart0 &= riceParamMask; + riceParamPart1 &= riceParamMask; + riceParamPart2 &= riceParamMask; + riceParamPart3 &= riceParamMask; + + riceParamPart0 |= (zeroCountPart0 << riceParam); + riceParamPart1 |= (zeroCountPart1 << riceParam); + riceParamPart2 |= (zeroCountPart2 << riceParam); + riceParamPart3 |= (zeroCountPart3 << riceParam); + + riceParamPart0 = (riceParamPart0 >> 1) ^ t[riceParamPart0 & 0x01]; + riceParamPart1 = (riceParamPart1 >> 1) ^ t[riceParamPart1 & 0x01]; + riceParamPart2 = (riceParamPart2 >> 1) ^ t[riceParamPart2 & 0x01]; + riceParamPart3 = (riceParamPart3 >> 1) ^ t[riceParamPart3 & 0x01]; + + pSamplesOut[0] = riceParamPart0 + drflac__calculate_prediction_32(lpcOrder, lpcShift, coefficients, pSamplesOut + 0); + pSamplesOut[1] = riceParamPart1 + drflac__calculate_prediction_32(lpcOrder, lpcShift, coefficients, pSamplesOut + 1); + pSamplesOut[2] = riceParamPart2 + drflac__calculate_prediction_32(lpcOrder, lpcShift, coefficients, pSamplesOut + 2); + pSamplesOut[3] = riceParamPart3 + drflac__calculate_prediction_32(lpcOrder, lpcShift, coefficients, pSamplesOut + 3); + + pSamplesOut += 4; + } + } + + i = (count & ~3); + while (i < count) { + /* Rice extraction. */ + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountPart0, &riceParamPart0)) { + return DRFLAC_FALSE; + } + + /* Rice reconstruction. */ + riceParamPart0 &= riceParamMask; + riceParamPart0 |= (zeroCountPart0 << riceParam); + riceParamPart0 = (riceParamPart0 >> 1) ^ t[riceParamPart0 & 0x01]; + /*riceParamPart0 = (riceParamPart0 >> 1) ^ (~(riceParamPart0 & 0x01) + 1);*/ + + /* Sample reconstruction. */ + if (drflac__use_64_bit_prediction(bitsPerSample, lpcOrder, lpcPrecision)) { + pSamplesOut[0] = riceParamPart0 + drflac__calculate_prediction_64(lpcOrder, lpcShift, coefficients, pSamplesOut + 0); + } else { + pSamplesOut[0] = riceParamPart0 + drflac__calculate_prediction_32(lpcOrder, lpcShift, coefficients, pSamplesOut + 0); + } + + i += 1; + pSamplesOut += 1; + } + + return DRFLAC_TRUE; +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE __m128i drflac__mm_packs_interleaved_epi32(__m128i a, __m128i b) +{ + __m128i r; + + /* Pack. */ + r = _mm_packs_epi32(a, b); + + /* a3a2 a1a0 b3b2 b1b0 -> a3a2 b3b2 a1a0 b1b0 */ + r = _mm_shuffle_epi32(r, _MM_SHUFFLE(3, 1, 2, 0)); + + /* a3a2 b3b2 a1a0 b1b0 -> a3b3 a2b2 a1b1 a0b0 */ + r = _mm_shufflehi_epi16(r, _MM_SHUFFLE(3, 1, 2, 0)); + r = _mm_shufflelo_epi16(r, _MM_SHUFFLE(3, 1, 2, 0)); + + return r; +} +#endif + +#if defined(DRFLAC_SUPPORT_SSE41) +static DRFLAC_INLINE __m128i drflac__mm_not_si128(__m128i a) +{ + return _mm_xor_si128(a, _mm_cmpeq_epi32(_mm_setzero_si128(), _mm_setzero_si128())); +} + +static DRFLAC_INLINE __m128i drflac__mm_hadd_epi32(__m128i x) +{ + __m128i x64 = _mm_add_epi32(x, _mm_shuffle_epi32(x, _MM_SHUFFLE(1, 0, 3, 2))); + __m128i x32 = _mm_shufflelo_epi16(x64, _MM_SHUFFLE(1, 0, 3, 2)); + return _mm_add_epi32(x64, x32); +} + +static DRFLAC_INLINE __m128i drflac__mm_hadd_epi64(__m128i x) +{ + return _mm_add_epi64(x, _mm_shuffle_epi32(x, _MM_SHUFFLE(1, 0, 3, 2))); +} + +static DRFLAC_INLINE __m128i drflac__mm_srai_epi64(__m128i x, int count) +{ + /* + To simplify this we are assuming count < 32. This restriction allows us to work on a low side and a high side. The low side + is shifted with zero bits, whereas the right side is shifted with sign bits. + */ + __m128i lo = _mm_srli_epi64(x, count); + __m128i hi = _mm_srai_epi32(x, count); + + hi = _mm_and_si128(hi, _mm_set_epi32(0xFFFFFFFF, 0, 0xFFFFFFFF, 0)); /* The high part needs to have the low part cleared. */ + + return _mm_or_si128(lo, hi); +} + +static drflac_bool32 drflac__decode_samples_with_residual__rice__sse41_32(drflac_bs* bs, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 order, drflac_int32 shift, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + int i; + drflac_uint32 riceParamMask; + drflac_int32* pDecodedSamples = pSamplesOut; + drflac_int32* pDecodedSamplesEnd = pSamplesOut + (count & ~3); + drflac_uint32 zeroCountParts0 = 0; + drflac_uint32 zeroCountParts1 = 0; + drflac_uint32 zeroCountParts2 = 0; + drflac_uint32 zeroCountParts3 = 0; + drflac_uint32 riceParamParts0 = 0; + drflac_uint32 riceParamParts1 = 0; + drflac_uint32 riceParamParts2 = 0; + drflac_uint32 riceParamParts3 = 0; + __m128i coefficients128_0; + __m128i coefficients128_4; + __m128i coefficients128_8; + __m128i samples128_0; + __m128i samples128_4; + __m128i samples128_8; + __m128i riceParamMask128; + + const drflac_uint32 t[2] = {0x00000000, 0xFFFFFFFF}; + + riceParamMask = (drflac_uint32)~((~0UL) << riceParam); + riceParamMask128 = _mm_set1_epi32(riceParamMask); + + /* Pre-load. */ + coefficients128_0 = _mm_setzero_si128(); + coefficients128_4 = _mm_setzero_si128(); + coefficients128_8 = _mm_setzero_si128(); + + samples128_0 = _mm_setzero_si128(); + samples128_4 = _mm_setzero_si128(); + samples128_8 = _mm_setzero_si128(); + + /* + Pre-loading the coefficients and prior samples is annoying because we need to ensure we don't try reading more than + what's available in the input buffers. It would be convenient to use a fall-through switch to do this, but this results + in strict aliasing warnings with GCC. To work around this I'm just doing something hacky. This feels a bit convoluted + so I think there's opportunity for this to be simplified. + */ +#if 1 + { + int runningOrder = order; + + /* 0 - 3. */ + if (runningOrder >= 4) { + coefficients128_0 = _mm_loadu_si128((const __m128i*)(coefficients + 0)); + samples128_0 = _mm_loadu_si128((const __m128i*)(pSamplesOut - 4)); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: coefficients128_0 = _mm_set_epi32(0, coefficients[2], coefficients[1], coefficients[0]); samples128_0 = _mm_set_epi32(pSamplesOut[-1], pSamplesOut[-2], pSamplesOut[-3], 0); break; + case 2: coefficients128_0 = _mm_set_epi32(0, 0, coefficients[1], coefficients[0]); samples128_0 = _mm_set_epi32(pSamplesOut[-1], pSamplesOut[-2], 0, 0); break; + case 1: coefficients128_0 = _mm_set_epi32(0, 0, 0, coefficients[0]); samples128_0 = _mm_set_epi32(pSamplesOut[-1], 0, 0, 0); break; + } + runningOrder = 0; + } + + /* 4 - 7 */ + if (runningOrder >= 4) { + coefficients128_4 = _mm_loadu_si128((const __m128i*)(coefficients + 4)); + samples128_4 = _mm_loadu_si128((const __m128i*)(pSamplesOut - 8)); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: coefficients128_4 = _mm_set_epi32(0, coefficients[6], coefficients[5], coefficients[4]); samples128_4 = _mm_set_epi32(pSamplesOut[-5], pSamplesOut[-6], pSamplesOut[-7], 0); break; + case 2: coefficients128_4 = _mm_set_epi32(0, 0, coefficients[5], coefficients[4]); samples128_4 = _mm_set_epi32(pSamplesOut[-5], pSamplesOut[-6], 0, 0); break; + case 1: coefficients128_4 = _mm_set_epi32(0, 0, 0, coefficients[4]); samples128_4 = _mm_set_epi32(pSamplesOut[-5], 0, 0, 0); break; + } + runningOrder = 0; + } + + /* 8 - 11 */ + if (runningOrder == 4) { + coefficients128_8 = _mm_loadu_si128((const __m128i*)(coefficients + 8)); + samples128_8 = _mm_loadu_si128((const __m128i*)(pSamplesOut - 12)); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: coefficients128_8 = _mm_set_epi32(0, coefficients[10], coefficients[9], coefficients[8]); samples128_8 = _mm_set_epi32(pSamplesOut[-9], pSamplesOut[-10], pSamplesOut[-11], 0); break; + case 2: coefficients128_8 = _mm_set_epi32(0, 0, coefficients[9], coefficients[8]); samples128_8 = _mm_set_epi32(pSamplesOut[-9], pSamplesOut[-10], 0, 0); break; + case 1: coefficients128_8 = _mm_set_epi32(0, 0, 0, coefficients[8]); samples128_8 = _mm_set_epi32(pSamplesOut[-9], 0, 0, 0); break; + } + runningOrder = 0; + } + + /* Coefficients need to be shuffled for our streaming algorithm below to work. Samples are already in the correct order from the loading routine above. */ + coefficients128_0 = _mm_shuffle_epi32(coefficients128_0, _MM_SHUFFLE(0, 1, 2, 3)); + coefficients128_4 = _mm_shuffle_epi32(coefficients128_4, _MM_SHUFFLE(0, 1, 2, 3)); + coefficients128_8 = _mm_shuffle_epi32(coefficients128_8, _MM_SHUFFLE(0, 1, 2, 3)); + } +#else + /* This causes strict-aliasing warnings with GCC. */ + switch (order) + { + case 12: ((drflac_int32*)&coefficients128_8)[0] = coefficients[11]; ((drflac_int32*)&samples128_8)[0] = pDecodedSamples[-12]; + case 11: ((drflac_int32*)&coefficients128_8)[1] = coefficients[10]; ((drflac_int32*)&samples128_8)[1] = pDecodedSamples[-11]; + case 10: ((drflac_int32*)&coefficients128_8)[2] = coefficients[ 9]; ((drflac_int32*)&samples128_8)[2] = pDecodedSamples[-10]; + case 9: ((drflac_int32*)&coefficients128_8)[3] = coefficients[ 8]; ((drflac_int32*)&samples128_8)[3] = pDecodedSamples[- 9]; + case 8: ((drflac_int32*)&coefficients128_4)[0] = coefficients[ 7]; ((drflac_int32*)&samples128_4)[0] = pDecodedSamples[- 8]; + case 7: ((drflac_int32*)&coefficients128_4)[1] = coefficients[ 6]; ((drflac_int32*)&samples128_4)[1] = pDecodedSamples[- 7]; + case 6: ((drflac_int32*)&coefficients128_4)[2] = coefficients[ 5]; ((drflac_int32*)&samples128_4)[2] = pDecodedSamples[- 6]; + case 5: ((drflac_int32*)&coefficients128_4)[3] = coefficients[ 4]; ((drflac_int32*)&samples128_4)[3] = pDecodedSamples[- 5]; + case 4: ((drflac_int32*)&coefficients128_0)[0] = coefficients[ 3]; ((drflac_int32*)&samples128_0)[0] = pDecodedSamples[- 4]; + case 3: ((drflac_int32*)&coefficients128_0)[1] = coefficients[ 2]; ((drflac_int32*)&samples128_0)[1] = pDecodedSamples[- 3]; + case 2: ((drflac_int32*)&coefficients128_0)[2] = coefficients[ 1]; ((drflac_int32*)&samples128_0)[2] = pDecodedSamples[- 2]; + case 1: ((drflac_int32*)&coefficients128_0)[3] = coefficients[ 0]; ((drflac_int32*)&samples128_0)[3] = pDecodedSamples[- 1]; + } +#endif + + /* For this version we are doing one sample at a time. */ + while (pDecodedSamples < pDecodedSamplesEnd) { + __m128i prediction128; + __m128i zeroCountPart128; + __m128i riceParamPart128; + + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts0, &riceParamParts0) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts1, &riceParamParts1) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts2, &riceParamParts2) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts3, &riceParamParts3)) { + return DRFLAC_FALSE; + } + + zeroCountPart128 = _mm_set_epi32(zeroCountParts3, zeroCountParts2, zeroCountParts1, zeroCountParts0); + riceParamPart128 = _mm_set_epi32(riceParamParts3, riceParamParts2, riceParamParts1, riceParamParts0); + + riceParamPart128 = _mm_and_si128(riceParamPart128, riceParamMask128); + riceParamPart128 = _mm_or_si128(riceParamPart128, _mm_slli_epi32(zeroCountPart128, riceParam)); + riceParamPart128 = _mm_xor_si128(_mm_srli_epi32(riceParamPart128, 1), _mm_add_epi32(drflac__mm_not_si128(_mm_and_si128(riceParamPart128, _mm_set1_epi32(0x01))), _mm_set1_epi32(0x01))); /* <-- SSE2 compatible */ + /*riceParamPart128 = _mm_xor_si128(_mm_srli_epi32(riceParamPart128, 1), _mm_mullo_epi32(_mm_and_si128(riceParamPart128, _mm_set1_epi32(0x01)), _mm_set1_epi32(0xFFFFFFFF)));*/ /* <-- Only supported from SSE4.1 and is slower in my testing... */ + + if (order <= 4) { + for (i = 0; i < 4; i += 1) { + prediction128 = _mm_mullo_epi32(coefficients128_0, samples128_0); + + /* Horizontal add and shift. */ + prediction128 = drflac__mm_hadd_epi32(prediction128); + prediction128 = _mm_srai_epi32(prediction128, shift); + prediction128 = _mm_add_epi32(riceParamPart128, prediction128); + + samples128_0 = _mm_alignr_epi8(prediction128, samples128_0, 4); + riceParamPart128 = _mm_alignr_epi8(_mm_setzero_si128(), riceParamPart128, 4); + } + } else if (order <= 8) { + for (i = 0; i < 4; i += 1) { + prediction128 = _mm_mullo_epi32(coefficients128_4, samples128_4); + prediction128 = _mm_add_epi32(prediction128, _mm_mullo_epi32(coefficients128_0, samples128_0)); + + /* Horizontal add and shift. */ + prediction128 = drflac__mm_hadd_epi32(prediction128); + prediction128 = _mm_srai_epi32(prediction128, shift); + prediction128 = _mm_add_epi32(riceParamPart128, prediction128); + + samples128_4 = _mm_alignr_epi8(samples128_0, samples128_4, 4); + samples128_0 = _mm_alignr_epi8(prediction128, samples128_0, 4); + riceParamPart128 = _mm_alignr_epi8(_mm_setzero_si128(), riceParamPart128, 4); + } + } else { + for (i = 0; i < 4; i += 1) { + prediction128 = _mm_mullo_epi32(coefficients128_8, samples128_8); + prediction128 = _mm_add_epi32(prediction128, _mm_mullo_epi32(coefficients128_4, samples128_4)); + prediction128 = _mm_add_epi32(prediction128, _mm_mullo_epi32(coefficients128_0, samples128_0)); + + /* Horizontal add and shift. */ + prediction128 = drflac__mm_hadd_epi32(prediction128); + prediction128 = _mm_srai_epi32(prediction128, shift); + prediction128 = _mm_add_epi32(riceParamPart128, prediction128); + + samples128_8 = _mm_alignr_epi8(samples128_4, samples128_8, 4); + samples128_4 = _mm_alignr_epi8(samples128_0, samples128_4, 4); + samples128_0 = _mm_alignr_epi8(prediction128, samples128_0, 4); + riceParamPart128 = _mm_alignr_epi8(_mm_setzero_si128(), riceParamPart128, 4); + } + } + + /* We store samples in groups of 4. */ + _mm_storeu_si128((__m128i*)pDecodedSamples, samples128_0); + pDecodedSamples += 4; + } + + /* Make sure we process the last few samples. */ + i = (count & ~3); + while (i < (int)count) { + /* Rice extraction. */ + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts0, &riceParamParts0)) { + return DRFLAC_FALSE; + } + + /* Rice reconstruction. */ + riceParamParts0 &= riceParamMask; + riceParamParts0 |= (zeroCountParts0 << riceParam); + riceParamParts0 = (riceParamParts0 >> 1) ^ t[riceParamParts0 & 0x01]; + + /* Sample reconstruction. */ + pDecodedSamples[0] = riceParamParts0 + drflac__calculate_prediction_32(order, shift, coefficients, pDecodedSamples); + + i += 1; + pDecodedSamples += 1; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_samples_with_residual__rice__sse41_64(drflac_bs* bs, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 order, drflac_int32 shift, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + int i; + drflac_uint32 riceParamMask; + drflac_int32* pDecodedSamples = pSamplesOut; + drflac_int32* pDecodedSamplesEnd = pSamplesOut + (count & ~3); + drflac_uint32 zeroCountParts0 = 0; + drflac_uint32 zeroCountParts1 = 0; + drflac_uint32 zeroCountParts2 = 0; + drflac_uint32 zeroCountParts3 = 0; + drflac_uint32 riceParamParts0 = 0; + drflac_uint32 riceParamParts1 = 0; + drflac_uint32 riceParamParts2 = 0; + drflac_uint32 riceParamParts3 = 0; + __m128i coefficients128_0; + __m128i coefficients128_4; + __m128i coefficients128_8; + __m128i samples128_0; + __m128i samples128_4; + __m128i samples128_8; + __m128i prediction128; + __m128i riceParamMask128; + + const drflac_uint32 t[2] = {0x00000000, 0xFFFFFFFF}; + + DRFLAC_ASSERT(order <= 12); + + riceParamMask = (drflac_uint32)~((~0UL) << riceParam); + riceParamMask128 = _mm_set1_epi32(riceParamMask); + + prediction128 = _mm_setzero_si128(); + + /* Pre-load. */ + coefficients128_0 = _mm_setzero_si128(); + coefficients128_4 = _mm_setzero_si128(); + coefficients128_8 = _mm_setzero_si128(); + + samples128_0 = _mm_setzero_si128(); + samples128_4 = _mm_setzero_si128(); + samples128_8 = _mm_setzero_si128(); + +#if 1 + { + int runningOrder = order; + + /* 0 - 3. */ + if (runningOrder >= 4) { + coefficients128_0 = _mm_loadu_si128((const __m128i*)(coefficients + 0)); + samples128_0 = _mm_loadu_si128((const __m128i*)(pSamplesOut - 4)); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: coefficients128_0 = _mm_set_epi32(0, coefficients[2], coefficients[1], coefficients[0]); samples128_0 = _mm_set_epi32(pSamplesOut[-1], pSamplesOut[-2], pSamplesOut[-3], 0); break; + case 2: coefficients128_0 = _mm_set_epi32(0, 0, coefficients[1], coefficients[0]); samples128_0 = _mm_set_epi32(pSamplesOut[-1], pSamplesOut[-2], 0, 0); break; + case 1: coefficients128_0 = _mm_set_epi32(0, 0, 0, coefficients[0]); samples128_0 = _mm_set_epi32(pSamplesOut[-1], 0, 0, 0); break; + } + runningOrder = 0; + } + + /* 4 - 7 */ + if (runningOrder >= 4) { + coefficients128_4 = _mm_loadu_si128((const __m128i*)(coefficients + 4)); + samples128_4 = _mm_loadu_si128((const __m128i*)(pSamplesOut - 8)); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: coefficients128_4 = _mm_set_epi32(0, coefficients[6], coefficients[5], coefficients[4]); samples128_4 = _mm_set_epi32(pSamplesOut[-5], pSamplesOut[-6], pSamplesOut[-7], 0); break; + case 2: coefficients128_4 = _mm_set_epi32(0, 0, coefficients[5], coefficients[4]); samples128_4 = _mm_set_epi32(pSamplesOut[-5], pSamplesOut[-6], 0, 0); break; + case 1: coefficients128_4 = _mm_set_epi32(0, 0, 0, coefficients[4]); samples128_4 = _mm_set_epi32(pSamplesOut[-5], 0, 0, 0); break; + } + runningOrder = 0; + } + + /* 8 - 11 */ + if (runningOrder == 4) { + coefficients128_8 = _mm_loadu_si128((const __m128i*)(coefficients + 8)); + samples128_8 = _mm_loadu_si128((const __m128i*)(pSamplesOut - 12)); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: coefficients128_8 = _mm_set_epi32(0, coefficients[10], coefficients[9], coefficients[8]); samples128_8 = _mm_set_epi32(pSamplesOut[-9], pSamplesOut[-10], pSamplesOut[-11], 0); break; + case 2: coefficients128_8 = _mm_set_epi32(0, 0, coefficients[9], coefficients[8]); samples128_8 = _mm_set_epi32(pSamplesOut[-9], pSamplesOut[-10], 0, 0); break; + case 1: coefficients128_8 = _mm_set_epi32(0, 0, 0, coefficients[8]); samples128_8 = _mm_set_epi32(pSamplesOut[-9], 0, 0, 0); break; + } + runningOrder = 0; + } + + /* Coefficients need to be shuffled for our streaming algorithm below to work. Samples are already in the correct order from the loading routine above. */ + coefficients128_0 = _mm_shuffle_epi32(coefficients128_0, _MM_SHUFFLE(0, 1, 2, 3)); + coefficients128_4 = _mm_shuffle_epi32(coefficients128_4, _MM_SHUFFLE(0, 1, 2, 3)); + coefficients128_8 = _mm_shuffle_epi32(coefficients128_8, _MM_SHUFFLE(0, 1, 2, 3)); + } +#else + switch (order) + { + case 12: ((drflac_int32*)&coefficients128_8)[0] = coefficients[11]; ((drflac_int32*)&samples128_8)[0] = pDecodedSamples[-12]; + case 11: ((drflac_int32*)&coefficients128_8)[1] = coefficients[10]; ((drflac_int32*)&samples128_8)[1] = pDecodedSamples[-11]; + case 10: ((drflac_int32*)&coefficients128_8)[2] = coefficients[ 9]; ((drflac_int32*)&samples128_8)[2] = pDecodedSamples[-10]; + case 9: ((drflac_int32*)&coefficients128_8)[3] = coefficients[ 8]; ((drflac_int32*)&samples128_8)[3] = pDecodedSamples[- 9]; + case 8: ((drflac_int32*)&coefficients128_4)[0] = coefficients[ 7]; ((drflac_int32*)&samples128_4)[0] = pDecodedSamples[- 8]; + case 7: ((drflac_int32*)&coefficients128_4)[1] = coefficients[ 6]; ((drflac_int32*)&samples128_4)[1] = pDecodedSamples[- 7]; + case 6: ((drflac_int32*)&coefficients128_4)[2] = coefficients[ 5]; ((drflac_int32*)&samples128_4)[2] = pDecodedSamples[- 6]; + case 5: ((drflac_int32*)&coefficients128_4)[3] = coefficients[ 4]; ((drflac_int32*)&samples128_4)[3] = pDecodedSamples[- 5]; + case 4: ((drflac_int32*)&coefficients128_0)[0] = coefficients[ 3]; ((drflac_int32*)&samples128_0)[0] = pDecodedSamples[- 4]; + case 3: ((drflac_int32*)&coefficients128_0)[1] = coefficients[ 2]; ((drflac_int32*)&samples128_0)[1] = pDecodedSamples[- 3]; + case 2: ((drflac_int32*)&coefficients128_0)[2] = coefficients[ 1]; ((drflac_int32*)&samples128_0)[2] = pDecodedSamples[- 2]; + case 1: ((drflac_int32*)&coefficients128_0)[3] = coefficients[ 0]; ((drflac_int32*)&samples128_0)[3] = pDecodedSamples[- 1]; + } +#endif + + /* For this version we are doing one sample at a time. */ + while (pDecodedSamples < pDecodedSamplesEnd) { + __m128i zeroCountPart128; + __m128i riceParamPart128; + + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts0, &riceParamParts0) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts1, &riceParamParts1) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts2, &riceParamParts2) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts3, &riceParamParts3)) { + return DRFLAC_FALSE; + } + + zeroCountPart128 = _mm_set_epi32(zeroCountParts3, zeroCountParts2, zeroCountParts1, zeroCountParts0); + riceParamPart128 = _mm_set_epi32(riceParamParts3, riceParamParts2, riceParamParts1, riceParamParts0); + + riceParamPart128 = _mm_and_si128(riceParamPart128, riceParamMask128); + riceParamPart128 = _mm_or_si128(riceParamPart128, _mm_slli_epi32(zeroCountPart128, riceParam)); + riceParamPart128 = _mm_xor_si128(_mm_srli_epi32(riceParamPart128, 1), _mm_add_epi32(drflac__mm_not_si128(_mm_and_si128(riceParamPart128, _mm_set1_epi32(1))), _mm_set1_epi32(1))); + + for (i = 0; i < 4; i += 1) { + prediction128 = _mm_xor_si128(prediction128, prediction128); /* Reset to 0. */ + + switch (order) + { + case 12: + case 11: prediction128 = _mm_add_epi64(prediction128, _mm_mul_epi32(_mm_shuffle_epi32(coefficients128_8, _MM_SHUFFLE(1, 1, 0, 0)), _mm_shuffle_epi32(samples128_8, _MM_SHUFFLE(1, 1, 0, 0)))); + case 10: + case 9: prediction128 = _mm_add_epi64(prediction128, _mm_mul_epi32(_mm_shuffle_epi32(coefficients128_8, _MM_SHUFFLE(3, 3, 2, 2)), _mm_shuffle_epi32(samples128_8, _MM_SHUFFLE(3, 3, 2, 2)))); + case 8: + case 7: prediction128 = _mm_add_epi64(prediction128, _mm_mul_epi32(_mm_shuffle_epi32(coefficients128_4, _MM_SHUFFLE(1, 1, 0, 0)), _mm_shuffle_epi32(samples128_4, _MM_SHUFFLE(1, 1, 0, 0)))); + case 6: + case 5: prediction128 = _mm_add_epi64(prediction128, _mm_mul_epi32(_mm_shuffle_epi32(coefficients128_4, _MM_SHUFFLE(3, 3, 2, 2)), _mm_shuffle_epi32(samples128_4, _MM_SHUFFLE(3, 3, 2, 2)))); + case 4: + case 3: prediction128 = _mm_add_epi64(prediction128, _mm_mul_epi32(_mm_shuffle_epi32(coefficients128_0, _MM_SHUFFLE(1, 1, 0, 0)), _mm_shuffle_epi32(samples128_0, _MM_SHUFFLE(1, 1, 0, 0)))); + case 2: + case 1: prediction128 = _mm_add_epi64(prediction128, _mm_mul_epi32(_mm_shuffle_epi32(coefficients128_0, _MM_SHUFFLE(3, 3, 2, 2)), _mm_shuffle_epi32(samples128_0, _MM_SHUFFLE(3, 3, 2, 2)))); + } + + /* Horizontal add and shift. */ + prediction128 = drflac__mm_hadd_epi64(prediction128); + prediction128 = drflac__mm_srai_epi64(prediction128, shift); + prediction128 = _mm_add_epi32(riceParamPart128, prediction128); + + /* Our value should be sitting in prediction128[0]. We need to combine this with our SSE samples. */ + samples128_8 = _mm_alignr_epi8(samples128_4, samples128_8, 4); + samples128_4 = _mm_alignr_epi8(samples128_0, samples128_4, 4); + samples128_0 = _mm_alignr_epi8(prediction128, samples128_0, 4); + + /* Slide our rice parameter down so that the value in position 0 contains the next one to process. */ + riceParamPart128 = _mm_alignr_epi8(_mm_setzero_si128(), riceParamPart128, 4); + } + + /* We store samples in groups of 4. */ + _mm_storeu_si128((__m128i*)pDecodedSamples, samples128_0); + pDecodedSamples += 4; + } + + /* Make sure we process the last few samples. */ + i = (count & ~3); + while (i < (int)count) { + /* Rice extraction. */ + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts0, &riceParamParts0)) { + return DRFLAC_FALSE; + } + + /* Rice reconstruction. */ + riceParamParts0 &= riceParamMask; + riceParamParts0 |= (zeroCountParts0 << riceParam); + riceParamParts0 = (riceParamParts0 >> 1) ^ t[riceParamParts0 & 0x01]; + + /* Sample reconstruction. */ + pDecodedSamples[0] = riceParamParts0 + drflac__calculate_prediction_64(order, shift, coefficients, pDecodedSamples); + + i += 1; + pDecodedSamples += 1; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_samples_with_residual__rice__sse41(drflac_bs* bs, drflac_uint32 bitsPerSample, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 lpcOrder, drflac_int32 lpcShift, drflac_uint32 lpcPrecision, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pSamplesOut != NULL); + + /* In my testing the order is rarely > 12, so in this case I'm going to simplify the SSE implementation by only handling order <= 12. */ + if (lpcOrder > 0 && lpcOrder <= 12) { + if (drflac__use_64_bit_prediction(bitsPerSample, lpcOrder, lpcPrecision)) { + return drflac__decode_samples_with_residual__rice__sse41_64(bs, count, riceParam, lpcOrder, lpcShift, coefficients, pSamplesOut); + } else { + return drflac__decode_samples_with_residual__rice__sse41_32(bs, count, riceParam, lpcOrder, lpcShift, coefficients, pSamplesOut); + } + } else { + return drflac__decode_samples_with_residual__rice__scalar(bs, bitsPerSample, count, riceParam, lpcOrder, lpcShift, lpcPrecision, coefficients, pSamplesOut); + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac__vst2q_s32(drflac_int32* p, int32x4x2_t x) +{ + vst1q_s32(p+0, x.val[0]); + vst1q_s32(p+4, x.val[1]); +} + +static DRFLAC_INLINE void drflac__vst2q_u32(drflac_uint32* p, uint32x4x2_t x) +{ + vst1q_u32(p+0, x.val[0]); + vst1q_u32(p+4, x.val[1]); +} + +static DRFLAC_INLINE void drflac__vst2q_f32(float* p, float32x4x2_t x) +{ + vst1q_f32(p+0, x.val[0]); + vst1q_f32(p+4, x.val[1]); +} + +static DRFLAC_INLINE void drflac__vst2q_s16(drflac_int16* p, int16x4x2_t x) +{ + vst1q_s16(p, vcombine_s16(x.val[0], x.val[1])); +} + +static DRFLAC_INLINE void drflac__vst2q_u16(drflac_uint16* p, uint16x4x2_t x) +{ + vst1q_u16(p, vcombine_u16(x.val[0], x.val[1])); +} + +static DRFLAC_INLINE int32x4_t drflac__vdupq_n_s32x4(drflac_int32 x3, drflac_int32 x2, drflac_int32 x1, drflac_int32 x0) +{ + drflac_int32 x[4]; + x[3] = x3; + x[2] = x2; + x[1] = x1; + x[0] = x0; + return vld1q_s32(x); +} + +static DRFLAC_INLINE int32x4_t drflac__valignrq_s32_1(int32x4_t a, int32x4_t b) +{ + /* Equivalent to SSE's _mm_alignr_epi8(a, b, 4) */ + + /* Reference */ + /*return drflac__vdupq_n_s32x4( + vgetq_lane_s32(a, 0), + vgetq_lane_s32(b, 3), + vgetq_lane_s32(b, 2), + vgetq_lane_s32(b, 1) + );*/ + + return vextq_s32(b, a, 1); +} + +static DRFLAC_INLINE uint32x4_t drflac__valignrq_u32_1(uint32x4_t a, uint32x4_t b) +{ + /* Equivalent to SSE's _mm_alignr_epi8(a, b, 4) */ + + /* Reference */ + /*return drflac__vdupq_n_s32x4( + vgetq_lane_s32(a, 0), + vgetq_lane_s32(b, 3), + vgetq_lane_s32(b, 2), + vgetq_lane_s32(b, 1) + );*/ + + return vextq_u32(b, a, 1); +} + +static DRFLAC_INLINE int32x2_t drflac__vhaddq_s32(int32x4_t x) +{ + /* The sum must end up in position 0. */ + + /* Reference */ + /*return vdupq_n_s32( + vgetq_lane_s32(x, 3) + + vgetq_lane_s32(x, 2) + + vgetq_lane_s32(x, 1) + + vgetq_lane_s32(x, 0) + );*/ + + int32x2_t r = vadd_s32(vget_high_s32(x), vget_low_s32(x)); + return vpadd_s32(r, r); +} + +static DRFLAC_INLINE int64x1_t drflac__vhaddq_s64(int64x2_t x) +{ + return vadd_s64(vget_high_s64(x), vget_low_s64(x)); +} + +static DRFLAC_INLINE int32x4_t drflac__vrevq_s32(int32x4_t x) +{ + /* Reference */ + /*return drflac__vdupq_n_s32x4( + vgetq_lane_s32(x, 0), + vgetq_lane_s32(x, 1), + vgetq_lane_s32(x, 2), + vgetq_lane_s32(x, 3) + );*/ + + return vrev64q_s32(vcombine_s32(vget_high_s32(x), vget_low_s32(x))); +} + +static DRFLAC_INLINE int32x4_t drflac__vnotq_s32(int32x4_t x) +{ + return veorq_s32(x, vdupq_n_s32(0xFFFFFFFF)); +} + +static DRFLAC_INLINE uint32x4_t drflac__vnotq_u32(uint32x4_t x) +{ + return veorq_u32(x, vdupq_n_u32(0xFFFFFFFF)); +} + +static drflac_bool32 drflac__decode_samples_with_residual__rice__neon_32(drflac_bs* bs, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 order, drflac_int32 shift, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + int i; + drflac_uint32 riceParamMask; + drflac_int32* pDecodedSamples = pSamplesOut; + drflac_int32* pDecodedSamplesEnd = pSamplesOut + (count & ~3); + drflac_uint32 zeroCountParts[4]; + drflac_uint32 riceParamParts[4]; + int32x4_t coefficients128_0; + int32x4_t coefficients128_4; + int32x4_t coefficients128_8; + int32x4_t samples128_0; + int32x4_t samples128_4; + int32x4_t samples128_8; + uint32x4_t riceParamMask128; + int32x4_t riceParam128; + int32x2_t shift64; + uint32x4_t one128; + + const drflac_uint32 t[2] = {0x00000000, 0xFFFFFFFF}; + + riceParamMask = (drflac_uint32)~((~0UL) << riceParam); + riceParamMask128 = vdupq_n_u32(riceParamMask); + + riceParam128 = vdupq_n_s32(riceParam); + shift64 = vdup_n_s32(-shift); /* Negate the shift because we'll be doing a variable shift using vshlq_s32(). */ + one128 = vdupq_n_u32(1); + + /* + Pre-loading the coefficients and prior samples is annoying because we need to ensure we don't try reading more than + what's available in the input buffers. It would be conenient to use a fall-through switch to do this, but this results + in strict aliasing warnings with GCC. To work around this I'm just doing something hacky. This feels a bit convoluted + so I think there's opportunity for this to be simplified. + */ + { + int runningOrder = order; + drflac_int32 tempC[4] = {0, 0, 0, 0}; + drflac_int32 tempS[4] = {0, 0, 0, 0}; + + /* 0 - 3. */ + if (runningOrder >= 4) { + coefficients128_0 = vld1q_s32(coefficients + 0); + samples128_0 = vld1q_s32(pSamplesOut - 4); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: tempC[2] = coefficients[2]; tempS[1] = pSamplesOut[-3]; /* fallthrough */ + case 2: tempC[1] = coefficients[1]; tempS[2] = pSamplesOut[-2]; /* fallthrough */ + case 1: tempC[0] = coefficients[0]; tempS[3] = pSamplesOut[-1]; /* fallthrough */ + } + + coefficients128_0 = vld1q_s32(tempC); + samples128_0 = vld1q_s32(tempS); + runningOrder = 0; + } + + /* 4 - 7 */ + if (runningOrder >= 4) { + coefficients128_4 = vld1q_s32(coefficients + 4); + samples128_4 = vld1q_s32(pSamplesOut - 8); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: tempC[2] = coefficients[6]; tempS[1] = pSamplesOut[-7]; /* fallthrough */ + case 2: tempC[1] = coefficients[5]; tempS[2] = pSamplesOut[-6]; /* fallthrough */ + case 1: tempC[0] = coefficients[4]; tempS[3] = pSamplesOut[-5]; /* fallthrough */ + } + + coefficients128_4 = vld1q_s32(tempC); + samples128_4 = vld1q_s32(tempS); + runningOrder = 0; + } + + /* 8 - 11 */ + if (runningOrder == 4) { + coefficients128_8 = vld1q_s32(coefficients + 8); + samples128_8 = vld1q_s32(pSamplesOut - 12); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: tempC[2] = coefficients[10]; tempS[1] = pSamplesOut[-11]; /* fallthrough */ + case 2: tempC[1] = coefficients[ 9]; tempS[2] = pSamplesOut[-10]; /* fallthrough */ + case 1: tempC[0] = coefficients[ 8]; tempS[3] = pSamplesOut[- 9]; /* fallthrough */ + } + + coefficients128_8 = vld1q_s32(tempC); + samples128_8 = vld1q_s32(tempS); + runningOrder = 0; + } + + /* Coefficients need to be shuffled for our streaming algorithm below to work. Samples are already in the correct order from the loading routine above. */ + coefficients128_0 = drflac__vrevq_s32(coefficients128_0); + coefficients128_4 = drflac__vrevq_s32(coefficients128_4); + coefficients128_8 = drflac__vrevq_s32(coefficients128_8); + } + + /* For this version we are doing one sample at a time. */ + while (pDecodedSamples < pDecodedSamplesEnd) { + int32x4_t prediction128; + int32x2_t prediction64; + uint32x4_t zeroCountPart128; + uint32x4_t riceParamPart128; + + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[0], &riceParamParts[0]) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[1], &riceParamParts[1]) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[2], &riceParamParts[2]) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[3], &riceParamParts[3])) { + return DRFLAC_FALSE; + } + + zeroCountPart128 = vld1q_u32(zeroCountParts); + riceParamPart128 = vld1q_u32(riceParamParts); + + riceParamPart128 = vandq_u32(riceParamPart128, riceParamMask128); + riceParamPart128 = vorrq_u32(riceParamPart128, vshlq_u32(zeroCountPart128, riceParam128)); + riceParamPart128 = veorq_u32(vshrq_n_u32(riceParamPart128, 1), vaddq_u32(drflac__vnotq_u32(vandq_u32(riceParamPart128, one128)), one128)); + + if (order <= 4) { + for (i = 0; i < 4; i += 1) { + prediction128 = vmulq_s32(coefficients128_0, samples128_0); + + /* Horizontal add and shift. */ + prediction64 = drflac__vhaddq_s32(prediction128); + prediction64 = vshl_s32(prediction64, shift64); + prediction64 = vadd_s32(prediction64, vget_low_s32(vreinterpretq_s32_u32(riceParamPart128))); + + samples128_0 = drflac__valignrq_s32_1(vcombine_s32(prediction64, vdup_n_s32(0)), samples128_0); + riceParamPart128 = drflac__valignrq_u32_1(vdupq_n_u32(0), riceParamPart128); + } + } else if (order <= 8) { + for (i = 0; i < 4; i += 1) { + prediction128 = vmulq_s32(coefficients128_4, samples128_4); + prediction128 = vmlaq_s32(prediction128, coefficients128_0, samples128_0); + + /* Horizontal add and shift. */ + prediction64 = drflac__vhaddq_s32(prediction128); + prediction64 = vshl_s32(prediction64, shift64); + prediction64 = vadd_s32(prediction64, vget_low_s32(vreinterpretq_s32_u32(riceParamPart128))); + + samples128_4 = drflac__valignrq_s32_1(samples128_0, samples128_4); + samples128_0 = drflac__valignrq_s32_1(vcombine_s32(prediction64, vdup_n_s32(0)), samples128_0); + riceParamPart128 = drflac__valignrq_u32_1(vdupq_n_u32(0), riceParamPart128); + } + } else { + for (i = 0; i < 4; i += 1) { + prediction128 = vmulq_s32(coefficients128_8, samples128_8); + prediction128 = vmlaq_s32(prediction128, coefficients128_4, samples128_4); + prediction128 = vmlaq_s32(prediction128, coefficients128_0, samples128_0); + + /* Horizontal add and shift. */ + prediction64 = drflac__vhaddq_s32(prediction128); + prediction64 = vshl_s32(prediction64, shift64); + prediction64 = vadd_s32(prediction64, vget_low_s32(vreinterpretq_s32_u32(riceParamPart128))); + + samples128_8 = drflac__valignrq_s32_1(samples128_4, samples128_8); + samples128_4 = drflac__valignrq_s32_1(samples128_0, samples128_4); + samples128_0 = drflac__valignrq_s32_1(vcombine_s32(prediction64, vdup_n_s32(0)), samples128_0); + riceParamPart128 = drflac__valignrq_u32_1(vdupq_n_u32(0), riceParamPart128); + } + } + + /* We store samples in groups of 4. */ + vst1q_s32(pDecodedSamples, samples128_0); + pDecodedSamples += 4; + } + + /* Make sure we process the last few samples. */ + i = (count & ~3); + while (i < (int)count) { + /* Rice extraction. */ + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[0], &riceParamParts[0])) { + return DRFLAC_FALSE; + } + + /* Rice reconstruction. */ + riceParamParts[0] &= riceParamMask; + riceParamParts[0] |= (zeroCountParts[0] << riceParam); + riceParamParts[0] = (riceParamParts[0] >> 1) ^ t[riceParamParts[0] & 0x01]; + + /* Sample reconstruction. */ + pDecodedSamples[0] = riceParamParts[0] + drflac__calculate_prediction_32(order, shift, coefficients, pDecodedSamples); + + i += 1; + pDecodedSamples += 1; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_samples_with_residual__rice__neon_64(drflac_bs* bs, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 order, drflac_int32 shift, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + int i; + drflac_uint32 riceParamMask; + drflac_int32* pDecodedSamples = pSamplesOut; + drflac_int32* pDecodedSamplesEnd = pSamplesOut + (count & ~3); + drflac_uint32 zeroCountParts[4]; + drflac_uint32 riceParamParts[4]; + int32x4_t coefficients128_0; + int32x4_t coefficients128_4; + int32x4_t coefficients128_8; + int32x4_t samples128_0; + int32x4_t samples128_4; + int32x4_t samples128_8; + uint32x4_t riceParamMask128; + int32x4_t riceParam128; + int64x1_t shift64; + uint32x4_t one128; + int64x2_t prediction128; + uint32x4_t zeroCountPart128; + uint32x4_t riceParamPart128; + + const drflac_uint32 t[2] = {0x00000000, 0xFFFFFFFF}; + + riceParamMask = (drflac_uint32)~((~0UL) << riceParam); + riceParamMask128 = vdupq_n_u32(riceParamMask); + + riceParam128 = vdupq_n_s32(riceParam); + shift64 = vdup_n_s64(-shift); /* Negate the shift because we'll be doing a variable shift using vshlq_s32(). */ + one128 = vdupq_n_u32(1); + + /* + Pre-loading the coefficients and prior samples is annoying because we need to ensure we don't try reading more than + what's available in the input buffers. It would be convenient to use a fall-through switch to do this, but this results + in strict aliasing warnings with GCC. To work around this I'm just doing something hacky. This feels a bit convoluted + so I think there's opportunity for this to be simplified. + */ + { + int runningOrder = order; + drflac_int32 tempC[4] = {0, 0, 0, 0}; + drflac_int32 tempS[4] = {0, 0, 0, 0}; + + /* 0 - 3. */ + if (runningOrder >= 4) { + coefficients128_0 = vld1q_s32(coefficients + 0); + samples128_0 = vld1q_s32(pSamplesOut - 4); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: tempC[2] = coefficients[2]; tempS[1] = pSamplesOut[-3]; /* fallthrough */ + case 2: tempC[1] = coefficients[1]; tempS[2] = pSamplesOut[-2]; /* fallthrough */ + case 1: tempC[0] = coefficients[0]; tempS[3] = pSamplesOut[-1]; /* fallthrough */ + } + + coefficients128_0 = vld1q_s32(tempC); + samples128_0 = vld1q_s32(tempS); + runningOrder = 0; + } + + /* 4 - 7 */ + if (runningOrder >= 4) { + coefficients128_4 = vld1q_s32(coefficients + 4); + samples128_4 = vld1q_s32(pSamplesOut - 8); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: tempC[2] = coefficients[6]; tempS[1] = pSamplesOut[-7]; /* fallthrough */ + case 2: tempC[1] = coefficients[5]; tempS[2] = pSamplesOut[-6]; /* fallthrough */ + case 1: tempC[0] = coefficients[4]; tempS[3] = pSamplesOut[-5]; /* fallthrough */ + } + + coefficients128_4 = vld1q_s32(tempC); + samples128_4 = vld1q_s32(tempS); + runningOrder = 0; + } + + /* 8 - 11 */ + if (runningOrder == 4) { + coefficients128_8 = vld1q_s32(coefficients + 8); + samples128_8 = vld1q_s32(pSamplesOut - 12); + runningOrder -= 4; + } else { + switch (runningOrder) { + case 3: tempC[2] = coefficients[10]; tempS[1] = pSamplesOut[-11]; /* fallthrough */ + case 2: tempC[1] = coefficients[ 9]; tempS[2] = pSamplesOut[-10]; /* fallthrough */ + case 1: tempC[0] = coefficients[ 8]; tempS[3] = pSamplesOut[- 9]; /* fallthrough */ + } + + coefficients128_8 = vld1q_s32(tempC); + samples128_8 = vld1q_s32(tempS); + runningOrder = 0; + } + + /* Coefficients need to be shuffled for our streaming algorithm below to work. Samples are already in the correct order from the loading routine above. */ + coefficients128_0 = drflac__vrevq_s32(coefficients128_0); + coefficients128_4 = drflac__vrevq_s32(coefficients128_4); + coefficients128_8 = drflac__vrevq_s32(coefficients128_8); + } + + /* For this version we are doing one sample at a time. */ + while (pDecodedSamples < pDecodedSamplesEnd) { + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[0], &riceParamParts[0]) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[1], &riceParamParts[1]) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[2], &riceParamParts[2]) || + !drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[3], &riceParamParts[3])) { + return DRFLAC_FALSE; + } + + zeroCountPart128 = vld1q_u32(zeroCountParts); + riceParamPart128 = vld1q_u32(riceParamParts); + + riceParamPart128 = vandq_u32(riceParamPart128, riceParamMask128); + riceParamPart128 = vorrq_u32(riceParamPart128, vshlq_u32(zeroCountPart128, riceParam128)); + riceParamPart128 = veorq_u32(vshrq_n_u32(riceParamPart128, 1), vaddq_u32(drflac__vnotq_u32(vandq_u32(riceParamPart128, one128)), one128)); + + for (i = 0; i < 4; i += 1) { + int64x1_t prediction64; + + prediction128 = vdupq_n_s64(0); + switch (order) + { + case 12: + case 11: prediction128 = vaddq_s64(prediction128, vmull_s32(vget_low_s32(coefficients128_8), vget_low_s32(samples128_8))); + case 10: + case 9: prediction128 = vaddq_s64(prediction128, vmull_s32(vget_high_s32(coefficients128_8), vget_high_s32(samples128_8))); + case 8: + case 7: prediction128 = vaddq_s64(prediction128, vmull_s32(vget_low_s32(coefficients128_4), vget_low_s32(samples128_4))); + case 6: + case 5: prediction128 = vaddq_s64(prediction128, vmull_s32(vget_high_s32(coefficients128_4), vget_high_s32(samples128_4))); + case 4: + case 3: prediction128 = vaddq_s64(prediction128, vmull_s32(vget_low_s32(coefficients128_0), vget_low_s32(samples128_0))); + case 2: + case 1: prediction128 = vaddq_s64(prediction128, vmull_s32(vget_high_s32(coefficients128_0), vget_high_s32(samples128_0))); + } + + /* Horizontal add and shift. */ + prediction64 = drflac__vhaddq_s64(prediction128); + prediction64 = vshl_s64(prediction64, shift64); + prediction64 = vadd_s64(prediction64, vdup_n_s64(vgetq_lane_u32(riceParamPart128, 0))); + + /* Our value should be sitting in prediction64[0]. We need to combine this with our SSE samples. */ + samples128_8 = drflac__valignrq_s32_1(samples128_4, samples128_8); + samples128_4 = drflac__valignrq_s32_1(samples128_0, samples128_4); + samples128_0 = drflac__valignrq_s32_1(vcombine_s32(vreinterpret_s32_s64(prediction64), vdup_n_s32(0)), samples128_0); + + /* Slide our rice parameter down so that the value in position 0 contains the next one to process. */ + riceParamPart128 = drflac__valignrq_u32_1(vdupq_n_u32(0), riceParamPart128); + } + + /* We store samples in groups of 4. */ + vst1q_s32(pDecodedSamples, samples128_0); + pDecodedSamples += 4; + } + + /* Make sure we process the last few samples. */ + i = (count & ~3); + while (i < (int)count) { + /* Rice extraction. */ + if (!drflac__read_rice_parts_x1(bs, riceParam, &zeroCountParts[0], &riceParamParts[0])) { + return DRFLAC_FALSE; + } + + /* Rice reconstruction. */ + riceParamParts[0] &= riceParamMask; + riceParamParts[0] |= (zeroCountParts[0] << riceParam); + riceParamParts[0] = (riceParamParts[0] >> 1) ^ t[riceParamParts[0] & 0x01]; + + /* Sample reconstruction. */ + pDecodedSamples[0] = riceParamParts[0] + drflac__calculate_prediction_64(order, shift, coefficients, pDecodedSamples); + + i += 1; + pDecodedSamples += 1; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_samples_with_residual__rice__neon(drflac_bs* bs, drflac_uint32 bitsPerSample, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 lpcOrder, drflac_int32 lpcShift, drflac_uint32 lpcPrecision, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(pSamplesOut != NULL); + + /* In my testing the order is rarely > 12, so in this case I'm going to simplify the NEON implementation by only handling order <= 12. */ + if (lpcOrder > 0 && lpcOrder <= 12) { + if (drflac__use_64_bit_prediction(bitsPerSample, lpcOrder, lpcPrecision)) { + return drflac__decode_samples_with_residual__rice__neon_64(bs, count, riceParam, lpcOrder, lpcShift, coefficients, pSamplesOut); + } else { + return drflac__decode_samples_with_residual__rice__neon_32(bs, count, riceParam, lpcOrder, lpcShift, coefficients, pSamplesOut); + } + } else { + return drflac__decode_samples_with_residual__rice__scalar(bs, bitsPerSample, count, riceParam, lpcOrder, lpcShift, lpcPrecision, coefficients, pSamplesOut); + } +} +#endif + +static drflac_bool32 drflac__decode_samples_with_residual__rice(drflac_bs* bs, drflac_uint32 bitsPerSample, drflac_uint32 count, drflac_uint8 riceParam, drflac_uint32 lpcOrder, drflac_int32 lpcShift, drflac_uint32 lpcPrecision, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ +#if defined(DRFLAC_SUPPORT_SSE41) + if (drflac__gIsSSE41Supported) { + return drflac__decode_samples_with_residual__rice__sse41(bs, bitsPerSample, count, riceParam, lpcOrder, lpcShift, lpcPrecision, coefficients, pSamplesOut); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported) { + return drflac__decode_samples_with_residual__rice__neon(bs, bitsPerSample, count, riceParam, lpcOrder, lpcShift, lpcPrecision, coefficients, pSamplesOut); + } else +#endif + { + /* Scalar fallback. */ + #if 0 + return drflac__decode_samples_with_residual__rice__reference(bs, bitsPerSample, count, riceParam, lpcOrder, lpcShift, lpcPrecision, coefficients, pSamplesOut); + #else + return drflac__decode_samples_with_residual__rice__scalar(bs, bitsPerSample, count, riceParam, lpcOrder, lpcShift, lpcPrecision, coefficients, pSamplesOut); + #endif + } +} + +/* Reads and seeks past a string of residual values as Rice codes. The decoder should be sitting on the first bit of the Rice codes. */ +static drflac_bool32 drflac__read_and_seek_residual__rice(drflac_bs* bs, drflac_uint32 count, drflac_uint8 riceParam) +{ + drflac_uint32 i; + + DRFLAC_ASSERT(bs != NULL); + + for (i = 0; i < count; ++i) { + if (!drflac__seek_rice_parts(bs, riceParam)) { + return DRFLAC_FALSE; + } + } + + return DRFLAC_TRUE; +} + +#if defined(__clang__) +__attribute__((no_sanitize("signed-integer-overflow"))) +#endif +static drflac_bool32 drflac__decode_samples_with_residual__unencoded(drflac_bs* bs, drflac_uint32 bitsPerSample, drflac_uint32 count, drflac_uint8 unencodedBitsPerSample, drflac_uint32 lpcOrder, drflac_int32 lpcShift, drflac_uint32 lpcPrecision, const drflac_int32* coefficients, drflac_int32* pSamplesOut) +{ + drflac_uint32 i; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(unencodedBitsPerSample <= 31); /* <-- unencodedBitsPerSample is a 5 bit number, so cannot exceed 31. */ + DRFLAC_ASSERT(pSamplesOut != NULL); + + for (i = 0; i < count; ++i) { + if (unencodedBitsPerSample > 0) { + if (!drflac__read_int32(bs, unencodedBitsPerSample, pSamplesOut + i)) { + return DRFLAC_FALSE; + } + } else { + pSamplesOut[i] = 0; + } + + if (drflac__use_64_bit_prediction(bitsPerSample, lpcOrder, lpcPrecision)) { + pSamplesOut[i] += drflac__calculate_prediction_64(lpcOrder, lpcShift, coefficients, pSamplesOut + i); + } else { + pSamplesOut[i] += drflac__calculate_prediction_32(lpcOrder, lpcShift, coefficients, pSamplesOut + i); + } + } + + return DRFLAC_TRUE; +} + + +/* +Reads and decodes the residual for the sub-frame the decoder is currently sitting on. This function should be called +when the decoder is sitting at the very start of the RESIDUAL block. The first <order> residuals will be ignored. The +<blockSize> and <order> parameters are used to determine how many residual values need to be decoded. +*/ +static drflac_bool32 drflac__decode_samples_with_residual(drflac_bs* bs, drflac_uint32 bitsPerSample, drflac_uint32 blockSize, drflac_uint32 lpcOrder, drflac_int32 lpcShift, drflac_uint32 lpcPrecision, const drflac_int32* coefficients, drflac_int32* pDecodedSamples) +{ + drflac_uint8 residualMethod; + drflac_uint8 partitionOrder; + drflac_uint32 samplesInPartition; + drflac_uint32 partitionsRemaining; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(blockSize != 0); + DRFLAC_ASSERT(pDecodedSamples != NULL); /* <-- Should we allow NULL, in which case we just seek past the residual rather than do a full decode? */ + + if (!drflac__read_uint8(bs, 2, &residualMethod)) { + return DRFLAC_FALSE; + } + + if (residualMethod != DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE && residualMethod != DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE2) { + return DRFLAC_FALSE; /* Unknown or unsupported residual coding method. */ + } + + /* Ignore the first <order> values. */ + pDecodedSamples += lpcOrder; + + if (!drflac__read_uint8(bs, 4, &partitionOrder)) { + return DRFLAC_FALSE; + } + + /* + From the FLAC spec: + The Rice partition order in a Rice-coded residual section must be less than or equal to 8. + */ + if (partitionOrder > 8) { + return DRFLAC_FALSE; + } + + /* Validation check. */ + if ((blockSize / (1 << partitionOrder)) < lpcOrder) { + return DRFLAC_FALSE; + } + + samplesInPartition = (blockSize / (1 << partitionOrder)) - lpcOrder; + partitionsRemaining = (1 << partitionOrder); + for (;;) { + drflac_uint8 riceParam = 0; + if (residualMethod == DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE) { + if (!drflac__read_uint8(bs, 4, &riceParam)) { + return DRFLAC_FALSE; + } + if (riceParam == 15) { + riceParam = 0xFF; + } + } else if (residualMethod == DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE2) { + if (!drflac__read_uint8(bs, 5, &riceParam)) { + return DRFLAC_FALSE; + } + if (riceParam == 31) { + riceParam = 0xFF; + } + } + + if (riceParam != 0xFF) { + if (!drflac__decode_samples_with_residual__rice(bs, bitsPerSample, samplesInPartition, riceParam, lpcOrder, lpcShift, lpcPrecision, coefficients, pDecodedSamples)) { + return DRFLAC_FALSE; + } + } else { + drflac_uint8 unencodedBitsPerSample = 0; + if (!drflac__read_uint8(bs, 5, &unencodedBitsPerSample)) { + return DRFLAC_FALSE; + } + + if (!drflac__decode_samples_with_residual__unencoded(bs, bitsPerSample, samplesInPartition, unencodedBitsPerSample, lpcOrder, lpcShift, lpcPrecision, coefficients, pDecodedSamples)) { + return DRFLAC_FALSE; + } + } + + pDecodedSamples += samplesInPartition; + + if (partitionsRemaining == 1) { + break; + } + + partitionsRemaining -= 1; + + if (partitionOrder != 0) { + samplesInPartition = blockSize / (1 << partitionOrder); + } + } + + return DRFLAC_TRUE; +} + +/* +Reads and seeks past the residual for the sub-frame the decoder is currently sitting on. This function should be called +when the decoder is sitting at the very start of the RESIDUAL block. The first <order> residuals will be set to 0. The +<blockSize> and <order> parameters are used to determine how many residual values need to be decoded. +*/ +static drflac_bool32 drflac__read_and_seek_residual(drflac_bs* bs, drflac_uint32 blockSize, drflac_uint32 order) +{ + drflac_uint8 residualMethod; + drflac_uint8 partitionOrder; + drflac_uint32 samplesInPartition; + drflac_uint32 partitionsRemaining; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(blockSize != 0); + + if (!drflac__read_uint8(bs, 2, &residualMethod)) { + return DRFLAC_FALSE; + } + + if (residualMethod != DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE && residualMethod != DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE2) { + return DRFLAC_FALSE; /* Unknown or unsupported residual coding method. */ + } + + if (!drflac__read_uint8(bs, 4, &partitionOrder)) { + return DRFLAC_FALSE; + } + + /* + From the FLAC spec: + The Rice partition order in a Rice-coded residual section must be less than or equal to 8. + */ + if (partitionOrder > 8) { + return DRFLAC_FALSE; + } + + /* Validation check. */ + if ((blockSize / (1 << partitionOrder)) < order) { + return DRFLAC_FALSE; + } + + samplesInPartition = (blockSize / (1 << partitionOrder)) - order; + partitionsRemaining = (1 << partitionOrder); + for (;;) + { + drflac_uint8 riceParam = 0; + if (residualMethod == DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE) { + if (!drflac__read_uint8(bs, 4, &riceParam)) { + return DRFLAC_FALSE; + } + if (riceParam == 15) { + riceParam = 0xFF; + } + } else if (residualMethod == DRFLAC_RESIDUAL_CODING_METHOD_PARTITIONED_RICE2) { + if (!drflac__read_uint8(bs, 5, &riceParam)) { + return DRFLAC_FALSE; + } + if (riceParam == 31) { + riceParam = 0xFF; + } + } + + if (riceParam != 0xFF) { + if (!drflac__read_and_seek_residual__rice(bs, samplesInPartition, riceParam)) { + return DRFLAC_FALSE; + } + } else { + drflac_uint8 unencodedBitsPerSample = 0; + if (!drflac__read_uint8(bs, 5, &unencodedBitsPerSample)) { + return DRFLAC_FALSE; + } + + if (!drflac__seek_bits(bs, unencodedBitsPerSample * samplesInPartition)) { + return DRFLAC_FALSE; + } + } + + + if (partitionsRemaining == 1) { + break; + } + + partitionsRemaining -= 1; + samplesInPartition = blockSize / (1 << partitionOrder); + } + + return DRFLAC_TRUE; +} + + +static drflac_bool32 drflac__decode_samples__constant(drflac_bs* bs, drflac_uint32 blockSize, drflac_uint32 subframeBitsPerSample, drflac_int32* pDecodedSamples) +{ + drflac_uint32 i; + + /* Only a single sample needs to be decoded here. */ + drflac_int32 sample; + if (!drflac__read_int32(bs, subframeBitsPerSample, &sample)) { + return DRFLAC_FALSE; + } + + /* + We don't really need to expand this, but it does simplify the process of reading samples. If this becomes a performance issue (unlikely) + we'll want to look at a more efficient way. + */ + for (i = 0; i < blockSize; ++i) { + pDecodedSamples[i] = sample; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_samples__verbatim(drflac_bs* bs, drflac_uint32 blockSize, drflac_uint32 subframeBitsPerSample, drflac_int32* pDecodedSamples) +{ + drflac_uint32 i; + + for (i = 0; i < blockSize; ++i) { + drflac_int32 sample; + if (!drflac__read_int32(bs, subframeBitsPerSample, &sample)) { + return DRFLAC_FALSE; + } + + pDecodedSamples[i] = sample; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_samples__fixed(drflac_bs* bs, drflac_uint32 blockSize, drflac_uint32 subframeBitsPerSample, drflac_uint8 lpcOrder, drflac_int32* pDecodedSamples) +{ + drflac_uint32 i; + + static drflac_int32 lpcCoefficientsTable[5][4] = { + {0, 0, 0, 0}, + {1, 0, 0, 0}, + {2, -1, 0, 0}, + {3, -3, 1, 0}, + {4, -6, 4, -1} + }; + + /* Warm up samples and coefficients. */ + for (i = 0; i < lpcOrder; ++i) { + drflac_int32 sample; + if (!drflac__read_int32(bs, subframeBitsPerSample, &sample)) { + return DRFLAC_FALSE; + } + + pDecodedSamples[i] = sample; + } + + if (!drflac__decode_samples_with_residual(bs, subframeBitsPerSample, blockSize, lpcOrder, 0, 4, lpcCoefficientsTable[lpcOrder], pDecodedSamples)) { + return DRFLAC_FALSE; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_samples__lpc(drflac_bs* bs, drflac_uint32 blockSize, drflac_uint32 bitsPerSample, drflac_uint8 lpcOrder, drflac_int32* pDecodedSamples) +{ + drflac_uint8 i; + drflac_uint8 lpcPrecision; + drflac_int8 lpcShift; + drflac_int32 coefficients[32]; + + /* Warm up samples. */ + for (i = 0; i < lpcOrder; ++i) { + drflac_int32 sample; + if (!drflac__read_int32(bs, bitsPerSample, &sample)) { + return DRFLAC_FALSE; + } + + pDecodedSamples[i] = sample; + } + + if (!drflac__read_uint8(bs, 4, &lpcPrecision)) { + return DRFLAC_FALSE; + } + if (lpcPrecision == 15) { + return DRFLAC_FALSE; /* Invalid. */ + } + lpcPrecision += 1; + + if (!drflac__read_int8(bs, 5, &lpcShift)) { + return DRFLAC_FALSE; + } + + /* + From the FLAC specification: + + Quantized linear predictor coefficient shift needed in bits (NOTE: this number is signed two's-complement) + + Emphasis on the "signed two's-complement". In practice there does not seem to be any encoders nor decoders supporting negative shifts. For now dr_flac is + not going to support negative shifts as I don't have any reference files. However, when a reference file comes through I will consider adding support. + */ + if (lpcShift < 0) { + return DRFLAC_FALSE; + } + + DRFLAC_ZERO_MEMORY(coefficients, sizeof(coefficients)); + for (i = 0; i < lpcOrder; ++i) { + if (!drflac__read_int32(bs, lpcPrecision, coefficients + i)) { + return DRFLAC_FALSE; + } + } + + if (!drflac__decode_samples_with_residual(bs, bitsPerSample, blockSize, lpcOrder, lpcShift, lpcPrecision, coefficients, pDecodedSamples)) { + return DRFLAC_FALSE; + } + + return DRFLAC_TRUE; +} + + +static drflac_bool32 drflac__read_next_flac_frame_header(drflac_bs* bs, drflac_uint8 streaminfoBitsPerSample, drflac_frame_header* header) +{ + const drflac_uint32 sampleRateTable[12] = {0, 88200, 176400, 192000, 8000, 16000, 22050, 24000, 32000, 44100, 48000, 96000}; + const drflac_uint8 bitsPerSampleTable[8] = {0, 8, 12, (drflac_uint8)-1, 16, 20, 24, (drflac_uint8)-1}; /* -1 = reserved. */ + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(header != NULL); + + /* Keep looping until we find a valid sync code. */ + for (;;) { + drflac_uint8 crc8 = 0xCE; /* 0xCE = drflac_crc8(0, 0x3FFE, 14); */ + drflac_uint8 reserved = 0; + drflac_uint8 blockingStrategy = 0; + drflac_uint8 blockSize = 0; + drflac_uint8 sampleRate = 0; + drflac_uint8 channelAssignment = 0; + drflac_uint8 bitsPerSample = 0; + drflac_bool32 isVariableBlockSize; + + if (!drflac__find_and_seek_to_next_sync_code(bs)) { + return DRFLAC_FALSE; + } + + if (!drflac__read_uint8(bs, 1, &reserved)) { + return DRFLAC_FALSE; + } + if (reserved == 1) { + continue; + } + crc8 = drflac_crc8(crc8, reserved, 1); + + if (!drflac__read_uint8(bs, 1, &blockingStrategy)) { + return DRFLAC_FALSE; + } + crc8 = drflac_crc8(crc8, blockingStrategy, 1); + + if (!drflac__read_uint8(bs, 4, &blockSize)) { + return DRFLAC_FALSE; + } + if (blockSize == 0) { + continue; + } + crc8 = drflac_crc8(crc8, blockSize, 4); + + if (!drflac__read_uint8(bs, 4, &sampleRate)) { + return DRFLAC_FALSE; + } + crc8 = drflac_crc8(crc8, sampleRate, 4); + + if (!drflac__read_uint8(bs, 4, &channelAssignment)) { + return DRFLAC_FALSE; + } + if (channelAssignment > 10) { + continue; + } + crc8 = drflac_crc8(crc8, channelAssignment, 4); + + if (!drflac__read_uint8(bs, 3, &bitsPerSample)) { + return DRFLAC_FALSE; + } + if (bitsPerSample == 3 || bitsPerSample == 7) { + continue; + } + crc8 = drflac_crc8(crc8, bitsPerSample, 3); + + + if (!drflac__read_uint8(bs, 1, &reserved)) { + return DRFLAC_FALSE; + } + if (reserved == 1) { + continue; + } + crc8 = drflac_crc8(crc8, reserved, 1); + + + isVariableBlockSize = blockingStrategy == 1; + if (isVariableBlockSize) { + drflac_uint64 pcmFrameNumber; + drflac_result result = drflac__read_utf8_coded_number(bs, &pcmFrameNumber, &crc8); + if (result != DRFLAC_SUCCESS) { + if (result == DRFLAC_AT_END) { + return DRFLAC_FALSE; + } else { + continue; + } + } + header->flacFrameNumber = 0; + header->pcmFrameNumber = pcmFrameNumber; + } else { + drflac_uint64 flacFrameNumber = 0; + drflac_result result = drflac__read_utf8_coded_number(bs, &flacFrameNumber, &crc8); + if (result != DRFLAC_SUCCESS) { + if (result == DRFLAC_AT_END) { + return DRFLAC_FALSE; + } else { + continue; + } + } + header->flacFrameNumber = (drflac_uint32)flacFrameNumber; /* <-- Safe cast. */ + header->pcmFrameNumber = 0; + } + + + DRFLAC_ASSERT(blockSize > 0); + if (blockSize == 1) { + header->blockSizeInPCMFrames = 192; + } else if (blockSize <= 5) { + DRFLAC_ASSERT(blockSize >= 2); + header->blockSizeInPCMFrames = 576 * (1 << (blockSize - 2)); + } else if (blockSize == 6) { + if (!drflac__read_uint16(bs, 8, &header->blockSizeInPCMFrames)) { + return DRFLAC_FALSE; + } + crc8 = drflac_crc8(crc8, header->blockSizeInPCMFrames, 8); + header->blockSizeInPCMFrames += 1; + } else if (blockSize == 7) { + if (!drflac__read_uint16(bs, 16, &header->blockSizeInPCMFrames)) { + return DRFLAC_FALSE; + } + crc8 = drflac_crc8(crc8, header->blockSizeInPCMFrames, 16); + if (header->blockSizeInPCMFrames == 0xFFFF) { + return DRFLAC_FALSE; /* Frame is too big. This is the size of the frame minus 1. The STREAMINFO block defines the max block size which is 16-bits. Adding one will make it 17 bits and therefore too big. */ + } + header->blockSizeInPCMFrames += 1; + } else { + DRFLAC_ASSERT(blockSize >= 8); + header->blockSizeInPCMFrames = 256 * (1 << (blockSize - 8)); + } + + + if (sampleRate <= 11) { + header->sampleRate = sampleRateTable[sampleRate]; + } else if (sampleRate == 12) { + if (!drflac__read_uint32(bs, 8, &header->sampleRate)) { + return DRFLAC_FALSE; + } + crc8 = drflac_crc8(crc8, header->sampleRate, 8); + header->sampleRate *= 1000; + } else if (sampleRate == 13) { + if (!drflac__read_uint32(bs, 16, &header->sampleRate)) { + return DRFLAC_FALSE; + } + crc8 = drflac_crc8(crc8, header->sampleRate, 16); + } else if (sampleRate == 14) { + if (!drflac__read_uint32(bs, 16, &header->sampleRate)) { + return DRFLAC_FALSE; + } + crc8 = drflac_crc8(crc8, header->sampleRate, 16); + header->sampleRate *= 10; + } else { + continue; /* Invalid. Assume an invalid block. */ + } + + + header->channelAssignment = channelAssignment; + + header->bitsPerSample = bitsPerSampleTable[bitsPerSample]; + if (header->bitsPerSample == 0) { + header->bitsPerSample = streaminfoBitsPerSample; + } + + if (header->bitsPerSample != streaminfoBitsPerSample) { + /* If this subframe has a different bitsPerSample then streaminfo or the first frame, reject it */ + return DRFLAC_FALSE; + } + + if (!drflac__read_uint8(bs, 8, &header->crc8)) { + return DRFLAC_FALSE; + } + +#ifndef DR_FLAC_NO_CRC + if (header->crc8 != crc8) { + continue; /* CRC mismatch. Loop back to the top and find the next sync code. */ + } +#endif + return DRFLAC_TRUE; + } +} + +static drflac_bool32 drflac__read_subframe_header(drflac_bs* bs, drflac_subframe* pSubframe) +{ + drflac_uint8 header; + int type; + + if (!drflac__read_uint8(bs, 8, &header)) { + return DRFLAC_FALSE; + } + + /* First bit should always be 0. */ + if ((header & 0x80) != 0) { + return DRFLAC_FALSE; + } + + /* + Default to 0 for the LPC order. It's important that we always set this to 0 for non LPC + and FIXED subframes because we'll be using it in a generic validation check later. + */ + pSubframe->lpcOrder = 0; + + type = (header & 0x7E) >> 1; + if (type == 0) { + pSubframe->subframeType = DRFLAC_SUBFRAME_CONSTANT; + } else if (type == 1) { + pSubframe->subframeType = DRFLAC_SUBFRAME_VERBATIM; + } else { + if ((type & 0x20) != 0) { + pSubframe->subframeType = DRFLAC_SUBFRAME_LPC; + pSubframe->lpcOrder = (drflac_uint8)(type & 0x1F) + 1; + } else if ((type & 0x08) != 0) { + pSubframe->subframeType = DRFLAC_SUBFRAME_FIXED; + pSubframe->lpcOrder = (drflac_uint8)(type & 0x07); + if (pSubframe->lpcOrder > 4) { + pSubframe->subframeType = DRFLAC_SUBFRAME_RESERVED; + pSubframe->lpcOrder = 0; + } + } else { + pSubframe->subframeType = DRFLAC_SUBFRAME_RESERVED; + } + } + + if (pSubframe->subframeType == DRFLAC_SUBFRAME_RESERVED) { + return DRFLAC_FALSE; + } + + /* Wasted bits per sample. */ + pSubframe->wastedBitsPerSample = 0; + if ((header & 0x01) == 1) { + unsigned int wastedBitsPerSample; + if (!drflac__seek_past_next_set_bit(bs, &wastedBitsPerSample)) { + return DRFLAC_FALSE; + } + pSubframe->wastedBitsPerSample = (drflac_uint8)wastedBitsPerSample + 1; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_subframe(drflac_bs* bs, drflac_frame* frame, int subframeIndex, drflac_int32* pDecodedSamplesOut) +{ + drflac_subframe* pSubframe; + drflac_uint32 subframeBitsPerSample; + drflac_bool32 decodeResult; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(frame != NULL); + + pSubframe = frame->subframes + subframeIndex; + if (!drflac__read_subframe_header(bs, pSubframe)) { + return DRFLAC_FALSE; + } + + /* Side channels require an extra bit per sample. Took a while to figure that one out... */ + subframeBitsPerSample = frame->header.bitsPerSample; + if ((frame->header.channelAssignment == DRFLAC_CHANNEL_ASSIGNMENT_LEFT_SIDE || frame->header.channelAssignment == DRFLAC_CHANNEL_ASSIGNMENT_MID_SIDE) && subframeIndex == 1) { + subframeBitsPerSample += 1; + } else if (frame->header.channelAssignment == DRFLAC_CHANNEL_ASSIGNMENT_RIGHT_SIDE && subframeIndex == 0) { + subframeBitsPerSample += 1; + } + + if (subframeBitsPerSample > 32) { + /* libFLAC and ffmpeg reject 33-bit subframes as well */ + return DRFLAC_FALSE; + } + + /* Need to handle wasted bits per sample. */ + if (pSubframe->wastedBitsPerSample >= subframeBitsPerSample) { + return DRFLAC_FALSE; + } + subframeBitsPerSample -= pSubframe->wastedBitsPerSample; + + pSubframe->pSamplesS32 = pDecodedSamplesOut; + + /* + pDecodedSamplesOut will be pointing to a buffer that was allocated with enough memory to store + maxBlockSizeInPCMFrames samples (as specified in the FLAC header). We need to guard against an + overflow here. At a higher level we are checking maxBlockSizeInPCMFrames from the header, but + here we need to do an additional check to ensure this frame's block size fully encompasses any + warmup samples which is determined by the LPC order. For non LPC and FIXED subframes, the LPC + order will be have been set to 0 in drflac__read_subframe_header(). + */ + if (frame->header.blockSizeInPCMFrames < pSubframe->lpcOrder) { + return DRFLAC_FALSE; + } + + switch (pSubframe->subframeType) + { + case DRFLAC_SUBFRAME_CONSTANT: + { + decodeResult = drflac__decode_samples__constant(bs, frame->header.blockSizeInPCMFrames, subframeBitsPerSample, pSubframe->pSamplesS32); + } break; + + case DRFLAC_SUBFRAME_VERBATIM: + { + decodeResult = drflac__decode_samples__verbatim(bs, frame->header.blockSizeInPCMFrames, subframeBitsPerSample, pSubframe->pSamplesS32); + } break; + + case DRFLAC_SUBFRAME_FIXED: + { + decodeResult = drflac__decode_samples__fixed(bs, frame->header.blockSizeInPCMFrames, subframeBitsPerSample, pSubframe->lpcOrder, pSubframe->pSamplesS32); + } break; + + case DRFLAC_SUBFRAME_LPC: + { + decodeResult = drflac__decode_samples__lpc(bs, frame->header.blockSizeInPCMFrames, subframeBitsPerSample, pSubframe->lpcOrder, pSubframe->pSamplesS32); + } break; + + default: decodeResult = DRFLAC_FALSE; + } + + return decodeResult; +} + +static drflac_bool32 drflac__seek_subframe(drflac_bs* bs, drflac_frame* frame, int subframeIndex) +{ + drflac_subframe* pSubframe; + drflac_uint32 subframeBitsPerSample; + + DRFLAC_ASSERT(bs != NULL); + DRFLAC_ASSERT(frame != NULL); + + pSubframe = frame->subframes + subframeIndex; + if (!drflac__read_subframe_header(bs, pSubframe)) { + return DRFLAC_FALSE; + } + + /* Side channels require an extra bit per sample. Took a while to figure that one out... */ + subframeBitsPerSample = frame->header.bitsPerSample; + if ((frame->header.channelAssignment == DRFLAC_CHANNEL_ASSIGNMENT_LEFT_SIDE || frame->header.channelAssignment == DRFLAC_CHANNEL_ASSIGNMENT_MID_SIDE) && subframeIndex == 1) { + subframeBitsPerSample += 1; + } else if (frame->header.channelAssignment == DRFLAC_CHANNEL_ASSIGNMENT_RIGHT_SIDE && subframeIndex == 0) { + subframeBitsPerSample += 1; + } + + /* Need to handle wasted bits per sample. */ + if (pSubframe->wastedBitsPerSample >= subframeBitsPerSample) { + return DRFLAC_FALSE; + } + subframeBitsPerSample -= pSubframe->wastedBitsPerSample; + + pSubframe->pSamplesS32 = NULL; + + switch (pSubframe->subframeType) + { + case DRFLAC_SUBFRAME_CONSTANT: + { + if (!drflac__seek_bits(bs, subframeBitsPerSample)) { + return DRFLAC_FALSE; + } + } break; + + case DRFLAC_SUBFRAME_VERBATIM: + { + unsigned int bitsToSeek = frame->header.blockSizeInPCMFrames * subframeBitsPerSample; + if (!drflac__seek_bits(bs, bitsToSeek)) { + return DRFLAC_FALSE; + } + } break; + + case DRFLAC_SUBFRAME_FIXED: + { + unsigned int bitsToSeek = pSubframe->lpcOrder * subframeBitsPerSample; + if (!drflac__seek_bits(bs, bitsToSeek)) { + return DRFLAC_FALSE; + } + + if (!drflac__read_and_seek_residual(bs, frame->header.blockSizeInPCMFrames, pSubframe->lpcOrder)) { + return DRFLAC_FALSE; + } + } break; + + case DRFLAC_SUBFRAME_LPC: + { + drflac_uint8 lpcPrecision; + + unsigned int bitsToSeek = pSubframe->lpcOrder * subframeBitsPerSample; + if (!drflac__seek_bits(bs, bitsToSeek)) { + return DRFLAC_FALSE; + } + + if (!drflac__read_uint8(bs, 4, &lpcPrecision)) { + return DRFLAC_FALSE; + } + if (lpcPrecision == 15) { + return DRFLAC_FALSE; /* Invalid. */ + } + lpcPrecision += 1; + + + bitsToSeek = (pSubframe->lpcOrder * lpcPrecision) + 5; /* +5 for shift. */ + if (!drflac__seek_bits(bs, bitsToSeek)) { + return DRFLAC_FALSE; + } + + if (!drflac__read_and_seek_residual(bs, frame->header.blockSizeInPCMFrames, pSubframe->lpcOrder)) { + return DRFLAC_FALSE; + } + } break; + + default: return DRFLAC_FALSE; + } + + return DRFLAC_TRUE; +} + + +static DRFLAC_INLINE drflac_uint8 drflac__get_channel_count_from_channel_assignment(drflac_int8 channelAssignment) +{ + drflac_uint8 lookup[] = {1, 2, 3, 4, 5, 6, 7, 8, 2, 2, 2}; + + DRFLAC_ASSERT(channelAssignment <= 10); + return lookup[channelAssignment]; +} + +static drflac_result drflac__decode_flac_frame(drflac* pFlac) +{ + int channelCount; + int i; + drflac_uint8 paddingSizeInBits; + drflac_uint16 desiredCRC16; +#ifndef DR_FLAC_NO_CRC + drflac_uint16 actualCRC16; +#endif + + /* This function should be called while the stream is sitting on the first byte after the frame header. */ + pFlac->currentFLACFrame.pcmFramesRemaining = 0; + DRFLAC_ZERO_MEMORY(pFlac->currentFLACFrame.subframes, sizeof(pFlac->currentFLACFrame.subframes)); + + /* The frame block size must never be larger than the maximum block size defined by the FLAC stream. */ + if (pFlac->currentFLACFrame.header.blockSizeInPCMFrames > pFlac->maxBlockSizeInPCMFrames) { + return DRFLAC_ERROR; + } + + /* The number of channels in the frame must match the channel count from the STREAMINFO block. */ + channelCount = drflac__get_channel_count_from_channel_assignment(pFlac->currentFLACFrame.header.channelAssignment); + if (channelCount != (int)pFlac->channels) { + return DRFLAC_ERROR; + } + + for (i = 0; i < channelCount; ++i) { + if (!drflac__decode_subframe(&pFlac->bs, &pFlac->currentFLACFrame, i, pFlac->pDecodedSamples + (pFlac->currentFLACFrame.header.blockSizeInPCMFrames * i))) { + return DRFLAC_ERROR; + } + } + + paddingSizeInBits = (drflac_uint8)(DRFLAC_CACHE_L1_BITS_REMAINING(&pFlac->bs) & 7); + if (paddingSizeInBits > 0) { + drflac_uint8 padding = 0; + if (!drflac__read_uint8(&pFlac->bs, paddingSizeInBits, &padding)) { + return DRFLAC_AT_END; + } + } + +#ifndef DR_FLAC_NO_CRC + actualCRC16 = drflac__flush_crc16(&pFlac->bs); +#endif + if (!drflac__read_uint16(&pFlac->bs, 16, &desiredCRC16)) { + return DRFLAC_AT_END; + } + +#ifndef DR_FLAC_NO_CRC + if (actualCRC16 != desiredCRC16) { + return DRFLAC_CRC_MISMATCH; /* CRC mismatch. */ + } +#endif + + pFlac->currentFLACFrame.pcmFramesRemaining = pFlac->currentFLACFrame.header.blockSizeInPCMFrames; + + return DRFLAC_SUCCESS; +} + +static drflac_result drflac__seek_flac_frame(drflac* pFlac) +{ + drflac_result result; + int channelCount; + int i; + drflac_uint16 desiredCRC16; +#ifndef DR_FLAC_NO_CRC + drflac_uint16 actualCRC16; +#endif + + pFlac->currentFLACFrame.pcmFramesRemaining = 0; + + channelCount = drflac__get_channel_count_from_channel_assignment(pFlac->currentFLACFrame.header.channelAssignment); + for (i = 0; i < channelCount; ++i) { + if (!drflac__seek_subframe(&pFlac->bs, &pFlac->currentFLACFrame, i)) { + result = DRFLAC_ERROR; + goto error; + } + } + + /* Padding. */ + if (!drflac__seek_bits(&pFlac->bs, DRFLAC_CACHE_L1_BITS_REMAINING(&pFlac->bs) & 7)) { + result = DRFLAC_ERROR; + goto error; + } + + /* CRC. */ +#ifndef DR_FLAC_NO_CRC + actualCRC16 = drflac__flush_crc16(&pFlac->bs); +#endif + if (!drflac__read_uint16(&pFlac->bs, 16, &desiredCRC16)) { + result = DRFLAC_AT_END; + goto error; + } + +#ifndef DR_FLAC_NO_CRC + if (actualCRC16 != desiredCRC16) { + result = DRFLAC_CRC_MISMATCH; /* CRC mismatch. */ + goto error; + } +#endif + + return DRFLAC_SUCCESS; + +error: + DRFLAC_ZERO_MEMORY(pFlac->currentFLACFrame.subframes, sizeof(pFlac->currentFLACFrame.subframes)); + return result; +} + +static drflac_bool32 drflac__read_and_decode_next_flac_frame(drflac* pFlac) +{ + DRFLAC_ASSERT(pFlac != NULL); + + for (;;) { + drflac_result result; + + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + + result = drflac__decode_flac_frame(pFlac); + if (result != DRFLAC_SUCCESS) { + if (result == DRFLAC_CRC_MISMATCH) { + continue; /* CRC mismatch. Skip to the next frame. */ + } else { + return DRFLAC_FALSE; + } + } + + return DRFLAC_TRUE; + } +} + +static void drflac__get_pcm_frame_range_of_current_flac_frame(drflac* pFlac, drflac_uint64* pFirstPCMFrame, drflac_uint64* pLastPCMFrame) +{ + drflac_uint64 firstPCMFrame; + drflac_uint64 lastPCMFrame; + + DRFLAC_ASSERT(pFlac != NULL); + + firstPCMFrame = pFlac->currentFLACFrame.header.pcmFrameNumber; + if (firstPCMFrame == 0) { + firstPCMFrame = ((drflac_uint64)pFlac->currentFLACFrame.header.flacFrameNumber) * pFlac->maxBlockSizeInPCMFrames; + } + + lastPCMFrame = firstPCMFrame + pFlac->currentFLACFrame.header.blockSizeInPCMFrames; + if (lastPCMFrame > 0) { + lastPCMFrame -= 1; /* Needs to be zero based. */ + } + + if (pFirstPCMFrame) { + *pFirstPCMFrame = firstPCMFrame; + } + if (pLastPCMFrame) { + *pLastPCMFrame = lastPCMFrame; + } +} + +static drflac_bool32 drflac__seek_to_first_frame(drflac* pFlac) +{ + drflac_bool32 result; + + DRFLAC_ASSERT(pFlac != NULL); + + result = drflac__seek_to_byte(&pFlac->bs, pFlac->firstFLACFramePosInBytes); + + DRFLAC_ZERO_MEMORY(&pFlac->currentFLACFrame, sizeof(pFlac->currentFLACFrame)); + pFlac->currentPCMFrame = 0; + + return result; +} + +static DRFLAC_INLINE drflac_result drflac__seek_to_next_flac_frame(drflac* pFlac) +{ + /* This function should only ever be called while the decoder is sitting on the first byte past the FRAME_HEADER section. */ + DRFLAC_ASSERT(pFlac != NULL); + return drflac__seek_flac_frame(pFlac); +} + + +static drflac_uint64 drflac__seek_forward_by_pcm_frames(drflac* pFlac, drflac_uint64 pcmFramesToSeek) +{ + drflac_uint64 pcmFramesRead = 0; + while (pcmFramesToSeek > 0) { + if (pFlac->currentFLACFrame.pcmFramesRemaining == 0) { + if (!drflac__read_and_decode_next_flac_frame(pFlac)) { + break; /* Couldn't read the next frame, so just break from the loop and return. */ + } + } else { + if (pFlac->currentFLACFrame.pcmFramesRemaining > pcmFramesToSeek) { + pcmFramesRead += pcmFramesToSeek; + pFlac->currentFLACFrame.pcmFramesRemaining -= (drflac_uint32)pcmFramesToSeek; /* <-- Safe cast. Will always be < currentFrame.pcmFramesRemaining < 65536. */ + pcmFramesToSeek = 0; + } else { + pcmFramesRead += pFlac->currentFLACFrame.pcmFramesRemaining; + pcmFramesToSeek -= pFlac->currentFLACFrame.pcmFramesRemaining; + pFlac->currentFLACFrame.pcmFramesRemaining = 0; + } + } + } + + pFlac->currentPCMFrame += pcmFramesRead; + return pcmFramesRead; +} + + +static drflac_bool32 drflac__seek_to_pcm_frame__brute_force(drflac* pFlac, drflac_uint64 pcmFrameIndex) +{ + drflac_bool32 isMidFrame = DRFLAC_FALSE; + drflac_uint64 runningPCMFrameCount; + + DRFLAC_ASSERT(pFlac != NULL); + + /* If we are seeking forward we start from the current position. Otherwise we need to start all the way from the start of the file. */ + if (pcmFrameIndex >= pFlac->currentPCMFrame) { + /* Seeking forward. Need to seek from the current position. */ + runningPCMFrameCount = pFlac->currentPCMFrame; + + /* The frame header for the first frame may not yet have been read. We need to do that if necessary. */ + if (pFlac->currentPCMFrame == 0 && pFlac->currentFLACFrame.pcmFramesRemaining == 0) { + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + } else { + isMidFrame = DRFLAC_TRUE; + } + } else { + /* Seeking backwards. Need to seek from the start of the file. */ + runningPCMFrameCount = 0; + + /* Move back to the start. */ + if (!drflac__seek_to_first_frame(pFlac)) { + return DRFLAC_FALSE; + } + + /* Decode the first frame in preparation for sample-exact seeking below. */ + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + } + + /* + We need to as quickly as possible find the frame that contains the target sample. To do this, we iterate over each frame and inspect its + header. If based on the header we can determine that the frame contains the sample, we do a full decode of that frame. + */ + for (;;) { + drflac_uint64 pcmFrameCountInThisFLACFrame; + drflac_uint64 firstPCMFrameInFLACFrame = 0; + drflac_uint64 lastPCMFrameInFLACFrame = 0; + + drflac__get_pcm_frame_range_of_current_flac_frame(pFlac, &firstPCMFrameInFLACFrame, &lastPCMFrameInFLACFrame); + + pcmFrameCountInThisFLACFrame = (lastPCMFrameInFLACFrame - firstPCMFrameInFLACFrame) + 1; + if (pcmFrameIndex < (runningPCMFrameCount + pcmFrameCountInThisFLACFrame)) { + /* + The sample should be in this frame. We need to fully decode it, however if it's an invalid frame (a CRC mismatch), we need to pretend + it never existed and keep iterating. + */ + drflac_uint64 pcmFramesToDecode = pcmFrameIndex - runningPCMFrameCount; + + if (!isMidFrame) { + drflac_result result = drflac__decode_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + /* The frame is valid. We just need to skip over some samples to ensure it's sample-exact. */ + return drflac__seek_forward_by_pcm_frames(pFlac, pcmFramesToDecode) == pcmFramesToDecode; /* <-- If this fails, something bad has happened (it should never fail). */ + } else { + if (result == DRFLAC_CRC_MISMATCH) { + goto next_iteration; /* CRC mismatch. Pretend this frame never existed. */ + } else { + return DRFLAC_FALSE; + } + } + } else { + /* We started seeking mid-frame which means we need to skip the frame decoding part. */ + return drflac__seek_forward_by_pcm_frames(pFlac, pcmFramesToDecode) == pcmFramesToDecode; + } + } else { + /* + It's not in this frame. We need to seek past the frame, but check if there was a CRC mismatch. If so, we pretend this + frame never existed and leave the running sample count untouched. + */ + if (!isMidFrame) { + drflac_result result = drflac__seek_to_next_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + runningPCMFrameCount += pcmFrameCountInThisFLACFrame; + } else { + if (result == DRFLAC_CRC_MISMATCH) { + goto next_iteration; /* CRC mismatch. Pretend this frame never existed. */ + } else { + return DRFLAC_FALSE; + } + } + } else { + /* + We started seeking mid-frame which means we need to seek by reading to the end of the frame instead of with + drflac__seek_to_next_flac_frame() which only works if the decoder is sitting on the byte just after the frame header. + */ + runningPCMFrameCount += pFlac->currentFLACFrame.pcmFramesRemaining; + pFlac->currentFLACFrame.pcmFramesRemaining = 0; + isMidFrame = DRFLAC_FALSE; + } + + /* The target can be a frame boundary, including EOF when the total frame count is unknown. */ + if (runningPCMFrameCount == pcmFrameIndex) { + return DRFLAC_TRUE; + } + } + + next_iteration: + /* Grab the next frame in preparation for the next iteration. */ + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + } +} + + +#if !defined(DR_FLAC_NO_CRC) +/* +We use an average compression ratio to determine our approximate start location. FLAC files are generally about 50%-70% the size of their +uncompressed counterparts so we'll use this as a basis. I'm going to split the middle and use a factor of 0.6 to determine the starting +location. +*/ +#define DRFLAC_BINARY_SEARCH_APPROX_COMPRESSION_RATIO 0.6f + +static drflac_bool32 drflac__read_and_decode_next_flac_frame_for_seek(drflac* pFlac) +{ + DRFLAC_ASSERT(pFlac != NULL); + + for (;;) { + drflac_result result; + + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + + result = drflac__decode_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + return DRFLAC_TRUE; + } + + if (result == DRFLAC_AT_END) { + return DRFLAC_FALSE; + } + } +} + +static drflac_bool32 drflac__seek_to_approximate_flac_frame_to_byte(drflac* pFlac, drflac_uint64 targetByte, drflac_uint64 rangeLo, drflac_uint64 rangeHi, drflac_uint64* pLastSuccessfulSeekOffset) +{ + DRFLAC_ASSERT(pFlac != NULL); + DRFLAC_ASSERT(pLastSuccessfulSeekOffset != NULL); + DRFLAC_ASSERT(targetByte >= rangeLo); + DRFLAC_ASSERT(targetByte <= rangeHi); + + *pLastSuccessfulSeekOffset = pFlac->firstFLACFramePosInBytes; + + for (;;) { + /* When seeking to a byte, failure probably means we've attempted to seek beyond the end of the stream. To counter this we just halve it each attempt. */ + if (!drflac__seek_to_byte(&pFlac->bs, targetByte)) { + /* If we couldn't even seek to the first byte in the stream we have a problem. Just abandon the whole thing. */ + if (targetByte == 0) { + drflac__seek_to_first_frame(pFlac); /* Try to recover. */ + return DRFLAC_FALSE; + } + + rangeHi = targetByte; + if (rangeHi <= rangeLo) { + return DRFLAC_FALSE; + } + + /* Halve the reduced byte range and continue. */ + targetByte = rangeLo + ((rangeHi - rangeLo)/2); + } else { + /* Getting here should mean that we have seeked to an appropriate byte. */ + + /* Clear the details of the FLAC frame so we don't misreport data. */ + DRFLAC_ZERO_MEMORY(&pFlac->currentFLACFrame, sizeof(pFlac->currentFLACFrame)); + + /* + Now seek to the next FLAC frame. We need to decode the entire frame (not just the header) because it's possible for the header to incorrectly pass the + CRC check and return bad data. We need to decode the entire frame to be more certain. Although this seems unlikely, this has happened to me in testing + so it needs to stay this way for now. + */ + #if 1 + { + if (!drflac__read_and_decode_next_flac_frame_for_seek(pFlac)) { + rangeHi = targetByte; + if (rangeHi <= rangeLo) { + return DRFLAC_FALSE; + } + + /* Halve the byte location and continue. */ + targetByte = rangeLo + ((rangeHi - rangeLo)/2); + } else { + break; + } + } + #else + { + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + /* Halve the byte location and continue. */ + targetByte = rangeLo + ((rangeHi - rangeLo)/2); + rangeHi = targetByte; + } else { + break; + } + } + #endif + } + } + + /* The current PCM frame needs to be updated based on the frame we just seeked to. */ + drflac__get_pcm_frame_range_of_current_flac_frame(pFlac, &pFlac->currentPCMFrame, NULL); + + DRFLAC_ASSERT(targetByte <= rangeHi); + + *pLastSuccessfulSeekOffset = targetByte; + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__decode_flac_frame_and_seek_forward_by_pcm_frames(drflac* pFlac, drflac_uint64 offset) +{ + /* This section of code would be used if we were only decoding the FLAC frame header when calling drflac__seek_to_approximate_flac_frame_to_byte(). */ +#if 0 + if (drflac__decode_flac_frame(pFlac) != DRFLAC_SUCCESS) { + /* We failed to decode this frame which may be due to it being corrupt. We'll just use the next valid FLAC frame. */ + if (drflac__read_and_decode_next_flac_frame(pFlac) == DRFLAC_FALSE) { + return DRFLAC_FALSE; + } + } +#endif + + return drflac__seek_forward_by_pcm_frames(pFlac, offset) == offset; +} + + +static drflac_bool32 drflac__seek_to_pcm_frame__binary_search_internal(drflac* pFlac, drflac_uint64 pcmFrameIndex, drflac_uint64 byteRangeLo, drflac_uint64 byteRangeHi) +{ + /* This assumes pFlac->currentPCMFrame is sitting on byteRangeLo upon entry. */ + + drflac_uint64 targetByte; + drflac_uint64 pcmRangeLo = pFlac->totalPCMFrameCount; + drflac_uint64 pcmRangeHi = 0; + drflac_uint64 lastSuccessfulSeekOffset = (drflac_uint64)-1; + drflac_uint64 closestSeekOffsetBeforeTargetPCMFrame = byteRangeLo; + drflac_uint32 seekForwardThreshold = (pFlac->maxBlockSizeInPCMFrames != 0) ? pFlac->maxBlockSizeInPCMFrames*2 : 4096; + + if (byteRangeLo > byteRangeHi || pFlac->currentPCMFrame > pcmFrameIndex) { + return DRFLAC_FALSE; + } + + /* Clamp the estimated offset before adding the lower bound to avoid wrapping the byte position. */ + targetByte = (drflac_uint64)(((drflac_int64)((pcmFrameIndex - pFlac->currentPCMFrame) * pFlac->channels * pFlac->bitsPerSample)/8.0f) * DRFLAC_BINARY_SEARCH_APPROX_COMPRESSION_RATIO); + if (targetByte > byteRangeHi - byteRangeLo) { + targetByte = byteRangeHi; + } else { + targetByte += byteRangeLo; + } + + for (;;) { + drflac_uint64 previousByteRangeLo; + drflac_uint64 previousByteRangeHi; + + previousByteRangeLo = byteRangeLo; + previousByteRangeHi = byteRangeHi; + + /* + If only two adjacent byte offsets remain, binary search cannot narrow the range any further. Seek to the closest frame before the target and decode + forward from there. + */ + if ((byteRangeHi - byteRangeLo) <= 1) { + if (!drflac__seek_to_approximate_flac_frame_to_byte(pFlac, closestSeekOffsetBeforeTargetPCMFrame, closestSeekOffsetBeforeTargetPCMFrame, byteRangeHi, &lastSuccessfulSeekOffset)) { + break; + } + + if (pFlac->currentPCMFrame <= pcmFrameIndex && drflac__decode_flac_frame_and_seek_forward_by_pcm_frames(pFlac, pcmFrameIndex - pFlac->currentPCMFrame)) { + return DRFLAC_TRUE; + } + + break; + } + + if (drflac__seek_to_approximate_flac_frame_to_byte(pFlac, targetByte, byteRangeLo, byteRangeHi, &lastSuccessfulSeekOffset)) { + /* We found a FLAC frame. We need to check if it contains the sample we're looking for. */ + drflac_uint64 newPCMRangeLo; + drflac_uint64 newPCMRangeHi; + drflac__get_pcm_frame_range_of_current_flac_frame(pFlac, &newPCMRangeLo, &newPCMRangeHi); + + /* If we selected the same frame, it means we should be pretty close. Just decode the rest. */ + if (pcmRangeLo == newPCMRangeLo) { + if (!drflac__seek_to_approximate_flac_frame_to_byte(pFlac, closestSeekOffsetBeforeTargetPCMFrame, closestSeekOffsetBeforeTargetPCMFrame, byteRangeHi, &lastSuccessfulSeekOffset)) { + break; /* Failed to seek to closest frame. */ + } + + if (drflac__decode_flac_frame_and_seek_forward_by_pcm_frames(pFlac, pcmFrameIndex - pFlac->currentPCMFrame)) { + return DRFLAC_TRUE; + } else { + break; /* Failed to seek forward. */ + } + } + + pcmRangeLo = newPCMRangeLo; + pcmRangeHi = newPCMRangeHi; + + if (pcmRangeLo <= pcmFrameIndex && pcmRangeHi >= pcmFrameIndex) { + /* The target PCM frame is in this FLAC frame. */ + if (drflac__decode_flac_frame_and_seek_forward_by_pcm_frames(pFlac, pcmFrameIndex - pFlac->currentPCMFrame) ) { + return DRFLAC_TRUE; + } else { + break; /* Failed to seek to FLAC frame. */ + } + } else { + if (pcmRangeLo > pcmFrameIndex) { + /* We seeked too far forward. We need to move our target byte backward and try again. */ + byteRangeHi = lastSuccessfulSeekOffset; + if (byteRangeLo > byteRangeHi) { + byteRangeLo = byteRangeHi; + } + + targetByte = byteRangeLo + ((byteRangeHi - byteRangeLo) / 2); + if (targetByte < byteRangeLo) { + targetByte = byteRangeLo; + } + } else /*if (pcmRangeHi < pcmFrameIndex)*/ { + /* We didn't seek far enough. We need to move our target byte forward and try again. */ + + /* If we're close enough we can just seek forward. */ + if ((pcmFrameIndex - pcmRangeLo) < seekForwardThreshold) { + if (drflac__decode_flac_frame_and_seek_forward_by_pcm_frames(pFlac, pcmFrameIndex - pFlac->currentPCMFrame)) { + return DRFLAC_TRUE; + } else { + break; /* Failed to seek to FLAC frame. */ + } + } else { + const double approxCompressionRatio = (drflac_int64)(lastSuccessfulSeekOffset - pFlac->firstFLACFramePosInBytes) / ((drflac_int64)(pcmRangeLo * pFlac->channels * pFlac->bitsPerSample)/8.0); + + byteRangeLo = lastSuccessfulSeekOffset; + if (byteRangeHi < byteRangeLo) { + byteRangeHi = byteRangeLo; + } + + targetByte = lastSuccessfulSeekOffset + (drflac_uint64)(((drflac_int64)((pcmFrameIndex-pcmRangeLo) * pFlac->channels * pFlac->bitsPerSample)/8.0) * approxCompressionRatio); + if (targetByte > byteRangeHi) { + targetByte = byteRangeHi; + } + + if (closestSeekOffsetBeforeTargetPCMFrame < lastSuccessfulSeekOffset) { + closestSeekOffsetBeforeTargetPCMFrame = lastSuccessfulSeekOffset; + } + } + } + } + + if (byteRangeLo == previousByteRangeLo && byteRangeHi == previousByteRangeHi) { + if (!drflac__seek_to_approximate_flac_frame_to_byte(pFlac, closestSeekOffsetBeforeTargetPCMFrame, closestSeekOffsetBeforeTargetPCMFrame, byteRangeHi, &lastSuccessfulSeekOffset)) { + break; + } + + if (pFlac->currentPCMFrame <= pcmFrameIndex && drflac__decode_flac_frame_and_seek_forward_by_pcm_frames(pFlac, pcmFrameIndex - pFlac->currentPCMFrame)) { + return DRFLAC_TRUE; + } + + break; + } + + if ((byteRangeHi - byteRangeLo) > 1 && (targetByte <= byteRangeLo || targetByte >= byteRangeHi)) { + targetByte = byteRangeLo + ((byteRangeHi - byteRangeLo) / 2); + } + } else { + /* Getting here is really bad. We just recover as best we can, but moving to the first frame in the stream, and then abort. */ + break; + } + } + + drflac__seek_to_first_frame(pFlac); /* <-- Try to recover. */ + return DRFLAC_FALSE; +} + +static drflac_bool32 drflac__seek_to_pcm_frame__binary_search(drflac* pFlac, drflac_uint64 pcmFrameIndex) +{ + drflac_uint64 byteRangeLo; + drflac_uint64 byteRangeHi; + drflac_uint32 seekForwardThreshold = (pFlac->maxBlockSizeInPCMFrames != 0) ? pFlac->maxBlockSizeInPCMFrames*2 : 4096; + + /* Our algorithm currently assumes the FLAC stream is currently sitting at the start. */ + if (drflac__seek_to_first_frame(pFlac) == DRFLAC_FALSE) { + return DRFLAC_FALSE; + } + + /* If we're close enough to the start, just move to the start and seek forward. */ + if (pcmFrameIndex < seekForwardThreshold) { + return drflac__seek_forward_by_pcm_frames(pFlac, pcmFrameIndex) == pcmFrameIndex; + } + + /* + Our starting byte range is the byte position of the first FLAC frame and the approximate end of the file as if it were completely uncompressed. This ensures + the entire file is included, even though most of the time it'll exceed the end of the actual stream. This is OK as the frame searching logic will handle it. + */ + byteRangeLo = pFlac->firstFLACFramePosInBytes; + byteRangeHi = pFlac->firstFLACFramePosInBytes + (drflac_uint64)((drflac_int64)(pFlac->totalPCMFrameCount * pFlac->channels * pFlac->bitsPerSample)/8.0f); + + return drflac__seek_to_pcm_frame__binary_search_internal(pFlac, pcmFrameIndex, byteRangeLo, byteRangeHi); +} +#endif /* !DR_FLAC_NO_CRC */ + +static drflac_bool32 drflac__seek_to_pcm_frame__seek_table(drflac* pFlac, drflac_uint64 pcmFrameIndex) +{ + drflac_uint32 iClosestSeekpoint = 0; + drflac_bool32 isMidFrame = DRFLAC_FALSE; + drflac_uint64 runningPCMFrameCount; + drflac_uint32 iSeekpoint; + + + DRFLAC_ASSERT(pFlac != NULL); + + if (pFlac->pSeekpoints == NULL || pFlac->seekpointCount == 0) { + return DRFLAC_FALSE; + } + + /* Do not use the seektable if pcmFramIndex is not coverd by it. */ + if (pFlac->pSeekpoints[0].firstPCMFrame > pcmFrameIndex) { + return DRFLAC_FALSE; + } + + for (iSeekpoint = 0; iSeekpoint < pFlac->seekpointCount; ++iSeekpoint) { + if (pFlac->pSeekpoints[iSeekpoint].firstPCMFrame >= pcmFrameIndex) { + break; + } + + iClosestSeekpoint = iSeekpoint; + } + + /* There's been cases where the seek table contains only zeros. We need to do some basic validation on the closest seekpoint. */ + if (pFlac->pSeekpoints[iClosestSeekpoint].pcmFrameCount == 0 || pFlac->pSeekpoints[iClosestSeekpoint].pcmFrameCount > pFlac->maxBlockSizeInPCMFrames) { + return DRFLAC_FALSE; + } + if (pFlac->pSeekpoints[iClosestSeekpoint].firstPCMFrame > pFlac->totalPCMFrameCount && pFlac->totalPCMFrameCount > 0) { + return DRFLAC_FALSE; + } + + if (pFlac->pSeekpoints[iClosestSeekpoint].flacFrameOffset > (drflac_uint64)-1 - pFlac->firstFLACFramePosInBytes) { + return DRFLAC_FALSE; + } + +#if !defined(DR_FLAC_NO_CRC) + /* At this point we should know the closest seek point. We can use a binary search for this. We need to know the total sample count for this. */ + if (pFlac->totalPCMFrameCount > 0) { + drflac_uint64 byteRangeLo; + drflac_uint64 byteRangeHi; + + byteRangeHi = pFlac->firstFLACFramePosInBytes + (drflac_uint64)((drflac_int64)(pFlac->totalPCMFrameCount * pFlac->channels * pFlac->bitsPerSample)/8.0f); + byteRangeLo = pFlac->firstFLACFramePosInBytes + pFlac->pSeekpoints[iClosestSeekpoint].flacFrameOffset; + + /* + If our closest seek point is not the last one, we only need to search between it and the next one. The section below calculates an appropriate starting + value for byteRangeHi which will clamp it appropriately. + + Note that the next seekpoint must have an offset greater than the closest seekpoint because otherwise our binary search algorithm will break down. There + have been cases where a seektable consists of seek points where every byte offset is set to 0 which causes problems. If this happens we need to abort. + */ + if (iClosestSeekpoint < pFlac->seekpointCount-1) { + drflac_uint32 iNextSeekpoint = iClosestSeekpoint + 1; + + if (pFlac->pSeekpoints[iNextSeekpoint].firstPCMFrame != (((drflac_uint64)0xFFFFFFFF << 32) | 0xFFFFFFFF)) { /* Make sure it's not a placeholder seekpoint. */ + /* Basic validation on the seekpoints to ensure they're usable. */ + if (pFlac->pSeekpoints[iClosestSeekpoint].flacFrameOffset >= pFlac->pSeekpoints[iNextSeekpoint].flacFrameOffset || pFlac->pSeekpoints[iNextSeekpoint].pcmFrameCount == 0) { + return DRFLAC_FALSE; /* The next seekpoint doesn't look right. The seek table cannot be trusted from here. Abort. */ + } + + if (pFlac->pSeekpoints[iNextSeekpoint].flacFrameOffset > (drflac_uint64)-1 - pFlac->firstFLACFramePosInBytes) { + return DRFLAC_FALSE; + } + + byteRangeHi = pFlac->firstFLACFramePosInBytes + pFlac->pSeekpoints[iNextSeekpoint].flacFrameOffset - 1; /* byteRangeHi must be zero based. */ + } + } + + if (byteRangeLo > byteRangeHi) { + return DRFLAC_FALSE; + } + + if (drflac__seek_to_byte(&pFlac->bs, pFlac->firstFLACFramePosInBytes + pFlac->pSeekpoints[iClosestSeekpoint].flacFrameOffset)) { + /* The cached samples no longer describe the frame at this byte position. */ + DRFLAC_ZERO_MEMORY(&pFlac->currentFLACFrame, sizeof(pFlac->currentFLACFrame)); + + if (drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + drflac__get_pcm_frame_range_of_current_flac_frame(pFlac, &pFlac->currentPCMFrame, NULL); + + if (pFlac->currentPCMFrame > pcmFrameIndex) { + return DRFLAC_FALSE; /* The seekpoint does not precede the target as expected. */ + } + + if (drflac__seek_to_pcm_frame__binary_search_internal(pFlac, pcmFrameIndex, byteRangeLo, byteRangeHi)) { + return DRFLAC_TRUE; + } + } + } + } +#endif /* !DR_FLAC_NO_CRC */ + + /* Getting here means we need to use a slower algorithm because the binary search method failed or cannot be used. */ + + /* + If we are seeking forward and the closest seekpoint is _before_ the current sample, we just seek forward from where we are. Otherwise we start seeking + from the seekpoint's first sample. + */ + if (pcmFrameIndex >= pFlac->currentPCMFrame && pFlac->pSeekpoints[iClosestSeekpoint].firstPCMFrame <= pFlac->currentPCMFrame) { + /* Optimized case. Just seek forward from where we are. */ + runningPCMFrameCount = pFlac->currentPCMFrame; + + /* The frame header for the first frame may not yet have been read. We need to do that if necessary. */ + if (pFlac->currentPCMFrame == 0 && pFlac->currentFLACFrame.pcmFramesRemaining == 0) { + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + } else { + isMidFrame = DRFLAC_TRUE; + } + } else { + /* Slower case. Seek to the start of the seekpoint and then seek forward from there. */ + runningPCMFrameCount = pFlac->pSeekpoints[iClosestSeekpoint].firstPCMFrame; + + if (!drflac__seek_to_byte(&pFlac->bs, pFlac->firstFLACFramePosInBytes + pFlac->pSeekpoints[iClosestSeekpoint].flacFrameOffset)) { + return DRFLAC_FALSE; + } + + /* Grab the frame the seekpoint is sitting on in preparation for the sample-exact seeking below. */ + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + } + + for (;;) { + drflac_uint64 pcmFrameCountInThisFLACFrame; + drflac_uint64 firstPCMFrameInFLACFrame = 0; + drflac_uint64 lastPCMFrameInFLACFrame = 0; + + drflac__get_pcm_frame_range_of_current_flac_frame(pFlac, &firstPCMFrameInFLACFrame, &lastPCMFrameInFLACFrame); + + pcmFrameCountInThisFLACFrame = (lastPCMFrameInFLACFrame - firstPCMFrameInFLACFrame) + 1; + if (pcmFrameIndex < (runningPCMFrameCount + pcmFrameCountInThisFLACFrame)) { + /* + The sample should be in this frame. We need to fully decode it, but if it's an invalid frame (a CRC mismatch) we need to pretend + it never existed and keep iterating. + */ + drflac_uint64 pcmFramesToDecode = pcmFrameIndex - runningPCMFrameCount; + + if (!isMidFrame) { + drflac_result result = drflac__decode_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + /* The frame is valid. We just need to skip over some samples to ensure it's sample-exact. */ + return drflac__seek_forward_by_pcm_frames(pFlac, pcmFramesToDecode) == pcmFramesToDecode; /* <-- If this fails, something bad has happened (it should never fail). */ + } else { + if (result == DRFLAC_CRC_MISMATCH) { + goto next_iteration; /* CRC mismatch. Pretend this frame never existed. */ + } else { + return DRFLAC_FALSE; + } + } + } else { + /* We started seeking mid-frame which means we need to skip the frame decoding part. */ + return drflac__seek_forward_by_pcm_frames(pFlac, pcmFramesToDecode) == pcmFramesToDecode; + } + } else { + /* + It's not in this frame. We need to seek past the frame, but check if there was a CRC mismatch. If so, we pretend this + frame never existed and leave the running sample count untouched. + */ + if (!isMidFrame) { + drflac_result result = drflac__seek_to_next_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + runningPCMFrameCount += pcmFrameCountInThisFLACFrame; + } else { + if (result == DRFLAC_CRC_MISMATCH) { + goto next_iteration; /* CRC mismatch. Pretend this frame never existed. */ + } else { + return DRFLAC_FALSE; + } + } + } else { + /* + We started seeking mid-frame which means we need to seek by reading to the end of the frame instead of with + drflac__seek_to_next_flac_frame() which only works if the decoder is sitting on the byte just after the frame header. + */ + runningPCMFrameCount += pFlac->currentFLACFrame.pcmFramesRemaining; + pFlac->currentFLACFrame.pcmFramesRemaining = 0; + isMidFrame = DRFLAC_FALSE; + } + + /* The target can be a frame boundary, including EOF when the total frame count is unknown. */ + if (runningPCMFrameCount == pcmFrameIndex) { + return DRFLAC_TRUE; + } + } + + next_iteration: + /* Grab the next frame in preparation for the next iteration. */ + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + } +} + + +#ifndef DR_FLAC_NO_OGG +typedef struct +{ + drflac_uint8 capturePattern[4]; /* Should be "OggS" */ + drflac_uint8 structureVersion; /* Always 0. */ + drflac_uint8 headerType; + drflac_uint64 granulePosition; + drflac_uint32 serialNumber; + drflac_uint32 sequenceNumber; + drflac_uint32 checksum; + drflac_uint8 segmentCount; + drflac_uint8 segmentTable[255]; +} drflac_ogg_page_header; +#endif + +typedef struct +{ + drflac_read_proc onRead; + drflac_seek_proc onSeek; + drflac_tell_proc onTell; + drflac_meta_proc onMeta; + drflac_container container; + void* pUserData; + void* pUserDataMD; + drflac_uint32 sampleRate; + drflac_uint8 channels; + drflac_uint8 bitsPerSample; + drflac_uint64 totalPCMFrameCount; + drflac_uint16 maxBlockSizeInPCMFrames; + drflac_uint64 runningFilePos; + drflac_bool32 hasStreamInfoBlock; + drflac_bool32 hasMetadataBlocks; + drflac_bs bs; /* <-- A bit streamer is required for loading data during initialization. */ + drflac_frame_header firstFrameHeader; /* <-- The header of the first frame that was read during relaxed initalization. Only set if there is no STREAMINFO block. */ + +#ifndef DR_FLAC_NO_OGG + drflac_uint32 oggSerial; + drflac_uint64 oggFirstBytePos; + drflac_ogg_page_header oggBosHeader; +#endif +} drflac_init_info; + +static DRFLAC_INLINE void drflac__decode_block_header(drflac_uint32 blockHeader, drflac_uint8* isLastBlock, drflac_uint8* blockType, drflac_uint32* blockSize) +{ + blockHeader = drflac__be2host_32(blockHeader); + *isLastBlock = (drflac_uint8)((blockHeader & 0x80000000UL) >> 31); + *blockType = (drflac_uint8)((blockHeader & 0x7F000000UL) >> 24); + *blockSize = (blockHeader & 0x00FFFFFFUL); +} + +static DRFLAC_INLINE drflac_bool32 drflac__read_and_decode_block_header(drflac_read_proc onRead, void* pUserData, drflac_uint8* isLastBlock, drflac_uint8* blockType, drflac_uint32* blockSize) +{ + drflac_uint32 blockHeader; + + *blockSize = 0; + if (onRead(pUserData, &blockHeader, 4) != 4) { + return DRFLAC_FALSE; + } + + drflac__decode_block_header(blockHeader, isLastBlock, blockType, blockSize); + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__read_streaminfo(drflac_read_proc onRead, void* pUserData, drflac_streaminfo* pStreamInfo) +{ + drflac_uint32 blockSizes; + drflac_uint64 frameSizes = 0; + drflac_uint64 importantProps; + drflac_uint8 md5[16]; + + /* min/max block size. */ + if (onRead(pUserData, &blockSizes, 4) != 4) { + return DRFLAC_FALSE; + } + + /* min/max frame size. */ + if (onRead(pUserData, &frameSizes, 6) != 6) { + return DRFLAC_FALSE; + } + + /* Sample rate, channels, bits per sample and total sample count. */ + if (onRead(pUserData, &importantProps, 8) != 8) { + return DRFLAC_FALSE; + } + + /* MD5 */ + if (onRead(pUserData, md5, sizeof(md5)) != sizeof(md5)) { + return DRFLAC_FALSE; + } + + blockSizes = drflac__be2host_32(blockSizes); + frameSizes = drflac__be2host_64(frameSizes); + importantProps = drflac__be2host_64(importantProps); + + pStreamInfo->minBlockSizeInPCMFrames = (drflac_uint16)((blockSizes & 0xFFFF0000) >> 16); + pStreamInfo->maxBlockSizeInPCMFrames = (drflac_uint16) (blockSizes & 0x0000FFFF); + pStreamInfo->minFrameSizeInPCMFrames = (drflac_uint32)((frameSizes & (((drflac_uint64)0x00FFFFFF << 16) << 24)) >> 40); + pStreamInfo->maxFrameSizeInPCMFrames = (drflac_uint32)((frameSizes & (((drflac_uint64)0x00FFFFFF << 16) << 0)) >> 16); + pStreamInfo->sampleRate = (drflac_uint32)((importantProps & (((drflac_uint64)0x000FFFFF << 16) << 28)) >> 44); + pStreamInfo->channels = (drflac_uint8 )((importantProps & (((drflac_uint64)0x0000000E << 16) << 24)) >> 41) + 1; + pStreamInfo->bitsPerSample = (drflac_uint8 )((importantProps & (((drflac_uint64)0x0000001F << 16) << 20)) >> 36) + 1; + pStreamInfo->totalPCMFrameCount = ((importantProps & ((((drflac_uint64)0x0000000F << 16) << 16) | 0xFFFFFFFF))); + DRFLAC_COPY_MEMORY(pStreamInfo->md5, md5, sizeof(md5)); + + return DRFLAC_TRUE; +} + + +static void* drflac__malloc_default(size_t sz, void* pUserData) +{ + (void)pUserData; + return DRFLAC_MALLOC(sz); +} + +static void* drflac__realloc_default(void* p, size_t sz, void* pUserData) +{ + (void)pUserData; + return DRFLAC_REALLOC(p, sz); +} + +static void drflac__free_default(void* p, void* pUserData) +{ + (void)pUserData; + DRFLAC_FREE(p); +} + + +static void* drflac__malloc_from_callbacks(size_t sz, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + if (pAllocationCallbacks == NULL) { + return NULL; + } + + if (pAllocationCallbacks->onMalloc != NULL) { + return pAllocationCallbacks->onMalloc(sz, pAllocationCallbacks->pUserData); + } + + /* Try using realloc(). */ + if (pAllocationCallbacks->onRealloc != NULL) { + return pAllocationCallbacks->onRealloc(NULL, sz, pAllocationCallbacks->pUserData); + } + + return NULL; +} + +static void* drflac__realloc_from_callbacks(void* p, size_t szNew, size_t szOld, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + if (pAllocationCallbacks == NULL) { + return NULL; + } + + if (pAllocationCallbacks->onRealloc != NULL) { + return pAllocationCallbacks->onRealloc(p, szNew, pAllocationCallbacks->pUserData); + } + + /* Try emulating realloc() in terms of malloc()/free(). */ + if (pAllocationCallbacks->onMalloc != NULL && pAllocationCallbacks->onFree != NULL) { + void* p2; + + p2 = pAllocationCallbacks->onMalloc(szNew, pAllocationCallbacks->pUserData); + if (p2 == NULL) { + return NULL; + } + + if (p != NULL) { + DRFLAC_COPY_MEMORY(p2, p, DRFLAC_MIN(szNew, szOld)); + pAllocationCallbacks->onFree(p, pAllocationCallbacks->pUserData); + } + + return p2; + } + + return NULL; +} + +static void drflac__free_from_callbacks(void* p, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + if (p == NULL || pAllocationCallbacks == NULL) { + return; + } + + if (pAllocationCallbacks->onFree != NULL) { + pAllocationCallbacks->onFree(p, pAllocationCallbacks->pUserData); + } +} + + +static drflac_bool32 drflac__read_and_decode_metadata(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_meta_proc onMeta, void* pUserData, void* pUserDataMD, drflac_uint64* pFirstFramePos, drflac_uint64* pSeektablePos, drflac_uint32* pSeekpointCount, drflac_allocation_callbacks* pAllocationCallbacks) +{ + /* + We want to keep track of the byte position in the stream of the seektable. At the time of calling this function we know that + we'll be sitting on byte 42. + */ + drflac_uint64 runningFilePos = 42; + drflac_uint64 seektablePos = 0; + drflac_uint32 seektableSize = 0; + drflac_int64 fileSize = 0; + drflac_bool32 hasKnownFileSize = DRFLAC_FALSE; + + /* We'll be doing some memory allocations here against untrusted data. We'll do a basic validation check that they don't exceed the size of the file. */ + if (onTell != NULL && onSeek != NULL) { + if (onSeek(pUserData, 0, DRFLAC_SEEK_END)) { + if (onTell(pUserData, &fileSize)) { + hasKnownFileSize = DRFLAC_TRUE; + } + + onSeek(pUserData, (int)runningFilePos, DRFLAC_SEEK_SET); /* Safe cast because runningFilePos should always be 42 at this point. */ + } + } + + for (;;) { + drflac_metadata metadata; + drflac_uint8 isLastBlock = 0; + drflac_uint8 blockType = 0; + drflac_uint32 blockSize; + if (drflac__read_and_decode_block_header(onRead, pUserData, &isLastBlock, &blockType, &blockSize) == DRFLAC_FALSE) { + return DRFLAC_FALSE; + } + + if (hasKnownFileSize && (blockSize > ((drflac_uint64)fileSize - runningFilePos))) { + return DRFLAC_FALSE; /* Block size exceeds the size of the file. */ + } + + runningFilePos += 4; + + metadata.type = blockType; + metadata.rawDataSize = 0; + metadata.rawDataOffset = runningFilePos; + metadata.pRawData = NULL; + + switch (blockType) + { + case DRFLAC_METADATA_BLOCK_TYPE_APPLICATION: + { + if (blockSize < 4) { + return DRFLAC_FALSE; + } + + if (onMeta) { + void* pRawData = drflac__malloc_from_callbacks(blockSize, pAllocationCallbacks); + if (pRawData == NULL) { + return DRFLAC_FALSE; + } + + if (onRead(pUserData, pRawData, blockSize) != blockSize) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + + metadata.pRawData = pRawData; + metadata.rawDataSize = blockSize; + metadata.data.application.id = drflac__be2host_32(*(drflac_uint32*)pRawData); + metadata.data.application.pData = (const void*)((drflac_uint8*)pRawData + sizeof(drflac_uint32)); + metadata.data.application.dataSize = blockSize - sizeof(drflac_uint32); + onMeta(pUserDataMD, &metadata); + + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + } + } break; + + case DRFLAC_METADATA_BLOCK_TYPE_SEEKTABLE: + { + seektablePos = runningFilePos; + seektableSize = blockSize; + + if (onMeta) { + drflac_uint32 seekpointCount; + drflac_uint32 iSeekpoint; + void* pRawData; + size_t rawDataSize; + + seekpointCount = blockSize/DRFLAC_SEEKPOINT_SIZE_IN_BYTES; + rawDataSize = seekpointCount * sizeof(drflac_seekpoint); + + pRawData = drflac__malloc_from_callbacks(rawDataSize, pAllocationCallbacks); + if (pRawData == NULL) { + return DRFLAC_FALSE; + } + + /* We need to read seekpoint by seekpoint and do some processing. */ + for (iSeekpoint = 0; iSeekpoint < seekpointCount; ++iSeekpoint) { + drflac_seekpoint* pSeekpoint = (drflac_seekpoint*)pRawData + iSeekpoint; + + if (onRead(pUserData, pSeekpoint, DRFLAC_SEEKPOINT_SIZE_IN_BYTES) != DRFLAC_SEEKPOINT_SIZE_IN_BYTES) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + + /* Endian swap. */ + pSeekpoint->firstPCMFrame = drflac__be2host_64(pSeekpoint->firstPCMFrame); + pSeekpoint->flacFrameOffset = drflac__be2host_64(pSeekpoint->flacFrameOffset); + pSeekpoint->pcmFrameCount = drflac__be2host_16(pSeekpoint->pcmFrameCount); + } + + metadata.pRawData = pRawData; + metadata.rawDataSize = (drflac_uint32)rawDataSize; + metadata.data.seektable.seekpointCount = seekpointCount; + metadata.data.seektable.pSeekpoints = (const drflac_seekpoint*)pRawData; + + onMeta(pUserDataMD, &metadata); + + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + } + } break; + + case DRFLAC_METADATA_BLOCK_TYPE_VORBIS_COMMENT: + { + if (blockSize < 8) { + return DRFLAC_FALSE; + } + + if (onMeta) { + void* pRawData; + const char* pRunningData; + const char* pRunningDataEnd; + drflac_uint32 i; + + pRawData = drflac__malloc_from_callbacks(blockSize, pAllocationCallbacks); + if (pRawData == NULL) { + return DRFLAC_FALSE; + } + + if (onRead(pUserData, pRawData, blockSize) != blockSize) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + + metadata.pRawData = pRawData; + metadata.rawDataSize = blockSize; + + pRunningData = (const char*)pRawData; + pRunningDataEnd = (const char*)pRawData + blockSize; + + metadata.data.vorbis_comment.vendorLength = drflac__le2host_32_ptr_unaligned(pRunningData); pRunningData += 4; + + /* Need space for the rest of the block */ + if ((pRunningDataEnd - pRunningData) - 4 < (drflac_int64)metadata.data.vorbis_comment.vendorLength) { /* <-- Note the order of operations to avoid overflow to a valid value */ + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + metadata.data.vorbis_comment.vendor = pRunningData; pRunningData += metadata.data.vorbis_comment.vendorLength; + metadata.data.vorbis_comment.commentCount = drflac__le2host_32_ptr_unaligned(pRunningData); pRunningData += 4; + + /* Need space for 'commentCount' comments after the block, which at minimum is a drflac_uint32 per comment */ + if ((pRunningDataEnd - pRunningData) / sizeof(drflac_uint32) < metadata.data.vorbis_comment.commentCount) { /* <-- Note the order of operations to avoid overflow to a valid value */ + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + metadata.data.vorbis_comment.pComments = pRunningData; + + /* Check that the comments section is valid before passing it to the callback */ + for (i = 0; i < metadata.data.vorbis_comment.commentCount; ++i) { + drflac_uint32 commentLength; + + if (pRunningDataEnd - pRunningData < 4) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + + commentLength = drflac__le2host_32_ptr_unaligned(pRunningData); pRunningData += 4; + if (pRunningDataEnd - pRunningData < (drflac_int64)commentLength) { /* <-- Note the order of operations to avoid overflow to a valid value */ + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + pRunningData += commentLength; + } + + onMeta(pUserDataMD, &metadata); + + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + } + } break; + + case DRFLAC_METADATA_BLOCK_TYPE_CUESHEET: + { + if (blockSize < 396) { + return DRFLAC_FALSE; + } + + if (onMeta) { + void* pRawData; + const char* pRunningData; + const char* pRunningDataEnd; + size_t bufferSize; + drflac_uint8 iTrack; + drflac_uint8 iIndex; + void* pTrackData; + + /* + This needs to be loaded in two passes. The first pass is used to calculate the size of the memory allocation + we need for storing the necessary data. The second pass will fill that buffer with usable data. + */ + pRawData = drflac__malloc_from_callbacks(blockSize, pAllocationCallbacks); + if (pRawData == NULL) { + return DRFLAC_FALSE; + } + + if (onRead(pUserData, pRawData, blockSize) != blockSize) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + + metadata.pRawData = pRawData; + metadata.rawDataSize = blockSize; + + pRunningData = (const char*)pRawData; + pRunningDataEnd = (const char*)pRawData + blockSize; + + DRFLAC_COPY_MEMORY(metadata.data.cuesheet.catalog, pRunningData, 128); pRunningData += 128; + metadata.data.cuesheet.leadInSampleCount = drflac__be2host_64(*(const drflac_uint64*)pRunningData); pRunningData += 8; + metadata.data.cuesheet.isCD = (pRunningData[0] & 0x80) != 0; pRunningData += 259; + metadata.data.cuesheet.trackCount = pRunningData[0]; pRunningData += 1; + metadata.data.cuesheet.pTrackData = NULL; /* Will be filled later. */ + + /* Pass 1: Calculate the size of the buffer for the track data. */ + { + const char* pRunningDataSaved = pRunningData; /* Will be restored at the end in preparation for the second pass. */ + + bufferSize = metadata.data.cuesheet.trackCount * DRFLAC_CUESHEET_TRACK_PADDED_SIZE_IN_BYTES; + + for (iTrack = 0; iTrack < metadata.data.cuesheet.trackCount; ++iTrack) { + drflac_uint8 indexCount; + drflac_uint32 indexPointSize; + + if (pRunningDataEnd - pRunningData < DRFLAC_CUESHEET_TRACK_SIZE_IN_BYTES) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + + /* Skip to the index point count */ + pRunningData += 35; + + indexCount = pRunningData[0]; + pRunningData += 1; + + bufferSize += indexCount * sizeof(drflac_cuesheet_track_index); + + /* Quick validation check. */ + indexPointSize = indexCount * DRFLAC_CUESHEET_TRACK_INDEX_SIZE_IN_BYTES; + if (pRunningDataEnd - pRunningData < (drflac_int64)indexPointSize) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + + pRunningData += indexPointSize; + } + + pRunningData = pRunningDataSaved; + } + + /* Pass 2: Allocate a buffer and fill the data. Validation was done in the step above so can be skipped. */ + { + char* pRunningTrackData; + + pTrackData = drflac__malloc_from_callbacks(bufferSize, pAllocationCallbacks); + if (pTrackData == NULL) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + + pRunningTrackData = (char*)pTrackData; + + for (iTrack = 0; iTrack < metadata.data.cuesheet.trackCount; ++iTrack) { + drflac_uint8 indexCount; + + DRFLAC_COPY_MEMORY(pRunningTrackData, pRunningData, DRFLAC_CUESHEET_TRACK_SIZE_IN_BYTES); + pRunningData += DRFLAC_CUESHEET_TRACK_SIZE_IN_BYTES-1; /* Skip forward, but not beyond the last byte in the CUESHEET_TRACK block which is the index count. */ + pRunningTrackData += DRFLAC_CUESHEET_TRACK_SIZE_IN_BYTES-1; + + /* Grab the index count for the next part. */ + indexCount = pRunningData[0]; + pRunningData += 1; + pRunningTrackData += 1 + (DRFLAC_CUESHEET_TRACK_PADDED_SIZE_IN_BYTES - DRFLAC_CUESHEET_TRACK_SIZE_IN_BYTES); + + /* Extract each track index. */ + for (iIndex = 0; iIndex < indexCount; ++iIndex) { + drflac_cuesheet_track_index* pTrackIndex = (drflac_cuesheet_track_index*)pRunningTrackData; + + DRFLAC_COPY_MEMORY(pRunningTrackData, pRunningData, DRFLAC_CUESHEET_TRACK_INDEX_SIZE_IN_BYTES); + pRunningData += DRFLAC_CUESHEET_TRACK_INDEX_SIZE_IN_BYTES; + pRunningTrackData += sizeof(drflac_cuesheet_track_index); + + pTrackIndex->offset = drflac__be2host_64(pTrackIndex->offset); + } + } + + metadata.data.cuesheet.pTrackData = pTrackData; + } + + onMeta(pUserDataMD, &metadata); + + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + pRawData = NULL; + drflac__free_from_callbacks(pTrackData, pAllocationCallbacks); + pTrackData = NULL; + } + } break; + + case DRFLAC_METADATA_BLOCK_TYPE_PICTURE: + { + if (blockSize < 32) { + return DRFLAC_FALSE; + } + + if (onMeta) { + drflac_bool32 result = DRFLAC_TRUE; + drflac_uint32 blockSizeRemaining = blockSize; + char* pMime = NULL; + char* pDescription = NULL; + void* pPictureData = NULL; + + if (blockSizeRemaining < 4 || onRead(pUserData, &metadata.data.picture.type, 4) != 4) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= 4; + metadata.data.picture.type = drflac__be2host_32(metadata.data.picture.type); + + + if (blockSizeRemaining < 4 || onRead(pUserData, &metadata.data.picture.mimeLength, 4) != 4) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= 4; + metadata.data.picture.mimeLength = drflac__be2host_32(metadata.data.picture.mimeLength); + + if (blockSizeRemaining < metadata.data.picture.mimeLength) { + result = DRFLAC_FALSE; + goto done_flac; + } + + pMime = (char*)drflac__malloc_from_callbacks(metadata.data.picture.mimeLength + 1, pAllocationCallbacks); /* +1 for null terminator. */ + if (pMime == NULL) { + result = DRFLAC_FALSE; + goto done_flac; + } + + if (onRead(pUserData, pMime, metadata.data.picture.mimeLength) != metadata.data.picture.mimeLength) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= metadata.data.picture.mimeLength; + pMime[metadata.data.picture.mimeLength] = '\0'; /* Null terminate for safety. */ + metadata.data.picture.mime = (const char*)pMime; + + + if (blockSizeRemaining < 4 || onRead(pUserData, &metadata.data.picture.descriptionLength, 4) != 4) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= 4; + metadata.data.picture.descriptionLength = drflac__be2host_32(metadata.data.picture.descriptionLength); + + if (blockSizeRemaining < metadata.data.picture.descriptionLength) { + result = DRFLAC_FALSE; + goto done_flac; + } + + pDescription = (char*)drflac__malloc_from_callbacks(metadata.data.picture.descriptionLength + 1, pAllocationCallbacks); /* +1 for null terminator. */ + if (pDescription == NULL) { + result = DRFLAC_FALSE; + goto done_flac; + } + + if (onRead(pUserData, pDescription, metadata.data.picture.descriptionLength) != metadata.data.picture.descriptionLength) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= metadata.data.picture.descriptionLength; + pDescription[metadata.data.picture.descriptionLength] = '\0'; /* Null terminate for safety. */ + metadata.data.picture.description = (const char*)pDescription; + + + if (blockSizeRemaining < 4 || onRead(pUserData, &metadata.data.picture.width, 4) != 4) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= 4; + metadata.data.picture.width = drflac__be2host_32(metadata.data.picture.width); + + if (blockSizeRemaining < 4 || onRead(pUserData, &metadata.data.picture.height, 4) != 4) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= 4; + metadata.data.picture.height = drflac__be2host_32(metadata.data.picture.height); + + if (blockSizeRemaining < 4 || onRead(pUserData, &metadata.data.picture.colorDepth, 4) != 4) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= 4; + metadata.data.picture.colorDepth = drflac__be2host_32(metadata.data.picture.colorDepth); + + if (blockSizeRemaining < 4 || onRead(pUserData, &metadata.data.picture.indexColorCount, 4) != 4) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= 4; + metadata.data.picture.indexColorCount = drflac__be2host_32(metadata.data.picture.indexColorCount); + + + /* Picture data. */ + if (blockSizeRemaining < 4 || onRead(pUserData, &metadata.data.picture.pictureDataSize, 4) != 4) { + result = DRFLAC_FALSE; + goto done_flac; + } + blockSizeRemaining -= 4; + metadata.data.picture.pictureDataSize = drflac__be2host_32(metadata.data.picture.pictureDataSize); + + if (blockSizeRemaining < metadata.data.picture.pictureDataSize) { + result = DRFLAC_FALSE; + goto done_flac; + } + + /* For the actual image data we want to store the offset to the start of the stream. */ + metadata.data.picture.pictureDataOffset = runningFilePos + (blockSize - blockSizeRemaining); + + /* + For the allocation of image data, we can allow memory allocation to fail, in which case we just leave + the pointer as null. If it fails, we need to fall back to seeking past the image data. + */ + #ifndef DR_FLAC_NO_PICTURE_METADATA_MALLOC + pPictureData = drflac__malloc_from_callbacks(metadata.data.picture.pictureDataSize, pAllocationCallbacks); + if (pPictureData != NULL) { + if (onRead(pUserData, pPictureData, metadata.data.picture.pictureDataSize) != metadata.data.picture.pictureDataSize) { + result = DRFLAC_FALSE; + goto done_flac; + } + } else + #endif + { + /* Allocation failed. We need to seek past the picture data. */ + if (!onSeek(pUserData, metadata.data.picture.pictureDataSize, DRFLAC_SEEK_CUR)) { + result = DRFLAC_FALSE; + goto done_flac; + } + } + + blockSizeRemaining -= metadata.data.picture.pictureDataSize; + (void)blockSizeRemaining; + + metadata.data.picture.pPictureData = (const drflac_uint8*)pPictureData; + + + /* Only fire the callback if we actually have a way to read the image data. We must have either a valid offset, or a valid data pointer. */ + if (metadata.data.picture.pictureDataOffset != 0 || metadata.data.picture.pPictureData != NULL) { + onMeta(pUserDataMD, &metadata); + } else { + /* Don't have a valid offset or data pointer, so just pretend we don't have a picture metadata. */ + } + + done_flac: + drflac__free_from_callbacks(pMime, pAllocationCallbacks); + drflac__free_from_callbacks(pDescription, pAllocationCallbacks); + drflac__free_from_callbacks(pPictureData, pAllocationCallbacks); + + if (result != DRFLAC_TRUE) { + return DRFLAC_FALSE; + } + } + } break; + + case DRFLAC_METADATA_BLOCK_TYPE_PADDING: + { + if (onMeta) { + metadata.data.padding.unused = 0; + + /* Padding doesn't have anything meaningful in it, so just skip over it, but make sure the caller is aware of it by firing the callback. */ + if (!onSeek(pUserData, blockSize, DRFLAC_SEEK_CUR)) { + isLastBlock = DRFLAC_TRUE; /* An error occurred while seeking. Attempt to recover by treating this as the last block which will in turn terminate the loop. */ + } else { + onMeta(pUserDataMD, &metadata); + } + } + } break; + + case DRFLAC_METADATA_BLOCK_TYPE_INVALID: + { + /* Invalid chunk. Just skip over this one. */ + if (onMeta) { + if (!onSeek(pUserData, blockSize, DRFLAC_SEEK_CUR)) { + isLastBlock = DRFLAC_TRUE; /* An error occurred while seeking. Attempt to recover by treating this as the last block which will in turn terminate the loop. */ + } + } + } break; + + default: + { + /* + It's an unknown chunk, but not necessarily invalid. There's a chance more metadata blocks might be defined later on, so we + can at the very least report the chunk to the application and let it look at the raw data. + */ + if (onMeta) { + void* pRawData = drflac__malloc_from_callbacks(blockSize, pAllocationCallbacks); + if (pRawData != NULL) { + if (onRead(pUserData, pRawData, blockSize) != blockSize) { + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + return DRFLAC_FALSE; + } + } else { + /* Allocation failed. We need to seek past the block. */ + if (!onSeek(pUserData, blockSize, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + } + + metadata.pRawData = pRawData; + metadata.rawDataSize = blockSize; + onMeta(pUserDataMD, &metadata); + + drflac__free_from_callbacks(pRawData, pAllocationCallbacks); + } + } break; + } + + /* If we're not handling metadata, just skip over the block. If we are, it will have been handled earlier in the switch statement above. */ + if (onMeta == NULL && blockSize > 0) { + if (!onSeek(pUserData, blockSize, DRFLAC_SEEK_CUR)) { + isLastBlock = DRFLAC_TRUE; + } + } + + runningFilePos += blockSize; + if (isLastBlock) { + break; + } + } + + *pSeektablePos = seektablePos; + *pSeekpointCount = seektableSize / DRFLAC_SEEKPOINT_SIZE_IN_BYTES; + *pFirstFramePos = runningFilePos; + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__init_private__native(drflac_init_info* pInit, drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_meta_proc onMeta, void* pUserData, void* pUserDataMD, drflac_bool32 relaxed) +{ + /* Pre Condition: The bit stream should be sitting just past the 4-byte id header. */ + + drflac_uint8 isLastBlock; + drflac_uint8 blockType; + drflac_uint32 blockSize; + + (void)onSeek; + + pInit->container = drflac_container_native; + + /* The first metadata block should be the STREAMINFO block. */ + if (!drflac__read_and_decode_block_header(onRead, pUserData, &isLastBlock, &blockType, &blockSize)) { + return DRFLAC_FALSE; + } + + if (blockType != DRFLAC_METADATA_BLOCK_TYPE_STREAMINFO || blockSize != 34) { + if (!relaxed) { + /* We're opening in strict mode and the first block is not the STREAMINFO block. Error. */ + return DRFLAC_FALSE; + } else { + /* + Relaxed mode. To open from here we need to just find the first frame and set the sample rate, etc. to whatever is defined + for that frame. + */ + pInit->hasStreamInfoBlock = DRFLAC_FALSE; + pInit->hasMetadataBlocks = DRFLAC_FALSE; + + if (!drflac__read_next_flac_frame_header(&pInit->bs, 0, &pInit->firstFrameHeader)) { + return DRFLAC_FALSE; /* Couldn't find a frame. */ + } + + if (pInit->firstFrameHeader.bitsPerSample == 0) { + return DRFLAC_FALSE; /* Failed to initialize because the first frame depends on the STREAMINFO block, which does not exist. */ + } + + pInit->sampleRate = pInit->firstFrameHeader.sampleRate; + pInit->channels = drflac__get_channel_count_from_channel_assignment(pInit->firstFrameHeader.channelAssignment); + pInit->bitsPerSample = pInit->firstFrameHeader.bitsPerSample; + pInit->maxBlockSizeInPCMFrames = 65535; /* <-- See notes here: https://xiph.org/flac/format.html#metadata_block_streaminfo */ + return DRFLAC_TRUE; + } + } else { + drflac_streaminfo streaminfo; + if (!drflac__read_streaminfo(onRead, pUserData, &streaminfo)) { + return DRFLAC_FALSE; + } + + pInit->hasStreamInfoBlock = DRFLAC_TRUE; + pInit->sampleRate = streaminfo.sampleRate; + pInit->channels = streaminfo.channels; + pInit->bitsPerSample = streaminfo.bitsPerSample; + pInit->totalPCMFrameCount = streaminfo.totalPCMFrameCount; + pInit->maxBlockSizeInPCMFrames = streaminfo.maxBlockSizeInPCMFrames; /* Don't care about the min block size - only the max (used for determining the size of the memory allocation). */ + pInit->hasMetadataBlocks = !isLastBlock; + + if (onMeta) { + drflac_metadata metadata; + metadata.type = DRFLAC_METADATA_BLOCK_TYPE_STREAMINFO; + metadata.pRawData = NULL; + metadata.rawDataSize = 0; + metadata.data.streaminfo = streaminfo; + onMeta(pUserDataMD, &metadata); + } + + return DRFLAC_TRUE; + } +} + +#ifndef DR_FLAC_NO_OGG +#define DRFLAC_OGG_MAX_PAGE_SIZE 65307 +#define DRFLAC_OGG_CAPTURE_PATTERN_CRC32 1605413199 /* CRC-32 of "OggS". */ + +typedef enum +{ + drflac_ogg_recover_on_crc_mismatch, + drflac_ogg_fail_on_crc_mismatch +} drflac_ogg_crc_mismatch_recovery; + +#ifndef DR_FLAC_NO_CRC +static drflac_uint32 drflac__crc32_table[] = { + 0x00000000L, 0x04C11DB7L, 0x09823B6EL, 0x0D4326D9L, + 0x130476DCL, 0x17C56B6BL, 0x1A864DB2L, 0x1E475005L, + 0x2608EDB8L, 0x22C9F00FL, 0x2F8AD6D6L, 0x2B4BCB61L, + 0x350C9B64L, 0x31CD86D3L, 0x3C8EA00AL, 0x384FBDBDL, + 0x4C11DB70L, 0x48D0C6C7L, 0x4593E01EL, 0x4152FDA9L, + 0x5F15ADACL, 0x5BD4B01BL, 0x569796C2L, 0x52568B75L, + 0x6A1936C8L, 0x6ED82B7FL, 0x639B0DA6L, 0x675A1011L, + 0x791D4014L, 0x7DDC5DA3L, 0x709F7B7AL, 0x745E66CDL, + 0x9823B6E0L, 0x9CE2AB57L, 0x91A18D8EL, 0x95609039L, + 0x8B27C03CL, 0x8FE6DD8BL, 0x82A5FB52L, 0x8664E6E5L, + 0xBE2B5B58L, 0xBAEA46EFL, 0xB7A96036L, 0xB3687D81L, + 0xAD2F2D84L, 0xA9EE3033L, 0xA4AD16EAL, 0xA06C0B5DL, + 0xD4326D90L, 0xD0F37027L, 0xDDB056FEL, 0xD9714B49L, + 0xC7361B4CL, 0xC3F706FBL, 0xCEB42022L, 0xCA753D95L, + 0xF23A8028L, 0xF6FB9D9FL, 0xFBB8BB46L, 0xFF79A6F1L, + 0xE13EF6F4L, 0xE5FFEB43L, 0xE8BCCD9AL, 0xEC7DD02DL, + 0x34867077L, 0x30476DC0L, 0x3D044B19L, 0x39C556AEL, + 0x278206ABL, 0x23431B1CL, 0x2E003DC5L, 0x2AC12072L, + 0x128E9DCFL, 0x164F8078L, 0x1B0CA6A1L, 0x1FCDBB16L, + 0x018AEB13L, 0x054BF6A4L, 0x0808D07DL, 0x0CC9CDCAL, + 0x7897AB07L, 0x7C56B6B0L, 0x71159069L, 0x75D48DDEL, + 0x6B93DDDBL, 0x6F52C06CL, 0x6211E6B5L, 0x66D0FB02L, + 0x5E9F46BFL, 0x5A5E5B08L, 0x571D7DD1L, 0x53DC6066L, + 0x4D9B3063L, 0x495A2DD4L, 0x44190B0DL, 0x40D816BAL, + 0xACA5C697L, 0xA864DB20L, 0xA527FDF9L, 0xA1E6E04EL, + 0xBFA1B04BL, 0xBB60ADFCL, 0xB6238B25L, 0xB2E29692L, + 0x8AAD2B2FL, 0x8E6C3698L, 0x832F1041L, 0x87EE0DF6L, + 0x99A95DF3L, 0x9D684044L, 0x902B669DL, 0x94EA7B2AL, + 0xE0B41DE7L, 0xE4750050L, 0xE9362689L, 0xEDF73B3EL, + 0xF3B06B3BL, 0xF771768CL, 0xFA325055L, 0xFEF34DE2L, + 0xC6BCF05FL, 0xC27DEDE8L, 0xCF3ECB31L, 0xCBFFD686L, + 0xD5B88683L, 0xD1799B34L, 0xDC3ABDEDL, 0xD8FBA05AL, + 0x690CE0EEL, 0x6DCDFD59L, 0x608EDB80L, 0x644FC637L, + 0x7A089632L, 0x7EC98B85L, 0x738AAD5CL, 0x774BB0EBL, + 0x4F040D56L, 0x4BC510E1L, 0x46863638L, 0x42472B8FL, + 0x5C007B8AL, 0x58C1663DL, 0x558240E4L, 0x51435D53L, + 0x251D3B9EL, 0x21DC2629L, 0x2C9F00F0L, 0x285E1D47L, + 0x36194D42L, 0x32D850F5L, 0x3F9B762CL, 0x3B5A6B9BL, + 0x0315D626L, 0x07D4CB91L, 0x0A97ED48L, 0x0E56F0FFL, + 0x1011A0FAL, 0x14D0BD4DL, 0x19939B94L, 0x1D528623L, + 0xF12F560EL, 0xF5EE4BB9L, 0xF8AD6D60L, 0xFC6C70D7L, + 0xE22B20D2L, 0xE6EA3D65L, 0xEBA91BBCL, 0xEF68060BL, + 0xD727BBB6L, 0xD3E6A601L, 0xDEA580D8L, 0xDA649D6FL, + 0xC423CD6AL, 0xC0E2D0DDL, 0xCDA1F604L, 0xC960EBB3L, + 0xBD3E8D7EL, 0xB9FF90C9L, 0xB4BCB610L, 0xB07DABA7L, + 0xAE3AFBA2L, 0xAAFBE615L, 0xA7B8C0CCL, 0xA379DD7BL, + 0x9B3660C6L, 0x9FF77D71L, 0x92B45BA8L, 0x9675461FL, + 0x8832161AL, 0x8CF30BADL, 0x81B02D74L, 0x857130C3L, + 0x5D8A9099L, 0x594B8D2EL, 0x5408ABF7L, 0x50C9B640L, + 0x4E8EE645L, 0x4A4FFBF2L, 0x470CDD2BL, 0x43CDC09CL, + 0x7B827D21L, 0x7F436096L, 0x7200464FL, 0x76C15BF8L, + 0x68860BFDL, 0x6C47164AL, 0x61043093L, 0x65C52D24L, + 0x119B4BE9L, 0x155A565EL, 0x18197087L, 0x1CD86D30L, + 0x029F3D35L, 0x065E2082L, 0x0B1D065BL, 0x0FDC1BECL, + 0x3793A651L, 0x3352BBE6L, 0x3E119D3FL, 0x3AD08088L, + 0x2497D08DL, 0x2056CD3AL, 0x2D15EBE3L, 0x29D4F654L, + 0xC5A92679L, 0xC1683BCEL, 0xCC2B1D17L, 0xC8EA00A0L, + 0xD6AD50A5L, 0xD26C4D12L, 0xDF2F6BCBL, 0xDBEE767CL, + 0xE3A1CBC1L, 0xE760D676L, 0xEA23F0AFL, 0xEEE2ED18L, + 0xF0A5BD1DL, 0xF464A0AAL, 0xF9278673L, 0xFDE69BC4L, + 0x89B8FD09L, 0x8D79E0BEL, 0x803AC667L, 0x84FBDBD0L, + 0x9ABC8BD5L, 0x9E7D9662L, 0x933EB0BBL, 0x97FFAD0CL, + 0xAFB010B1L, 0xAB710D06L, 0xA6322BDFL, 0xA2F33668L, + 0xBCB4666DL, 0xB8757BDAL, 0xB5365D03L, 0xB1F740B4L +}; +#endif + +static DRFLAC_INLINE drflac_uint32 drflac_crc32_byte(drflac_uint32 crc32, drflac_uint8 data) +{ +#ifndef DR_FLAC_NO_CRC + return (crc32 << 8) ^ drflac__crc32_table[(drflac_uint8)((crc32 >> 24) & 0xFF) ^ data]; +#else + (void)data; + return crc32; +#endif +} + +#if 0 +static DRFLAC_INLINE drflac_uint32 drflac_crc32_uint32(drflac_uint32 crc32, drflac_uint32 data) +{ + crc32 = drflac_crc32_byte(crc32, (drflac_uint8)((data >> 24) & 0xFF)); + crc32 = drflac_crc32_byte(crc32, (drflac_uint8)((data >> 16) & 0xFF)); + crc32 = drflac_crc32_byte(crc32, (drflac_uint8)((data >> 8) & 0xFF)); + crc32 = drflac_crc32_byte(crc32, (drflac_uint8)((data >> 0) & 0xFF)); + return crc32; +} + +static DRFLAC_INLINE drflac_uint32 drflac_crc32_uint64(drflac_uint32 crc32, drflac_uint64 data) +{ + crc32 = drflac_crc32_uint32(crc32, (drflac_uint32)((data >> 32) & 0xFFFFFFFF)); + crc32 = drflac_crc32_uint32(crc32, (drflac_uint32)((data >> 0) & 0xFFFFFFFF)); + return crc32; +} +#endif + +static DRFLAC_INLINE drflac_uint32 drflac_crc32_buffer(drflac_uint32 crc32, drflac_uint8* pData, drflac_uint32 dataSize) +{ + /* This can be optimized. */ + drflac_uint32 i; + for (i = 0; i < dataSize; ++i) { + crc32 = drflac_crc32_byte(crc32, pData[i]); + } + return crc32; +} + + +static DRFLAC_INLINE drflac_bool32 drflac_ogg__is_capture_pattern(drflac_uint8 pattern[4]) +{ + return pattern[0] == 'O' && pattern[1] == 'g' && pattern[2] == 'g' && pattern[3] == 'S'; +} + +static DRFLAC_INLINE drflac_uint32 drflac_ogg__get_page_header_size(drflac_ogg_page_header* pHeader) +{ + return 27 + pHeader->segmentCount; +} + +static DRFLAC_INLINE drflac_uint32 drflac_ogg__get_page_body_size(drflac_ogg_page_header* pHeader) +{ + drflac_uint32 pageBodySize = 0; + int i; + + for (i = 0; i < pHeader->segmentCount; ++i) { + pageBodySize += pHeader->segmentTable[i]; + } + + return pageBodySize; +} + +static drflac_result drflac_ogg__read_page_header_after_capture_pattern(drflac_read_proc onRead, void* pUserData, drflac_ogg_page_header* pHeader, drflac_uint32* pBytesRead, drflac_uint32* pCRC32) +{ + drflac_uint8 data[23]; + drflac_uint32 i; + + DRFLAC_ASSERT(*pCRC32 == DRFLAC_OGG_CAPTURE_PATTERN_CRC32); + + if (onRead(pUserData, data, 23) != 23) { + return DRFLAC_AT_END; + } + *pBytesRead += 23; + + /* + It's not actually used, but set the capture pattern to 'OggS' for completeness. Not doing this will cause static analysers to complain about + us trying to access uninitialized data. We could alternatively just comment out this member of the drflac_ogg_page_header structure, but I + like to have it map to the structure of the underlying data. + */ + pHeader->capturePattern[0] = 'O'; + pHeader->capturePattern[1] = 'g'; + pHeader->capturePattern[2] = 'g'; + pHeader->capturePattern[3] = 'S'; + + pHeader->structureVersion = data[0]; + pHeader->headerType = data[1]; + DRFLAC_COPY_MEMORY(&pHeader->granulePosition, &data[ 2], 8); + DRFLAC_COPY_MEMORY(&pHeader->serialNumber, &data[10], 4); + DRFLAC_COPY_MEMORY(&pHeader->sequenceNumber, &data[14], 4); + DRFLAC_COPY_MEMORY(&pHeader->checksum, &data[18], 4); + pHeader->segmentCount = data[22]; + + /* Calculate the CRC. Note that for the calculation the checksum part of the page needs to be set to 0. */ + data[18] = 0; + data[19] = 0; + data[20] = 0; + data[21] = 0; + + for (i = 0; i < 23; ++i) { + *pCRC32 = drflac_crc32_byte(*pCRC32, data[i]); + } + + + if (onRead(pUserData, pHeader->segmentTable, pHeader->segmentCount) != pHeader->segmentCount) { + return DRFLAC_AT_END; + } + *pBytesRead += pHeader->segmentCount; + + for (i = 0; i < pHeader->segmentCount; ++i) { + *pCRC32 = drflac_crc32_byte(*pCRC32, pHeader->segmentTable[i]); + } + + return DRFLAC_SUCCESS; +} + +static drflac_result drflac_ogg__read_page_header(drflac_read_proc onRead, void* pUserData, drflac_ogg_page_header* pHeader, drflac_uint32* pBytesRead, drflac_uint32* pCRC32) +{ + drflac_uint8 id[4]; + + *pBytesRead = 0; + + if (onRead(pUserData, id, 4) != 4) { + return DRFLAC_AT_END; + } + *pBytesRead += 4; + + /* We need to read byte-by-byte until we find the OggS capture pattern. */ + for (;;) { + if (drflac_ogg__is_capture_pattern(id)) { + drflac_result result; + + *pCRC32 = DRFLAC_OGG_CAPTURE_PATTERN_CRC32; + + result = drflac_ogg__read_page_header_after_capture_pattern(onRead, pUserData, pHeader, pBytesRead, pCRC32); + if (result == DRFLAC_SUCCESS) { + return DRFLAC_SUCCESS; + } else { + if (result == DRFLAC_CRC_MISMATCH) { + continue; + } else { + return result; + } + } + } else { + /* The first 4 bytes did not equal the capture pattern. Read the next byte and try again. */ + id[0] = id[1]; + id[1] = id[2]; + id[2] = id[3]; + if (onRead(pUserData, &id[3], 1) != 1) { + return DRFLAC_AT_END; + } + *pBytesRead += 1; + } + } +} + + +/* +The main part of the Ogg encapsulation is the conversion from the physical Ogg bitstream to the native FLAC bitstream. It works +in three general stages: Ogg Physical Bitstream -> Ogg/FLAC Logical Bitstream -> FLAC Native Bitstream. dr_flac is designed +in such a way that the core sections assume everything is delivered in native format. Therefore, for each encapsulation type +dr_flac is supporting there needs to be a layer sitting on top of the onRead and onSeek callbacks that ensures the bits read from +the physical Ogg bitstream are converted and delivered in native FLAC format. +*/ +typedef struct +{ + drflac_read_proc onRead; /* The original onRead callback from drflac_open() and family. */ + drflac_seek_proc onSeek; /* The original onSeek callback from drflac_open() and family. */ + drflac_tell_proc onTell; /* The original onTell callback from drflac_open() and family. */ + void* pUserData; /* The user data passed on onRead and onSeek. This is the user data that was passed on drflac_open() and family. */ + drflac_uint64 currentBytePos; /* The position of the byte we are sitting on in the physical byte stream. Used for efficient seeking. */ + drflac_uint64 firstBytePos; /* The position of the first byte in the physical bitstream. Points to the start of the "OggS" identifier of the FLAC bos page. */ + drflac_uint32 serialNumber; /* The serial number of the FLAC audio pages. This is determined by the initial header page that was read during initialization. */ + drflac_ogg_page_header bosPageHeader; /* Used for seeking. */ + drflac_ogg_page_header currentPageHeader; + drflac_uint32 bytesRemainingInPage; + drflac_uint32 pageDataSize; + drflac_uint8 pageData[DRFLAC_OGG_MAX_PAGE_SIZE]; +} drflac_oggbs; /* oggbs = Ogg Bitstream */ + +static size_t drflac_oggbs__read_physical(drflac_oggbs* oggbs, void* bufferOut, size_t bytesToRead) +{ + size_t bytesActuallyRead = oggbs->onRead(oggbs->pUserData, bufferOut, bytesToRead); + oggbs->currentBytePos += bytesActuallyRead; + + return bytesActuallyRead; +} + +static drflac_bool32 drflac_oggbs__seek_physical(drflac_oggbs* oggbs, drflac_uint64 offset, drflac_seek_origin origin) +{ + if (origin == DRFLAC_SEEK_SET) { + if (offset <= 0x7FFFFFFF) { + if (!oggbs->onSeek(oggbs->pUserData, (int)offset, DRFLAC_SEEK_SET)) { + return DRFLAC_FALSE; + } + oggbs->currentBytePos = offset; + + return DRFLAC_TRUE; + } else { + if (!oggbs->onSeek(oggbs->pUserData, 0x7FFFFFFF, DRFLAC_SEEK_SET)) { + return DRFLAC_FALSE; + } + oggbs->currentBytePos = offset; + + return drflac_oggbs__seek_physical(oggbs, offset - 0x7FFFFFFF, DRFLAC_SEEK_CUR); + } + } else { + while (offset > 0x7FFFFFFF) { + if (!oggbs->onSeek(oggbs->pUserData, 0x7FFFFFFF, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + oggbs->currentBytePos += 0x7FFFFFFF; + offset -= 0x7FFFFFFF; + } + + if (!oggbs->onSeek(oggbs->pUserData, (int)offset, DRFLAC_SEEK_CUR)) { /* <-- Safe cast thanks to the loop above. */ + return DRFLAC_FALSE; + } + oggbs->currentBytePos += offset; + + return DRFLAC_TRUE; + } +} + +static drflac_bool32 drflac_oggbs__goto_next_page(drflac_oggbs* oggbs, drflac_ogg_crc_mismatch_recovery recoveryMethod) +{ + drflac_ogg_page_header header; + for (;;) { + drflac_uint32 crc32 = 0; + drflac_uint32 bytesRead; + drflac_uint32 pageBodySize; +#ifndef DR_FLAC_NO_CRC + drflac_uint32 actualCRC32; +#endif + + if (drflac_ogg__read_page_header(oggbs->onRead, oggbs->pUserData, &header, &bytesRead, &crc32) != DRFLAC_SUCCESS) { + return DRFLAC_FALSE; + } + oggbs->currentBytePos += bytesRead; + + pageBodySize = drflac_ogg__get_page_body_size(&header); + if (pageBodySize > DRFLAC_OGG_MAX_PAGE_SIZE) { + continue; /* Invalid page size. Assume it's corrupted and just move to the next page. */ + } + + if (header.serialNumber != oggbs->serialNumber) { + /* It's not a FLAC page. Skip it. */ + if (pageBodySize > 0 && !drflac_oggbs__seek_physical(oggbs, pageBodySize, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + continue; + } + + + /* We need to read the entire page and then do a CRC check on it. If there's a CRC mismatch we need to skip this page. */ + if (drflac_oggbs__read_physical(oggbs, oggbs->pageData, pageBodySize) != pageBodySize) { + return DRFLAC_FALSE; + } + oggbs->pageDataSize = pageBodySize; + +#ifndef DR_FLAC_NO_CRC + actualCRC32 = drflac_crc32_buffer(crc32, oggbs->pageData, oggbs->pageDataSize); + if (actualCRC32 != header.checksum) { + if (recoveryMethod == drflac_ogg_recover_on_crc_mismatch) { + continue; /* CRC mismatch. Skip this page. */ + } else { + /* + Even though we are failing on a CRC mismatch, we still want our stream to be in a good state. Therefore we + go to the next valid page to ensure we're in a good state, but return false to let the caller know that the + seek did not fully complete. + */ + drflac_oggbs__goto_next_page(oggbs, drflac_ogg_recover_on_crc_mismatch); + return DRFLAC_FALSE; + } + } +#else + (void)recoveryMethod; /* <-- Silence a warning. */ +#endif + + oggbs->currentPageHeader = header; + oggbs->bytesRemainingInPage = pageBodySize; + return DRFLAC_TRUE; + } +} + +/* Function below is unused at the moment, but I might be re-adding it later. */ +#if 0 +static drflac_uint8 drflac_oggbs__get_current_segment_index(drflac_oggbs* oggbs, drflac_uint8* pBytesRemainingInSeg) +{ + drflac_uint32 bytesConsumedInPage = drflac_ogg__get_page_body_size(&oggbs->currentPageHeader) - oggbs->bytesRemainingInPage; + drflac_uint8 iSeg = 0; + drflac_uint32 iByte = 0; + while (iByte < bytesConsumedInPage) { + drflac_uint8 segmentSize = oggbs->currentPageHeader.segmentTable[iSeg]; + if (iByte + segmentSize > bytesConsumedInPage) { + break; + } else { + iSeg += 1; + iByte += segmentSize; + } + } + + *pBytesRemainingInSeg = oggbs->currentPageHeader.segmentTable[iSeg] - (drflac_uint8)(bytesConsumedInPage - iByte); + return iSeg; +} + +static drflac_bool32 drflac_oggbs__seek_to_next_packet(drflac_oggbs* oggbs) +{ + /* The current packet ends when we get to the segment with a lacing value of < 255 which is not at the end of a page. */ + for (;;) { + drflac_bool32 atEndOfPage = DRFLAC_FALSE; + + drflac_uint8 bytesRemainingInSeg; + drflac_uint8 iFirstSeg = drflac_oggbs__get_current_segment_index(oggbs, &bytesRemainingInSeg); + + drflac_uint32 bytesToEndOfPacketOrPage = bytesRemainingInSeg; + for (drflac_uint8 iSeg = iFirstSeg; iSeg < oggbs->currentPageHeader.segmentCount; ++iSeg) { + drflac_uint8 segmentSize = oggbs->currentPageHeader.segmentTable[iSeg]; + if (segmentSize < 255) { + if (iSeg == oggbs->currentPageHeader.segmentCount-1) { + atEndOfPage = DRFLAC_TRUE; + } + + break; + } + + bytesToEndOfPacketOrPage += segmentSize; + } + + /* + At this point we will have found either the packet or the end of the page. If were at the end of the page we'll + want to load the next page and keep searching for the end of the packet. + */ + drflac_oggbs__seek_physical(oggbs, bytesToEndOfPacketOrPage, DRFLAC_SEEK_CUR); + oggbs->bytesRemainingInPage -= bytesToEndOfPacketOrPage; + + if (atEndOfPage) { + /* + We're potentially at the next packet, but we need to check the next page first to be sure because the packet may + straddle pages. + */ + if (!drflac_oggbs__goto_next_page(oggbs)) { + return DRFLAC_FALSE; + } + + /* If it's a fresh packet it most likely means we're at the next packet. */ + if ((oggbs->currentPageHeader.headerType & 0x01) == 0) { + return DRFLAC_TRUE; + } + } else { + /* We're at the next packet. */ + return DRFLAC_TRUE; + } + } +} + +static drflac_bool32 drflac_oggbs__seek_to_next_frame(drflac_oggbs* oggbs) +{ + /* The bitstream should be sitting on the first byte just after the header of the frame. */ + + /* What we're actually doing here is seeking to the start of the next packet. */ + return drflac_oggbs__seek_to_next_packet(oggbs); +} +#endif + +static size_t drflac__on_read_ogg(void* pUserData, void* bufferOut, size_t bytesToRead) +{ + drflac_oggbs* oggbs = (drflac_oggbs*)pUserData; + drflac_uint8* pRunningBufferOut = (drflac_uint8*)bufferOut; + size_t bytesRead = 0; + + DRFLAC_ASSERT(oggbs != NULL); + DRFLAC_ASSERT(pRunningBufferOut != NULL); + + /* Reading is done page-by-page. If we've run out of bytes in the page we need to move to the next one. */ + while (bytesRead < bytesToRead) { + size_t bytesRemainingToRead = bytesToRead - bytesRead; + + if (oggbs->bytesRemainingInPage >= bytesRemainingToRead) { + DRFLAC_COPY_MEMORY(pRunningBufferOut, oggbs->pageData + (oggbs->pageDataSize - oggbs->bytesRemainingInPage), bytesRemainingToRead); + bytesRead += bytesRemainingToRead; + oggbs->bytesRemainingInPage -= (drflac_uint32)bytesRemainingToRead; + break; + } + + /* If we get here it means some of the requested data is contained in the next pages. */ + if (oggbs->bytesRemainingInPage > 0) { + DRFLAC_COPY_MEMORY(pRunningBufferOut, oggbs->pageData + (oggbs->pageDataSize - oggbs->bytesRemainingInPage), oggbs->bytesRemainingInPage); + bytesRead += oggbs->bytesRemainingInPage; + pRunningBufferOut += oggbs->bytesRemainingInPage; + oggbs->bytesRemainingInPage = 0; + } + + DRFLAC_ASSERT(bytesRemainingToRead > 0); + if (!drflac_oggbs__goto_next_page(oggbs, drflac_ogg_recover_on_crc_mismatch)) { + break; /* Failed to go to the next page. Might have simply hit the end of the stream. */ + } + } + + return bytesRead; +} + +static drflac_bool32 drflac__on_seek_ogg(void* pUserData, int offset, drflac_seek_origin origin) +{ + drflac_oggbs* oggbs = (drflac_oggbs*)pUserData; + int bytesSeeked = 0; + + DRFLAC_ASSERT(oggbs != NULL); + DRFLAC_ASSERT(offset >= 0); /* <-- Never seek backwards. */ + + /* Seeking is always forward which makes things a lot simpler. */ + if (origin == DRFLAC_SEEK_SET) { + if (!drflac_oggbs__seek_physical(oggbs, (int)oggbs->firstBytePos, DRFLAC_SEEK_SET)) { + return DRFLAC_FALSE; + } + + if (!drflac_oggbs__goto_next_page(oggbs, drflac_ogg_fail_on_crc_mismatch)) { + return DRFLAC_FALSE; + } + + return drflac__on_seek_ogg(pUserData, offset, DRFLAC_SEEK_CUR); + } else if (origin == DRFLAC_SEEK_CUR) { + while (bytesSeeked < offset) { + int bytesRemainingToSeek = offset - bytesSeeked; + DRFLAC_ASSERT(bytesRemainingToSeek >= 0); + + if (oggbs->bytesRemainingInPage >= (size_t)bytesRemainingToSeek) { + bytesSeeked += bytesRemainingToSeek; + (void)bytesSeeked; /* <-- Silence a dead store warning emitted by Clang Static Analyzer. */ + oggbs->bytesRemainingInPage -= bytesRemainingToSeek; + break; + } + + /* If we get here it means some of the requested data is contained in the next pages. */ + if (oggbs->bytesRemainingInPage > 0) { + bytesSeeked += (int)oggbs->bytesRemainingInPage; + oggbs->bytesRemainingInPage = 0; + } + + DRFLAC_ASSERT(bytesRemainingToSeek > 0); + if (!drflac_oggbs__goto_next_page(oggbs, drflac_ogg_fail_on_crc_mismatch)) { + /* Failed to go to the next page. We either hit the end of the stream or had a CRC mismatch. */ + return DRFLAC_FALSE; + } + } + } else if (origin == DRFLAC_SEEK_END) { + /* Seeking to the end is not supported. */ + return DRFLAC_FALSE; + } + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__on_tell_ogg(void* pUserData, drflac_int64* pCursor) +{ + /* + Not implemented for Ogg containers because we don't currently track the byte position of the logical bitstream. To support this, we'll need + to track the position in drflac__on_read_ogg and drflac__on_seek_ogg. + */ + (void)pUserData; + (void)pCursor; + return DRFLAC_FALSE; +} + + +static drflac_bool32 drflac_ogg__seek_to_pcm_frame(drflac* pFlac, drflac_uint64 pcmFrameIndex) +{ + drflac_oggbs* oggbs = (drflac_oggbs*)pFlac->_oggbs; + drflac_uint64 originalBytePos; + drflac_uint64 runningGranulePosition; + drflac_uint64 runningFrameBytePos = 0; + drflac_uint64 runningPCMFrameCount = 0; + drflac_bool32 hasFramePage = DRFLAC_FALSE; + + DRFLAC_ASSERT(oggbs != NULL); + + originalBytePos = oggbs->currentBytePos; /* For recovery. Points to the OggS identifier. */ + + /* First seek to the first frame. */ + if (!drflac__seek_to_byte(&pFlac->bs, pFlac->firstFLACFramePosInBytes)) { + return DRFLAC_FALSE; + } + oggbs->bytesRemainingInPage = 0; + + runningGranulePosition = 0; + for (;;) { + if (!drflac_oggbs__goto_next_page(oggbs, drflac_ogg_recover_on_crc_mismatch)) { + drflac_oggbs__seek_physical(oggbs, originalBytePos, DRFLAC_SEEK_SET); + return DRFLAC_FALSE; /* Never did find that sample... */ + } + + /* + Keep a page where decoding can start. The page containing the target might begin with a continued packet, + in which case we must start at an earlier page so the complete FLAC frame is available. + */ + if ((oggbs->currentPageHeader.headerType & 0x01) == 0) { /* <-- Is it a fresh page? */ + if (oggbs->currentPageHeader.segmentTable[0] >= 2) { + drflac_uint8 firstBytesInPage[2]; + firstBytesInPage[0] = oggbs->pageData[0]; + firstBytesInPage[1] = oggbs->pageData[1]; + + if ((firstBytesInPage[0] == 0xFF) && (firstBytesInPage[1] & 0xFC) == 0xF8) { /* <-- Does the page begin with a frame's sync code? */ + runningFrameBytePos = oggbs->currentBytePos - drflac_ogg__get_page_header_size(&oggbs->currentPageHeader) - oggbs->pageDataSize; + runningPCMFrameCount = runningGranulePosition; + hasFramePage = DRFLAC_TRUE; + } + } + } + + /* A page with no completed packets has no granule position. */ + if (oggbs->currentPageHeader.granulePosition != (drflac_uint64)-1) { + if (hasFramePage && oggbs->currentPageHeader.granulePosition >= pcmFrameIndex) { + break; + } + + runningGranulePosition = oggbs->currentPageHeader.granulePosition; + } + } + + /* + We found the page that that is closest to the sample, so now we need to find it. The first thing to do is seek to the + start of that page. In the loop above we checked that it was a fresh page which means this page is also the start of + a new frame. This property means that after we've seeked to the page we can immediately start looping over frames until + we find the one containing the target sample. + */ + if (!drflac_oggbs__seek_physical(oggbs, runningFrameBytePos, DRFLAC_SEEK_SET)) { + return DRFLAC_FALSE; + } + if (!drflac_oggbs__goto_next_page(oggbs, drflac_ogg_recover_on_crc_mismatch)) { + return DRFLAC_FALSE; + } + + /* + At this point we'll be sitting on the first byte of the frame header of the first frame in the page. We just keep + looping over these frames until we find the one containing the sample we're after. + */ + for (;;) { + /* + There are two ways to find the sample and seek past irrelevant frames: + 1) Use the native FLAC decoder. + 2) Use Ogg's framing system. + + Both of these options have their own pros and cons. Using the native FLAC decoder is slower because it needs to + do a full decode of the frame. Using Ogg's framing system is faster, but more complicated and involves some code + duplication for the decoding of frame headers. + + Another thing to consider is that using the Ogg framing system will perform direct seeking of the physical Ogg + bitstream. This is important to consider because it means we cannot read data from the drflac_bs object using the + standard drflac__*() APIs because that will read in extra data for its own internal caching which in turn breaks + the positioning of the read pointer of the physical Ogg bitstream. Therefore, anything that would normally be read + using the native FLAC decoding APIs, such as drflac__read_next_flac_frame_header(), need to be re-implemented so as to + avoid the use of the drflac_bs object. + + Considering these issues, I have decided to use the slower native FLAC decoding method for the following reasons: + 1) Seeking is already partially accelerated using Ogg's paging system in the code block above. + 2) Seeking in an Ogg encapsulated FLAC stream is probably quite uncommon. + 3) Simplicity. + */ + drflac_uint64 firstPCMFrameInFLACFrame = 0; + drflac_uint64 lastPCMFrameInFLACFrame = 0; + drflac_uint64 pcmFrameCountInThisFrame; + + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + return DRFLAC_FALSE; + } + + drflac__get_pcm_frame_range_of_current_flac_frame(pFlac, &firstPCMFrameInFLACFrame, &lastPCMFrameInFLACFrame); + + pcmFrameCountInThisFrame = (lastPCMFrameInFLACFrame - firstPCMFrameInFLACFrame) + 1; + + /* The target can be a frame boundary, including EOF when the total frame count is unknown. */ + if ((runningPCMFrameCount + pcmFrameCountInThisFrame) == pcmFrameIndex) { + drflac_result result = drflac__decode_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + pFlac->currentPCMFrame = pcmFrameIndex; + pFlac->currentFLACFrame.pcmFramesRemaining = 0; + return DRFLAC_TRUE; + } else { + return DRFLAC_FALSE; + } + } + + if (pcmFrameIndex < (runningPCMFrameCount + pcmFrameCountInThisFrame)) { + /* + The sample should be in this FLAC frame. We need to fully decode it, however if it's an invalid frame (a CRC mismatch), we need to pretend + it never existed and keep iterating. + */ + drflac_result result = drflac__decode_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + /* The frame is valid. We just need to skip over some samples to ensure it's sample-exact. */ + drflac_uint64 pcmFramesToDecode = (size_t)(pcmFrameIndex - runningPCMFrameCount); /* <-- Safe cast because the maximum number of samples in a frame is 65535. */ + if (pcmFramesToDecode == 0) { + return DRFLAC_TRUE; + } + + pFlac->currentPCMFrame = runningPCMFrameCount; + + return drflac__seek_forward_by_pcm_frames(pFlac, pcmFramesToDecode) == pcmFramesToDecode; /* <-- If this fails, something bad has happened (it should never fail). */ + } else { + if (result == DRFLAC_CRC_MISMATCH) { + continue; /* CRC mismatch. Pretend this frame never existed. */ + } else { + return DRFLAC_FALSE; + } + } + } else { + /* + It's not in this frame. We need to seek past the frame, but check if there was a CRC mismatch. If so, we pretend this + frame never existed and leave the running sample count untouched. + */ + drflac_result result = drflac__seek_to_next_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + runningPCMFrameCount += pcmFrameCountInThisFrame; + } else { + if (result == DRFLAC_CRC_MISMATCH) { + continue; /* CRC mismatch. Pretend this frame never existed. */ + } else { + return DRFLAC_FALSE; + } + } + } + } +} + + + +static drflac_bool32 drflac__init_private__ogg(drflac_init_info* pInit, drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_meta_proc onMeta, void* pUserData, void* pUserDataMD, drflac_bool32 relaxed) +{ + drflac_ogg_page_header header; + drflac_uint32 crc32 = DRFLAC_OGG_CAPTURE_PATTERN_CRC32; + drflac_uint32 bytesRead = 0; + + /* Pre Condition: The bit stream should be sitting just past the 4-byte OggS capture pattern. */ + (void)relaxed; + + pInit->container = drflac_container_ogg; + pInit->oggFirstBytePos = 0; + + /* + We'll get here if the first 4 bytes of the stream were the OggS capture pattern, however it doesn't necessarily mean the + stream includes FLAC encoded audio. To check for this we need to scan the beginning-of-stream page markers and check if + any match the FLAC specification. Important to keep in mind that the stream may be multiplexed. + */ + if (drflac_ogg__read_page_header_after_capture_pattern(onRead, pUserData, &header, &bytesRead, &crc32) != DRFLAC_SUCCESS) { + return DRFLAC_FALSE; + } + pInit->runningFilePos += bytesRead; + + for (;;) { + int pageBodySize; + + /* Break if we're past the beginning of stream page. */ + if ((header.headerType & 0x02) == 0) { + return DRFLAC_FALSE; + } + + /* Check if it's a FLAC header. */ + pageBodySize = drflac_ogg__get_page_body_size(&header); + if (pageBodySize == 51) { /* 51 = the lacing value of the FLAC header packet. */ + /* It could be a FLAC page... */ + drflac_uint32 bytesRemainingInPage = pageBodySize; + drflac_uint8 packetType; + + if (onRead(pUserData, &packetType, 1) != 1) { + return DRFLAC_FALSE; + } + + bytesRemainingInPage -= 1; + if (packetType == 0x7F) { + /* Increasingly more likely to be a FLAC page... */ + drflac_uint8 sig[4]; + if (onRead(pUserData, sig, 4) != 4) { + return DRFLAC_FALSE; + } + + bytesRemainingInPage -= 4; + if (sig[0] == 'F' && sig[1] == 'L' && sig[2] == 'A' && sig[3] == 'C') { + /* Almost certainly a FLAC page... */ + drflac_uint8 mappingVersion[2]; + if (onRead(pUserData, mappingVersion, 2) != 2) { + return DRFLAC_FALSE; + } + + if (mappingVersion[0] != 1) { + return DRFLAC_FALSE; /* Only supporting version 1.x of the Ogg mapping. */ + } + + /* + The next 2 bytes are the non-audio packets, not including this one. We don't care about this because we're going to + be handling it in a generic way based on the serial number and packet types. + */ + if (!onSeek(pUserData, 2, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + + /* Expecting the native FLAC signature "fLaC". */ + if (onRead(pUserData, sig, 4) != 4) { + return DRFLAC_FALSE; + } + + if (sig[0] == 'f' && sig[1] == 'L' && sig[2] == 'a' && sig[3] == 'C') { + /* The remaining data in the page should be the STREAMINFO block. */ + drflac_streaminfo streaminfo; + drflac_uint8 isLastBlock; + drflac_uint8 blockType; + drflac_uint32 blockSize; + if (!drflac__read_and_decode_block_header(onRead, pUserData, &isLastBlock, &blockType, &blockSize)) { + return DRFLAC_FALSE; + } + + if (blockType != DRFLAC_METADATA_BLOCK_TYPE_STREAMINFO || blockSize != 34) { + return DRFLAC_FALSE; /* Invalid block type. First block must be the STREAMINFO block. */ + } + + if (drflac__read_streaminfo(onRead, pUserData, &streaminfo)) { + /* Success! */ + pInit->hasStreamInfoBlock = DRFLAC_TRUE; + pInit->sampleRate = streaminfo.sampleRate; + pInit->channels = streaminfo.channels; + pInit->bitsPerSample = streaminfo.bitsPerSample; + pInit->totalPCMFrameCount = streaminfo.totalPCMFrameCount; + pInit->maxBlockSizeInPCMFrames = streaminfo.maxBlockSizeInPCMFrames; + pInit->hasMetadataBlocks = !isLastBlock; + + if (onMeta) { + drflac_metadata metadata; + metadata.type = DRFLAC_METADATA_BLOCK_TYPE_STREAMINFO; + metadata.pRawData = NULL; + metadata.rawDataSize = 0; + metadata.data.streaminfo = streaminfo; + onMeta(pUserDataMD, &metadata); + } + + pInit->runningFilePos += pageBodySize; + pInit->oggFirstBytePos = pInit->runningFilePos - 79; /* Subtracting 79 will place us right on top of the "OggS" identifier of the FLAC bos page. */ + pInit->oggSerial = header.serialNumber; + pInit->oggBosHeader = header; + break; + } else { + /* Failed to read STREAMINFO block. Aww, so close... */ + return DRFLAC_FALSE; + } + } else { + /* Invalid file. */ + return DRFLAC_FALSE; + } + } else { + /* Not a FLAC header. Skip it. */ + if (!onSeek(pUserData, bytesRemainingInPage, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + } + } else { + /* Not a FLAC header. Seek past the entire page and move on to the next. */ + if (!onSeek(pUserData, bytesRemainingInPage, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + } + } else { + if (!onSeek(pUserData, pageBodySize, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; + } + } + + pInit->runningFilePos += pageBodySize; + + + /* Read the header of the next page. */ + if (drflac_ogg__read_page_header(onRead, pUserData, &header, &bytesRead, &crc32) != DRFLAC_SUCCESS) { + return DRFLAC_FALSE; + } + pInit->runningFilePos += bytesRead; + } + + /* + If we get here it means we found a FLAC audio stream. We should be sitting on the first byte of the header of the next page. The next + packets in the FLAC logical stream contain the metadata. The only thing left to do in the initialization phase for Ogg is to create the + Ogg bistream object. + */ + pInit->hasMetadataBlocks = DRFLAC_TRUE; /* <-- Always have at least VORBIS_COMMENT metadata block. */ + return DRFLAC_TRUE; +} +#endif + +static drflac_bool32 drflac__init_private(drflac_init_info* pInit, drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_meta_proc onMeta, drflac_container container, void* pUserData, void* pUserDataMD) +{ + drflac_bool32 relaxed; + drflac_uint8 id[4]; + + if (pInit == NULL || onRead == NULL || onSeek == NULL) { /* <-- onTell is optional. */ + return DRFLAC_FALSE; + } + + DRFLAC_ZERO_MEMORY(pInit, sizeof(*pInit)); + pInit->onRead = onRead; + pInit->onSeek = onSeek; + pInit->onTell = onTell; + pInit->onMeta = onMeta; + pInit->container = container; + pInit->pUserData = pUserData; + pInit->pUserDataMD = pUserDataMD; + + pInit->bs.onRead = onRead; + pInit->bs.onSeek = onSeek; + pInit->bs.onTell = onTell; + pInit->bs.pUserData = pUserData; + drflac__reset_cache(&pInit->bs); + + + /* If the container is explicitly defined then we can try opening in relaxed mode. */ + relaxed = container != drflac_container_unknown; + + /* Skip over any ID3 tags. */ + for (;;) { + if (onRead(pUserData, id, 4) != 4) { + return DRFLAC_FALSE; /* Ran out of data. */ + } + pInit->runningFilePos += 4; + + if (id[0] == 'I' && id[1] == 'D' && id[2] == '3') { + drflac_uint8 header[6]; + drflac_uint8 flags; + drflac_uint32 headerSize; + + if (onRead(pUserData, header, 6) != 6) { + return DRFLAC_FALSE; /* Ran out of data. */ + } + pInit->runningFilePos += 6; + + flags = header[1]; + + DRFLAC_COPY_MEMORY(&headerSize, header+2, 4); + headerSize = drflac__unsynchsafe_32(drflac__be2host_32(headerSize)); + if (flags & 0x10) { + headerSize += 10; + } + + if (!onSeek(pUserData, headerSize, DRFLAC_SEEK_CUR)) { + return DRFLAC_FALSE; /* Failed to seek past the tag. */ + } + pInit->runningFilePos += headerSize; + } else { + break; + } + } + + if (id[0] == 'f' && id[1] == 'L' && id[2] == 'a' && id[3] == 'C') { + return drflac__init_private__native(pInit, onRead, onSeek, onMeta, pUserData, pUserDataMD, relaxed); + } +#ifndef DR_FLAC_NO_OGG + if (id[0] == 'O' && id[1] == 'g' && id[2] == 'g' && id[3] == 'S') { + return drflac__init_private__ogg(pInit, onRead, onSeek, onMeta, pUserData, pUserDataMD, relaxed); + } +#endif + + /* If we get here it means we likely don't have a header. Try opening in relaxed mode, if applicable. */ + if (relaxed) { + if (container == drflac_container_native) { + return drflac__init_private__native(pInit, onRead, onSeek, onMeta, pUserData, pUserDataMD, relaxed); + } +#ifndef DR_FLAC_NO_OGG + if (container == drflac_container_ogg) { + return drflac__init_private__ogg(pInit, onRead, onSeek, onMeta, pUserData, pUserDataMD, relaxed); + } +#endif + } + + /* Unsupported container. */ + return DRFLAC_FALSE; +} + +static void drflac__init_from_info(drflac* pFlac, const drflac_init_info* pInit) +{ + DRFLAC_ASSERT(pFlac != NULL); + DRFLAC_ASSERT(pInit != NULL); + + DRFLAC_ZERO_MEMORY(pFlac, sizeof(*pFlac)); + pFlac->bs = pInit->bs; + pFlac->onMeta = pInit->onMeta; + pFlac->pUserDataMD = pInit->pUserDataMD; + pFlac->maxBlockSizeInPCMFrames = pInit->maxBlockSizeInPCMFrames; + pFlac->sampleRate = pInit->sampleRate; + pFlac->channels = (drflac_uint8)pInit->channels; + pFlac->bitsPerSample = (drflac_uint8)pInit->bitsPerSample; + pFlac->totalPCMFrameCount = pInit->totalPCMFrameCount; + pFlac->container = pInit->container; +} + + +static drflac* drflac_open_with_metadata_private(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_meta_proc onMeta, drflac_container container, void* pUserData, void* pUserDataMD, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac_init_info init; + drflac_uint32 allocationSize; + drflac_uint32 wholeSIMDVectorCountPerChannel; + drflac_uint32 decodedSamplesAllocationSize; +#ifndef DR_FLAC_NO_OGG + drflac_oggbs* pOggbs = NULL; +#endif + drflac_uint64 firstFramePos; + drflac_uint64 seektablePos; + drflac_uint32 seekpointCount; + drflac_allocation_callbacks allocationCallbacks; + drflac* pFlac; + + /* CPU support first. */ + drflac__init_cpu_caps(); + + if (!drflac__init_private(&init, onRead, onSeek, onTell, onMeta, container, pUserData, pUserDataMD)) { + return NULL; + } + + if (pAllocationCallbacks != NULL) { + allocationCallbacks = *pAllocationCallbacks; + if (allocationCallbacks.onFree == NULL || (allocationCallbacks.onMalloc == NULL && allocationCallbacks.onRealloc == NULL)) { + return NULL; /* Invalid allocation callbacks. */ + } + } else { + allocationCallbacks.pUserData = NULL; + allocationCallbacks.onMalloc = drflac__malloc_default; + allocationCallbacks.onRealloc = drflac__realloc_default; + allocationCallbacks.onFree = drflac__free_default; + } + + + /* + The size of the allocation for the drflac object needs to be large enough to fit the following: + 1) The main members of the drflac structure + 2) A block of memory large enough to store the decoded samples of the largest frame in the stream + 3) If the container is Ogg, a drflac_oggbs object + + The complicated part of the allocation is making sure there's enough room the decoded samples, taking into consideration + the different SIMD instruction sets. + */ + allocationSize = sizeof(drflac); + + /* + The allocation size for decoded frames depends on the number of 32-bit integers that fit inside the largest SIMD vector + we are supporting. + */ + if ((init.maxBlockSizeInPCMFrames % (DRFLAC_MAX_SIMD_VECTOR_SIZE / sizeof(drflac_int32))) == 0) { + wholeSIMDVectorCountPerChannel = (init.maxBlockSizeInPCMFrames / (DRFLAC_MAX_SIMD_VECTOR_SIZE / sizeof(drflac_int32))); + } else { + wholeSIMDVectorCountPerChannel = (init.maxBlockSizeInPCMFrames / (DRFLAC_MAX_SIMD_VECTOR_SIZE / sizeof(drflac_int32))) + 1; + } + + decodedSamplesAllocationSize = wholeSIMDVectorCountPerChannel * DRFLAC_MAX_SIMD_VECTOR_SIZE * init.channels; + + allocationSize += decodedSamplesAllocationSize; + allocationSize += DRFLAC_MAX_SIMD_VECTOR_SIZE; /* Allocate extra bytes to ensure we have enough for alignment. */ + +#ifndef DR_FLAC_NO_OGG + /* There's additional data required for Ogg streams. */ + if (init.container == drflac_container_ogg) { + allocationSize += sizeof(drflac_oggbs); + + pOggbs = (drflac_oggbs*)drflac__malloc_from_callbacks(sizeof(*pOggbs), &allocationCallbacks); + if (pOggbs == NULL) { + return NULL; /*DRFLAC_OUT_OF_MEMORY;*/ + } + + DRFLAC_ZERO_MEMORY(pOggbs, sizeof(*pOggbs)); + pOggbs->onRead = onRead; + pOggbs->onSeek = onSeek; + pOggbs->onTell = onTell; + pOggbs->pUserData = pUserData; + pOggbs->currentBytePos = init.oggFirstBytePos; + pOggbs->firstBytePos = init.oggFirstBytePos; + pOggbs->serialNumber = init.oggSerial; + pOggbs->bosPageHeader = init.oggBosHeader; + pOggbs->bytesRemainingInPage = 0; + } +#endif + + /* + This part is a bit awkward. We need to load the seektable so that it can be referenced in-memory, but I want the drflac object to + consist of only a single heap allocation. To this, the size of the seek table needs to be known, which we determine when reading + and decoding the metadata. + */ + firstFramePos = 42; /* <-- We know we are at byte 42 at this point. */ + seektablePos = 0; + seekpointCount = 0; + if (init.hasMetadataBlocks) { + drflac_read_proc onReadOverride = onRead; + drflac_seek_proc onSeekOverride = onSeek; + drflac_tell_proc onTellOverride = onTell; + void* pUserDataOverride = pUserData; + +#ifndef DR_FLAC_NO_OGG + if (init.container == drflac_container_ogg) { + onReadOverride = drflac__on_read_ogg; + onSeekOverride = drflac__on_seek_ogg; + onTellOverride = drflac__on_tell_ogg; + pUserDataOverride = (void*)pOggbs; + } +#endif + + if (!drflac__read_and_decode_metadata(onReadOverride, onSeekOverride, onTellOverride, onMeta, pUserDataOverride, pUserDataMD, &firstFramePos, &seektablePos, &seekpointCount, &allocationCallbacks)) { + #ifndef DR_FLAC_NO_OGG + drflac__free_from_callbacks(pOggbs, &allocationCallbacks); + #endif + return NULL; + } + + if ((0xFFFFFFFF - (seekpointCount * sizeof(drflac_seekpoint))) < allocationSize) { + #ifndef DR_FLAC_NO_OGG + drflac__free_from_callbacks(pOggbs, &allocationCallbacks); + #endif + return NULL; + } + + allocationSize += seekpointCount * sizeof(drflac_seekpoint); + } + + pFlac = (drflac*)drflac__malloc_from_callbacks((size_t)allocationSize, &allocationCallbacks); + if (pFlac == NULL) { + #ifndef DR_FLAC_NO_OGG + drflac__free_from_callbacks(pOggbs, &allocationCallbacks); + #endif + return NULL; + } + + drflac__init_from_info(pFlac, &init); + pFlac->allocationCallbacks = allocationCallbacks; + pFlac->pDecodedSamples = (drflac_int32*)drflac_align((size_t)pFlac->pExtraData, DRFLAC_MAX_SIMD_VECTOR_SIZE); + +#ifndef DR_FLAC_NO_OGG + if (init.container == drflac_container_ogg) { + drflac_oggbs* pInternalOggbs = (drflac_oggbs*)((drflac_uint8*)pFlac->pDecodedSamples + decodedSamplesAllocationSize + (seekpointCount * sizeof(drflac_seekpoint))); + DRFLAC_COPY_MEMORY(pInternalOggbs, pOggbs, sizeof(*pOggbs)); + + /* At this point the pOggbs object has been handed over to pInternalOggbs and can be freed. */ + drflac__free_from_callbacks(pOggbs, &allocationCallbacks); + pOggbs = NULL; + + /* The Ogg bistream needs to be layered on top of the original bitstream. */ + pFlac->bs.onRead = drflac__on_read_ogg; + pFlac->bs.onSeek = drflac__on_seek_ogg; + pFlac->bs.onTell = drflac__on_tell_ogg; + pFlac->bs.pUserData = (void*)pInternalOggbs; + pFlac->_oggbs = (void*)pInternalOggbs; + } +#endif + + pFlac->firstFLACFramePosInBytes = firstFramePos; + + /* NOTE: Seektables are not currently compatible with Ogg encapsulation (Ogg has its own accelerated seeking system). I may change this later, so I'm leaving this here for now. */ +#ifndef DR_FLAC_NO_OGG + if (init.container == drflac_container_ogg) + { + pFlac->pSeekpoints = NULL; + pFlac->seekpointCount = 0; + } + else +#endif + { + /* If we have a seektable we need to load it now, making sure we move back to where we were previously. */ + if (seektablePos != 0) { + pFlac->seekpointCount = seekpointCount; + pFlac->pSeekpoints = (drflac_seekpoint*)((drflac_uint8*)pFlac->pDecodedSamples + decodedSamplesAllocationSize); + + DRFLAC_ASSERT(pFlac->bs.onSeek != NULL); + DRFLAC_ASSERT(pFlac->bs.onRead != NULL); + + /* Seek to the seektable, then just read directly into our seektable buffer. */ + if (pFlac->bs.onSeek(pFlac->bs.pUserData, (int)seektablePos, DRFLAC_SEEK_SET)) { + drflac_uint32 iSeekpoint; + + for (iSeekpoint = 0; iSeekpoint < seekpointCount; iSeekpoint += 1) { + if (pFlac->bs.onRead(pFlac->bs.pUserData, pFlac->pSeekpoints + iSeekpoint, DRFLAC_SEEKPOINT_SIZE_IN_BYTES) == DRFLAC_SEEKPOINT_SIZE_IN_BYTES) { + /* Endian swap. */ + pFlac->pSeekpoints[iSeekpoint].firstPCMFrame = drflac__be2host_64(pFlac->pSeekpoints[iSeekpoint].firstPCMFrame); + pFlac->pSeekpoints[iSeekpoint].flacFrameOffset = drflac__be2host_64(pFlac->pSeekpoints[iSeekpoint].flacFrameOffset); + pFlac->pSeekpoints[iSeekpoint].pcmFrameCount = drflac__be2host_16(pFlac->pSeekpoints[iSeekpoint].pcmFrameCount); + } else { + /* Failed to read the seektable. Pretend we don't have one. */ + pFlac->pSeekpoints = NULL; + pFlac->seekpointCount = 0; + break; + } + } + + /* We need to seek back to where we were. If this fails it's a critical error. */ + if (!pFlac->bs.onSeek(pFlac->bs.pUserData, (int)pFlac->firstFLACFramePosInBytes, DRFLAC_SEEK_SET)) { + drflac__free_from_callbacks(pFlac, &allocationCallbacks); + return NULL; + } + } else { + /* Failed to seek to the seektable. Ominous sign, but for now we can just pretend we don't have one. */ + pFlac->pSeekpoints = NULL; + pFlac->seekpointCount = 0; + } + } + } + + + /* + If we get here, but don't have a STREAMINFO block, it means we've opened the stream in relaxed mode and need to decode + the first frame. + */ + if (!init.hasStreamInfoBlock) { + pFlac->currentFLACFrame.header = init.firstFrameHeader; + for (;;) { + drflac_result result = drflac__decode_flac_frame(pFlac); + if (result == DRFLAC_SUCCESS) { + break; + } else { + if (result == DRFLAC_CRC_MISMATCH) { + if (!drflac__read_next_flac_frame_header(&pFlac->bs, pFlac->bitsPerSample, &pFlac->currentFLACFrame.header)) { + drflac__free_from_callbacks(pFlac, &allocationCallbacks); + return NULL; + } + continue; + } else { + drflac__free_from_callbacks(pFlac, &allocationCallbacks); + return NULL; + } + } + } + } + + return pFlac; +} + + + +#ifndef DR_FLAC_NO_STDIO +#include <stdio.h> +#ifndef DR_FLAC_NO_WCHAR +#include <wchar.h> /* For wcslen(), wcsrtombs() */ +#endif + +/* Errno */ +/* drflac_result_from_errno() is only used for fopen() and wfopen() so putting it inside DR_WAV_NO_STDIO for now. If something else needs this later we can move it out. */ +#include <errno.h> +static drflac_result drflac_result_from_errno(int e) +{ + switch (e) + { + case 0: return DRFLAC_SUCCESS; + #ifdef EPERM + case EPERM: return DRFLAC_INVALID_OPERATION; + #endif + #ifdef ENOENT + case ENOENT: return DRFLAC_DOES_NOT_EXIST; + #endif + #ifdef ESRCH + case ESRCH: return DRFLAC_DOES_NOT_EXIST; + #endif + #ifdef EINTR + case EINTR: return DRFLAC_INTERRUPT; + #endif + #ifdef EIO + case EIO: return DRFLAC_IO_ERROR; + #endif + #ifdef ENXIO + case ENXIO: return DRFLAC_DOES_NOT_EXIST; + #endif + #ifdef E2BIG + case E2BIG: return DRFLAC_INVALID_ARGS; + #endif + #ifdef ENOEXEC + case ENOEXEC: return DRFLAC_INVALID_FILE; + #endif + #ifdef EBADF + case EBADF: return DRFLAC_INVALID_FILE; + #endif + #ifdef ECHILD + case ECHILD: return DRFLAC_ERROR; + #endif + #ifdef EAGAIN + case EAGAIN: return DRFLAC_UNAVAILABLE; + #endif + #ifdef ENOMEM + case ENOMEM: return DRFLAC_OUT_OF_MEMORY; + #endif + #ifdef EACCES + case EACCES: return DRFLAC_ACCESS_DENIED; + #endif + #ifdef EFAULT + case EFAULT: return DRFLAC_BAD_ADDRESS; + #endif + #ifdef ENOTBLK + case ENOTBLK: return DRFLAC_ERROR; + #endif + #ifdef EBUSY + case EBUSY: return DRFLAC_BUSY; + #endif + #ifdef EEXIST + case EEXIST: return DRFLAC_ALREADY_EXISTS; + #endif + #ifdef EXDEV + case EXDEV: return DRFLAC_ERROR; + #endif + #ifdef ENODEV + case ENODEV: return DRFLAC_DOES_NOT_EXIST; + #endif + #ifdef ENOTDIR + case ENOTDIR: return DRFLAC_NOT_DIRECTORY; + #endif + #ifdef EISDIR + case EISDIR: return DRFLAC_IS_DIRECTORY; + #endif + #ifdef EINVAL + case EINVAL: return DRFLAC_INVALID_ARGS; + #endif + #ifdef ENFILE + case ENFILE: return DRFLAC_TOO_MANY_OPEN_FILES; + #endif + #ifdef EMFILE + case EMFILE: return DRFLAC_TOO_MANY_OPEN_FILES; + #endif + #ifdef ENOTTY + case ENOTTY: return DRFLAC_INVALID_OPERATION; + #endif + #ifdef ETXTBSY + case ETXTBSY: return DRFLAC_BUSY; + #endif + #ifdef EFBIG + case EFBIG: return DRFLAC_TOO_BIG; + #endif + #ifdef ENOSPC + case ENOSPC: return DRFLAC_NO_SPACE; + #endif + #ifdef ESPIPE + case ESPIPE: return DRFLAC_BAD_SEEK; + #endif + #ifdef EROFS + case EROFS: return DRFLAC_ACCESS_DENIED; + #endif + #ifdef EMLINK + case EMLINK: return DRFLAC_TOO_MANY_LINKS; + #endif + #ifdef EPIPE + case EPIPE: return DRFLAC_BAD_PIPE; + #endif + #ifdef EDOM + case EDOM: return DRFLAC_OUT_OF_RANGE; + #endif + #ifdef ERANGE + case ERANGE: return DRFLAC_OUT_OF_RANGE; + #endif + #ifdef EDEADLK + case EDEADLK: return DRFLAC_DEADLOCK; + #endif + #ifdef ENAMETOOLONG + case ENAMETOOLONG: return DRFLAC_PATH_TOO_LONG; + #endif + #ifdef ENOLCK + case ENOLCK: return DRFLAC_ERROR; + #endif + #ifdef ENOSYS + case ENOSYS: return DRFLAC_NOT_IMPLEMENTED; + #endif + #if defined(ENOTEMPTY) && ENOTEMPTY != EEXIST /* In AIX, ENOTEMPTY and EEXIST use the same value. */ + case ENOTEMPTY: return DRFLAC_DIRECTORY_NOT_EMPTY; + #endif + #ifdef ELOOP + case ELOOP: return DRFLAC_TOO_MANY_LINKS; + #endif + #ifdef ENOMSG + case ENOMSG: return DRFLAC_NO_MESSAGE; + #endif + #ifdef EIDRM + case EIDRM: return DRFLAC_ERROR; + #endif + #ifdef ECHRNG + case ECHRNG: return DRFLAC_ERROR; + #endif + #ifdef EL2NSYNC + case EL2NSYNC: return DRFLAC_ERROR; + #endif + #ifdef EL3HLT + case EL3HLT: return DRFLAC_ERROR; + #endif + #ifdef EL3RST + case EL3RST: return DRFLAC_ERROR; + #endif + #ifdef ELNRNG + case ELNRNG: return DRFLAC_OUT_OF_RANGE; + #endif + #ifdef EUNATCH + case EUNATCH: return DRFLAC_ERROR; + #endif + #ifdef ENOCSI + case ENOCSI: return DRFLAC_ERROR; + #endif + #ifdef EL2HLT + case EL2HLT: return DRFLAC_ERROR; + #endif + #ifdef EBADE + case EBADE: return DRFLAC_ERROR; + #endif + #ifdef EBADR + case EBADR: return DRFLAC_ERROR; + #endif + #ifdef EXFULL + case EXFULL: return DRFLAC_ERROR; + #endif + #ifdef ENOANO + case ENOANO: return DRFLAC_ERROR; + #endif + #ifdef EBADRQC + case EBADRQC: return DRFLAC_ERROR; + #endif + #ifdef EBADSLT + case EBADSLT: return DRFLAC_ERROR; + #endif + #ifdef EBFONT + case EBFONT: return DRFLAC_INVALID_FILE; + #endif + #ifdef ENOSTR + case ENOSTR: return DRFLAC_ERROR; + #endif + #ifdef ENODATA + case ENODATA: return DRFLAC_NO_DATA_AVAILABLE; + #endif + #ifdef ETIME + case ETIME: return DRFLAC_TIMEOUT; + #endif + #ifdef ENOSR + case ENOSR: return DRFLAC_NO_DATA_AVAILABLE; + #endif + #ifdef ENONET + case ENONET: return DRFLAC_NO_NETWORK; + #endif + #ifdef ENOPKG + case ENOPKG: return DRFLAC_ERROR; + #endif + #ifdef EREMOTE + case EREMOTE: return DRFLAC_ERROR; + #endif + #ifdef ENOLINK + case ENOLINK: return DRFLAC_ERROR; + #endif + #ifdef EADV + case EADV: return DRFLAC_ERROR; + #endif + #ifdef ESRMNT + case ESRMNT: return DRFLAC_ERROR; + #endif + #ifdef ECOMM + case ECOMM: return DRFLAC_ERROR; + #endif + #ifdef EPROTO + case EPROTO: return DRFLAC_ERROR; + #endif + #ifdef EMULTIHOP + case EMULTIHOP: return DRFLAC_ERROR; + #endif + #ifdef EDOTDOT + case EDOTDOT: return DRFLAC_ERROR; + #endif + #ifdef EBADMSG + case EBADMSG: return DRFLAC_BAD_MESSAGE; + #endif + #ifdef EOVERFLOW + case EOVERFLOW: return DRFLAC_TOO_BIG; + #endif + #ifdef ENOTUNIQ + case ENOTUNIQ: return DRFLAC_NOT_UNIQUE; + #endif + #ifdef EBADFD + case EBADFD: return DRFLAC_ERROR; + #endif + #ifdef EREMCHG + case EREMCHG: return DRFLAC_ERROR; + #endif + #ifdef ELIBACC + case ELIBACC: return DRFLAC_ACCESS_DENIED; + #endif + #ifdef ELIBBAD + case ELIBBAD: return DRFLAC_INVALID_FILE; + #endif + #ifdef ELIBSCN + case ELIBSCN: return DRFLAC_INVALID_FILE; + #endif + #ifdef ELIBMAX + case ELIBMAX: return DRFLAC_ERROR; + #endif + #ifdef ELIBEXEC + case ELIBEXEC: return DRFLAC_ERROR; + #endif + #ifdef EILSEQ + case EILSEQ: return DRFLAC_INVALID_DATA; + #endif + #ifdef ERESTART + case ERESTART: return DRFLAC_ERROR; + #endif + #ifdef ESTRPIPE + case ESTRPIPE: return DRFLAC_ERROR; + #endif + #ifdef EUSERS + case EUSERS: return DRFLAC_ERROR; + #endif + #ifdef ENOTSOCK + case ENOTSOCK: return DRFLAC_NOT_SOCKET; + #endif + #ifdef EDESTADDRREQ + case EDESTADDRREQ: return DRFLAC_NO_ADDRESS; + #endif + #ifdef EMSGSIZE + case EMSGSIZE: return DRFLAC_TOO_BIG; + #endif + #ifdef EPROTOTYPE + case EPROTOTYPE: return DRFLAC_BAD_PROTOCOL; + #endif + #ifdef ENOPROTOOPT + case ENOPROTOOPT: return DRFLAC_PROTOCOL_UNAVAILABLE; + #endif + #ifdef EPROTONOSUPPORT + case EPROTONOSUPPORT: return DRFLAC_PROTOCOL_NOT_SUPPORTED; + #endif + #ifdef ESOCKTNOSUPPORT + case ESOCKTNOSUPPORT: return DRFLAC_SOCKET_NOT_SUPPORTED; + #endif + #ifdef EOPNOTSUPP + case EOPNOTSUPP: return DRFLAC_INVALID_OPERATION; + #endif + #ifdef EPFNOSUPPORT + case EPFNOSUPPORT: return DRFLAC_PROTOCOL_FAMILY_NOT_SUPPORTED; + #endif + #ifdef EAFNOSUPPORT + case EAFNOSUPPORT: return DRFLAC_ADDRESS_FAMILY_NOT_SUPPORTED; + #endif + #ifdef EADDRINUSE + case EADDRINUSE: return DRFLAC_ALREADY_IN_USE; + #endif + #ifdef EADDRNOTAVAIL + case EADDRNOTAVAIL: return DRFLAC_ERROR; + #endif + #ifdef ENETDOWN + case ENETDOWN: return DRFLAC_NO_NETWORK; + #endif + #ifdef ENETUNREACH + case ENETUNREACH: return DRFLAC_NO_NETWORK; + #endif + #ifdef ENETRESET + case ENETRESET: return DRFLAC_NO_NETWORK; + #endif + #ifdef ECONNABORTED + case ECONNABORTED: return DRFLAC_NO_NETWORK; + #endif + #ifdef ECONNRESET + case ECONNRESET: return DRFLAC_CONNECTION_RESET; + #endif + #ifdef ENOBUFS + case ENOBUFS: return DRFLAC_NO_SPACE; + #endif + #ifdef EISCONN + case EISCONN: return DRFLAC_ALREADY_CONNECTED; + #endif + #ifdef ENOTCONN + case ENOTCONN: return DRFLAC_NOT_CONNECTED; + #endif + #ifdef ESHUTDOWN + case ESHUTDOWN: return DRFLAC_ERROR; + #endif + #ifdef ETOOMANYREFS + case ETOOMANYREFS: return DRFLAC_ERROR; + #endif + #ifdef ETIMEDOUT + case ETIMEDOUT: return DRFLAC_TIMEOUT; + #endif + #ifdef ECONNREFUSED + case ECONNREFUSED: return DRFLAC_CONNECTION_REFUSED; + #endif + #ifdef EHOSTDOWN + case EHOSTDOWN: return DRFLAC_NO_HOST; + #endif + #ifdef EHOSTUNREACH + case EHOSTUNREACH: return DRFLAC_NO_HOST; + #endif + #ifdef EALREADY + case EALREADY: return DRFLAC_IN_PROGRESS; + #endif + #ifdef EINPROGRESS + case EINPROGRESS: return DRFLAC_IN_PROGRESS; + #endif + #ifdef ESTALE + case ESTALE: return DRFLAC_INVALID_FILE; + #endif + #ifdef EUCLEAN + case EUCLEAN: return DRFLAC_ERROR; + #endif + #ifdef ENOTNAM + case ENOTNAM: return DRFLAC_ERROR; + #endif + #ifdef ENAVAIL + case ENAVAIL: return DRFLAC_ERROR; + #endif + #ifdef EISNAM + case EISNAM: return DRFLAC_ERROR; + #endif + #ifdef EREMOTEIO + case EREMOTEIO: return DRFLAC_IO_ERROR; + #endif + #ifdef EDQUOT + case EDQUOT: return DRFLAC_NO_SPACE; + #endif + #ifdef ENOMEDIUM + case ENOMEDIUM: return DRFLAC_DOES_NOT_EXIST; + #endif + #ifdef EMEDIUMTYPE + case EMEDIUMTYPE: return DRFLAC_ERROR; + #endif + #ifdef ECANCELED + case ECANCELED: return DRFLAC_CANCELLED; + #endif + #ifdef ENOKEY + case ENOKEY: return DRFLAC_ERROR; + #endif + #ifdef EKEYEXPIRED + case EKEYEXPIRED: return DRFLAC_ERROR; + #endif + #ifdef EKEYREVOKED + case EKEYREVOKED: return DRFLAC_ERROR; + #endif + #ifdef EKEYREJECTED + case EKEYREJECTED: return DRFLAC_ERROR; + #endif + #ifdef EOWNERDEAD + case EOWNERDEAD: return DRFLAC_ERROR; + #endif + #ifdef ENOTRECOVERABLE + case ENOTRECOVERABLE: return DRFLAC_ERROR; + #endif + #ifdef ERFKILL + case ERFKILL: return DRFLAC_ERROR; + #endif + #ifdef EHWPOISON + case EHWPOISON: return DRFLAC_ERROR; + #endif + default: return DRFLAC_ERROR; + } +} +/* End Errno */ + +/* fopen */ +static drflac_result drflac_fopen(FILE** ppFile, const char* pFilePath, const char* pOpenMode) +{ +#if defined(_MSC_VER) && _MSC_VER >= 1400 + errno_t err; +#endif + + if (ppFile != NULL) { + *ppFile = NULL; /* Safety. */ + } + + if (pFilePath == NULL || pOpenMode == NULL || ppFile == NULL) { + return DRFLAC_INVALID_ARGS; + } + +#if defined(_MSC_VER) && _MSC_VER >= 1400 + err = fopen_s(ppFile, pFilePath, pOpenMode); + if (err != 0) { + return drflac_result_from_errno(err); + } +#else +#if defined(_WIN32) || defined(__APPLE__) + *ppFile = fopen(pFilePath, pOpenMode); +#else + #if defined(_FILE_OFFSET_BITS) && _FILE_OFFSET_BITS == 64 && defined(_LARGEFILE64_SOURCE) + *ppFile = fopen64(pFilePath, pOpenMode); + #else + *ppFile = fopen(pFilePath, pOpenMode); + #endif +#endif + if (*ppFile == NULL) { + drflac_result result = drflac_result_from_errno(errno); + if (result == DRFLAC_SUCCESS) { + result = DRFLAC_ERROR; /* Just a safety check to make sure we never ever return success when pFile == NULL. */ + } + + return result; + } +#endif + + return DRFLAC_SUCCESS; +} + +/* +_wfopen() isn't always available in all compilation environments. + + * Windows only. + * MSVC seems to support it universally as far back as VC6 from what I can tell (haven't checked further back). + * MinGW-64 (both 32- and 64-bit) seems to support it. + * MinGW wraps it in !defined(__STRICT_ANSI__). + * OpenWatcom wraps it in !defined(_NO_EXT_KEYS). + +This can be reviewed as compatibility issues arise. The preference is to use _wfopen_s() and _wfopen() as opposed to the wcsrtombs() +fallback, so if you notice your compiler not detecting this properly I'm happy to look at adding support. +*/ +#if defined(_WIN32) + #if defined(_MSC_VER) || defined(__MINGW64__) || (!defined(__STRICT_ANSI__) && !defined(_NO_EXT_KEYS)) + #define DRFLAC_HAS_WFOPEN + #endif +#endif + +#ifndef DR_FLAC_NO_WCHAR +static drflac_result drflac_wfopen(FILE** ppFile, const wchar_t* pFilePath, const wchar_t* pOpenMode, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + if (ppFile != NULL) { + *ppFile = NULL; /* Safety. */ + } + + if (pFilePath == NULL || pOpenMode == NULL || ppFile == NULL) { + return DRFLAC_INVALID_ARGS; + } + +#if defined(DRFLAC_HAS_WFOPEN) + { + /* Use _wfopen() on Windows. */ + #if defined(_MSC_VER) && _MSC_VER >= 1400 + errno_t err = _wfopen_s(ppFile, pFilePath, pOpenMode); + if (err != 0) { + return drflac_result_from_errno(err); + } + #else + *ppFile = _wfopen(pFilePath, pOpenMode); + if (*ppFile == NULL) { + return drflac_result_from_errno(errno); + } + #endif + (void)pAllocationCallbacks; + } +#else + /* + Use fopen() on anything other than Windows. Requires a conversion. This is annoying because + fopen() is locale specific. The only real way I can think of to do this is with wcsrtombs(). Note + that wcstombs() is apparently not thread-safe because it uses a static global mbstate_t object for + maintaining state. I've checked this with -std=c89 and it works, but if somebody get's a compiler + error I'll look into improving compatibility. + */ + + /* + Some compilers don't support wchar_t or wcsrtombs() which we're using below. In this case we just + need to abort with an error. If you encounter a compiler lacking such support, add it to this list + and submit a bug report and it'll be added to the library upstream. + */ + #if defined(__DJGPP__) + { + /* Nothing to do here. This will fall through to the error check below. */ + } + #else + { + mbstate_t mbs; + size_t lenMB; + const wchar_t* pFilePathTemp = pFilePath; + char* pFilePathMB = NULL; + char pOpenModeMB[32] = {0}; + + /* Get the length first. */ + DRFLAC_ZERO_OBJECT(&mbs); + lenMB = wcsrtombs(NULL, &pFilePathTemp, 0, &mbs); + if (lenMB == (size_t)-1) { + return drflac_result_from_errno(errno); + } + + pFilePathMB = (char*)drflac__malloc_from_callbacks(lenMB + 1, pAllocationCallbacks); + if (pFilePathMB == NULL) { + return DRFLAC_OUT_OF_MEMORY; + } + + pFilePathTemp = pFilePath; + DRFLAC_ZERO_OBJECT(&mbs); + wcsrtombs(pFilePathMB, &pFilePathTemp, lenMB + 1, &mbs); + + /* The open mode should always consist of ASCII characters so we should be able to do a trivial conversion. */ + { + size_t i = 0; + for (;;) { + if (pOpenMode[i] == 0) { + pOpenModeMB[i] = '\0'; + break; + } + + pOpenModeMB[i] = (char)pOpenMode[i]; + i += 1; + } + } + + *ppFile = fopen(pFilePathMB, pOpenModeMB); + + drflac__free_from_callbacks(pFilePathMB, pAllocationCallbacks); + } + #endif + + if (*ppFile == NULL) { + return DRFLAC_ERROR; + } +#endif + + return DRFLAC_SUCCESS; +} +#endif +/* End fopen */ + +static size_t drflac__on_read_stdio(void* pUserData, void* bufferOut, size_t bytesToRead) +{ + return fread(bufferOut, 1, bytesToRead, (FILE*)pUserData); +} + +static drflac_bool32 drflac__on_seek_stdio(void* pUserData, int offset, drflac_seek_origin origin) +{ + int whence = SEEK_SET; + if (origin == DRFLAC_SEEK_CUR) { + whence = SEEK_CUR; + } else if (origin == DRFLAC_SEEK_END) { + whence = SEEK_END; + } + + return fseek((FILE*)pUserData, offset, whence) == 0; +} + +static drflac_bool32 drflac__on_tell_stdio(void* pUserData, drflac_int64* pCursor) +{ + FILE* pFileStdio = (FILE*)pUserData; + drflac_int64 result; + + /* These were all validated at a higher level. */ + DRFLAC_ASSERT(pFileStdio != NULL); + DRFLAC_ASSERT(pCursor != NULL); + +#if defined(_WIN32) && !defined(NXDK) + #if defined(_MSC_VER) && _MSC_VER > 1200 + result = _ftelli64(pFileStdio); + #else + result = ftell(pFileStdio); + #endif +#else + result = ftell(pFileStdio); +#endif + + *pCursor = result; + + return DRFLAC_TRUE; +} + + + +DRFLAC_API drflac* drflac_open_file(const char* pFileName, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + FILE* pFile; + + if (drflac_fopen(&pFile, pFileName, "rb") != DRFLAC_SUCCESS) { + return NULL; + } + + pFlac = drflac_open(drflac__on_read_stdio, drflac__on_seek_stdio, drflac__on_tell_stdio, (void*)pFile, pAllocationCallbacks); + if (pFlac == NULL) { + fclose(pFile); + return NULL; + } + + return pFlac; +} + +#ifndef DR_FLAC_NO_WCHAR +DRFLAC_API drflac* drflac_open_file_w(const wchar_t* pFileName, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + FILE* pFile; + + if (drflac_wfopen(&pFile, pFileName, L"rb", pAllocationCallbacks) != DRFLAC_SUCCESS) { + return NULL; + } + + pFlac = drflac_open(drflac__on_read_stdio, drflac__on_seek_stdio, drflac__on_tell_stdio, (void*)pFile, pAllocationCallbacks); + if (pFlac == NULL) { + fclose(pFile); + return NULL; + } + + return pFlac; +} +#endif + +DRFLAC_API drflac* drflac_open_file_with_metadata(const char* pFileName, drflac_meta_proc onMeta, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + FILE* pFile; + + if (drflac_fopen(&pFile, pFileName, "rb") != DRFLAC_SUCCESS) { + return NULL; + } + + pFlac = drflac_open_with_metadata_private(drflac__on_read_stdio, drflac__on_seek_stdio, drflac__on_tell_stdio, onMeta, drflac_container_unknown, (void*)pFile, pUserData, pAllocationCallbacks); + if (pFlac == NULL) { + fclose(pFile); + return pFlac; + } + + return pFlac; +} + +#ifndef DR_FLAC_NO_WCHAR +DRFLAC_API drflac* drflac_open_file_with_metadata_w(const wchar_t* pFileName, drflac_meta_proc onMeta, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + FILE* pFile; + + if (drflac_wfopen(&pFile, pFileName, L"rb", pAllocationCallbacks) != DRFLAC_SUCCESS) { + return NULL; + } + + pFlac = drflac_open_with_metadata_private(drflac__on_read_stdio, drflac__on_seek_stdio, drflac__on_tell_stdio, onMeta, drflac_container_unknown, (void*)pFile, pUserData, pAllocationCallbacks); + if (pFlac == NULL) { + fclose(pFile); + return pFlac; + } + + return pFlac; +} +#endif +#endif /* DR_FLAC_NO_STDIO */ + +static size_t drflac__on_read_memory(void* pUserData, void* bufferOut, size_t bytesToRead) +{ + drflac__memory_stream* memoryStream = (drflac__memory_stream*)pUserData; + size_t bytesRemaining; + + DRFLAC_ASSERT(memoryStream != NULL); + DRFLAC_ASSERT(memoryStream->dataSize >= memoryStream->currentReadPos); + + bytesRemaining = memoryStream->dataSize - memoryStream->currentReadPos; + if (bytesToRead > bytesRemaining) { + bytesToRead = bytesRemaining; + } + + if (bytesToRead > 0) { + DRFLAC_COPY_MEMORY(bufferOut, memoryStream->data + memoryStream->currentReadPos, bytesToRead); + memoryStream->currentReadPos += bytesToRead; + } + + return bytesToRead; +} + +static drflac_bool32 drflac__on_seek_memory(void* pUserData, int offset, drflac_seek_origin origin) +{ + drflac__memory_stream* memoryStream = (drflac__memory_stream*)pUserData; + drflac_int64 newCursor; + + DRFLAC_ASSERT(memoryStream != NULL); + + if (origin == DRFLAC_SEEK_SET) { + newCursor = 0; + } else if (origin == DRFLAC_SEEK_CUR) { + newCursor = (drflac_int64)memoryStream->currentReadPos; + } else if (origin == DRFLAC_SEEK_END) { + newCursor = (drflac_int64)memoryStream->dataSize; + } else { + DRFLAC_ASSERT(!"Invalid seek origin"); + return DRFLAC_FALSE; + } + + newCursor += offset; + + if (newCursor < 0) { + return DRFLAC_FALSE; /* Trying to seek prior to the start of the buffer. */ + } + if ((size_t)newCursor > memoryStream->dataSize) { + return DRFLAC_FALSE; /* Trying to seek beyond the end of the buffer. */ + } + + memoryStream->currentReadPos = (size_t)newCursor; + + return DRFLAC_TRUE; +} + +static drflac_bool32 drflac__on_tell_memory(void* pUserData, drflac_int64* pCursor) +{ + drflac__memory_stream* memoryStream = (drflac__memory_stream*)pUserData; + + DRFLAC_ASSERT(memoryStream != NULL); + DRFLAC_ASSERT(pCursor != NULL); + + *pCursor = (drflac_int64)memoryStream->currentReadPos; + return DRFLAC_TRUE; +} + +DRFLAC_API drflac* drflac_open_memory(const void* pData, size_t dataSize, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac__memory_stream memoryStream; + drflac* pFlac; + + memoryStream.data = (const drflac_uint8*)pData; + memoryStream.dataSize = dataSize; + memoryStream.currentReadPos = 0; + pFlac = drflac_open(drflac__on_read_memory, drflac__on_seek_memory, drflac__on_tell_memory, &memoryStream, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + pFlac->memoryStream = memoryStream; + + /* This is an awful hack... */ +#ifndef DR_FLAC_NO_OGG + if (pFlac->container == drflac_container_ogg) + { + drflac_oggbs* oggbs = (drflac_oggbs*)pFlac->_oggbs; + oggbs->pUserData = &pFlac->memoryStream; + } + else +#endif + { + pFlac->bs.pUserData = &pFlac->memoryStream; + } + + return pFlac; +} + +DRFLAC_API drflac* drflac_open_memory_with_metadata(const void* pData, size_t dataSize, drflac_meta_proc onMeta, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac__memory_stream memoryStream; + drflac* pFlac; + + memoryStream.data = (const drflac_uint8*)pData; + memoryStream.dataSize = dataSize; + memoryStream.currentReadPos = 0; + pFlac = drflac_open_with_metadata_private(drflac__on_read_memory, drflac__on_seek_memory, drflac__on_tell_memory, onMeta, drflac_container_unknown, &memoryStream, pUserData, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + pFlac->memoryStream = memoryStream; + + /* This is an awful hack... */ +#ifndef DR_FLAC_NO_OGG + if (pFlac->container == drflac_container_ogg) + { + drflac_oggbs* oggbs = (drflac_oggbs*)pFlac->_oggbs; + oggbs->pUserData = &pFlac->memoryStream; + } + else +#endif + { + pFlac->bs.pUserData = &pFlac->memoryStream; + } + + return pFlac; +} + + + +DRFLAC_API drflac* drflac_open(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + return drflac_open_with_metadata_private(onRead, onSeek, onTell, NULL, drflac_container_unknown, pUserData, pUserData, pAllocationCallbacks); +} +DRFLAC_API drflac* drflac_open_relaxed(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_container container, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + return drflac_open_with_metadata_private(onRead, onSeek, onTell, NULL, container, pUserData, pUserData, pAllocationCallbacks); +} + +DRFLAC_API drflac* drflac_open_with_metadata(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_meta_proc onMeta, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + return drflac_open_with_metadata_private(onRead, onSeek, onTell, onMeta, drflac_container_unknown, pUserData, pUserData, pAllocationCallbacks); +} +DRFLAC_API drflac* drflac_open_with_metadata_relaxed(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, drflac_meta_proc onMeta, drflac_container container, void* pUserData, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + return drflac_open_with_metadata_private(onRead, onSeek, onTell, onMeta, container, pUserData, pUserData, pAllocationCallbacks); +} + +DRFLAC_API void drflac_close(drflac* pFlac) +{ + if (pFlac == NULL) { + return; + } + +#ifndef DR_FLAC_NO_STDIO + /* + If we opened the file with drflac_open_file() we will want to close the file handle. We can know whether or not drflac_open_file() + was used by looking at the callbacks. + */ + if (pFlac->bs.onRead == drflac__on_read_stdio) { + fclose((FILE*)pFlac->bs.pUserData); + } + +#ifndef DR_FLAC_NO_OGG + /* Need to clean up Ogg streams a bit differently due to the way the bit streaming is chained. */ + if (pFlac->container == drflac_container_ogg) { + drflac_oggbs* oggbs = (drflac_oggbs*)pFlac->_oggbs; + DRFLAC_ASSERT(pFlac->bs.onRead == drflac__on_read_ogg); + + if (oggbs->onRead == drflac__on_read_stdio) { + fclose((FILE*)oggbs->pUserData); + } + } +#endif +#endif + + drflac__free_from_callbacks(pFlac, &pFlac->allocationCallbacks); +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_left_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + for (i = 0; i < frameCount; ++i) { + drflac_uint32 left = (drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + drflac_uint32 side = (drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + drflac_uint32 right = left - side; + + pOutputSamples[i*2+0] = (drflac_int32)left; + pOutputSamples[i*2+1] = (drflac_int32)right; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_left_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 left0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 left1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 left2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 left3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << shift1; + + drflac_uint32 right0 = left0 - side0; + drflac_uint32 right1 = left1 - side1; + drflac_uint32 right2 = left2 - side2; + drflac_uint32 right3 = left3 - side3; + + pOutputSamples[i*8+0] = (drflac_int32)left0; + pOutputSamples[i*8+1] = (drflac_int32)right0; + pOutputSamples[i*8+2] = (drflac_int32)left1; + pOutputSamples[i*8+3] = (drflac_int32)right1; + pOutputSamples[i*8+4] = (drflac_int32)left2; + pOutputSamples[i*8+5] = (drflac_int32)right2; + pOutputSamples[i*8+6] = (drflac_int32)left3; + pOutputSamples[i*8+7] = (drflac_int32)right3; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + pOutputSamples[i*2+0] = (drflac_int32)left; + pOutputSamples[i*2+1] = (drflac_int32)right; + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_left_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + for (i = 0; i < frameCount4; ++i) { + __m128i left = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + __m128i side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + __m128i right = _mm_sub_epi32(left, side); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 0), _mm_unpacklo_epi32(left, right)); + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 4), _mm_unpackhi_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + pOutputSamples[i*2+0] = (drflac_int32)left; + pOutputSamples[i*2+1] = (drflac_int32)right; + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_left_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + int32x4_t shift0_4; + int32x4_t shift1_4; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + shift0_4 = vdupq_n_s32(shift0); + shift1_4 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + uint32x4_t left; + uint32x4_t side; + uint32x4_t right; + + left = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift0_4); + side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift1_4); + right = vsubq_u32(left, side); + + drflac__vst2q_u32((drflac_uint32*)pOutputSamples + i*8, vzipq_u32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + pOutputSamples[i*2+0] = (drflac_int32)left; + pOutputSamples[i*2+1] = (drflac_int32)right; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_duplicate_mono(drflac_uint64 frameCount, drflac_uint32 shift, const drflac_int32* pInputSamples, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + const drflac_uint32* pInputSamplesU32 = (const drflac_uint32*)pInputSamples; + + DRFLAC_ASSERT(shift < 32); + + for (i = 0; i < frameCount; ++i) { + drflac_int32 sample; + + sample = (drflac_int32)(pInputSamplesU32[i] << shift); + pOutputSamples[i*2+0] = sample; + pOutputSamples[i*2+1] = sample; + } +} + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_left_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + if (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample == 32) { + drflac_read_pcm_frames_s32__decode_duplicate_mono(frameCount, unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample, pInputSamples0, pOutputSamples); + return; + } + +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s32__decode_left_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s32__decode_left_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_s32__decode_left_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_s32__decode_left_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_right_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + for (i = 0; i < frameCount; ++i) { + drflac_uint32 side = (drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + drflac_uint32 right = (drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + drflac_uint32 left = right + side; + + pOutputSamples[i*2+0] = (drflac_int32)left; + pOutputSamples[i*2+1] = (drflac_int32)right; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_right_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 side0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 side1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 side2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 side3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 right0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 right1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 right2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 right3 = pInputSamples1U32[i*4+3] << shift1; + + drflac_uint32 left0 = right0 + side0; + drflac_uint32 left1 = right1 + side1; + drflac_uint32 left2 = right2 + side2; + drflac_uint32 left3 = right3 + side3; + + pOutputSamples[i*8+0] = (drflac_int32)left0; + pOutputSamples[i*8+1] = (drflac_int32)right0; + pOutputSamples[i*8+2] = (drflac_int32)left1; + pOutputSamples[i*8+3] = (drflac_int32)right1; + pOutputSamples[i*8+4] = (drflac_int32)left2; + pOutputSamples[i*8+5] = (drflac_int32)right2; + pOutputSamples[i*8+6] = (drflac_int32)left3; + pOutputSamples[i*8+7] = (drflac_int32)right3; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + pOutputSamples[i*2+0] = (drflac_int32)left; + pOutputSamples[i*2+1] = (drflac_int32)right; + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_right_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + for (i = 0; i < frameCount4; ++i) { + __m128i side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + __m128i right = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + __m128i left = _mm_add_epi32(right, side); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 0), _mm_unpacklo_epi32(left, right)); + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 4), _mm_unpackhi_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + pOutputSamples[i*2+0] = (drflac_int32)left; + pOutputSamples[i*2+1] = (drflac_int32)right; + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_right_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + int32x4_t shift0_4; + int32x4_t shift1_4; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + shift0_4 = vdupq_n_s32(shift0); + shift1_4 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + uint32x4_t side; + uint32x4_t right; + uint32x4_t left; + + side = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift0_4); + right = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift1_4); + left = vaddq_u32(right, side); + + drflac__vst2q_u32((drflac_uint32*)pOutputSamples + i*8, vzipq_u32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + pOutputSamples[i*2+0] = (drflac_int32)left; + pOutputSamples[i*2+1] = (drflac_int32)right; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_right_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + if (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample == 32) { + drflac_read_pcm_frames_s32__decode_duplicate_mono(frameCount, unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample, pInputSamples1, pOutputSamples); + return; + } + +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s32__decode_right_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s32__decode_right_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_s32__decode_right_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_s32__decode_right_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_mid_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + for (drflac_uint64 i = 0; i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)((drflac_uint32)((drflac_int32)(mid + side) >> 1) << unusedBitsPerSample); + pOutputSamples[i*2+1] = (drflac_int32)((drflac_uint32)((drflac_int32)(mid - side) >> 1) << unusedBitsPerSample); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_mid_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_int32 shift = unusedBitsPerSample; + + if (shift > 0) { + shift -= 1; + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 temp0L; + drflac_uint32 temp1L; + drflac_uint32 temp2L; + drflac_uint32 temp3L; + drflac_uint32 temp0R; + drflac_uint32 temp1R; + drflac_uint32 temp2R; + drflac_uint32 temp3R; + + drflac_uint32 mid0 = pInputSamples0U32[i*4+0] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid1 = pInputSamples0U32[i*4+1] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid2 = pInputSamples0U32[i*4+2] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid3 = pInputSamples0U32[i*4+3] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid0 = (mid0 << 1) | (side0 & 0x01); + mid1 = (mid1 << 1) | (side1 & 0x01); + mid2 = (mid2 << 1) | (side2 & 0x01); + mid3 = (mid3 << 1) | (side3 & 0x01); + + temp0L = (mid0 + side0) << shift; + temp1L = (mid1 + side1) << shift; + temp2L = (mid2 + side2) << shift; + temp3L = (mid3 + side3) << shift; + + temp0R = (mid0 - side0) << shift; + temp1R = (mid1 - side1) << shift; + temp2R = (mid2 - side2) << shift; + temp3R = (mid3 - side3) << shift; + + pOutputSamples[i*8+0] = (drflac_int32)temp0L; + pOutputSamples[i*8+1] = (drflac_int32)temp0R; + pOutputSamples[i*8+2] = (drflac_int32)temp1L; + pOutputSamples[i*8+3] = (drflac_int32)temp1R; + pOutputSamples[i*8+4] = (drflac_int32)temp2L; + pOutputSamples[i*8+5] = (drflac_int32)temp2R; + pOutputSamples[i*8+6] = (drflac_int32)temp3L; + pOutputSamples[i*8+7] = (drflac_int32)temp3R; + } + } else { + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 temp0L; + drflac_uint32 temp1L; + drflac_uint32 temp2L; + drflac_uint32 temp3L; + drflac_uint32 temp0R; + drflac_uint32 temp1R; + drflac_uint32 temp2R; + drflac_uint32 temp3R; + + drflac_uint32 mid0 = pInputSamples0U32[i*4+0] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid1 = pInputSamples0U32[i*4+1] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid2 = pInputSamples0U32[i*4+2] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid3 = pInputSamples0U32[i*4+3] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid0 = (mid0 << 1) | (side0 & 0x01); + mid1 = (mid1 << 1) | (side1 & 0x01); + mid2 = (mid2 << 1) | (side2 & 0x01); + mid3 = (mid3 << 1) | (side3 & 0x01); + + temp0L = (drflac_uint32)((drflac_int32)(mid0 + side0) >> 1); + temp1L = (drflac_uint32)((drflac_int32)(mid1 + side1) >> 1); + temp2L = (drflac_uint32)((drflac_int32)(mid2 + side2) >> 1); + temp3L = (drflac_uint32)((drflac_int32)(mid3 + side3) >> 1); + + temp0R = (drflac_uint32)((drflac_int32)(mid0 - side0) >> 1); + temp1R = (drflac_uint32)((drflac_int32)(mid1 - side1) >> 1); + temp2R = (drflac_uint32)((drflac_int32)(mid2 - side2) >> 1); + temp3R = (drflac_uint32)((drflac_int32)(mid3 - side3) >> 1); + + pOutputSamples[i*8+0] = (drflac_int32)temp0L; + pOutputSamples[i*8+1] = (drflac_int32)temp0R; + pOutputSamples[i*8+2] = (drflac_int32)temp1L; + pOutputSamples[i*8+3] = (drflac_int32)temp1R; + pOutputSamples[i*8+4] = (drflac_int32)temp2L; + pOutputSamples[i*8+5] = (drflac_int32)temp2R; + pOutputSamples[i*8+6] = (drflac_int32)temp3L; + pOutputSamples[i*8+7] = (drflac_int32)temp3R; + } + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)((drflac_uint32)((drflac_int32)(mid + side) >> 1) << unusedBitsPerSample); + pOutputSamples[i*2+1] = (drflac_int32)((drflac_uint32)((drflac_int32)(mid - side) >> 1) << unusedBitsPerSample); + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_mid_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_int32 shift = unusedBitsPerSample; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + if (shift == 0) { + for (i = 0; i < frameCount4; ++i) { + __m128i mid; + __m128i side; + __m128i left; + __m128i right; + + mid = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + + mid = _mm_or_si128(_mm_slli_epi32(mid, 1), _mm_and_si128(side, _mm_set1_epi32(0x01))); + + left = _mm_srai_epi32(_mm_add_epi32(mid, side), 1); + right = _mm_srai_epi32(_mm_sub_epi32(mid, side), 1); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 0), _mm_unpacklo_epi32(left, right)); + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 4), _mm_unpackhi_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)(mid + side) >> 1; + pOutputSamples[i*2+1] = (drflac_int32)(mid - side) >> 1; + } + } else { + shift -= 1; + for (i = 0; i < frameCount4; ++i) { + __m128i mid; + __m128i side; + __m128i left; + __m128i right; + + mid = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + + mid = _mm_or_si128(_mm_slli_epi32(mid, 1), _mm_and_si128(side, _mm_set1_epi32(0x01))); + + left = _mm_slli_epi32(_mm_add_epi32(mid, side), shift); + right = _mm_slli_epi32(_mm_sub_epi32(mid, side), shift); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 0), _mm_unpacklo_epi32(left, right)); + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 4), _mm_unpackhi_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)((mid + side) << shift); + pOutputSamples[i*2+1] = (drflac_int32)((mid - side) << shift); + } + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_mid_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_int32 shift = unusedBitsPerSample; + int32x4_t wbpsShift0_4; /* wbps = Wasted Bits Per Sample */ + int32x4_t wbpsShift1_4; /* wbps = Wasted Bits Per Sample */ + uint32x4_t one4; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + wbpsShift0_4 = vdupq_n_s32(pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + wbpsShift1_4 = vdupq_n_s32(pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + one4 = vdupq_n_u32(1); + + if (shift == 0) { + for (i = 0; i < frameCount4; ++i) { + uint32x4_t mid; + uint32x4_t side; + int32x4_t left; + int32x4_t right; + + mid = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), wbpsShift0_4); + side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), wbpsShift1_4); + + mid = vorrq_u32(vshlq_n_u32(mid, 1), vandq_u32(side, one4)); + + left = vshrq_n_s32(vreinterpretq_s32_u32(vaddq_u32(mid, side)), 1); + right = vshrq_n_s32(vreinterpretq_s32_u32(vsubq_u32(mid, side)), 1); + + drflac__vst2q_s32(pOutputSamples + i*8, vzipq_s32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)(mid + side) >> 1; + pOutputSamples[i*2+1] = (drflac_int32)(mid - side) >> 1; + } + } else { + int32x4_t shift4; + + shift -= 1; + shift4 = vdupq_n_s32(shift); + + for (i = 0; i < frameCount4; ++i) { + uint32x4_t mid; + uint32x4_t side; + int32x4_t left; + int32x4_t right; + + mid = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), wbpsShift0_4); + side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), wbpsShift1_4); + + mid = vorrq_u32(vshlq_n_u32(mid, 1), vandq_u32(side, one4)); + + left = vreinterpretq_s32_u32(vshlq_u32(vaddq_u32(mid, side), shift4)); + right = vreinterpretq_s32_u32(vshlq_u32(vsubq_u32(mid, side), shift4)); + + drflac__vst2q_s32(pOutputSamples + i*8, vzipq_s32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)((mid + side) << shift); + pOutputSamples[i*2+1] = (drflac_int32)((mid - side) << shift); + } + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_mid_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s32__decode_mid_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s32__decode_mid_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_s32__decode_mid_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_s32__decode_mid_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_independent_stereo__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + for (drflac_uint64 i = 0; i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int32)((drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample)); + pOutputSamples[i*2+1] = (drflac_int32)((drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample)); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_independent_stereo__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 tempL0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 tempL1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 tempL2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 tempL3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 tempR0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 tempR1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 tempR2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 tempR3 = pInputSamples1U32[i*4+3] << shift1; + + pOutputSamples[i*8+0] = (drflac_int32)tempL0; + pOutputSamples[i*8+1] = (drflac_int32)tempR0; + pOutputSamples[i*8+2] = (drflac_int32)tempL1; + pOutputSamples[i*8+3] = (drflac_int32)tempR1; + pOutputSamples[i*8+4] = (drflac_int32)tempL2; + pOutputSamples[i*8+5] = (drflac_int32)tempR2; + pOutputSamples[i*8+6] = (drflac_int32)tempL3; + pOutputSamples[i*8+7] = (drflac_int32)tempR3; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int32)(pInputSamples0U32[i] << shift0); + pOutputSamples[i*2+1] = (drflac_int32)(pInputSamples1U32[i] << shift1); + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_independent_stereo__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + for (i = 0; i < frameCount4; ++i) { + __m128i left = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + __m128i right = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 0), _mm_unpacklo_epi32(left, right)); + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8 + 4), _mm_unpackhi_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int32)(pInputSamples0U32[i] << shift0); + pOutputSamples[i*2+1] = (drflac_int32)(pInputSamples1U32[i] << shift1); + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_independent_stereo__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + int32x4_t shift4_0 = vdupq_n_s32(shift0); + int32x4_t shift4_1 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + int32x4_t left; + int32x4_t right; + + left = vreinterpretq_s32_u32(vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift4_0)); + right = vreinterpretq_s32_u32(vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift4_1)); + + drflac__vst2q_s32(pOutputSamples + i*8, vzipq_s32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int32)(pInputSamples0U32[i] << shift0); + pOutputSamples[i*2+1] = (drflac_int32)(pInputSamples1U32[i] << shift1); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s32__decode_independent_stereo(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int32* pOutputSamples) +{ +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s32__decode_independent_stereo__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s32__decode_independent_stereo__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_s32__decode_independent_stereo__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_s32__decode_independent_stereo__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +static drflac_bool32 drflac__is_current_flac_frame_valid(drflac* pFlac) +{ + drflac_uint32 iChannel; + drflac_uint32 channelCount; + + if (pFlac->currentFLACFrame.header.blockSizeInPCMFrames > pFlac->maxBlockSizeInPCMFrames || pFlac->currentFLACFrame.pcmFramesRemaining > pFlac->currentFLACFrame.header.blockSizeInPCMFrames) { + return DRFLAC_FALSE; + } + + /* The frame's channel count must agree with STREAMINFO, otherwise the subframes below may not have been decoded. */ + channelCount = drflac__get_channel_count_from_channel_assignment(pFlac->currentFLACFrame.header.channelAssignment); + if (channelCount != pFlac->channels) { + return DRFLAC_FALSE; + } + + for (iChannel = 0; iChannel < channelCount; iChannel += 1) { + if (pFlac->currentFLACFrame.subframes[iChannel].pSamplesS32 == NULL) { + return DRFLAC_FALSE; + } + } + + return DRFLAC_TRUE; +} + +DRFLAC_API drflac_uint64 drflac_read_pcm_frames_s32(drflac* pFlac, drflac_uint64 framesToRead, drflac_int32* pBufferOut) +{ + drflac_uint64 framesRead; + drflac_uint32 unusedBitsPerSample; + + if (pFlac == NULL || framesToRead == 0) { + return 0; + } + + if (pBufferOut == NULL) { + return drflac__seek_forward_by_pcm_frames(pFlac, framesToRead); + } + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 32); + unusedBitsPerSample = 32 - pFlac->bitsPerSample; + + framesRead = 0; + while (framesToRead > 0) { + /* If we've run out of samples in this frame, go to the next. */ + if (pFlac->currentFLACFrame.pcmFramesRemaining == 0) { + if (!drflac__read_and_decode_next_flac_frame(pFlac)) { + break; /* Couldn't read the next frame, so just break from the loop and return. */ + } + } else { + unsigned int channelCount = drflac__get_channel_count_from_channel_assignment(pFlac->currentFLACFrame.header.channelAssignment); + drflac_uint64 iFirstPCMFrame = pFlac->currentFLACFrame.header.blockSizeInPCMFrames - pFlac->currentFLACFrame.pcmFramesRemaining; + drflac_uint64 frameCountThisIteration = framesToRead; + + if (frameCountThisIteration > pFlac->currentFLACFrame.pcmFramesRemaining) { + frameCountThisIteration = pFlac->currentFLACFrame.pcmFramesRemaining; + } + + if (channelCount == 2) { + const drflac_int32* pDecodedSamples0 = pFlac->currentFLACFrame.subframes[0].pSamplesS32 + iFirstPCMFrame; + const drflac_int32* pDecodedSamples1 = pFlac->currentFLACFrame.subframes[1].pSamplesS32 + iFirstPCMFrame; + + switch (pFlac->currentFLACFrame.header.channelAssignment) + { + case DRFLAC_CHANNEL_ASSIGNMENT_LEFT_SIDE: + { + drflac_read_pcm_frames_s32__decode_left_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_RIGHT_SIDE: + { + drflac_read_pcm_frames_s32__decode_right_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_MID_SIDE: + { + drflac_read_pcm_frames_s32__decode_mid_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_INDEPENDENT: + default: + { + drflac_read_pcm_frames_s32__decode_independent_stereo(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + } + } else { + /* Generic interleaving. */ + drflac_uint64 i; + for (i = 0; i < frameCountThisIteration; ++i) { + unsigned int j; + for (j = 0; j < channelCount; ++j) { + pBufferOut[(i*channelCount)+j] = (drflac_int32)((drflac_uint32)(pFlac->currentFLACFrame.subframes[j].pSamplesS32[iFirstPCMFrame + i]) << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[j].wastedBitsPerSample)); + } + } + } + + framesRead += frameCountThisIteration; + pBufferOut += frameCountThisIteration * channelCount; + framesToRead -= frameCountThisIteration; + pFlac->currentPCMFrame += frameCountThisIteration; + pFlac->currentFLACFrame.pcmFramesRemaining -= (drflac_uint32)frameCountThisIteration; + } + } + + return framesRead; +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_left_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + for (i = 0; i < frameCount; ++i) { + drflac_uint32 left = (drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + drflac_uint32 side = (drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + drflac_uint32 right = left - side; + + left >>= 16; + right >>= 16; + + pOutputSamples[i*2+0] = (drflac_int16)left; + pOutputSamples[i*2+1] = (drflac_int16)right; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_left_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 left0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 left1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 left2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 left3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << shift1; + + drflac_uint32 right0 = left0 - side0; + drflac_uint32 right1 = left1 - side1; + drflac_uint32 right2 = left2 - side2; + drflac_uint32 right3 = left3 - side3; + + left0 >>= 16; + left1 >>= 16; + left2 >>= 16; + left3 >>= 16; + + right0 >>= 16; + right1 >>= 16; + right2 >>= 16; + right3 >>= 16; + + pOutputSamples[i*8+0] = (drflac_int16)left0; + pOutputSamples[i*8+1] = (drflac_int16)right0; + pOutputSamples[i*8+2] = (drflac_int16)left1; + pOutputSamples[i*8+3] = (drflac_int16)right1; + pOutputSamples[i*8+4] = (drflac_int16)left2; + pOutputSamples[i*8+5] = (drflac_int16)right2; + pOutputSamples[i*8+6] = (drflac_int16)left3; + pOutputSamples[i*8+7] = (drflac_int16)right3; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + left >>= 16; + right >>= 16; + + pOutputSamples[i*2+0] = (drflac_int16)left; + pOutputSamples[i*2+1] = (drflac_int16)right; + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_left_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + for (i = 0; i < frameCount4; ++i) { + __m128i left = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + __m128i side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + __m128i right = _mm_sub_epi32(left, side); + + left = _mm_srai_epi32(left, 16); + right = _mm_srai_epi32(right, 16); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8), drflac__mm_packs_interleaved_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + left >>= 16; + right >>= 16; + + pOutputSamples[i*2+0] = (drflac_int16)left; + pOutputSamples[i*2+1] = (drflac_int16)right; + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_left_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + int32x4_t shift0_4; + int32x4_t shift1_4; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + shift0_4 = vdupq_n_s32(shift0); + shift1_4 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + uint32x4_t left; + uint32x4_t side; + uint32x4_t right; + + left = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift0_4); + side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift1_4); + right = vsubq_u32(left, side); + + left = vshrq_n_u32(left, 16); + right = vshrq_n_u32(right, 16); + + drflac__vst2q_u16((drflac_uint16*)pOutputSamples + i*8, vzip_u16(vmovn_u32(left), vmovn_u32(right))); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + left >>= 16; + right >>= 16; + + pOutputSamples[i*2+0] = (drflac_int16)left; + pOutputSamples[i*2+1] = (drflac_int16)right; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_duplicate_mono(drflac_uint64 frameCount, drflac_uint32 shift, const drflac_int32* pInputSamples, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + const drflac_uint32* pInputSamplesU32 = (const drflac_uint32*)pInputSamples; + + DRFLAC_ASSERT(shift < 32); + + for (i = 0; i < frameCount; ++i) { + drflac_int16 sample; + + sample = (drflac_int16)((pInputSamplesU32[i] << shift) >> 16); + pOutputSamples[i*2+0] = sample; + pOutputSamples[i*2+1] = sample; + } +} + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_left_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + if (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample == 32) { + drflac_read_pcm_frames_s16__decode_duplicate_mono(frameCount, unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample, pInputSamples0, pOutputSamples); + return; + } + +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s16__decode_left_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s16__decode_left_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_s16__decode_left_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_s16__decode_left_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_right_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + for (i = 0; i < frameCount; ++i) { + drflac_uint32 side = (drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + drflac_uint32 right = (drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + drflac_uint32 left = right + side; + + left >>= 16; + right >>= 16; + + pOutputSamples[i*2+0] = (drflac_int16)left; + pOutputSamples[i*2+1] = (drflac_int16)right; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_right_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 side0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 side1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 side2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 side3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 right0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 right1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 right2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 right3 = pInputSamples1U32[i*4+3] << shift1; + + drflac_uint32 left0 = right0 + side0; + drflac_uint32 left1 = right1 + side1; + drflac_uint32 left2 = right2 + side2; + drflac_uint32 left3 = right3 + side3; + + left0 >>= 16; + left1 >>= 16; + left2 >>= 16; + left3 >>= 16; + + right0 >>= 16; + right1 >>= 16; + right2 >>= 16; + right3 >>= 16; + + pOutputSamples[i*8+0] = (drflac_int16)left0; + pOutputSamples[i*8+1] = (drflac_int16)right0; + pOutputSamples[i*8+2] = (drflac_int16)left1; + pOutputSamples[i*8+3] = (drflac_int16)right1; + pOutputSamples[i*8+4] = (drflac_int16)left2; + pOutputSamples[i*8+5] = (drflac_int16)right2; + pOutputSamples[i*8+6] = (drflac_int16)left3; + pOutputSamples[i*8+7] = (drflac_int16)right3; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + left >>= 16; + right >>= 16; + + pOutputSamples[i*2+0] = (drflac_int16)left; + pOutputSamples[i*2+1] = (drflac_int16)right; + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_right_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + for (i = 0; i < frameCount4; ++i) { + __m128i side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + __m128i right = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + __m128i left = _mm_add_epi32(right, side); + + left = _mm_srai_epi32(left, 16); + right = _mm_srai_epi32(right, 16); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8), drflac__mm_packs_interleaved_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + left >>= 16; + right >>= 16; + + pOutputSamples[i*2+0] = (drflac_int16)left; + pOutputSamples[i*2+1] = (drflac_int16)right; + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_right_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + int32x4_t shift0_4; + int32x4_t shift1_4; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + shift0_4 = vdupq_n_s32(shift0); + shift1_4 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + uint32x4_t side; + uint32x4_t right; + uint32x4_t left; + + side = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift0_4); + right = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift1_4); + left = vaddq_u32(right, side); + + left = vshrq_n_u32(left, 16); + right = vshrq_n_u32(right, 16); + + drflac__vst2q_u16((drflac_uint16*)pOutputSamples + i*8, vzip_u16(vmovn_u32(left), vmovn_u32(right))); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + left >>= 16; + right >>= 16; + + pOutputSamples[i*2+0] = (drflac_int16)left; + pOutputSamples[i*2+1] = (drflac_int16)right; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_right_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + if (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample == 32) { + drflac_read_pcm_frames_s16__decode_duplicate_mono(frameCount, unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample, pInputSamples1, pOutputSamples); + return; + } + +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s16__decode_right_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s16__decode_right_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_s16__decode_right_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_s16__decode_right_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_mid_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + for (drflac_uint64 i = 0; i < frameCount; ++i) { + drflac_uint32 mid = (drflac_uint32)pInputSamples0[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = (drflac_uint32)pInputSamples1[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int16)(((drflac_uint32)((drflac_int32)(mid + side) >> 1) << unusedBitsPerSample) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)(((drflac_uint32)((drflac_int32)(mid - side) >> 1) << unusedBitsPerSample) >> 16); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_mid_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift = unusedBitsPerSample; + + if (shift > 0) { + shift -= 1; + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 temp0L; + drflac_uint32 temp1L; + drflac_uint32 temp2L; + drflac_uint32 temp3L; + drflac_uint32 temp0R; + drflac_uint32 temp1R; + drflac_uint32 temp2R; + drflac_uint32 temp3R; + + drflac_uint32 mid0 = pInputSamples0U32[i*4+0] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid1 = pInputSamples0U32[i*4+1] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid2 = pInputSamples0U32[i*4+2] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid3 = pInputSamples0U32[i*4+3] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid0 = (mid0 << 1) | (side0 & 0x01); + mid1 = (mid1 << 1) | (side1 & 0x01); + mid2 = (mid2 << 1) | (side2 & 0x01); + mid3 = (mid3 << 1) | (side3 & 0x01); + + temp0L = (mid0 + side0) << shift; + temp1L = (mid1 + side1) << shift; + temp2L = (mid2 + side2) << shift; + temp3L = (mid3 + side3) << shift; + + temp0R = (mid0 - side0) << shift; + temp1R = (mid1 - side1) << shift; + temp2R = (mid2 - side2) << shift; + temp3R = (mid3 - side3) << shift; + + temp0L >>= 16; + temp1L >>= 16; + temp2L >>= 16; + temp3L >>= 16; + + temp0R >>= 16; + temp1R >>= 16; + temp2R >>= 16; + temp3R >>= 16; + + pOutputSamples[i*8+0] = (drflac_int16)temp0L; + pOutputSamples[i*8+1] = (drflac_int16)temp0R; + pOutputSamples[i*8+2] = (drflac_int16)temp1L; + pOutputSamples[i*8+3] = (drflac_int16)temp1R; + pOutputSamples[i*8+4] = (drflac_int16)temp2L; + pOutputSamples[i*8+5] = (drflac_int16)temp2R; + pOutputSamples[i*8+6] = (drflac_int16)temp3L; + pOutputSamples[i*8+7] = (drflac_int16)temp3R; + } + } else { + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 temp0L; + drflac_uint32 temp1L; + drflac_uint32 temp2L; + drflac_uint32 temp3L; + drflac_uint32 temp0R; + drflac_uint32 temp1R; + drflac_uint32 temp2R; + drflac_uint32 temp3R; + + drflac_uint32 mid0 = pInputSamples0U32[i*4+0] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid1 = pInputSamples0U32[i*4+1] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid2 = pInputSamples0U32[i*4+2] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid3 = pInputSamples0U32[i*4+3] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid0 = (mid0 << 1) | (side0 & 0x01); + mid1 = (mid1 << 1) | (side1 & 0x01); + mid2 = (mid2 << 1) | (side2 & 0x01); + mid3 = (mid3 << 1) | (side3 & 0x01); + + temp0L = ((drflac_int32)(mid0 + side0) >> 1); + temp1L = ((drflac_int32)(mid1 + side1) >> 1); + temp2L = ((drflac_int32)(mid2 + side2) >> 1); + temp3L = ((drflac_int32)(mid3 + side3) >> 1); + + temp0R = ((drflac_int32)(mid0 - side0) >> 1); + temp1R = ((drflac_int32)(mid1 - side1) >> 1); + temp2R = ((drflac_int32)(mid2 - side2) >> 1); + temp3R = ((drflac_int32)(mid3 - side3) >> 1); + + temp0L >>= 16; + temp1L >>= 16; + temp2L >>= 16; + temp3L >>= 16; + + temp0R >>= 16; + temp1R >>= 16; + temp2R >>= 16; + temp3R >>= 16; + + pOutputSamples[i*8+0] = (drflac_int16)temp0L; + pOutputSamples[i*8+1] = (drflac_int16)temp0R; + pOutputSamples[i*8+2] = (drflac_int16)temp1L; + pOutputSamples[i*8+3] = (drflac_int16)temp1R; + pOutputSamples[i*8+4] = (drflac_int16)temp2L; + pOutputSamples[i*8+5] = (drflac_int16)temp2R; + pOutputSamples[i*8+6] = (drflac_int16)temp3L; + pOutputSamples[i*8+7] = (drflac_int16)temp3R; + } + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int16)(((drflac_uint32)((drflac_int32)(mid + side) >> 1) << unusedBitsPerSample) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)(((drflac_uint32)((drflac_int32)(mid - side) >> 1) << unusedBitsPerSample) >> 16); + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_mid_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift = unusedBitsPerSample; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + if (shift == 0) { + for (i = 0; i < frameCount4; ++i) { + __m128i mid; + __m128i side; + __m128i left; + __m128i right; + + mid = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + + mid = _mm_or_si128(_mm_slli_epi32(mid, 1), _mm_and_si128(side, _mm_set1_epi32(0x01))); + + left = _mm_srai_epi32(_mm_add_epi32(mid, side), 1); + right = _mm_srai_epi32(_mm_sub_epi32(mid, side), 1); + + left = _mm_srai_epi32(left, 16); + right = _mm_srai_epi32(right, 16); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8), drflac__mm_packs_interleaved_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int16)(((drflac_int32)(mid + side) >> 1) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)(((drflac_int32)(mid - side) >> 1) >> 16); + } + } else { + shift -= 1; + for (i = 0; i < frameCount4; ++i) { + __m128i mid; + __m128i side; + __m128i left; + __m128i right; + + mid = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + + mid = _mm_or_si128(_mm_slli_epi32(mid, 1), _mm_and_si128(side, _mm_set1_epi32(0x01))); + + left = _mm_slli_epi32(_mm_add_epi32(mid, side), shift); + right = _mm_slli_epi32(_mm_sub_epi32(mid, side), shift); + + left = _mm_srai_epi32(left, 16); + right = _mm_srai_epi32(right, 16); + + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8), drflac__mm_packs_interleaved_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int16)(((mid + side) << shift) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)(((mid - side) << shift) >> 16); + } + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_mid_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift = unusedBitsPerSample; + int32x4_t wbpsShift0_4; /* wbps = Wasted Bits Per Sample */ + int32x4_t wbpsShift1_4; /* wbps = Wasted Bits Per Sample */ + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + wbpsShift0_4 = vdupq_n_s32(pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + wbpsShift1_4 = vdupq_n_s32(pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + + if (shift == 0) { + for (i = 0; i < frameCount4; ++i) { + uint32x4_t mid; + uint32x4_t side; + int32x4_t left; + int32x4_t right; + + mid = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), wbpsShift0_4); + side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), wbpsShift1_4); + + mid = vorrq_u32(vshlq_n_u32(mid, 1), vandq_u32(side, vdupq_n_u32(1))); + + left = vshrq_n_s32(vreinterpretq_s32_u32(vaddq_u32(mid, side)), 1); + right = vshrq_n_s32(vreinterpretq_s32_u32(vsubq_u32(mid, side)), 1); + + left = vshrq_n_s32(left, 16); + right = vshrq_n_s32(right, 16); + + drflac__vst2q_s16(pOutputSamples + i*8, vzip_s16(vmovn_s32(left), vmovn_s32(right))); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int16)(((drflac_int32)(mid + side) >> 1) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)(((drflac_int32)(mid - side) >> 1) >> 16); + } + } else { + int32x4_t shift4; + + shift -= 1; + shift4 = vdupq_n_s32(shift); + + for (i = 0; i < frameCount4; ++i) { + uint32x4_t mid; + uint32x4_t side; + int32x4_t left; + int32x4_t right; + + mid = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), wbpsShift0_4); + side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), wbpsShift1_4); + + mid = vorrq_u32(vshlq_n_u32(mid, 1), vandq_u32(side, vdupq_n_u32(1))); + + left = vreinterpretq_s32_u32(vshlq_u32(vaddq_u32(mid, side), shift4)); + right = vreinterpretq_s32_u32(vshlq_u32(vsubq_u32(mid, side), shift4)); + + left = vshrq_n_s32(left, 16); + right = vshrq_n_s32(right, 16); + + drflac__vst2q_s16(pOutputSamples + i*8, vzip_s16(vmovn_s32(left), vmovn_s32(right))); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int16)(((mid + side) << shift) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)(((mid - side) << shift) >> 16); + } + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_mid_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s16__decode_mid_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s16__decode_mid_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_s16__decode_mid_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_s16__decode_mid_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_independent_stereo__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + for (drflac_uint64 i = 0; i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int16)((drflac_int32)((drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample)) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)((drflac_int32)((drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample)) >> 16); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_independent_stereo__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 tempL0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 tempL1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 tempL2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 tempL3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 tempR0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 tempR1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 tempR2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 tempR3 = pInputSamples1U32[i*4+3] << shift1; + + tempL0 >>= 16; + tempL1 >>= 16; + tempL2 >>= 16; + tempL3 >>= 16; + + tempR0 >>= 16; + tempR1 >>= 16; + tempR2 >>= 16; + tempR3 >>= 16; + + pOutputSamples[i*8+0] = (drflac_int16)tempL0; + pOutputSamples[i*8+1] = (drflac_int16)tempR0; + pOutputSamples[i*8+2] = (drflac_int16)tempL1; + pOutputSamples[i*8+3] = (drflac_int16)tempR1; + pOutputSamples[i*8+4] = (drflac_int16)tempL2; + pOutputSamples[i*8+5] = (drflac_int16)tempR2; + pOutputSamples[i*8+6] = (drflac_int16)tempL3; + pOutputSamples[i*8+7] = (drflac_int16)tempR3; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int16)((pInputSamples0U32[i] << shift0) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)((pInputSamples1U32[i] << shift1) >> 16); + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_independent_stereo__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + for (i = 0; i < frameCount4; ++i) { + __m128i left = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + __m128i right = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + + left = _mm_srai_epi32(left, 16); + right = _mm_srai_epi32(right, 16); + + /* At this point we have results. We can now pack and interleave these into a single __m128i object and then store the in the output buffer. */ + _mm_storeu_si128((__m128i*)(pOutputSamples + i*8), drflac__mm_packs_interleaved_epi32(left, right)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int16)((pInputSamples0U32[i] << shift0) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)((pInputSamples1U32[i] << shift1) >> 16); + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_independent_stereo__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + int32x4_t shift0_4 = vdupq_n_s32(shift0); + int32x4_t shift1_4 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + int32x4_t left; + int32x4_t right; + + left = vreinterpretq_s32_u32(vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift0_4)); + right = vreinterpretq_s32_u32(vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift1_4)); + + left = vshrq_n_s32(left, 16); + right = vshrq_n_s32(right, 16); + + drflac__vst2q_s16(pOutputSamples + i*8, vzip_s16(vmovn_s32(left), vmovn_s32(right))); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int16)((pInputSamples0U32[i] << shift0) >> 16); + pOutputSamples[i*2+1] = (drflac_int16)((pInputSamples1U32[i] << shift1) >> 16); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_s16__decode_independent_stereo(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, drflac_int16* pOutputSamples) +{ +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s16__decode_independent_stereo__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_s16__decode_independent_stereo__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_s16__decode_independent_stereo__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_s16__decode_independent_stereo__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + +DRFLAC_API drflac_uint64 drflac_read_pcm_frames_s16(drflac* pFlac, drflac_uint64 framesToRead, drflac_int16* pBufferOut) +{ + drflac_uint64 framesRead; + drflac_uint32 unusedBitsPerSample; + + if (pFlac == NULL || framesToRead == 0) { + return 0; + } + + if (pBufferOut == NULL) { + return drflac__seek_forward_by_pcm_frames(pFlac, framesToRead); + } + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 32); + unusedBitsPerSample = 32 - pFlac->bitsPerSample; + + framesRead = 0; + while (framesToRead > 0) { + /* If we've run out of samples in this frame, go to the next. */ + if (pFlac->currentFLACFrame.pcmFramesRemaining == 0) { + if (!drflac__read_and_decode_next_flac_frame(pFlac)) { + break; /* Couldn't read the next frame, so just break from the loop and return. */ + } + } else { + unsigned int channelCount = drflac__get_channel_count_from_channel_assignment(pFlac->currentFLACFrame.header.channelAssignment); + drflac_uint64 iFirstPCMFrame = pFlac->currentFLACFrame.header.blockSizeInPCMFrames - pFlac->currentFLACFrame.pcmFramesRemaining; + drflac_uint64 frameCountThisIteration = framesToRead; + + if (frameCountThisIteration > pFlac->currentFLACFrame.pcmFramesRemaining) { + frameCountThisIteration = pFlac->currentFLACFrame.pcmFramesRemaining; + } + + if (channelCount == 2) { + const drflac_int32* pDecodedSamples0 = pFlac->currentFLACFrame.subframes[0].pSamplesS32 + iFirstPCMFrame; + const drflac_int32* pDecodedSamples1 = pFlac->currentFLACFrame.subframes[1].pSamplesS32 + iFirstPCMFrame; + + switch (pFlac->currentFLACFrame.header.channelAssignment) + { + case DRFLAC_CHANNEL_ASSIGNMENT_LEFT_SIDE: + { + drflac_read_pcm_frames_s16__decode_left_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_RIGHT_SIDE: + { + drflac_read_pcm_frames_s16__decode_right_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_MID_SIDE: + { + drflac_read_pcm_frames_s16__decode_mid_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_INDEPENDENT: + default: + { + drflac_read_pcm_frames_s16__decode_independent_stereo(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + } + } else { + /* Generic interleaving. */ + drflac_uint64 i; + for (i = 0; i < frameCountThisIteration; ++i) { + unsigned int j; + for (j = 0; j < channelCount; ++j) { + drflac_int32 sampleS32 = (drflac_int32)((drflac_uint32)(pFlac->currentFLACFrame.subframes[j].pSamplesS32[iFirstPCMFrame + i]) << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[j].wastedBitsPerSample)); + pBufferOut[(i*channelCount)+j] = (drflac_int16)(sampleS32 >> 16); + } + } + } + + framesRead += frameCountThisIteration; + pBufferOut += frameCountThisIteration * channelCount; + framesToRead -= frameCountThisIteration; + pFlac->currentPCMFrame += frameCountThisIteration; + pFlac->currentFLACFrame.pcmFramesRemaining -= (drflac_uint32)frameCountThisIteration; + } + } + + return framesRead; +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_left_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + for (i = 0; i < frameCount; ++i) { + drflac_uint32 left = (drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + drflac_uint32 side = (drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + drflac_uint32 right = left - side; + + pOutputSamples[i*2+0] = (float)((drflac_int32)left / 2147483648.0); + pOutputSamples[i*2+1] = (float)((drflac_int32)right / 2147483648.0); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_left_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + float factor = 1 / 2147483648.0; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 left0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 left1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 left2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 left3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << shift1; + + drflac_uint32 right0 = left0 - side0; + drflac_uint32 right1 = left1 - side1; + drflac_uint32 right2 = left2 - side2; + drflac_uint32 right3 = left3 - side3; + + pOutputSamples[i*8+0] = (drflac_int32)left0 * factor; + pOutputSamples[i*8+1] = (drflac_int32)right0 * factor; + pOutputSamples[i*8+2] = (drflac_int32)left1 * factor; + pOutputSamples[i*8+3] = (drflac_int32)right1 * factor; + pOutputSamples[i*8+4] = (drflac_int32)left2 * factor; + pOutputSamples[i*8+5] = (drflac_int32)right2 * factor; + pOutputSamples[i*8+6] = (drflac_int32)left3 * factor; + pOutputSamples[i*8+7] = (drflac_int32)right3 * factor; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + pOutputSamples[i*2+0] = (drflac_int32)left * factor; + pOutputSamples[i*2+1] = (drflac_int32)right * factor; + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_left_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample) - 8; + drflac_uint32 shift1 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample) - 8; + __m128 factor; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + factor = _mm_set1_ps(1.0f / 8388608.0f); + + for (i = 0; i < frameCount4; ++i) { + __m128i left = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + __m128i side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + __m128i right = _mm_sub_epi32(left, side); + __m128 leftf = _mm_mul_ps(_mm_cvtepi32_ps(left), factor); + __m128 rightf = _mm_mul_ps(_mm_cvtepi32_ps(right), factor); + + _mm_storeu_ps(pOutputSamples + i*8 + 0, _mm_unpacklo_ps(leftf, rightf)); + _mm_storeu_ps(pOutputSamples + i*8 + 4, _mm_unpackhi_ps(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + pOutputSamples[i*2+0] = (drflac_int32)left / 8388608.0f; + pOutputSamples[i*2+1] = (drflac_int32)right / 8388608.0f; + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_left_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample) - 8; + drflac_uint32 shift1 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample) - 8; + float32x4_t factor4; + int32x4_t shift0_4; + int32x4_t shift1_4; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + factor4 = vdupq_n_f32(1.0f / 8388608.0f); + shift0_4 = vdupq_n_s32(shift0); + shift1_4 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + uint32x4_t left; + uint32x4_t side; + uint32x4_t right; + float32x4_t leftf; + float32x4_t rightf; + + left = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift0_4); + side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift1_4); + right = vsubq_u32(left, side); + leftf = vmulq_f32(vcvtq_f32_s32(vreinterpretq_s32_u32(left)), factor4); + rightf = vmulq_f32(vcvtq_f32_s32(vreinterpretq_s32_u32(right)), factor4); + + drflac__vst2q_f32(pOutputSamples + i*8, vzipq_f32(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 left = pInputSamples0U32[i] << shift0; + drflac_uint32 side = pInputSamples1U32[i] << shift1; + drflac_uint32 right = left - side; + + pOutputSamples[i*2+0] = (drflac_int32)left / 8388608.0f; + pOutputSamples[i*2+1] = (drflac_int32)right / 8388608.0f; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_duplicate_mono(drflac_uint64 frameCount, drflac_uint32 shift, const drflac_int32* pInputSamples, float* pOutputSamples) +{ + drflac_uint64 i; + const drflac_uint32* pInputSamplesU32 = (const drflac_uint32*)pInputSamples; + float factor; + + DRFLAC_ASSERT(shift < 32); + + factor = 1 / 2147483648.0f; + + for (i = 0; i < frameCount; ++i) { + float sample; + + sample = (drflac_int32)(pInputSamplesU32[i] << shift) * factor; + pOutputSamples[i*2+0] = sample; + pOutputSamples[i*2+1] = sample; + } +} + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_left_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + if (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample == 32) { + drflac_read_pcm_frames_f32__decode_duplicate_mono(frameCount, unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample, pInputSamples0, pOutputSamples); + return; + } + +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_f32__decode_left_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_f32__decode_left_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_f32__decode_left_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_f32__decode_left_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_right_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + for (i = 0; i < frameCount; ++i) { + drflac_uint32 side = (drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + drflac_uint32 right = (drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + drflac_uint32 left = right + side; + + pOutputSamples[i*2+0] = (float)((drflac_int32)left / 2147483648.0); + pOutputSamples[i*2+1] = (float)((drflac_int32)right / 2147483648.0); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_right_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + float factor = 1 / 2147483648.0; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 side0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 side1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 side2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 side3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 right0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 right1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 right2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 right3 = pInputSamples1U32[i*4+3] << shift1; + + drflac_uint32 left0 = right0 + side0; + drflac_uint32 left1 = right1 + side1; + drflac_uint32 left2 = right2 + side2; + drflac_uint32 left3 = right3 + side3; + + pOutputSamples[i*8+0] = (drflac_int32)left0 * factor; + pOutputSamples[i*8+1] = (drflac_int32)right0 * factor; + pOutputSamples[i*8+2] = (drflac_int32)left1 * factor; + pOutputSamples[i*8+3] = (drflac_int32)right1 * factor; + pOutputSamples[i*8+4] = (drflac_int32)left2 * factor; + pOutputSamples[i*8+5] = (drflac_int32)right2 * factor; + pOutputSamples[i*8+6] = (drflac_int32)left3 * factor; + pOutputSamples[i*8+7] = (drflac_int32)right3 * factor; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + pOutputSamples[i*2+0] = (drflac_int32)left * factor; + pOutputSamples[i*2+1] = (drflac_int32)right * factor; + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_right_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample) - 8; + drflac_uint32 shift1 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample) - 8; + __m128 factor; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + factor = _mm_set1_ps(1.0f / 8388608.0f); + + for (i = 0; i < frameCount4; ++i) { + __m128i side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + __m128i right = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + __m128i left = _mm_add_epi32(right, side); + __m128 leftf = _mm_mul_ps(_mm_cvtepi32_ps(left), factor); + __m128 rightf = _mm_mul_ps(_mm_cvtepi32_ps(right), factor); + + _mm_storeu_ps(pOutputSamples + i*8 + 0, _mm_unpacklo_ps(leftf, rightf)); + _mm_storeu_ps(pOutputSamples + i*8 + 4, _mm_unpackhi_ps(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + pOutputSamples[i*2+0] = (drflac_int32)left / 8388608.0f; + pOutputSamples[i*2+1] = (drflac_int32)right / 8388608.0f; + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_right_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample) - 8; + drflac_uint32 shift1 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample) - 8; + float32x4_t factor4; + int32x4_t shift0_4; + int32x4_t shift1_4; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + factor4 = vdupq_n_f32(1.0f / 8388608.0f); + shift0_4 = vdupq_n_s32(shift0); + shift1_4 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + uint32x4_t side; + uint32x4_t right; + uint32x4_t left; + float32x4_t leftf; + float32x4_t rightf; + + side = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift0_4); + right = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift1_4); + left = vaddq_u32(right, side); + leftf = vmulq_f32(vcvtq_f32_s32(vreinterpretq_s32_u32(left)), factor4); + rightf = vmulq_f32(vcvtq_f32_s32(vreinterpretq_s32_u32(right)), factor4); + + drflac__vst2q_f32(pOutputSamples + i*8, vzipq_f32(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 side = pInputSamples0U32[i] << shift0; + drflac_uint32 right = pInputSamples1U32[i] << shift1; + drflac_uint32 left = right + side; + + pOutputSamples[i*2+0] = (drflac_int32)left / 8388608.0f; + pOutputSamples[i*2+1] = (drflac_int32)right / 8388608.0f; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_right_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + if (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample == 32) { + drflac_read_pcm_frames_f32__decode_duplicate_mono(frameCount, unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample, pInputSamples1, pOutputSamples); + return; + } + +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_f32__decode_right_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_f32__decode_right_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_f32__decode_right_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_f32__decode_right_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_mid_side__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + for (drflac_uint64 i = 0; i < frameCount; ++i) { + drflac_uint32 mid = (drflac_uint32)pInputSamples0[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = (drflac_uint32)pInputSamples1[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (float)((((drflac_int32)(mid + side) >> 1) << (unusedBitsPerSample)) / 2147483648.0); + pOutputSamples[i*2+1] = (float)((((drflac_int32)(mid - side) >> 1) << (unusedBitsPerSample)) / 2147483648.0); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_mid_side__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift = unusedBitsPerSample; + float factor = 1 / 2147483648.0; + + if (shift > 0) { + shift -= 1; + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 temp0L; + drflac_uint32 temp1L; + drflac_uint32 temp2L; + drflac_uint32 temp3L; + drflac_uint32 temp0R; + drflac_uint32 temp1R; + drflac_uint32 temp2R; + drflac_uint32 temp3R; + + drflac_uint32 mid0 = pInputSamples0U32[i*4+0] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid1 = pInputSamples0U32[i*4+1] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid2 = pInputSamples0U32[i*4+2] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid3 = pInputSamples0U32[i*4+3] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid0 = (mid0 << 1) | (side0 & 0x01); + mid1 = (mid1 << 1) | (side1 & 0x01); + mid2 = (mid2 << 1) | (side2 & 0x01); + mid3 = (mid3 << 1) | (side3 & 0x01); + + temp0L = (mid0 + side0) << shift; + temp1L = (mid1 + side1) << shift; + temp2L = (mid2 + side2) << shift; + temp3L = (mid3 + side3) << shift; + + temp0R = (mid0 - side0) << shift; + temp1R = (mid1 - side1) << shift; + temp2R = (mid2 - side2) << shift; + temp3R = (mid3 - side3) << shift; + + pOutputSamples[i*8+0] = (drflac_int32)temp0L * factor; + pOutputSamples[i*8+1] = (drflac_int32)temp0R * factor; + pOutputSamples[i*8+2] = (drflac_int32)temp1L * factor; + pOutputSamples[i*8+3] = (drflac_int32)temp1R * factor; + pOutputSamples[i*8+4] = (drflac_int32)temp2L * factor; + pOutputSamples[i*8+5] = (drflac_int32)temp2R * factor; + pOutputSamples[i*8+6] = (drflac_int32)temp3L * factor; + pOutputSamples[i*8+7] = (drflac_int32)temp3R * factor; + } + } else { + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 temp0L; + drflac_uint32 temp1L; + drflac_uint32 temp2L; + drflac_uint32 temp3L; + drflac_uint32 temp0R; + drflac_uint32 temp1R; + drflac_uint32 temp2R; + drflac_uint32 temp3R; + + drflac_uint32 mid0 = pInputSamples0U32[i*4+0] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid1 = pInputSamples0U32[i*4+1] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid2 = pInputSamples0U32[i*4+2] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 mid3 = pInputSamples0U32[i*4+3] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + + drflac_uint32 side0 = pInputSamples1U32[i*4+0] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side1 = pInputSamples1U32[i*4+1] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side2 = pInputSamples1U32[i*4+2] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + drflac_uint32 side3 = pInputSamples1U32[i*4+3] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid0 = (mid0 << 1) | (side0 & 0x01); + mid1 = (mid1 << 1) | (side1 & 0x01); + mid2 = (mid2 << 1) | (side2 & 0x01); + mid3 = (mid3 << 1) | (side3 & 0x01); + + temp0L = (drflac_uint32)((drflac_int32)(mid0 + side0) >> 1); + temp1L = (drflac_uint32)((drflac_int32)(mid1 + side1) >> 1); + temp2L = (drflac_uint32)((drflac_int32)(mid2 + side2) >> 1); + temp3L = (drflac_uint32)((drflac_int32)(mid3 + side3) >> 1); + + temp0R = (drflac_uint32)((drflac_int32)(mid0 - side0) >> 1); + temp1R = (drflac_uint32)((drflac_int32)(mid1 - side1) >> 1); + temp2R = (drflac_uint32)((drflac_int32)(mid2 - side2) >> 1); + temp3R = (drflac_uint32)((drflac_int32)(mid3 - side3) >> 1); + + pOutputSamples[i*8+0] = (drflac_int32)temp0L * factor; + pOutputSamples[i*8+1] = (drflac_int32)temp0R * factor; + pOutputSamples[i*8+2] = (drflac_int32)temp1L * factor; + pOutputSamples[i*8+3] = (drflac_int32)temp1R * factor; + pOutputSamples[i*8+4] = (drflac_int32)temp2L * factor; + pOutputSamples[i*8+5] = (drflac_int32)temp2R * factor; + pOutputSamples[i*8+6] = (drflac_int32)temp3L * factor; + pOutputSamples[i*8+7] = (drflac_int32)temp3R * factor; + } + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)((drflac_uint32)((drflac_int32)(mid + side) >> 1) << unusedBitsPerSample) * factor; + pOutputSamples[i*2+1] = (drflac_int32)((drflac_uint32)((drflac_int32)(mid - side) >> 1) << unusedBitsPerSample) * factor; + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_mid_side__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift = unusedBitsPerSample - 8; + float factor; + __m128 factor128; + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + factor = 1.0f / 8388608.0f; + factor128 = _mm_set1_ps(factor); + + if (shift == 0) { + for (i = 0; i < frameCount4; ++i) { + __m128i mid; + __m128i side; + __m128i tempL; + __m128i tempR; + __m128 leftf; + __m128 rightf; + + mid = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + + mid = _mm_or_si128(_mm_slli_epi32(mid, 1), _mm_and_si128(side, _mm_set1_epi32(0x01))); + + tempL = _mm_srai_epi32(_mm_add_epi32(mid, side), 1); + tempR = _mm_srai_epi32(_mm_sub_epi32(mid, side), 1); + + leftf = _mm_mul_ps(_mm_cvtepi32_ps(tempL), factor128); + rightf = _mm_mul_ps(_mm_cvtepi32_ps(tempR), factor128); + + _mm_storeu_ps(pOutputSamples + i*8 + 0, _mm_unpacklo_ps(leftf, rightf)); + _mm_storeu_ps(pOutputSamples + i*8 + 4, _mm_unpackhi_ps(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = ((drflac_int32)(mid + side) >> 1) * factor; + pOutputSamples[i*2+1] = ((drflac_int32)(mid - side) >> 1) * factor; + } + } else { + shift -= 1; + for (i = 0; i < frameCount4; ++i) { + __m128i mid; + __m128i side; + __m128i tempL; + __m128i tempR; + __m128 leftf; + __m128 rightf; + + mid = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + side = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + + mid = _mm_or_si128(_mm_slli_epi32(mid, 1), _mm_and_si128(side, _mm_set1_epi32(0x01))); + + tempL = _mm_slli_epi32(_mm_add_epi32(mid, side), shift); + tempR = _mm_slli_epi32(_mm_sub_epi32(mid, side), shift); + + leftf = _mm_mul_ps(_mm_cvtepi32_ps(tempL), factor128); + rightf = _mm_mul_ps(_mm_cvtepi32_ps(tempR), factor128); + + _mm_storeu_ps(pOutputSamples + i*8 + 0, _mm_unpacklo_ps(leftf, rightf)); + _mm_storeu_ps(pOutputSamples + i*8 + 4, _mm_unpackhi_ps(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)((mid + side) << shift) * factor; + pOutputSamples[i*2+1] = (drflac_int32)((mid - side) << shift) * factor; + } + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_mid_side__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift = unusedBitsPerSample - 8; + float factor; + float32x4_t factor4; + int32x4_t shift4; + int32x4_t wbps0_4; /* Wasted Bits Per Sample */ + int32x4_t wbps1_4; /* Wasted Bits Per Sample */ + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 24); + + factor = 1.0f / 8388608.0f; + factor4 = vdupq_n_f32(factor); + wbps0_4 = vdupq_n_s32(pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample); + wbps1_4 = vdupq_n_s32(pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample); + + if (shift == 0) { + for (i = 0; i < frameCount4; ++i) { + int32x4_t lefti; + int32x4_t righti; + float32x4_t leftf; + float32x4_t rightf; + + uint32x4_t mid = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), wbps0_4); + uint32x4_t side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), wbps1_4); + + mid = vorrq_u32(vshlq_n_u32(mid, 1), vandq_u32(side, vdupq_n_u32(1))); + + lefti = vshrq_n_s32(vreinterpretq_s32_u32(vaddq_u32(mid, side)), 1); + righti = vshrq_n_s32(vreinterpretq_s32_u32(vsubq_u32(mid, side)), 1); + + leftf = vmulq_f32(vcvtq_f32_s32(lefti), factor4); + rightf = vmulq_f32(vcvtq_f32_s32(righti), factor4); + + drflac__vst2q_f32(pOutputSamples + i*8, vzipq_f32(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = ((drflac_int32)(mid + side) >> 1) * factor; + pOutputSamples[i*2+1] = ((drflac_int32)(mid - side) >> 1) * factor; + } + } else { + shift -= 1; + shift4 = vdupq_n_s32(shift); + for (i = 0; i < frameCount4; ++i) { + uint32x4_t mid; + uint32x4_t side; + int32x4_t lefti; + int32x4_t righti; + float32x4_t leftf; + float32x4_t rightf; + + mid = vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), wbps0_4); + side = vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), wbps1_4); + + mid = vorrq_u32(vshlq_n_u32(mid, 1), vandq_u32(side, vdupq_n_u32(1))); + + lefti = vreinterpretq_s32_u32(vshlq_u32(vaddq_u32(mid, side), shift4)); + righti = vreinterpretq_s32_u32(vshlq_u32(vsubq_u32(mid, side), shift4)); + + leftf = vmulq_f32(vcvtq_f32_s32(lefti), factor4); + rightf = vmulq_f32(vcvtq_f32_s32(righti), factor4); + + drflac__vst2q_f32(pOutputSamples + i*8, vzipq_f32(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + drflac_uint32 mid = pInputSamples0U32[i] << pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 side = pInputSamples1U32[i] << pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + + mid = (mid << 1) | (side & 0x01); + + pOutputSamples[i*2+0] = (drflac_int32)((mid + side) << shift) * factor; + pOutputSamples[i*2+1] = (drflac_int32)((mid - side) << shift) * factor; + } + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_mid_side(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_f32__decode_mid_side__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_f32__decode_mid_side__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_f32__decode_mid_side__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_f32__decode_mid_side__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + +#if 0 +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_independent_stereo__reference(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + for (drflac_uint64 i = 0; i < frameCount; ++i) { + pOutputSamples[i*2+0] = (float)((drflac_int32)((drflac_uint32)pInputSamples0[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample)) / 2147483648.0); + pOutputSamples[i*2+1] = (float)((drflac_int32)((drflac_uint32)pInputSamples1[i] << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample)) / 2147483648.0); + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_independent_stereo__scalar(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample; + drflac_uint32 shift1 = unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample; + float factor = 1 / 2147483648.0; + + for (i = 0; i < frameCount4; ++i) { + drflac_uint32 tempL0 = pInputSamples0U32[i*4+0] << shift0; + drflac_uint32 tempL1 = pInputSamples0U32[i*4+1] << shift0; + drflac_uint32 tempL2 = pInputSamples0U32[i*4+2] << shift0; + drflac_uint32 tempL3 = pInputSamples0U32[i*4+3] << shift0; + + drflac_uint32 tempR0 = pInputSamples1U32[i*4+0] << shift1; + drflac_uint32 tempR1 = pInputSamples1U32[i*4+1] << shift1; + drflac_uint32 tempR2 = pInputSamples1U32[i*4+2] << shift1; + drflac_uint32 tempR3 = pInputSamples1U32[i*4+3] << shift1; + + pOutputSamples[i*8+0] = (drflac_int32)tempL0 * factor; + pOutputSamples[i*8+1] = (drflac_int32)tempR0 * factor; + pOutputSamples[i*8+2] = (drflac_int32)tempL1 * factor; + pOutputSamples[i*8+3] = (drflac_int32)tempR1 * factor; + pOutputSamples[i*8+4] = (drflac_int32)tempL2 * factor; + pOutputSamples[i*8+5] = (drflac_int32)tempR2 * factor; + pOutputSamples[i*8+6] = (drflac_int32)tempL3 * factor; + pOutputSamples[i*8+7] = (drflac_int32)tempR3 * factor; + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int32)(pInputSamples0U32[i] << shift0) * factor; + pOutputSamples[i*2+1] = (drflac_int32)(pInputSamples1U32[i] << shift1) * factor; + } +} + +#if defined(DRFLAC_SUPPORT_SSE2) +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_independent_stereo__sse2(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample) - 8; + drflac_uint32 shift1 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample) - 8; + + float factor = 1.0f / 8388608.0f; + __m128 factor128 = _mm_set1_ps(factor); + + for (i = 0; i < frameCount4; ++i) { + __m128i lefti; + __m128i righti; + __m128 leftf; + __m128 rightf; + + lefti = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples0 + i), shift0); + righti = _mm_slli_epi32(_mm_loadu_si128((const __m128i*)pInputSamples1 + i), shift1); + + leftf = _mm_mul_ps(_mm_cvtepi32_ps(lefti), factor128); + rightf = _mm_mul_ps(_mm_cvtepi32_ps(righti), factor128); + + _mm_storeu_ps(pOutputSamples + i*8 + 0, _mm_unpacklo_ps(leftf, rightf)); + _mm_storeu_ps(pOutputSamples + i*8 + 4, _mm_unpackhi_ps(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int32)(pInputSamples0U32[i] << shift0) * factor; + pOutputSamples[i*2+1] = (drflac_int32)(pInputSamples1U32[i] << shift1) * factor; + } +} +#endif + +#if defined(DRFLAC_SUPPORT_NEON) +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_independent_stereo__neon(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ + drflac_uint64 i; + drflac_uint64 frameCount4 = frameCount >> 2; + const drflac_uint32* pInputSamples0U32 = (const drflac_uint32*)pInputSamples0; + const drflac_uint32* pInputSamples1U32 = (const drflac_uint32*)pInputSamples1; + drflac_uint32 shift0 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[0].wastedBitsPerSample) - 8; + drflac_uint32 shift1 = (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[1].wastedBitsPerSample) - 8; + + float factor = 1.0f / 8388608.0f; + float32x4_t factor4 = vdupq_n_f32(factor); + int32x4_t shift0_4 = vdupq_n_s32(shift0); + int32x4_t shift1_4 = vdupq_n_s32(shift1); + + for (i = 0; i < frameCount4; ++i) { + int32x4_t lefti; + int32x4_t righti; + float32x4_t leftf; + float32x4_t rightf; + + lefti = vreinterpretq_s32_u32(vshlq_u32(vld1q_u32(pInputSamples0U32 + i*4), shift0_4)); + righti = vreinterpretq_s32_u32(vshlq_u32(vld1q_u32(pInputSamples1U32 + i*4), shift1_4)); + + leftf = vmulq_f32(vcvtq_f32_s32(lefti), factor4); + rightf = vmulq_f32(vcvtq_f32_s32(righti), factor4); + + drflac__vst2q_f32(pOutputSamples + i*8, vzipq_f32(leftf, rightf)); + } + + for (i = (frameCount4 << 2); i < frameCount; ++i) { + pOutputSamples[i*2+0] = (drflac_int32)(pInputSamples0U32[i] << shift0) * factor; + pOutputSamples[i*2+1] = (drflac_int32)(pInputSamples1U32[i] << shift1) * factor; + } +} +#endif + +static DRFLAC_INLINE void drflac_read_pcm_frames_f32__decode_independent_stereo(drflac* pFlac, drflac_uint64 frameCount, drflac_uint32 unusedBitsPerSample, const drflac_int32* pInputSamples0, const drflac_int32* pInputSamples1, float* pOutputSamples) +{ +#if defined(DRFLAC_SUPPORT_SSE2) + if (drflac__gIsSSE2Supported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_f32__decode_independent_stereo__sse2(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#elif defined(DRFLAC_SUPPORT_NEON) + if (drflac__gIsNEONSupported && pFlac->bitsPerSample <= 24) { + drflac_read_pcm_frames_f32__decode_independent_stereo__neon(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); + } else +#endif + { + /* Scalar fallback. */ +#if 0 + drflac_read_pcm_frames_f32__decode_independent_stereo__reference(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#else + drflac_read_pcm_frames_f32__decode_independent_stereo__scalar(pFlac, frameCount, unusedBitsPerSample, pInputSamples0, pInputSamples1, pOutputSamples); +#endif + } +} + +DRFLAC_API drflac_uint64 drflac_read_pcm_frames_f32(drflac* pFlac, drflac_uint64 framesToRead, float* pBufferOut) +{ + drflac_uint64 framesRead; + drflac_uint32 unusedBitsPerSample; + + if (pFlac == NULL || framesToRead == 0) { + return 0; + } + + if (pBufferOut == NULL) { + return drflac__seek_forward_by_pcm_frames(pFlac, framesToRead); + } + + DRFLAC_ASSERT(pFlac->bitsPerSample <= 32); + unusedBitsPerSample = 32 - pFlac->bitsPerSample; + + framesRead = 0; + while (framesToRead > 0) { + /* If we've run out of samples in this frame, go to the next. */ + if (pFlac->currentFLACFrame.pcmFramesRemaining == 0) { + if (!drflac__read_and_decode_next_flac_frame(pFlac)) { + break; /* Couldn't read the next frame, so just break from the loop and return. */ + } + } else { + unsigned int channelCount = drflac__get_channel_count_from_channel_assignment(pFlac->currentFLACFrame.header.channelAssignment); + drflac_uint64 iFirstPCMFrame = pFlac->currentFLACFrame.header.blockSizeInPCMFrames - pFlac->currentFLACFrame.pcmFramesRemaining; + drflac_uint64 frameCountThisIteration = framesToRead; + + if (frameCountThisIteration > pFlac->currentFLACFrame.pcmFramesRemaining) { + frameCountThisIteration = pFlac->currentFLACFrame.pcmFramesRemaining; + } + + if (channelCount == 2) { + const drflac_int32* pDecodedSamples0 = pFlac->currentFLACFrame.subframes[0].pSamplesS32 + iFirstPCMFrame; + const drflac_int32* pDecodedSamples1 = pFlac->currentFLACFrame.subframes[1].pSamplesS32 + iFirstPCMFrame; + + switch (pFlac->currentFLACFrame.header.channelAssignment) + { + case DRFLAC_CHANNEL_ASSIGNMENT_LEFT_SIDE: + { + drflac_read_pcm_frames_f32__decode_left_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_RIGHT_SIDE: + { + drflac_read_pcm_frames_f32__decode_right_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_MID_SIDE: + { + drflac_read_pcm_frames_f32__decode_mid_side(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + + case DRFLAC_CHANNEL_ASSIGNMENT_INDEPENDENT: + default: + { + drflac_read_pcm_frames_f32__decode_independent_stereo(pFlac, frameCountThisIteration, unusedBitsPerSample, pDecodedSamples0, pDecodedSamples1, pBufferOut); + } break; + } + } else { + /* Generic interleaving. */ + drflac_uint64 i; + for (i = 0; i < frameCountThisIteration; ++i) { + unsigned int j; + for (j = 0; j < channelCount; ++j) { + drflac_int32 sampleS32 = (drflac_int32)((drflac_uint32)(pFlac->currentFLACFrame.subframes[j].pSamplesS32[iFirstPCMFrame + i]) << (unusedBitsPerSample + pFlac->currentFLACFrame.subframes[j].wastedBitsPerSample)); + pBufferOut[(i*channelCount)+j] = (float)(sampleS32 / 2147483648.0); + } + } + } + + framesRead += frameCountThisIteration; + pBufferOut += frameCountThisIteration * channelCount; + framesToRead -= frameCountThisIteration; + pFlac->currentPCMFrame += frameCountThisIteration; + pFlac->currentFLACFrame.pcmFramesRemaining -= (unsigned int)frameCountThisIteration; + } + } + + return framesRead; +} + + +DRFLAC_API drflac_bool32 drflac_seek_to_pcm_frame(drflac* pFlac, drflac_uint64 pcmFrameIndex) +{ + if (pFlac == NULL) { + return DRFLAC_FALSE; + } + + /* Don't do anything if we're already on the seek point. */ + if (pFlac->currentPCMFrame == pcmFrameIndex) { + return DRFLAC_TRUE; + } + + /* + If we don't know where the first frame begins then we can't seek. This will happen when the STREAMINFO block was not present + when the decoder was opened. + */ + if (pFlac->firstFLACFramePosInBytes == 0) { + return DRFLAC_FALSE; + } + + if (pcmFrameIndex == 0) { + pFlac->currentPCMFrame = 0; + return drflac__seek_to_first_frame(pFlac); + } else { + drflac_bool32 wasSuccessful = DRFLAC_FALSE; + drflac_uint64 originalPCMFrame = pFlac->currentPCMFrame; + + /* Clamp the sample to the end only when the length is known. */ + if (pFlac->totalPCMFrameCount != 0 && pcmFrameIndex > pFlac->totalPCMFrameCount) { + pcmFrameIndex = pFlac->totalPCMFrameCount; + } + + /* If the target sample and the current sample are in the same frame we just move the position forward. */ + if (drflac__is_current_flac_frame_valid(pFlac)) { + if (pcmFrameIndex > pFlac->currentPCMFrame) { + /* Forward. */ + drflac_uint64 offset = pcmFrameIndex - pFlac->currentPCMFrame; + if (pFlac->currentFLACFrame.pcmFramesRemaining > offset) { + pFlac->currentFLACFrame.pcmFramesRemaining -= (drflac_uint32)offset; + pFlac->currentPCMFrame = pcmFrameIndex; + return DRFLAC_TRUE; + } + } else { + /* Backward. */ + drflac_uint64 offsetAbs = pFlac->currentPCMFrame - pcmFrameIndex; + drflac_uint32 currentFLACFramePCMFrameCount = pFlac->currentFLACFrame.header.blockSizeInPCMFrames; + drflac_uint32 currentFLACFramePCMFramesConsumed = currentFLACFramePCMFrameCount - pFlac->currentFLACFrame.pcmFramesRemaining; + if (currentFLACFramePCMFramesConsumed > offsetAbs) { + pFlac->currentFLACFrame.pcmFramesRemaining += (drflac_uint32)offsetAbs; + pFlac->currentPCMFrame = pcmFrameIndex; + return DRFLAC_TRUE; + } + } + } + + /* + Different techniques depending on encapsulation. Using the native FLAC seektable with Ogg encapsulation is a bit awkward so + we'll instead use Ogg's natural seeking facility. + */ +#ifndef DR_FLAC_NO_OGG + if (pFlac->container == drflac_container_ogg) + { + wasSuccessful = drflac_ogg__seek_to_pcm_frame(pFlac, pcmFrameIndex); + } + else +#endif + { + /* First try seeking via the seek table. If this fails, fall back to a brute force seek which is much slower. */ + if (/*!wasSuccessful && */!pFlac->_noSeekTableSeek) { + wasSuccessful = drflac__seek_to_pcm_frame__seek_table(pFlac, pcmFrameIndex); + } + +#if !defined(DR_FLAC_NO_CRC) + /* Fall back to binary search if seek table seeking fails. This requires the length of the stream to be known. */ + if (!wasSuccessful && !pFlac->_noBinarySearchSeek && pFlac->totalPCMFrameCount > 0) { + wasSuccessful = drflac__seek_to_pcm_frame__binary_search(pFlac, pcmFrameIndex); + } +#endif + + /* Fall back to brute force if all else fails. */ + if (!wasSuccessful && !pFlac->_noBruteForceSeek) { + wasSuccessful = drflac__seek_to_pcm_frame__brute_force(pFlac, pcmFrameIndex); + } + } + + if (wasSuccessful) { + pFlac->currentPCMFrame = pcmFrameIndex; + } else { + /* Reset the stream before recovery because a failed seek may have moved it without updating currentPCMFrame. */ + drflac__seek_to_first_frame(pFlac); + if (originalPCMFrame != 0 && drflac_seek_to_pcm_frame(pFlac, originalPCMFrame) == DRFLAC_FALSE) { + /* Failed to seek back to the original PCM frame. Fall back to 0. */ + drflac__seek_to_first_frame(pFlac); + } + } + + return wasSuccessful; + } +} + + + +/* High Level APIs */ + +/* SIZE_MAX */ +#if defined(SIZE_MAX) + #define DRFLAC_SIZE_MAX SIZE_MAX +#else + #if defined(DRFLAC_64BIT) + #define DRFLAC_SIZE_MAX ((drflac_uint64)0xFFFFFFFFFFFFFFFF) + #else + #define DRFLAC_SIZE_MAX 0xFFFFFFFF + #endif +#endif +/* End SIZE_MAX */ + + +/* Using a macro as the definition of the drflac__full_decode_and_close_*() API family. Sue me. */ +#define DRFLAC_DEFINE_FULL_READ_AND_CLOSE(extension, type) \ +static type* drflac__full_read_and_close_ ## extension (drflac* pFlac, unsigned int* channelsOut, unsigned int* sampleRateOut, drflac_uint64* totalPCMFrameCountOut)\ +{ \ + type* pSampleData = NULL; \ + drflac_uint64 totalPCMFrameCount; \ + type buffer[4096]; \ + drflac_uint64 pcmFramesRead; \ + size_t sampleDataBufferSize = sizeof(buffer); \ + \ + DRFLAC_ASSERT(pFlac != NULL); \ + \ + totalPCMFrameCount = 0; \ + \ + pSampleData = (type*)drflac__malloc_from_callbacks(sampleDataBufferSize, &pFlac->allocationCallbacks); \ + if (pSampleData == NULL) { \ + goto on_error; \ + } \ + \ + while ((pcmFramesRead = (drflac_uint64)drflac_read_pcm_frames_##extension(pFlac, sizeof(buffer)/sizeof(buffer[0])/pFlac->channels, buffer)) > 0) { \ + if (((totalPCMFrameCount + pcmFramesRead) * pFlac->channels * sizeof(type)) > sampleDataBufferSize) { \ + type* pNewSampleData; \ + size_t newSampleDataBufferSize; \ + \ + newSampleDataBufferSize = sampleDataBufferSize * 2; \ + pNewSampleData = (type*)drflac__realloc_from_callbacks(pSampleData, newSampleDataBufferSize, sampleDataBufferSize, &pFlac->allocationCallbacks); \ + if (pNewSampleData == NULL) { \ + drflac__free_from_callbacks(pSampleData, &pFlac->allocationCallbacks); \ + goto on_error; \ + } \ + \ + sampleDataBufferSize = newSampleDataBufferSize; \ + pSampleData = pNewSampleData; \ + } \ + \ + DRFLAC_COPY_MEMORY(pSampleData + (totalPCMFrameCount*pFlac->channels), buffer, (size_t)(pcmFramesRead*pFlac->channels*sizeof(type))); \ + totalPCMFrameCount += pcmFramesRead; \ + } \ + \ + /* At this point everything should be decoded, but we just want to fill the unused part buffer with silence - need to \ + protect those ears from random noise! */ \ + DRFLAC_ZERO_MEMORY(pSampleData + (totalPCMFrameCount*pFlac->channels), (size_t)(sampleDataBufferSize - totalPCMFrameCount*pFlac->channels*sizeof(type))); \ + \ + if (sampleRateOut) *sampleRateOut = pFlac->sampleRate; \ + if (channelsOut) *channelsOut = pFlac->channels; \ + if (totalPCMFrameCountOut) *totalPCMFrameCountOut = totalPCMFrameCount; \ + \ + drflac_close(pFlac); \ + return pSampleData; \ + \ +on_error: \ + drflac_close(pFlac); \ + return NULL; \ +} + +DRFLAC_DEFINE_FULL_READ_AND_CLOSE(s32, drflac_int32) +DRFLAC_DEFINE_FULL_READ_AND_CLOSE(s16, drflac_int16) +DRFLAC_DEFINE_FULL_READ_AND_CLOSE(f32, float) + +DRFLAC_API drflac_int32* drflac_open_and_read_pcm_frames_s32(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drflac_uint64* totalPCMFrameCountOut, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalPCMFrameCountOut) { + *totalPCMFrameCountOut = 0; + } + + pFlac = drflac_open(onRead, onSeek, onTell, pUserData, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_s32(pFlac, channelsOut, sampleRateOut, totalPCMFrameCountOut); +} + +DRFLAC_API drflac_int16* drflac_open_and_read_pcm_frames_s16(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drflac_uint64* totalPCMFrameCountOut, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalPCMFrameCountOut) { + *totalPCMFrameCountOut = 0; + } + + pFlac = drflac_open(onRead, onSeek, onTell, pUserData, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_s16(pFlac, channelsOut, sampleRateOut, totalPCMFrameCountOut); +} + +DRFLAC_API float* drflac_open_and_read_pcm_frames_f32(drflac_read_proc onRead, drflac_seek_proc onSeek, drflac_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drflac_uint64* totalPCMFrameCountOut, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalPCMFrameCountOut) { + *totalPCMFrameCountOut = 0; + } + + pFlac = drflac_open(onRead, onSeek, onTell, pUserData, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_f32(pFlac, channelsOut, sampleRateOut, totalPCMFrameCountOut); +} + +#ifndef DR_FLAC_NO_STDIO +DRFLAC_API drflac_int32* drflac_open_file_and_read_pcm_frames_s32(const char* filename, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (sampleRate) { + *sampleRate = 0; + } + if (channels) { + *channels = 0; + } + if (totalPCMFrameCount) { + *totalPCMFrameCount = 0; + } + + pFlac = drflac_open_file(filename, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_s32(pFlac, channels, sampleRate, totalPCMFrameCount); +} + +DRFLAC_API drflac_int16* drflac_open_file_and_read_pcm_frames_s16(const char* filename, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (sampleRate) { + *sampleRate = 0; + } + if (channels) { + *channels = 0; + } + if (totalPCMFrameCount) { + *totalPCMFrameCount = 0; + } + + pFlac = drflac_open_file(filename, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_s16(pFlac, channels, sampleRate, totalPCMFrameCount); +} + +DRFLAC_API float* drflac_open_file_and_read_pcm_frames_f32(const char* filename, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (sampleRate) { + *sampleRate = 0; + } + if (channels) { + *channels = 0; + } + if (totalPCMFrameCount) { + *totalPCMFrameCount = 0; + } + + pFlac = drflac_open_file(filename, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_f32(pFlac, channels, sampleRate, totalPCMFrameCount); +} +#endif + +DRFLAC_API drflac_int32* drflac_open_memory_and_read_pcm_frames_s32(const void* data, size_t dataSize, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (sampleRate) { + *sampleRate = 0; + } + if (channels) { + *channels = 0; + } + if (totalPCMFrameCount) { + *totalPCMFrameCount = 0; + } + + pFlac = drflac_open_memory(data, dataSize, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_s32(pFlac, channels, sampleRate, totalPCMFrameCount); +} + +DRFLAC_API drflac_int16* drflac_open_memory_and_read_pcm_frames_s16(const void* data, size_t dataSize, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (sampleRate) { + *sampleRate = 0; + } + if (channels) { + *channels = 0; + } + if (totalPCMFrameCount) { + *totalPCMFrameCount = 0; + } + + pFlac = drflac_open_memory(data, dataSize, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_s16(pFlac, channels, sampleRate, totalPCMFrameCount); +} + +DRFLAC_API float* drflac_open_memory_and_read_pcm_frames_f32(const void* data, size_t dataSize, unsigned int* channels, unsigned int* sampleRate, drflac_uint64* totalPCMFrameCount, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + drflac* pFlac; + + if (sampleRate) { + *sampleRate = 0; + } + if (channels) { + *channels = 0; + } + if (totalPCMFrameCount) { + *totalPCMFrameCount = 0; + } + + pFlac = drflac_open_memory(data, dataSize, pAllocationCallbacks); + if (pFlac == NULL) { + return NULL; + } + + return drflac__full_read_and_close_f32(pFlac, channels, sampleRate, totalPCMFrameCount); +} + + +DRFLAC_API void drflac_free(void* p, const drflac_allocation_callbacks* pAllocationCallbacks) +{ + if (pAllocationCallbacks != NULL) { + drflac__free_from_callbacks(p, pAllocationCallbacks); + } else { + drflac__free_default(p, NULL); + } +} + + + + +DRFLAC_API void drflac_init_vorbis_comment_iterator(drflac_vorbis_comment_iterator* pIter, drflac_uint32 commentCount, const void* pComments) +{ + if (pIter == NULL) { + return; + } + + pIter->countRemaining = commentCount; + pIter->pRunningData = (const char*)pComments; +} + +DRFLAC_API const char* drflac_next_vorbis_comment(drflac_vorbis_comment_iterator* pIter, drflac_uint32* pCommentLengthOut) +{ + drflac_int32 length; + const char* pComment; + + /* Safety. */ + if (pCommentLengthOut) { + *pCommentLengthOut = 0; + } + + if (pIter == NULL || pIter->countRemaining == 0 || pIter->pRunningData == NULL) { + return NULL; + } + + length = drflac__le2host_32_ptr_unaligned(pIter->pRunningData); + pIter->pRunningData += 4; + + pComment = pIter->pRunningData; + pIter->pRunningData += length; + pIter->countRemaining -= 1; + + if (pCommentLengthOut) { + *pCommentLengthOut = length; + } + + return pComment; +} + + + + +DRFLAC_API void drflac_init_cuesheet_track_iterator(drflac_cuesheet_track_iterator* pIter, drflac_uint32 trackCount, const void* pTrackData) +{ + if (pIter == NULL) { + return; + } + + pIter->countRemaining = trackCount; + pIter->pRunningData = (const char*)pTrackData; +} + +DRFLAC_API drflac_bool32 drflac_next_cuesheet_track(drflac_cuesheet_track_iterator* pIter, drflac_cuesheet_track* pCuesheetTrack) +{ + drflac_cuesheet_track cuesheetTrack; + const char* pRunningData; + drflac_uint64 offsetHi; + drflac_uint64 offsetLo; + + if (pIter == NULL || pIter->countRemaining == 0 || pIter->pRunningData == NULL) { + return DRFLAC_FALSE; + } + + pRunningData = pIter->pRunningData; + + offsetHi = drflac__be2host_32(*(const drflac_uint32*)pRunningData); pRunningData += 4; + offsetLo = drflac__be2host_32(*(const drflac_uint32*)pRunningData); pRunningData += 4; + cuesheetTrack.offset = offsetLo | (offsetHi << 32); + cuesheetTrack.trackNumber = pRunningData[0]; pRunningData += 1; + DRFLAC_COPY_MEMORY(cuesheetTrack.ISRC, pRunningData, sizeof(cuesheetTrack.ISRC)); pRunningData += 12; + cuesheetTrack.isAudio = (pRunningData[0] & 0x80) != 0; + cuesheetTrack.preEmphasis = (pRunningData[0] & 0x40) != 0; pRunningData += 14; + cuesheetTrack.indexCount = pRunningData[0]; pRunningData += 1 + (DRFLAC_CUESHEET_TRACK_PADDED_SIZE_IN_BYTES - DRFLAC_CUESHEET_TRACK_SIZE_IN_BYTES); + cuesheetTrack.pIndexPoints = (const drflac_cuesheet_track_index*)pRunningData; pRunningData += cuesheetTrack.indexCount * sizeof(drflac_cuesheet_track_index); + + pIter->pRunningData = pRunningData; + pIter->countRemaining -= 1; + + if (pCuesheetTrack) { + *pCuesheetTrack = cuesheetTrack; + } + + return DRFLAC_TRUE; +} + +#if defined(__clang__) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))) + #pragma GCC diagnostic pop +#endif +#endif /* dr_flac_c */ +#endif /* DR_FLAC_IMPLEMENTATION */ + +/* +This software is available as a choice of the following licenses. Choose +whichever you prefer. + +=============================================================================== +ALTERNATIVE 1 - Public Domain (www.unlicense.org) +=============================================================================== +This is free and unencumbered software released into the public domain. + +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. + +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +For more information, please refer to <http://unlicense.org/> + +=============================================================================== +ALTERNATIVE 2 - MIT No Attribution +=============================================================================== +Copyright 2023 David Reid + +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +*/ diff --git a/vendor/dr_wav.h b/vendor/dr_wav.h new file mode 100644 index 0000000..0ae2fdb --- /dev/null +++ b/vendor/dr_wav.h @@ -0,0 +1,8630 @@ +/* +WAV audio loader and writer. Choice of public domain or MIT-0. See license statements at the end of this file. +dr_wav - v0.14.6 - TBD + +David Reid - [email protected] + +GitHub: https://github.com/mackron/dr_libs +*/ + +/* +Introduction +============ +This is a single file library. To use it, do something like the following in one .c file. + + ```c + #define DR_WAV_IMPLEMENTATION + #include "dr_wav.h" + ``` + +You can then #include this file in other parts of the program as you would with any other header file. Do something like the following to read audio data: + + ```c + drwav wav; + if (!drwav_init_file(&wav, "my_song.wav", NULL)) { + // Error opening WAV file. + } + + drwav_int32* pDecodedInterleavedPCMFrames = malloc(wav.totalPCMFrameCount * wav.channels * sizeof(drwav_int32)); + size_t numberOfSamplesActuallyDecoded = drwav_read_pcm_frames_s32(&wav, wav.totalPCMFrameCount, pDecodedInterleavedPCMFrames); + + ... + + drwav_uninit(&wav); + ``` + +If you just want to quickly open and read the audio data in a single operation you can do something like this: + + ```c + unsigned int channels; + unsigned int sampleRate; + drwav_uint64 totalPCMFrameCount; + float* pSampleData = drwav_open_file_and_read_pcm_frames_f32("my_song.wav", &channels, &sampleRate, &totalPCMFrameCount, NULL); + if (pSampleData == NULL) { + // Error opening and reading WAV file. + } + + ... + + drwav_free(pSampleData, NULL); + ``` + +The examples above use versions of the API that convert the audio data to a consistent format (32-bit signed PCM, in this case), but you can still output the +audio data in its internal format (see notes below for supported formats): + + ```c + size_t framesRead = drwav_read_pcm_frames(&wav, wav.totalPCMFrameCount, pDecodedInterleavedPCMFrames); + ``` + +You can also read the raw bytes of audio data, which could be useful if dr_wav does not have native support for a particular data format: + + ```c + size_t bytesRead = drwav_read_raw(&wav, bytesToRead, pRawDataBuffer); + ``` + +dr_wav can also be used to output WAV files. This does not currently support compressed formats. To use this, look at `drwav_init_write()`, +`drwav_init_file_write()`, etc. Use `drwav_write_pcm_frames()` to write samples, or `drwav_write_raw()` to write raw data in the "data" chunk. + + ```c + drwav_data_format format; + format.container = drwav_container_riff; // <-- drwav_container_riff = normal WAV files, drwav_container_w64 = Sony Wave64. + format.format = DR_WAVE_FORMAT_PCM; // <-- Any of the DR_WAVE_FORMAT_* codes. + format.channels = 2; + format.sampleRate = 44100; + format.bitsPerSample = 16; + drwav_init_file_write(&wav, "data/recording.wav", &format, NULL); + + ... + + drwav_uint64 framesWritten = drwav_write_pcm_frames(pWav, frameCount, pSamples); + ``` + +Note that writing to AIFF or RIFX is not supported. + +dr_wav has support for decoding from a number of different encapsulation formats. See below for details. + + +Build Options +============= +#define these options before including this file. + +#define DR_WAV_NO_CONVERSION_API + Disables conversion APIs such as `drwav_read_pcm_frames_f32()` and `drwav_s16_to_f32()`. + +#define DR_WAV_NO_STDIO + Disables APIs that initialize a decoder from a file such as `drwav_init_file()`, `drwav_init_file_write()`, etc. + +#define DR_WAV_NO_WCHAR + Disables all functions ending with `_w`. Use this if your compiler does not provide wchar.h. Not required if DR_WAV_NO_STDIO is also defined. + + +Supported Encapsulations +======================== +- RIFF (Regular WAV) +- RIFX (Big-Endian) +- AIFF (Does not currently support ADPCM) +- RF64 +- W64 + +Note that AIFF and RIFX do not support write mode, nor do they support reading of metadata. + + +Supported Encodings +=================== +- Unsigned 8-bit PCM +- Signed 12-bit PCM +- Signed 16-bit PCM +- Signed 24-bit PCM +- Signed 32-bit PCM +- IEEE 32-bit floating point +- IEEE 64-bit floating point +- A-law and u-law +- Microsoft ADPCM +- IMA ADPCM (DVI, format code 0x11) + +8-bit PCM encodings are always assumed to be unsigned. Signed 8-bit encoding can only be read with `drwav_read_raw()`. + +Note that ADPCM is not currently supported with AIFF. Contributions welcome. + + +Notes +===== +- Samples are always interleaved. +- The default read function does not do any data conversion. Use `drwav_read_pcm_frames_f32()`, `drwav_read_pcm_frames_s32()` and `drwav_read_pcm_frames_s16()` + to read and convert audio data to 32-bit floating point, signed 32-bit integer and signed 16-bit integer samples respectively. +- dr_wav will try to read the WAV file as best it can, even if it's not strictly conformant to the WAV format. +*/ + +#ifndef dr_wav_h +#define dr_wav_h + +#ifdef __cplusplus +extern "C" { +#endif + +#define DRWAV_STRINGIFY(x) #x +#define DRWAV_XSTRINGIFY(x) DRWAV_STRINGIFY(x) + +#define DRWAV_VERSION_MAJOR 0 +#define DRWAV_VERSION_MINOR 14 +#define DRWAV_VERSION_REVISION 6 +#define DRWAV_VERSION_STRING DRWAV_XSTRINGIFY(DRWAV_VERSION_MAJOR) "." DRWAV_XSTRINGIFY(DRWAV_VERSION_MINOR) "." DRWAV_XSTRINGIFY(DRWAV_VERSION_REVISION) + +#include <stddef.h> /* For size_t. */ + +/* Sized Types */ +typedef signed char drwav_int8; +typedef unsigned char drwav_uint8; +typedef signed short drwav_int16; +typedef unsigned short drwav_uint16; +typedef signed int drwav_int32; +typedef unsigned int drwav_uint32; +#if defined(_MSC_VER) && !defined(__clang__) + typedef signed __int64 drwav_int64; + typedef unsigned __int64 drwav_uint64; +#else + #if defined(__clang__) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))) + #pragma GCC diagnostic push + #pragma GCC diagnostic ignored "-Wlong-long" + #if defined(__clang__) + #pragma GCC diagnostic ignored "-Wc++11-long-long" + #endif + #endif + typedef signed long long drwav_int64; + typedef unsigned long long drwav_uint64; + #if defined(__clang__) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))) + #pragma GCC diagnostic pop + #endif +#endif +#if defined(__LP64__) || defined(_WIN64) || (defined(__x86_64__) && !defined(__ILP32__)) || defined(_M_X64) || defined(__ia64) || defined (_M_IA64) || defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC) || defined(__powerpc64__) + typedef drwav_uint64 drwav_uintptr; +#else + typedef drwav_uint32 drwav_uintptr; +#endif +typedef drwav_uint8 drwav_bool8; +typedef drwav_uint32 drwav_bool32; +#define DRWAV_TRUE 1 +#define DRWAV_FALSE 0 +/* End Sized Types */ + +/* Decorations */ +#if !defined(DRWAV_API) + #if defined(DRWAV_DLL) + #if defined(_WIN32) + #define DRWAV_DLL_IMPORT __declspec(dllimport) + #define DRWAV_DLL_EXPORT __declspec(dllexport) + #define DRWAV_DLL_PRIVATE static + #else + #if defined(__GNUC__) && __GNUC__ >= 4 + #define DRWAV_DLL_IMPORT __attribute__((visibility("default"))) + #define DRWAV_DLL_EXPORT __attribute__((visibility("default"))) + #define DRWAV_DLL_PRIVATE __attribute__((visibility("hidden"))) + #else + #define DRWAV_DLL_IMPORT + #define DRWAV_DLL_EXPORT + #define DRWAV_DLL_PRIVATE static + #endif + #endif + + #if defined(DR_WAV_IMPLEMENTATION) || defined(DRWAV_IMPLEMENTATION) + #define DRWAV_API DRWAV_DLL_EXPORT + #else + #define DRWAV_API DRWAV_DLL_IMPORT + #endif + #define DRWAV_PRIVATE DRWAV_DLL_PRIVATE + #else + #define DRWAV_API extern + #define DRWAV_PRIVATE static + #endif +#endif +/* End Decorations */ + +/* Result Codes */ +typedef drwav_int32 drwav_result; +#define DRWAV_SUCCESS 0 +#define DRWAV_ERROR -1 /* A generic error. */ +#define DRWAV_INVALID_ARGS -2 +#define DRWAV_INVALID_OPERATION -3 +#define DRWAV_OUT_OF_MEMORY -4 +#define DRWAV_OUT_OF_RANGE -5 +#define DRWAV_ACCESS_DENIED -6 +#define DRWAV_DOES_NOT_EXIST -7 +#define DRWAV_ALREADY_EXISTS -8 +#define DRWAV_TOO_MANY_OPEN_FILES -9 +#define DRWAV_INVALID_FILE -10 +#define DRWAV_TOO_BIG -11 +#define DRWAV_PATH_TOO_LONG -12 +#define DRWAV_NAME_TOO_LONG -13 +#define DRWAV_NOT_DIRECTORY -14 +#define DRWAV_IS_DIRECTORY -15 +#define DRWAV_DIRECTORY_NOT_EMPTY -16 +#define DRWAV_END_OF_FILE -17 +#define DRWAV_NO_SPACE -18 +#define DRWAV_BUSY -19 +#define DRWAV_IO_ERROR -20 +#define DRWAV_INTERRUPT -21 +#define DRWAV_UNAVAILABLE -22 +#define DRWAV_ALREADY_IN_USE -23 +#define DRWAV_BAD_ADDRESS -24 +#define DRWAV_BAD_SEEK -25 +#define DRWAV_BAD_PIPE -26 +#define DRWAV_DEADLOCK -27 +#define DRWAV_TOO_MANY_LINKS -28 +#define DRWAV_NOT_IMPLEMENTED -29 +#define DRWAV_NO_MESSAGE -30 +#define DRWAV_BAD_MESSAGE -31 +#define DRWAV_NO_DATA_AVAILABLE -32 +#define DRWAV_INVALID_DATA -33 +#define DRWAV_TIMEOUT -34 +#define DRWAV_NO_NETWORK -35 +#define DRWAV_NOT_UNIQUE -36 +#define DRWAV_NOT_SOCKET -37 +#define DRWAV_NO_ADDRESS -38 +#define DRWAV_BAD_PROTOCOL -39 +#define DRWAV_PROTOCOL_UNAVAILABLE -40 +#define DRWAV_PROTOCOL_NOT_SUPPORTED -41 +#define DRWAV_PROTOCOL_FAMILY_NOT_SUPPORTED -42 +#define DRWAV_ADDRESS_FAMILY_NOT_SUPPORTED -43 +#define DRWAV_SOCKET_NOT_SUPPORTED -44 +#define DRWAV_CONNECTION_RESET -45 +#define DRWAV_ALREADY_CONNECTED -46 +#define DRWAV_NOT_CONNECTED -47 +#define DRWAV_CONNECTION_REFUSED -48 +#define DRWAV_NO_HOST -49 +#define DRWAV_IN_PROGRESS -50 +#define DRWAV_CANCELLED -51 +#define DRWAV_MEMORY_ALREADY_MAPPED -52 +#define DRWAV_AT_END -53 +/* End Result Codes */ + +/* Common data formats. */ +#define DR_WAVE_FORMAT_PCM 0x1 +#define DR_WAVE_FORMAT_ADPCM 0x2 +#define DR_WAVE_FORMAT_IEEE_FLOAT 0x3 +#define DR_WAVE_FORMAT_ALAW 0x6 +#define DR_WAVE_FORMAT_MULAW 0x7 +#define DR_WAVE_FORMAT_DTS 0x8 +#define DR_WAVE_FORMAT_DVI_ADPCM 0x11 +#define DR_WAVE_FORMAT_EXTENSIBLE 0xFFFE + +/* Flags to pass into drwav_init_ex(), etc. */ +#define DRWAV_SEQUENTIAL 0x00000001 +#define DRWAV_WITH_METADATA 0x00000002 + +DRWAV_API void drwav_version(drwav_uint32* pMajor, drwav_uint32* pMinor, drwav_uint32* pRevision); +DRWAV_API const char* drwav_version_string(void); + +/* Allocation Callbacks */ +typedef struct +{ + void* pUserData; + void* (* onMalloc)(size_t sz, void* pUserData); + void* (* onRealloc)(void* p, size_t sz, void* pUserData); + void (* onFree)(void* p, void* pUserData); +} drwav_allocation_callbacks; +/* End Allocation Callbacks */ + +typedef enum +{ + DRWAV_SEEK_SET, + DRWAV_SEEK_CUR, + DRWAV_SEEK_END +} drwav_seek_origin; + +typedef enum +{ + drwav_container_riff, + drwav_container_rifx, + drwav_container_w64, + drwav_container_rf64, + drwav_container_aiff +} drwav_container; + +typedef struct +{ + union + { + drwav_uint8 fourcc[4]; + drwav_uint8 guid[16]; + } id; + + /* The size in bytes of the chunk. */ + drwav_uint64 sizeInBytes; + + /* + RIFF = 2 byte alignment. + W64 = 8 byte alignment. + */ + unsigned int paddingSize; +} drwav_chunk_header; + +typedef struct +{ + /* + The format tag exactly as specified in the wave file's "fmt" chunk. This can be used by applications + that require support for data formats not natively supported by dr_wav. + */ + drwav_uint16 formatTag; + + /* The number of channels making up the audio data. When this is set to 1 it is mono, 2 is stereo, etc. */ + drwav_uint16 channels; + + /* The sample rate. Usually set to something like 44100. */ + drwav_uint32 sampleRate; + + /* Average bytes per second. You probably don't need this, but it's left here for informational purposes. */ + drwav_uint32 avgBytesPerSec; + + /* Block align. This is equal to the number of channels * bytes per sample. */ + drwav_uint16 blockAlign; + + /* Bits per sample. */ + drwav_uint16 bitsPerSample; + + /* The size of the extended data. Only used internally for validation, but left here for informational purposes. */ + drwav_uint16 extendedSize; + + /* + The number of valid bits per sample. When <formatTag> is equal to WAVE_FORMAT_EXTENSIBLE, <bitsPerSample> + is always rounded up to the nearest multiple of 8. This variable contains information about exactly how + many bits are valid per sample. Mainly used for informational purposes. + */ + drwav_uint16 validBitsPerSample; + + /* The channel mask. Not used at the moment. */ + drwav_uint32 channelMask; + + /* The sub-format, exactly as specified by the wave file. */ + drwav_uint8 subFormat[16]; +} drwav_fmt; + +DRWAV_API drwav_uint16 drwav_fmt_get_format(const drwav_fmt* pFMT); + + +/* +Callback for when data is read. Return value is the number of bytes actually read. + +pUserData [in] The user data that was passed to drwav_init() and family. +pBufferOut [out] The output buffer. +bytesToRead [in] The number of bytes to read. + +Returns the number of bytes actually read. + +A return value of less than bytesToRead indicates the end of the stream. Do _not_ return from this callback until +either the entire bytesToRead is filled or you have reached the end of the stream. +*/ +typedef size_t (* drwav_read_proc)(void* pUserData, void* pBufferOut, size_t bytesToRead); + +/* +Callback for when data is written. Returns value is the number of bytes actually written. + +pUserData [in] The user data that was passed to drwav_init_write() and family. +pData [out] A pointer to the data to write. +bytesToWrite [in] The number of bytes to write. + +Returns the number of bytes actually written. + +If the return value differs from bytesToWrite, it indicates an error. +*/ +typedef size_t (* drwav_write_proc)(void* pUserData, const void* pData, size_t bytesToWrite); + +/* +Callback for when data needs to be seeked. + +pUserData [in] The user data that was passed to drwav_init() and family. +offset [in] The number of bytes to move, relative to the origin. Will never be negative. +origin [in] The origin of the seek - the current position or the start of the stream. + +Returns whether or not the seek was successful. + +Whether or not it is relative to the beginning or current position is determined by the "origin" parameter which will be either DRWAV_SEEK_SET or +DRWAV_SEEK_CUR. +*/ +typedef drwav_bool32 (* drwav_seek_proc)(void* pUserData, int offset, drwav_seek_origin origin); + +/* +Callback for when the current position in the stream needs to be retrieved. + +pUserData [in] The user data that was passed to drwav_init() and family. +pCursor [out] A pointer to a variable to receive the current position in the stream. + +Returns whether or not the operation was successful. +*/ +typedef drwav_bool32 (* drwav_tell_proc)(void* pUserData, drwav_int64* pCursor); + +/* +Callback for when drwav_init_ex() finds a chunk. + +pChunkUserData [in] The user data that was passed to the pChunkUserData parameter of drwav_init_ex() and family. +onRead [in] A pointer to the function to call when reading. +onSeek [in] A pointer to the function to call when seeking. +pReadSeekUserData [in] The user data that was passed to the pReadSeekUserData parameter of drwav_init_ex() and family. +pChunkHeader [in] A pointer to an object containing basic header information about the chunk. Use this to identify the chunk. +container [in] Whether or not the WAV file is a RIFF or Wave64 container. If you're unsure of the difference, assume RIFF. +pFMT [in] A pointer to the object containing the contents of the "fmt" chunk. + +Returns the number of bytes read + seeked. + +To read data from the chunk, call onRead(), passing in pReadSeekUserData as the first parameter. Do the same for seeking with onSeek(). The return value must +be the total number of bytes you have read _plus_ seeked. + +Use the `container` argument to discriminate the fields in `pChunkHeader->id`. If the container is `drwav_container_riff` or `drwav_container_rf64` you should +use `id.fourcc`, otherwise you should use `id.guid`. + +The `pFMT` parameter can be used to determine the data format of the wave file. Use `drwav_fmt_get_format()` to get the sample format, which will be one of the +`DR_WAVE_FORMAT_*` identifiers. + +The read pointer will be sitting on the first byte after the chunk's header. You must not attempt to read beyond the boundary of the chunk. +*/ +typedef drwav_uint64 (* drwav_chunk_proc)(void* pChunkUserData, drwav_read_proc onRead, drwav_seek_proc onSeek, void* pReadSeekUserData, const drwav_chunk_header* pChunkHeader, drwav_container container, const drwav_fmt* pFMT); + + +/* Structure for internal use. Only used for loaders opened with drwav_init_memory(). */ +typedef struct +{ + const drwav_uint8* data; + size_t dataSize; + size_t currentReadPos; +} drwav__memory_stream; + +/* Structure for internal use. Only used for writers opened with drwav_init_memory_write(). */ +typedef struct +{ + void** ppData; + size_t* pDataSize; + size_t dataSize; + size_t dataCapacity; + size_t currentWritePos; +} drwav__memory_stream_write; + +typedef struct +{ + drwav_container container; /* RIFF, W64. */ + drwav_uint32 format; /* DR_WAVE_FORMAT_* */ + drwav_uint32 channels; + drwav_uint32 sampleRate; + drwav_uint32 bitsPerSample; +} drwav_data_format; + +typedef enum +{ + drwav_metadata_type_none = 0, + + /* + Unknown simply means a chunk that drwav does not handle specifically. You can still ask to + receive these chunks as metadata objects. It is then up to you to interpret the chunk's data. + You can also write unknown metadata to a wav file. Be careful writing unknown chunks if you + have also edited the audio data. The unknown chunks could represent offsets/sizes that no + longer correctly correspond to the audio data. + */ + drwav_metadata_type_unknown = 1 << 0, + + /* Only 1 of each of these metadata items are allowed in a wav file. */ + drwav_metadata_type_smpl = 1 << 1, + drwav_metadata_type_inst = 1 << 2, + drwav_metadata_type_cue = 1 << 3, + drwav_metadata_type_acid = 1 << 4, + drwav_metadata_type_bext = 1 << 5, + + /* + Wav files often have a LIST chunk. This is a chunk that contains a set of subchunks. For this + higher-level metadata API, we don't make a distinction between a regular chunk and a LIST + subchunk. Instead, they are all just 'metadata' items. + + There can be multiple of these metadata items in a wav file. + */ + drwav_metadata_type_list_label = 1 << 6, + drwav_metadata_type_list_note = 1 << 7, + drwav_metadata_type_list_labelled_cue_region = 1 << 8, + + drwav_metadata_type_list_info_software = 1 << 9, + drwav_metadata_type_list_info_copyright = 1 << 10, + drwav_metadata_type_list_info_title = 1 << 11, + drwav_metadata_type_list_info_artist = 1 << 12, + drwav_metadata_type_list_info_comment = 1 << 13, + drwav_metadata_type_list_info_date = 1 << 14, + drwav_metadata_type_list_info_genre = 1 << 15, + drwav_metadata_type_list_info_album = 1 << 16, + drwav_metadata_type_list_info_tracknumber = 1 << 17, + drwav_metadata_type_list_info_location = 1 << 18, + drwav_metadata_type_list_info_organization = 1 << 19, + drwav_metadata_type_list_info_keywords = 1 << 20, + drwav_metadata_type_list_info_medium = 1 << 21, + drwav_metadata_type_list_info_description = 1 << 22, + + /* Other type constants for convenience. */ + drwav_metadata_type_list_all_info_strings = drwav_metadata_type_list_info_software + | drwav_metadata_type_list_info_copyright + | drwav_metadata_type_list_info_title + | drwav_metadata_type_list_info_artist + | drwav_metadata_type_list_info_comment + | drwav_metadata_type_list_info_date + | drwav_metadata_type_list_info_genre + | drwav_metadata_type_list_info_album + | drwav_metadata_type_list_info_tracknumber + | drwav_metadata_type_list_info_location + | drwav_metadata_type_list_info_organization + | drwav_metadata_type_list_info_keywords + | drwav_metadata_type_list_info_medium + | drwav_metadata_type_list_info_description, + + drwav_metadata_type_list_all_adtl = drwav_metadata_type_list_label + | drwav_metadata_type_list_note + | drwav_metadata_type_list_labelled_cue_region, + + drwav_metadata_type_all = -2, /*0xFFFFFFFF & ~drwav_metadata_type_unknown,*/ + drwav_metadata_type_all_including_unknown = -1 /*0xFFFFFFFF,*/ +} drwav_metadata_type; + +/* +Sampler Metadata + +The sampler chunk contains information about how a sound should be played in the context of a whole +audio production, and when used in a sampler. See https://en.wikipedia.org/wiki/Sample-based_synthesis. +*/ +typedef enum +{ + drwav_smpl_loop_type_forward = 0, + drwav_smpl_loop_type_pingpong = 1, + drwav_smpl_loop_type_backward = 2 +} drwav_smpl_loop_type; + +typedef struct +{ + /* The ID of the associated cue point, see drwav_cue and drwav_cue_point. As with all cue point IDs, this can correspond to a label chunk to give this loop a name, see drwav_list_label_or_note. */ + drwav_uint32 cuePointId; + + /* See drwav_smpl_loop_type. */ + drwav_uint32 type; + + /* The offset of the first sample to be played in the loop. */ + drwav_uint32 firstSampleOffset; + + /* The offset into the audio data of the last sample to be played in the loop. */ + drwav_uint32 lastSampleOffset; + + /* A value to represent that playback should occur at a point between samples. This value ranges from 0 to UINT32_MAX. Where a value of 0 means no fraction, and a value of (UINT32_MAX / 2) would mean half a sample. */ + drwav_uint32 sampleFraction; + + /* Number of times to play the loop. 0 means loop infinitely. */ + drwav_uint32 playCount; +} drwav_smpl_loop; + +typedef struct +{ + /* IDs for a particular MIDI manufacturer. 0 if not used. */ + drwav_uint32 manufacturerId; + drwav_uint32 productId; + + /* The period of 1 sample in nanoseconds. */ + drwav_uint32 samplePeriodNanoseconds; + + /* The MIDI root note of this file. 0 to 127. */ + drwav_uint32 midiUnityNote; + + /* The fraction of a semitone up from the given MIDI note. This is a value from 0 to UINT32_MAX, where 0 means no change and (UINT32_MAX / 2) is half a semitone (AKA 50 cents). */ + drwav_uint32 midiPitchFraction; + + /* Data relating to SMPTE standards which are used for syncing audio and video. 0 if not used. */ + drwav_uint32 smpteFormat; + drwav_uint32 smpteOffset; + + /* drwav_smpl_loop loops. */ + drwav_uint32 sampleLoopCount; + + /* Optional sampler-specific data. */ + drwav_uint32 samplerSpecificDataSizeInBytes; + + drwav_smpl_loop* pLoops; + drwav_uint8* pSamplerSpecificData; +} drwav_smpl; + +/* +Instrument Metadata + +The inst metadata contains data about how a sound should be played as part of an instrument. This +commonly read by samplers. See https://en.wikipedia.org/wiki/Sample-based_synthesis. +*/ +typedef struct +{ + drwav_int8 midiUnityNote; /* The root note of the audio as a MIDI note number. 0 to 127. */ + drwav_int8 fineTuneCents; /* -50 to +50 */ + drwav_int8 gainDecibels; /* -64 to +64 */ + drwav_int8 lowNote; /* 0 to 127 */ + drwav_int8 highNote; /* 0 to 127 */ + drwav_int8 lowVelocity; /* 1 to 127 */ + drwav_int8 highVelocity; /* 1 to 127 */ +} drwav_inst; + +/* +Cue Metadata + +Cue points are markers at specific points in the audio. They often come with an associated piece of +drwav_list_label_or_note metadata which contains the text for the marker. +*/ +typedef struct +{ + /* Unique identification value. */ + drwav_uint32 id; + + /* Set to 0. This is only relevant if there is a 'playlist' chunk - which is not supported by dr_wav. */ + drwav_uint32 playOrderPosition; + + /* Should always be "data". This represents the fourcc value of the chunk that this cue point corresponds to. dr_wav only supports a single data chunk so this should always be "data". */ + drwav_uint8 dataChunkId[4]; + + /* Set to 0. This is only relevant if there is a wave list chunk. dr_wav, like lots of readers/writers, do not support this. */ + drwav_uint32 chunkStart; + + /* Set to 0 for uncompressed formats. Else the last byte in compressed wave data where decompression can begin to find the value of the corresponding sample value. */ + drwav_uint32 blockStart; + + /* For uncompressed formats this is the offset of the cue point into the audio data. For compressed formats this is relative to the block specified with blockStart. */ + drwav_uint32 sampleOffset; +} drwav_cue_point; + +typedef struct +{ + drwav_uint32 cuePointCount; + drwav_cue_point *pCuePoints; +} drwav_cue; + +/* +Acid Metadata + +This chunk contains some information about the time signature and the tempo of the audio. +*/ +typedef enum +{ + drwav_acid_flag_one_shot = 1, /* If this is not set, then it is a loop instead of a one-shot. */ + drwav_acid_flag_root_note_set = 2, + drwav_acid_flag_stretch = 4, + drwav_acid_flag_disk_based = 8, + drwav_acid_flag_acidizer = 16 /* Not sure what this means. */ +} drwav_acid_flag; + +typedef struct +{ + /* A bit-field, see drwav_acid_flag. */ + drwav_uint32 flags; + + /* Valid if flags contains drwav_acid_flag_root_note_set. It represents the MIDI root note the file - a value from 0 to 127. */ + drwav_uint16 midiUnityNote; + + /* Reserved values that should probably be ignored. reserved1 seems to often be 128 and reserved2 is 0. */ + drwav_uint16 reserved1; + float reserved2; + + /* Number of beats. */ + drwav_uint32 numBeats; + + /* The time signature of the audio. */ + drwav_uint16 meterDenominator; + drwav_uint16 meterNumerator; + + /* Beats per minute of the track. Setting a value of 0 suggests that there is no tempo. */ + float tempo; +} drwav_acid; + +/* +Cue Label or Note metadata + +These are 2 different types of metadata, but they have the exact same format. Labels tend to be the +more common and represent a short name for a cue point. Notes might be used to represent a longer +comment. +*/ +typedef struct +{ + /* The ID of a cue point that this label or note corresponds to. */ + drwav_uint32 cuePointId; + + /* Size of the string not including any null terminator. */ + drwav_uint32 stringLength; + + /* The string. The *init_with_metadata functions null terminate this for convenience. */ + char* pString; +} drwav_list_label_or_note; + +/* +BEXT metadata, also known as Broadcast Wave Format (BWF) + +This metadata adds some extra description to an audio file. You must check the version field to +determine if the UMID or the loudness fields are valid. +*/ +typedef struct +{ + /* + These top 3 fields, and the umid field are actually defined in the standard as a statically + sized buffers. In order to reduce the size of this struct (and therefore the union in the + metadata struct), we instead store these as pointers. + */ + char* pDescription; /* Can be NULL or a null-terminated string, must be <= 256 characters. */ + char* pOriginatorName; /* Can be NULL or a null-terminated string, must be <= 32 characters. */ + char* pOriginatorReference; /* Can be NULL or a null-terminated string, must be <= 32 characters. */ + char pOriginationDate[10]; /* ASCII "yyyy:mm:dd". */ + char pOriginationTime[8]; /* ASCII "hh:mm:ss". */ + drwav_uint64 timeReference; /* First sample count since midnight. */ + drwav_uint16 version; /* Version of the BWF, check this to see if the fields below are valid. */ + + /* + Unrestricted ASCII characters containing a collection of strings terminated by CR/LF. Each + string shall contain a description of a coding process applied to the audio data. + */ + char* pCodingHistory; + drwav_uint32 codingHistorySize; + + /* Fields below this point are only valid if the version is 1 or above. */ + drwav_uint8* pUMID; /* Exactly 64 bytes of SMPTE UMID */ + + /* Fields below this point are only valid if the version is 2 or above. */ + drwav_uint16 loudnessValue; /* Integrated Loudness Value of the file in LUFS (multiplied by 100). */ + drwav_uint16 loudnessRange; /* Loudness Range of the file in LU (multiplied by 100). */ + drwav_uint16 maxTruePeakLevel; /* Maximum True Peak Level of the file expressed as dBTP (multiplied by 100). */ + drwav_uint16 maxMomentaryLoudness; /* Highest value of the Momentary Loudness Level of the file in LUFS (multiplied by 100). */ + drwav_uint16 maxShortTermLoudness; /* Highest value of the Short-Term Loudness Level of the file in LUFS (multiplied by 100). */ +} drwav_bext; + +/* +Info Text Metadata + +There a many different types of information text that can be saved in this format. This is where +things like the album name, the artists, the year it was produced, etc are saved. See +drwav_metadata_type for the full list of types that dr_wav supports. +*/ +typedef struct +{ + /* Size of the string not including any null terminator. */ + drwav_uint32 stringLength; + + /* The string. The *init_with_metadata functions null terminate this for convenience. */ + char* pString; +} drwav_list_info_text; + +/* +Labelled Cue Region Metadata + +The labelled cue region metadata is used to associate some region of audio with text. The region +starts at a cue point, and extends for the given number of samples. +*/ +typedef struct +{ + /* The ID of a cue point that this object corresponds to. */ + drwav_uint32 cuePointId; + + /* The number of samples from the cue point forwards that should be considered this region */ + drwav_uint32 sampleLength; + + /* Four characters used to say what the purpose of this region is. */ + drwav_uint8 purposeId[4]; + + /* Unsure of the exact meanings of these. It appears to be acceptable to set them all to 0. */ + drwav_uint16 country; + drwav_uint16 language; + drwav_uint16 dialect; + drwav_uint16 codePage; + + /* Size of the string not including any null terminator. */ + drwav_uint32 stringLength; + + /* The string. The *init_with_metadata functions null terminate this for convenience. */ + char* pString; +} drwav_list_labelled_cue_region; + +/* +Unknown Metadata + +This chunk just represents a type of chunk that dr_wav does not understand. + +Unknown metadata has a location attached to it. This is because wav files can have a LIST chunk +that contains subchunks. These LIST chunks can be one of two types. An adtl list, or an INFO +list. This enum is used to specify the location of a chunk that dr_wav currently doesn't support. +*/ +typedef enum +{ + drwav_metadata_location_invalid, + drwav_metadata_location_top_level, + drwav_metadata_location_inside_info_list, + drwav_metadata_location_inside_adtl_list +} drwav_metadata_location; + +typedef struct +{ + drwav_uint8 id[4]; + drwav_metadata_location chunkLocation; + drwav_uint32 dataSizeInBytes; + drwav_uint8* pData; +} drwav_unknown_metadata; + +/* +Metadata is saved as a union of all the supported types. +*/ +typedef struct +{ + /* Determines which item in the union is valid. */ + drwav_metadata_type type; + + union + { + drwav_cue cue; + drwav_smpl smpl; + drwav_acid acid; + drwav_inst inst; + drwav_bext bext; + drwav_list_label_or_note labelOrNote; /* List label or list note. */ + drwav_list_labelled_cue_region labelledCueRegion; + drwav_list_info_text infoText; /* Any of the list info types. */ + drwav_unknown_metadata unknown; + } data; +} drwav_metadata; + +typedef struct +{ + /* A pointer to the function to call when more data is needed. */ + drwav_read_proc onRead; + + /* A pointer to the function to call when data needs to be written. Only used when the drwav object is opened in write mode. */ + drwav_write_proc onWrite; + + /* A pointer to the function to call when the wav file needs to be seeked. */ + drwav_seek_proc onSeek; + + /* A pointer to the function to call when the position of the stream needs to be retrieved. */ + drwav_tell_proc onTell; + + /* The user data to pass to callbacks. */ + void* pUserData; + + /* Allocation callbacks. */ + drwav_allocation_callbacks allocationCallbacks; + + + /* Whether or not the WAV file is formatted as a standard RIFF file or W64. */ + drwav_container container; + + + /* Structure containing format information exactly as specified by the wav file. */ + drwav_fmt fmt; + + /* The sample rate. Will be set to something like 44100. */ + drwav_uint32 sampleRate; + + /* The number of channels. This will be set to 1 for monaural streams, 2 for stereo, etc. */ + drwav_uint16 channels; + + /* The bits per sample. Will be set to something like 16, 24, etc. */ + drwav_uint16 bitsPerSample; + + /* Equal to fmt.formatTag, or the value specified by fmt.subFormat if fmt.formatTag is equal to 65534 (WAVE_FORMAT_EXTENSIBLE). */ + drwav_uint16 translatedFormatTag; + + /* The total number of PCM frames making up the audio data. */ + drwav_uint64 totalPCMFrameCount; + + + /* The size in bytes of the data chunk. */ + drwav_uint64 dataChunkDataSize; + + /* The position in the stream of the first data byte of the data chunk. This is used for seeking. */ + drwav_uint64 dataChunkDataPos; + + /* The number of bytes remaining in the data chunk. */ + drwav_uint64 bytesRemaining; + + /* The current read position in PCM frames. */ + drwav_uint64 readCursorInPCMFrames; + + + /* + Only used in sequential write mode. Keeps track of the desired size of the "data" chunk at the point of initialization time. Always + set to 0 for non-sequential writes and when the drwav object is opened in read mode. Used for validation. + */ + drwav_uint64 dataChunkDataSizeTargetWrite; + + /* Keeps track of whether or not the wav writer was initialized in sequential mode. */ + drwav_bool32 isSequentialWrite; + + + /* A array of metadata. This is valid after the *init_with_metadata call returns. It will be valid until drwav_uninit() is called. You can take ownership of this data with drwav_take_ownership_of_metadata(). */ + drwav_metadata* pMetadata; + drwav_uint32 metadataCount; + + + /* A hack to avoid a DRWAV_MALLOC() when opening a decoder with drwav_init_memory(). */ + drwav__memory_stream memoryStream; + drwav__memory_stream_write memoryStreamWrite; + + + /* Microsoft ADPCM specific data. */ + struct + { + drwav_uint32 bytesRemainingInBlock; + drwav_uint16 predictor[2]; + drwav_int32 delta[2]; + drwav_int32 cachedFrames[4]; /* Samples are stored in this cache during decoding. */ + drwav_uint32 cachedFrameCount; + drwav_int32 prevFrames[2][2]; /* The previous 2 samples for each channel (2 channels at most). */ + } msadpcm; + + /* IMA ADPCM specific data. */ + struct + { + drwav_uint32 bytesRemainingInBlock; + drwav_int32 predictor[2]; + drwav_int32 stepIndex[2]; + drwav_int32 cachedFrames[16]; /* Samples are stored in this cache during decoding. */ + drwav_uint32 cachedFrameCount; + } ima; + + /* AIFF specific data. */ + struct + { + drwav_bool8 isLE; /* Will be set to true if the audio data is little-endian encoded. */ + drwav_bool8 isUnsigned; /* Only used for 8-bit samples. When set to true, will be treated as unsigned. */ + } aiff; +} drwav; + + +/* +Initializes a pre-allocated drwav object for reading. + +pWav [out] A pointer to the drwav object being initialized. +onRead [in] The function to call when data needs to be read from the client. +onSeek [in] The function to call when the read position of the client data needs to move. +onChunk [in, optional] The function to call when a chunk is enumerated at initialized time. +pUserData, pReadSeekUserData [in, optional] A pointer to application defined data that will be passed to onRead and onSeek. +pChunkUserData [in, optional] A pointer to application defined data that will be passed to onChunk. +flags [in, optional] A set of flags for controlling how things are loaded. + +Returns true if successful; false otherwise. + +Close the loader with drwav_uninit(). + +This is the lowest level function for initializing a WAV file. You can also use drwav_init_file() and drwav_init_memory() +to open the stream from a file or from a block of memory respectively. + +Possible values for flags: + DRWAV_SEQUENTIAL: Never perform a backwards seek while loading. This disables the chunk callback and will cause this function + to return as soon as the data chunk is found. Any chunks after the data chunk will be ignored. + +drwav_init() is equivalent to "drwav_init_ex(pWav, onRead, onSeek, NULL, pUserData, NULL, 0);". + +The onChunk callback is not called for the WAVE or FMT chunks. The contents of the FMT chunk can be read from pWav->fmt +after the function returns. + +See also: drwav_init_file(), drwav_init_memory(), drwav_uninit() +*/ +DRWAV_API drwav_bool32 drwav_init(drwav* pWav, drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_ex(drwav* pWav, drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, drwav_chunk_proc onChunk, void* pReadSeekTellUserData, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_with_metadata(drwav* pWav, drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks); + +/* +Initializes a pre-allocated drwav object for writing. + +onWrite [in] The function to call when data needs to be written. +onSeek [in] The function to call when the write position needs to move. +pUserData [in, optional] A pointer to application defined data that will be passed to onWrite and onSeek. +metadata, numMetadata [in, optional] An array of metadata objects that should be written to the file. The array is not edited. You are responsible for this metadata memory and it must maintain valid until drwav_uninit() is called. + +Returns true if successful; false otherwise. + +Close the writer with drwav_uninit(). + +This is the lowest level function for initializing a WAV file. You can also use drwav_init_file_write() and drwav_init_memory_write() +to open the stream from a file or from a block of memory respectively. + +If the total sample count is known, you can use drwav_init_write_sequential(). This avoids the need for dr_wav to perform +a post-processing step for storing the total sample count and the size of the data chunk which requires a backwards seek. + +See also: drwav_init_file_write(), drwav_init_memory_write(), drwav_uninit() +*/ +DRWAV_API drwav_bool32 drwav_init_write(drwav* pWav, const drwav_data_format* pFormat, drwav_write_proc onWrite, drwav_seek_proc onSeek, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_write_sequential(drwav* pWav, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, drwav_write_proc onWrite, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_write_sequential_pcm_frames(drwav* pWav, const drwav_data_format* pFormat, drwav_uint64 totalPCMFrameCount, drwav_write_proc onWrite, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_write_with_metadata(drwav* pWav, const drwav_data_format* pFormat, drwav_write_proc onWrite, drwav_seek_proc onSeek, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks, drwav_metadata* pMetadata, drwav_uint32 metadataCount); + +/* +Utility function to determine the target size of the entire data to be written (including all headers and chunks). + +Returns the target size in bytes. + +The metadata argument can be NULL meaning no metadata exists. + +Useful if the application needs to know the size to allocate. + +Only writing to the RIFF chunk and one data chunk is currently supported. + +See also: drwav_init_write(), drwav_init_file_write(), drwav_init_memory_write() +*/ +DRWAV_API drwav_uint64 drwav_target_write_size_bytes(const drwav_data_format* pFormat, drwav_uint64 totalFrameCount, drwav_metadata* pMetadata, drwav_uint32 metadataCount); + +/* +Take ownership of the metadata objects that were allocated via one of the init_with_metadata() function calls. The init_with_metdata functions perform a single heap allocation for this metadata. + +Useful if you want the data to persist beyond the lifetime of the drwav object. + +You must free the data returned from this function using drwav_free(). +*/ +DRWAV_API drwav_metadata* drwav_take_ownership_of_metadata(drwav* pWav); + +/* +Uninitializes the given drwav object. + +Use this only for objects initialized with drwav_init*() functions (drwav_init(), drwav_init_ex(), drwav_init_write(), drwav_init_write_sequential()). +*/ +DRWAV_API drwav_result drwav_uninit(drwav* pWav); + + +/* +Reads raw audio data. + +This is the lowest level function for reading audio data. It simply reads the given number of +bytes of the raw internal sample data. + +Consider using drwav_read_pcm_frames_s16(), drwav_read_pcm_frames_s32() or drwav_read_pcm_frames_f32() for +reading sample data in a consistent format. + +pBufferOut can be NULL in which case a seek will be performed. + +Returns the number of bytes actually read. +*/ +DRWAV_API size_t drwav_read_raw(drwav* pWav, size_t bytesToRead, void* pBufferOut); + +/* +Reads up to the specified number of PCM frames from the WAV file. + +The output data will be in the file's internal format, converted to native-endian byte order. Use +drwav_read_pcm_frames_s16/f32/s32() to read data in a specific format. + +If the return value is less than <framesToRead> it means the end of the file has been reached or +you have requested more PCM frames than can possibly fit in the output buffer. + +This function will only work when sample data is of a fixed size and uncompressed. If you are +using a compressed format consider using drwav_read_raw() or drwav_read_pcm_frames_s16/s32/f32(). + +pBufferOut can be NULL in which case a seek will be performed. +*/ +DRWAV_API drwav_uint64 drwav_read_pcm_frames(drwav* pWav, drwav_uint64 framesToRead, void* pBufferOut); +DRWAV_API drwav_uint64 drwav_read_pcm_frames_le(drwav* pWav, drwav_uint64 framesToRead, void* pBufferOut); +DRWAV_API drwav_uint64 drwav_read_pcm_frames_be(drwav* pWav, drwav_uint64 framesToRead, void* pBufferOut); + +/* +Seeks to the given PCM frame. + +Returns true if successful; false otherwise. +*/ +DRWAV_API drwav_bool32 drwav_seek_to_pcm_frame(drwav* pWav, drwav_uint64 targetFrameIndex); + +/* +Retrieves the current read position in pcm frames. +*/ +DRWAV_API drwav_result drwav_get_cursor_in_pcm_frames(drwav* pWav, drwav_uint64* pCursor); + +/* +Retrieves the length of the file. +*/ +DRWAV_API drwav_result drwav_get_length_in_pcm_frames(drwav* pWav, drwav_uint64* pLength); + + +/* +Writes raw audio data. + +Returns the number of bytes actually written. If this differs from bytesToWrite, it indicates an error. +*/ +DRWAV_API size_t drwav_write_raw(drwav* pWav, size_t bytesToWrite, const void* pData); + +/* +Writes PCM frames. + +Returns the number of PCM frames written. + +Input samples need to be in native-endian byte order. On big-endian architectures the input data will be converted to +little-endian. Use drwav_write_raw() to write raw audio data without performing any conversion. +*/ +DRWAV_API drwav_uint64 drwav_write_pcm_frames(drwav* pWav, drwav_uint64 framesToWrite, const void* pData); +DRWAV_API drwav_uint64 drwav_write_pcm_frames_le(drwav* pWav, drwav_uint64 framesToWrite, const void* pData); +DRWAV_API drwav_uint64 drwav_write_pcm_frames_be(drwav* pWav, drwav_uint64 framesToWrite, const void* pData); + +/* Conversion Utilities */ +#ifndef DR_WAV_NO_CONVERSION_API + +/* +Reads a chunk of audio data and converts it to signed 16-bit PCM samples. + +pBufferOut can be NULL in which case a seek will be performed. + +Returns the number of PCM frames actually read. + +If the return value is less than <framesToRead> it means the end of the file has been reached. +*/ +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s16(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut); +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s16le(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut); +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s16be(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut); + +/* Low-level function for converting unsigned 8-bit PCM samples to signed 16-bit PCM samples. */ +DRWAV_API void drwav_u8_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting signed 24-bit PCM samples to signed 16-bit PCM samples. */ +DRWAV_API void drwav_s24_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting signed 32-bit PCM samples to signed 16-bit PCM samples. */ +DRWAV_API void drwav_s32_to_s16(drwav_int16* pOut, const drwav_int32* pIn, size_t sampleCount); + +/* Low-level function for converting IEEE 32-bit floating point samples to signed 16-bit PCM samples. */ +DRWAV_API void drwav_f32_to_s16(drwav_int16* pOut, const float* pIn, size_t sampleCount); + +/* Low-level function for converting IEEE 64-bit floating point samples to signed 16-bit PCM samples. */ +DRWAV_API void drwav_f64_to_s16(drwav_int16* pOut, const double* pIn, size_t sampleCount); + +/* Low-level function for converting A-law samples to signed 16-bit PCM samples. */ +DRWAV_API void drwav_alaw_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting u-law samples to signed 16-bit PCM samples. */ +DRWAV_API void drwav_mulaw_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t sampleCount); + + +/* +Reads a chunk of audio data and converts it to IEEE 32-bit floating point samples. + +pBufferOut can be NULL in which case a seek will be performed. + +Returns the number of PCM frames actually read. + +If the return value is less than <framesToRead> it means the end of the file has been reached. +*/ +DRWAV_API drwav_uint64 drwav_read_pcm_frames_f32(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut); +DRWAV_API drwav_uint64 drwav_read_pcm_frames_f32le(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut); +DRWAV_API drwav_uint64 drwav_read_pcm_frames_f32be(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut); + +/* Low-level function for converting unsigned 8-bit PCM samples to IEEE 32-bit floating point samples. */ +DRWAV_API void drwav_u8_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting signed 16-bit PCM samples to IEEE 32-bit floating point samples. */ +DRWAV_API void drwav_s16_to_f32(float* pOut, const drwav_int16* pIn, size_t sampleCount); + +/* Low-level function for converting signed 24-bit PCM samples to IEEE 32-bit floating point samples. */ +DRWAV_API void drwav_s24_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting signed 32-bit PCM samples to IEEE 32-bit floating point samples. */ +DRWAV_API void drwav_s32_to_f32(float* pOut, const drwav_int32* pIn, size_t sampleCount); + +/* Low-level function for converting IEEE 64-bit floating point samples to IEEE 32-bit floating point samples. */ +DRWAV_API void drwav_f64_to_f32(float* pOut, const double* pIn, size_t sampleCount); + +/* Low-level function for converting A-law samples to IEEE 32-bit floating point samples. */ +DRWAV_API void drwav_alaw_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting u-law samples to IEEE 32-bit floating point samples. */ +DRWAV_API void drwav_mulaw_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount); + + +/* +Reads a chunk of audio data and converts it to signed 32-bit PCM samples. + +pBufferOut can be NULL in which case a seek will be performed. + +Returns the number of PCM frames actually read. + +If the return value is less than <framesToRead> it means the end of the file has been reached. +*/ +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s32(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut); +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s32le(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut); +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s32be(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut); + +/* Low-level function for converting unsigned 8-bit PCM samples to signed 32-bit PCM samples. */ +DRWAV_API void drwav_u8_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting signed 16-bit PCM samples to signed 32-bit PCM samples. */ +DRWAV_API void drwav_s16_to_s32(drwav_int32* pOut, const drwav_int16* pIn, size_t sampleCount); + +/* Low-level function for converting signed 24-bit PCM samples to signed 32-bit PCM samples. */ +DRWAV_API void drwav_s24_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting IEEE 32-bit floating point samples to signed 32-bit PCM samples. */ +DRWAV_API void drwav_f32_to_s32(drwav_int32* pOut, const float* pIn, size_t sampleCount); + +/* Low-level function for converting IEEE 64-bit floating point samples to signed 32-bit PCM samples. */ +DRWAV_API void drwav_f64_to_s32(drwav_int32* pOut, const double* pIn, size_t sampleCount); + +/* Low-level function for converting A-law samples to signed 32-bit PCM samples. */ +DRWAV_API void drwav_alaw_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t sampleCount); + +/* Low-level function for converting u-law samples to signed 32-bit PCM samples. */ +DRWAV_API void drwav_mulaw_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t sampleCount); + +#endif /* DR_WAV_NO_CONVERSION_API */ + + +/* High-Level Convenience Helpers */ + +#ifndef DR_WAV_NO_STDIO +/* +Helper for initializing a wave file for reading using stdio. + +This holds the internal FILE object until drwav_uninit() is called. Keep this in mind if you're caching drwav +objects because the operating system may restrict the number of file handles an application can have open at +any given time. +*/ +DRWAV_API drwav_bool32 drwav_init_file(drwav* pWav, const char* filename, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_ex(drwav* pWav, const char* filename, drwav_chunk_proc onChunk, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_w(drwav* pWav, const wchar_t* filename, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_ex_w(drwav* pWav, const wchar_t* filename, drwav_chunk_proc onChunk, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_with_metadata(drwav* pWav, const char* filename, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_with_metadata_w(drwav* pWav, const wchar_t* filename, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks); + + +/* +Helper for initializing a wave file for writing using stdio. + +This holds the internal FILE object until drwav_uninit() is called. Keep this in mind if you're caching drwav +objects because the operating system may restrict the number of file handles an application can have open at +any given time. +*/ +DRWAV_API drwav_bool32 drwav_init_file_write(drwav* pWav, const char* filename, const drwav_data_format* pFormat, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_write_sequential(drwav* pWav, const char* filename, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_write_sequential_pcm_frames(drwav* pWav, const char* filename, const drwav_data_format* pFormat, drwav_uint64 totalPCMFrameCount, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_write_w(drwav* pWav, const wchar_t* filename, const drwav_data_format* pFormat, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_write_sequential_w(drwav* pWav, const wchar_t* filename, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_file_write_sequential_pcm_frames_w(drwav* pWav, const wchar_t* filename, const drwav_data_format* pFormat, drwav_uint64 totalPCMFrameCount, const drwav_allocation_callbacks* pAllocationCallbacks); +#endif /* DR_WAV_NO_STDIO */ + +/* +Helper for initializing a loader from a pre-allocated memory buffer. + +This does not create a copy of the data. It is up to the application to ensure the buffer remains valid for +the lifetime of the drwav object. + +The buffer should contain the contents of the entire wave file, not just the sample data. +*/ +DRWAV_API drwav_bool32 drwav_init_memory(drwav* pWav, const void* data, size_t dataSize, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_memory_ex(drwav* pWav, const void* data, size_t dataSize, drwav_chunk_proc onChunk, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_memory_with_metadata(drwav* pWav, const void* data, size_t dataSize, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks); + +/* +Helper for initializing a writer which outputs data to a memory buffer. + +dr_wav will manage the memory allocations, however it is up to the caller to free the data with drwav_free(). + +The buffer will remain allocated even after drwav_uninit() is called. The buffer should not be considered valid +until after drwav_uninit() has been called. +*/ +DRWAV_API drwav_bool32 drwav_init_memory_write(drwav* pWav, void** ppData, size_t* pDataSize, const drwav_data_format* pFormat, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_memory_write_sequential(drwav* pWav, void** ppData, size_t* pDataSize, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_bool32 drwav_init_memory_write_sequential_pcm_frames(drwav* pWav, void** ppData, size_t* pDataSize, const drwav_data_format* pFormat, drwav_uint64 totalPCMFrameCount, const drwav_allocation_callbacks* pAllocationCallbacks); + + +#ifndef DR_WAV_NO_CONVERSION_API +/* +Opens and reads an entire wav file in a single operation. + +The return value is a heap-allocated buffer containing the audio data. Use drwav_free() to free the buffer. +*/ +DRWAV_API drwav_int16* drwav_open_and_read_pcm_frames_s16(drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API float* drwav_open_and_read_pcm_frames_f32(drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_int32* drwav_open_and_read_pcm_frames_s32(drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +#ifndef DR_WAV_NO_STDIO +/* +Opens and decodes an entire wav file in a single operation. + +The return value is a heap-allocated buffer containing the audio data. Use drwav_free() to free the buffer. +*/ +DRWAV_API drwav_int16* drwav_open_file_and_read_pcm_frames_s16(const char* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API float* drwav_open_file_and_read_pcm_frames_f32(const char* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_int32* drwav_open_file_and_read_pcm_frames_s32(const char* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_int16* drwav_open_file_and_read_pcm_frames_s16_w(const wchar_t* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API float* drwav_open_file_and_read_pcm_frames_f32_w(const wchar_t* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_int32* drwav_open_file_and_read_pcm_frames_s32_w(const wchar_t* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +#endif +/* +Opens and decodes an entire wav file from a block of memory in a single operation. + +The return value is a heap-allocated buffer containing the audio data. Use drwav_free() to free the buffer. +*/ +DRWAV_API drwav_int16* drwav_open_memory_and_read_pcm_frames_s16(const void* data, size_t dataSize, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API float* drwav_open_memory_and_read_pcm_frames_f32(const void* data, size_t dataSize, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +DRWAV_API drwav_int32* drwav_open_memory_and_read_pcm_frames_s32(const void* data, size_t dataSize, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks); +#endif + +/* Frees data that was allocated internally by dr_wav. */ +DRWAV_API void drwav_free(void* p, const drwav_allocation_callbacks* pAllocationCallbacks); + +/* Converts bytes from a wav stream to a sized type of native endian. */ +DRWAV_API drwav_uint16 drwav_bytes_to_u16(const drwav_uint8* data); +DRWAV_API drwav_int16 drwav_bytes_to_s16(const drwav_uint8* data); +DRWAV_API drwav_uint32 drwav_bytes_to_u32(const drwav_uint8* data); +DRWAV_API drwav_int32 drwav_bytes_to_s32(const drwav_uint8* data); +DRWAV_API drwav_uint64 drwav_bytes_to_u64(const drwav_uint8* data); +DRWAV_API drwav_int64 drwav_bytes_to_s64(const drwav_uint8* data); +DRWAV_API float drwav_bytes_to_f32(const drwav_uint8* data); + +/* Compares a GUID for the purpose of checking the type of a Wave64 chunk. */ +DRWAV_API drwav_bool32 drwav_guid_equal(const drwav_uint8 a[16], const drwav_uint8 b[16]); + +/* Compares a four-character-code for the purpose of checking the type of a RIFF chunk. */ +DRWAV_API drwav_bool32 drwav_fourcc_equal(const drwav_uint8* a, const char* b); + +#ifdef __cplusplus +} +#endif +#endif /* dr_wav_h */ + + +/************************************************************************************************************************************************************ + ************************************************************************************************************************************************************ + + IMPLEMENTATION + + ************************************************************************************************************************************************************ + ************************************************************************************************************************************************************/ +#if defined(DR_WAV_IMPLEMENTATION) || defined(DRWAV_IMPLEMENTATION) +#ifndef dr_wav_c +#define dr_wav_c + +#ifdef __MRC__ +/* MrC currently doesn't compile dr_wav correctly with any optimizations enabled. */ +#pragma options opt off +#endif + +#include <stdlib.h> +#include <string.h> +#include <limits.h> /* For INT_MAX */ + +#ifndef DR_WAV_NO_STDIO +#include <stdio.h> +#ifndef DR_WAV_NO_WCHAR +#include <wchar.h> +#endif +#endif + +/* Standard library stuff. */ +#ifndef DRWAV_ASSERT +#include <assert.h> +#define DRWAV_ASSERT(expression) assert(expression) +#endif +#ifndef DRWAV_MALLOC +#define DRWAV_MALLOC(sz) malloc((sz)) +#endif +#ifndef DRWAV_REALLOC +#define DRWAV_REALLOC(p, sz) realloc((p), (sz)) +#endif +#ifndef DRWAV_FREE +#define DRWAV_FREE(p) free((p)) +#endif +#ifndef DRWAV_COPY_MEMORY +#define DRWAV_COPY_MEMORY(dst, src, sz) memcpy((dst), (src), (sz)) +#endif +#ifndef DRWAV_ZERO_MEMORY +#define DRWAV_ZERO_MEMORY(p, sz) memset((p), 0, (sz)) +#endif +#ifndef DRWAV_ZERO_OBJECT +#define DRWAV_ZERO_OBJECT(p) DRWAV_ZERO_MEMORY((p), sizeof(*p)) +#endif + +#define drwav_countof(x) (sizeof(x) / sizeof(x[0])) +#define drwav_align(x, a) ((((x) + (a) - 1) / (a)) * (a)) +#define drwav_min(a, b) (((a) < (b)) ? (a) : (b)) +#define drwav_max(a, b) (((a) > (b)) ? (a) : (b)) +#define drwav_clamp(x, lo, hi) (drwav_max((lo), drwav_min((hi), (x)))) +#define drwav_offset_ptr(p, offset) (((drwav_uint8*)(p)) + (offset)) + +#define DRWAV_MAX_SIMD_VECTOR_SIZE 32 + +/* Architecture Detection */ +#if defined(__x86_64__) || (defined(_M_X64) && !defined(_M_ARM64EC)) + #define DRWAV_X64 +#elif defined(__i386) || defined(_M_IX86) + #define DRWAV_X86 +#elif defined(__arm__) || defined(_M_ARM) + #define DRWAV_ARM +#endif +/* End Architecture Detection */ + +/* Inline */ +#ifdef _MSC_VER + #define DRWAV_INLINE __forceinline +#elif defined(__GNUC__) + /* + I've had a bug report where GCC is emitting warnings about functions possibly not being inlineable. This warning happens when + the __attribute__((always_inline)) attribute is defined without an "inline" statement. I think therefore there must be some + case where "__inline__" is not always defined, thus the compiler emitting these warnings. When using -std=c89 or -ansi on the + command line, we cannot use the "inline" keyword and instead need to use "__inline__". In an attempt to work around this issue + I am using "__inline__" only when we're compiling in strict ANSI mode. + */ + #if defined(__STRICT_ANSI__) + #define DRWAV_GNUC_INLINE_HINT __inline__ + #else + #define DRWAV_GNUC_INLINE_HINT inline + #endif + + #if (__GNUC__ > 3 || (__GNUC__ == 3 && __GNUC_MINOR__ >= 2)) || defined(__clang__) + #define DRWAV_INLINE DRWAV_GNUC_INLINE_HINT __attribute__((always_inline)) + #else + #define DRWAV_INLINE DRWAV_GNUC_INLINE_HINT + #endif +#elif defined(__WATCOMC__) + #define DRWAV_INLINE __inline +#else + #define DRWAV_INLINE +#endif +/* End Inline */ + +/* SIZE_MAX */ +#if defined(SIZE_MAX) + #define DRWAV_SIZE_MAX SIZE_MAX +#else + #if defined(_WIN64) || defined(_LP64) || defined(__LP64__) + #define DRWAV_SIZE_MAX ((drwav_uint64)0xFFFFFFFFFFFFFFFF) + #else + #define DRWAV_SIZE_MAX 0xFFFFFFFF + #endif +#endif +/* End SIZE_MAX */ + +/* Weird bit manipulation is for C89 compatibility (no direct support for 64-bit integers). */ +#define DRWAV_INT64_MIN ((drwav_int64) ((drwav_uint64)0x80000000 << 32)) +#define DRWAV_INT64_MAX ((drwav_int64)(((drwav_uint64)0x7FFFFFFF << 32) | 0xFFFFFFFF)) + +#if defined(_MSC_VER) && _MSC_VER >= 1400 + #define DRWAV_HAS_BYTESWAP16_INTRINSIC + #define DRWAV_HAS_BYTESWAP32_INTRINSIC + #define DRWAV_HAS_BYTESWAP64_INTRINSIC +#elif defined(__clang__) + #if defined(__has_builtin) + #if __has_builtin(__builtin_bswap16) + #define DRWAV_HAS_BYTESWAP16_INTRINSIC + #endif + #if __has_builtin(__builtin_bswap32) + #define DRWAV_HAS_BYTESWAP32_INTRINSIC + #endif + #if __has_builtin(__builtin_bswap64) + #define DRWAV_HAS_BYTESWAP64_INTRINSIC + #endif + #endif +#elif defined(__GNUC__) + #if ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3)) + #define DRWAV_HAS_BYTESWAP32_INTRINSIC + #define DRWAV_HAS_BYTESWAP64_INTRINSIC + #endif + #if ((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) + #define DRWAV_HAS_BYTESWAP16_INTRINSIC + #endif +#endif + +DRWAV_API void drwav_version(drwav_uint32* pMajor, drwav_uint32* pMinor, drwav_uint32* pRevision) +{ + if (pMajor) { + *pMajor = DRWAV_VERSION_MAJOR; + } + + if (pMinor) { + *pMinor = DRWAV_VERSION_MINOR; + } + + if (pRevision) { + *pRevision = DRWAV_VERSION_REVISION; + } +} + +DRWAV_API const char* drwav_version_string(void) +{ + return DRWAV_VERSION_STRING; +} + +/* +These limits are used for basic validation when initializing the decoder. If you exceed these limits, first of all: what on Earth are +you doing?! (Let me know, I'd be curious!) Second, you can adjust these by #define-ing them before the dr_wav implementation. +*/ +#ifndef DRWAV_MAX_SAMPLE_RATE +#define DRWAV_MAX_SAMPLE_RATE 384000 +#endif +#ifndef DRWAV_MAX_CHANNELS +#define DRWAV_MAX_CHANNELS 256 +#endif +#ifndef DRWAV_MAX_BITS_PER_SAMPLE +#define DRWAV_MAX_BITS_PER_SAMPLE 64 +#endif + +static const drwav_uint8 drwavGUID_W64_RIFF[16] = {0x72,0x69,0x66,0x66, 0x2E,0x91, 0xCF,0x11, 0xA5,0xD6, 0x28,0xDB,0x04,0xC1,0x00,0x00}; /* 66666972-912E-11CF-A5D6-28DB04C10000 */ +static const drwav_uint8 drwavGUID_W64_WAVE[16] = {0x77,0x61,0x76,0x65, 0xF3,0xAC, 0xD3,0x11, 0x8C,0xD1, 0x00,0xC0,0x4F,0x8E,0xDB,0x8A}; /* 65766177-ACF3-11D3-8CD1-00C04F8EDB8A */ +/*static const drwav_uint8 drwavGUID_W64_JUNK[16] = {0x6A,0x75,0x6E,0x6B, 0xF3,0xAC, 0xD3,0x11, 0x8C,0xD1, 0x00,0xC0,0x4F,0x8E,0xDB,0x8A};*/ /* 6B6E756A-ACF3-11D3-8CD1-00C04F8EDB8A */ +static const drwav_uint8 drwavGUID_W64_FMT [16] = {0x66,0x6D,0x74,0x20, 0xF3,0xAC, 0xD3,0x11, 0x8C,0xD1, 0x00,0xC0,0x4F,0x8E,0xDB,0x8A}; /* 20746D66-ACF3-11D3-8CD1-00C04F8EDB8A */ +static const drwav_uint8 drwavGUID_W64_FACT[16] = {0x66,0x61,0x63,0x74, 0xF3,0xAC, 0xD3,0x11, 0x8C,0xD1, 0x00,0xC0,0x4F,0x8E,0xDB,0x8A}; /* 74636166-ACF3-11D3-8CD1-00C04F8EDB8A */ +static const drwav_uint8 drwavGUID_W64_DATA[16] = {0x64,0x61,0x74,0x61, 0xF3,0xAC, 0xD3,0x11, 0x8C,0xD1, 0x00,0xC0,0x4F,0x8E,0xDB,0x8A}; /* 61746164-ACF3-11D3-8CD1-00C04F8EDB8A */ +/*static const drwav_uint8 drwavGUID_W64_SMPL[16] = {0x73,0x6D,0x70,0x6C, 0xF3,0xAC, 0xD3,0x11, 0x8C,0xD1, 0x00,0xC0,0x4F,0x8E,0xDB,0x8A};*/ /* 6C706D73-ACF3-11D3-8CD1-00C04F8EDB8A */ + + +static DRWAV_INLINE int drwav__is_little_endian(void) +{ +#if defined(DRWAV_X86) || defined(DRWAV_X64) + return DRWAV_TRUE; +#elif defined(__BYTE_ORDER) && defined(__LITTLE_ENDIAN) && __BYTE_ORDER == __LITTLE_ENDIAN + return DRWAV_TRUE; +#else + int n = 1; + return (*(char*)&n) == 1; +#endif +} + + +static DRWAV_INLINE void drwav_bytes_to_guid(const drwav_uint8* data, drwav_uint8* guid) +{ + int i; + for (i = 0; i < 16; ++i) { + guid[i] = data[i]; + } +} + + +static DRWAV_INLINE drwav_uint16 drwav__bswap16(drwav_uint16 n) +{ +#ifdef DRWAV_HAS_BYTESWAP16_INTRINSIC + #if defined(_MSC_VER) + return _byteswap_ushort(n); + #elif defined(__GNUC__) || defined(__clang__) + return __builtin_bswap16(n); + #else + #error "This compiler does not support the byte swap intrinsic." + #endif +#else + return ((n & 0xFF00) >> 8) | + ((n & 0x00FF) << 8); +#endif +} + +static DRWAV_INLINE drwav_uint32 drwav__bswap32(drwav_uint32 n) +{ +#ifdef DRWAV_HAS_BYTESWAP32_INTRINSIC + #if defined(_MSC_VER) + return _byteswap_ulong(n); + #elif defined(__GNUC__) || defined(__clang__) + #if defined(DRWAV_ARM) && (defined(__ARM_ARCH) && __ARM_ARCH >= 6) && !defined(DRWAV_64BIT) /* <-- 64-bit inline assembly has not been tested, so disabling for now. */ + /* Inline assembly optimized implementation for ARM. In my testing, GCC does not generate optimized code with __builtin_bswap32(). */ + drwav_uint32 r; + __asm__ __volatile__ ( + #if defined(DRWAV_64BIT) + "rev %w[out], %w[in]" : [out]"=r"(r) : [in]"r"(n) /* <-- This is untested. If someone in the community could test this, that would be appreciated! */ + #else + "rev %[out], %[in]" : [out]"=r"(r) : [in]"r"(n) + #endif + ); + return r; + #else + return __builtin_bswap32(n); + #endif + #else + #error "This compiler does not support the byte swap intrinsic." + #endif +#else + return ((n & 0xFF000000) >> 24) | + ((n & 0x00FF0000) >> 8) | + ((n & 0x0000FF00) << 8) | + ((n & 0x000000FF) << 24); +#endif +} + +static DRWAV_INLINE drwav_uint64 drwav__bswap64(drwav_uint64 n) +{ +#ifdef DRWAV_HAS_BYTESWAP64_INTRINSIC + #if defined(_MSC_VER) + return _byteswap_uint64(n); + #elif defined(__GNUC__) || defined(__clang__) + return __builtin_bswap64(n); + #else + #error "This compiler does not support the byte swap intrinsic." + #endif +#else + /* Weird "<< 32" bitshift is required for C89 because it doesn't support 64-bit constants. Should be optimized out by a good compiler. */ + return ((n & ((drwav_uint64)0xFF000000 << 32)) >> 56) | + ((n & ((drwav_uint64)0x00FF0000 << 32)) >> 40) | + ((n & ((drwav_uint64)0x0000FF00 << 32)) >> 24) | + ((n & ((drwav_uint64)0x000000FF << 32)) >> 8) | + ((n & ((drwav_uint64)0xFF000000 )) << 8) | + ((n & ((drwav_uint64)0x00FF0000 )) << 24) | + ((n & ((drwav_uint64)0x0000FF00 )) << 40) | + ((n & ((drwav_uint64)0x000000FF )) << 56); +#endif +} + + +static DRWAV_INLINE drwav_int16 drwav__bswap_s16(drwav_int16 n) +{ + return (drwav_int16)drwav__bswap16((drwav_uint16)n); +} + +static DRWAV_INLINE void drwav__bswap_samples_s16(drwav_int16* pSamples, drwav_uint64 sampleCount) +{ + drwav_uint64 iSample; + for (iSample = 0; iSample < sampleCount; iSample += 1) { + pSamples[iSample] = drwav__bswap_s16(pSamples[iSample]); + } +} + + +static DRWAV_INLINE void drwav__bswap_s24(drwav_uint8* p) +{ + drwav_uint8 t; + t = p[0]; + p[0] = p[2]; + p[2] = t; +} + +static DRWAV_INLINE void drwav__bswap_samples_s24(drwav_uint8* pSamples, drwav_uint64 sampleCount) +{ + drwav_uint64 iSample; + for (iSample = 0; iSample < sampleCount; iSample += 1) { + drwav_uint8* pSample = pSamples + (iSample*3); + drwav__bswap_s24(pSample); + } +} + + +static DRWAV_INLINE drwav_int32 drwav__bswap_s32(drwav_int32 n) +{ + return (drwav_int32)drwav__bswap32((drwav_uint32)n); +} + +static DRWAV_INLINE void drwav__bswap_samples_s32(drwav_int32* pSamples, drwav_uint64 sampleCount) +{ + drwav_uint64 iSample; + for (iSample = 0; iSample < sampleCount; iSample += 1) { + pSamples[iSample] = drwav__bswap_s32(pSamples[iSample]); + } +} + + +static DRWAV_INLINE drwav_int64 drwav__bswap_s64(drwav_int64 n) +{ + return (drwav_int64)drwav__bswap64((drwav_uint64)n); +} + +static DRWAV_INLINE void drwav__bswap_samples_s64(drwav_int64* pSamples, drwav_uint64 sampleCount) +{ + drwav_uint64 iSample; + for (iSample = 0; iSample < sampleCount; iSample += 1) { + pSamples[iSample] = drwav__bswap_s64(pSamples[iSample]); + } +} + + +static DRWAV_INLINE float drwav__bswap_f32(float n) +{ + union { + drwav_uint32 i; + float f; + } x; + x.f = n; + x.i = drwav__bswap32(x.i); + + return x.f; +} + +static DRWAV_INLINE void drwav__bswap_samples_f32(float* pSamples, drwav_uint64 sampleCount) +{ + drwav_uint64 iSample; + for (iSample = 0; iSample < sampleCount; iSample += 1) { + pSamples[iSample] = drwav__bswap_f32(pSamples[iSample]); + } +} + + +static DRWAV_INLINE void drwav__bswap_samples(void* pSamples, drwav_uint64 sampleCount, drwav_uint32 bytesPerSample) +{ + switch (bytesPerSample) + { + case 1: + { + /* No-op. */ + } break; + case 2: + { + drwav__bswap_samples_s16((drwav_int16*)pSamples, sampleCount); + } break; + case 3: + { + drwav__bswap_samples_s24((drwav_uint8*)pSamples, sampleCount); + } break; + case 4: + { + drwav__bswap_samples_s32((drwav_int32*)pSamples, sampleCount); + } break; + case 8: + { + drwav__bswap_samples_s64((drwav_int64*)pSamples, sampleCount); + } break; + default: + { + drwav_uint64 iSample; + + for (iSample = 0; iSample < sampleCount; iSample += 1) { + drwav_uint8* pSample = (drwav_uint8*)pSamples + (iSample * bytesPerSample); + drwav_uint32 iByte; + + for (iByte = 0; iByte < bytesPerSample / 2; iByte += 1) { + drwav_uint8 temp = pSample[iByte]; + pSample[iByte] = pSample[bytesPerSample - iByte - 1]; + pSample[bytesPerSample - iByte - 1] = temp; + } + } + } break; + } +} + + + +DRWAV_PRIVATE DRWAV_INLINE drwav_bool32 drwav_is_container_be(drwav_container container) +{ + if (container == drwav_container_rifx || container == drwav_container_aiff) { + return DRWAV_TRUE; + } else { + return DRWAV_FALSE; + } +} + + +DRWAV_PRIVATE DRWAV_INLINE drwav_uint16 drwav_bytes_to_u16_le(const drwav_uint8* data) +{ + return ((drwav_uint16)data[0] << 0) | ((drwav_uint16)data[1] << 8); +} + +DRWAV_PRIVATE DRWAV_INLINE drwav_uint16 drwav_bytes_to_u16_be(const drwav_uint8* data) +{ + return ((drwav_uint16)data[1] << 0) | ((drwav_uint16)data[0] << 8); +} + +DRWAV_PRIVATE DRWAV_INLINE drwav_uint16 drwav_bytes_to_u16_ex(const drwav_uint8* data, drwav_container container) +{ + if (drwav_is_container_be(container)) { + return drwav_bytes_to_u16_be(data); + } else { + return drwav_bytes_to_u16_le(data); + } +} + + +DRWAV_PRIVATE DRWAV_INLINE drwav_uint32 drwav_bytes_to_u32_le(const drwav_uint8* data) +{ + return ((drwav_uint32)data[0] << 0) | ((drwav_uint32)data[1] << 8) | ((drwav_uint32)data[2] << 16) | ((drwav_uint32)data[3] << 24); +} + +DRWAV_PRIVATE DRWAV_INLINE drwav_uint32 drwav_bytes_to_u32_be(const drwav_uint8* data) +{ + return ((drwav_uint32)data[3] << 0) | ((drwav_uint32)data[2] << 8) | ((drwav_uint32)data[1] << 16) | ((drwav_uint32)data[0] << 24); +} + +DRWAV_PRIVATE DRWAV_INLINE drwav_uint32 drwav_bytes_to_u32_ex(const drwav_uint8* data, drwav_container container) +{ + if (drwav_is_container_be(container)) { + return drwav_bytes_to_u32_be(data); + } else { + return drwav_bytes_to_u32_le(data); + } +} + + + +DRWAV_PRIVATE drwav_int64 drwav_aiff_extented_to_s64(const drwav_uint8* data) +{ + drwav_uint32 exponent = ((drwav_uint32)data[0] << 8) | data[1]; + drwav_uint64 hi = ((drwav_uint64)data[2] << 24) | ((drwav_uint64)data[3] << 16) | ((drwav_uint64)data[4] << 8) | ((drwav_uint64)data[5] << 0); + drwav_uint64 lo = ((drwav_uint64)data[6] << 24) | ((drwav_uint64)data[7] << 16) | ((drwav_uint64)data[8] << 8) | ((drwav_uint64)data[9] << 0); + drwav_uint64 significand = (hi << 32) | lo; + int sign = exponent >> 15; + + /* Remove sign bit. */ + exponent &= 0x7FFF; + + /* Special cases. */ + if (exponent == 0 && significand == 0) { + return 0; + } else if (exponent == 0x7FFF) { + return sign ? DRWAV_INT64_MIN : DRWAV_INT64_MAX; /* Infinite. */ + } + + exponent -= 16383; + + if (exponent > 63) { + return sign ? DRWAV_INT64_MIN : DRWAV_INT64_MAX; /* Too big for a 64-bit integer. */ + } else if (exponent < 1) { + return 0; /* Number is less than 1, so rounds down to 0. */ + } + + significand >>= (63 - exponent); + + if (sign) { + return -(drwav_int64)significand; + } else { + return (drwav_int64)significand; + } +} + + +DRWAV_PRIVATE void* drwav__malloc_default(size_t sz, void* pUserData) +{ + (void)pUserData; + return DRWAV_MALLOC(sz); +} + +DRWAV_PRIVATE void* drwav__realloc_default(void* p, size_t sz, void* pUserData) +{ + (void)pUserData; + return DRWAV_REALLOC(p, sz); +} + +DRWAV_PRIVATE void drwav__free_default(void* p, void* pUserData) +{ + (void)pUserData; + DRWAV_FREE(p); +} + + +DRWAV_PRIVATE void* drwav__malloc_from_callbacks(size_t sz, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pAllocationCallbacks == NULL) { + return NULL; + } + + if (pAllocationCallbacks->onMalloc != NULL) { + return pAllocationCallbacks->onMalloc(sz, pAllocationCallbacks->pUserData); + } + + /* Try using realloc(). */ + if (pAllocationCallbacks->onRealloc != NULL) { + return pAllocationCallbacks->onRealloc(NULL, sz, pAllocationCallbacks->pUserData); + } + + return NULL; +} + +DRWAV_PRIVATE void* drwav__realloc_from_callbacks(void* p, size_t szNew, size_t szOld, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pAllocationCallbacks == NULL) { + return NULL; + } + + if (pAllocationCallbacks->onRealloc != NULL) { + return pAllocationCallbacks->onRealloc(p, szNew, pAllocationCallbacks->pUserData); + } + + /* Try emulating realloc() in terms of malloc()/free(). */ + if (pAllocationCallbacks->onMalloc != NULL && pAllocationCallbacks->onFree != NULL) { + void* p2; + + p2 = pAllocationCallbacks->onMalloc(szNew, pAllocationCallbacks->pUserData); + if (p2 == NULL) { + return NULL; + } + + if (p != NULL) { + DRWAV_COPY_MEMORY(p2, p, szOld); + pAllocationCallbacks->onFree(p, pAllocationCallbacks->pUserData); + } + + return p2; + } + + return NULL; +} + +DRWAV_PRIVATE void drwav__free_from_callbacks(void* p, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (p == NULL || pAllocationCallbacks == NULL) { + return; + } + + if (pAllocationCallbacks->onFree != NULL) { + pAllocationCallbacks->onFree(p, pAllocationCallbacks->pUserData); + } +} + + +DRWAV_PRIVATE drwav_allocation_callbacks drwav_copy_allocation_callbacks_or_defaults(const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pAllocationCallbacks != NULL) { + /* Copy. */ + return *pAllocationCallbacks; + } else { + /* Defaults. */ + drwav_allocation_callbacks allocationCallbacks; + allocationCallbacks.pUserData = NULL; + allocationCallbacks.onMalloc = drwav__malloc_default; + allocationCallbacks.onRealloc = drwav__realloc_default; + allocationCallbacks.onFree = drwav__free_default; + return allocationCallbacks; + } +} + + +static DRWAV_INLINE drwav_bool32 drwav__is_compressed_format_tag(drwav_uint16 formatTag) +{ + return + formatTag == DR_WAVE_FORMAT_ADPCM || + formatTag == DR_WAVE_FORMAT_DVI_ADPCM; +} + +DRWAV_PRIVATE unsigned int drwav__chunk_padding_size_riff(drwav_uint64 chunkSize) +{ + return (unsigned int)(chunkSize % 2); +} + +DRWAV_PRIVATE unsigned int drwav__chunk_padding_size_w64(drwav_uint64 chunkSize) +{ + return (unsigned int)(chunkSize % 8); +} + +DRWAV_PRIVATE unsigned int drwav_calculate_padding_size(drwav_container container, drwav_uint64 chunkSize) +{ + if (container == drwav_container_riff || container == drwav_container_rf64) { + return drwav__chunk_padding_size_riff(chunkSize); + } else { + return drwav__chunk_padding_size_w64(chunkSize); + } +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s16__msadpcm(drwav* pWav, drwav_uint64 samplesToRead, drwav_int16* pBufferOut); +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s16__ima(drwav* pWav, drwav_uint64 samplesToRead, drwav_int16* pBufferOut); +DRWAV_PRIVATE drwav_bool32 drwav_init_write__internal(drwav* pWav, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount); + +DRWAV_PRIVATE drwav_result drwav__read_chunk_header(drwav_read_proc onRead, void* pUserData, drwav_container container, drwav_uint64* pRunningBytesReadOut, drwav_chunk_header* pHeaderOut) +{ + if (container == drwav_container_riff || container == drwav_container_rifx || container == drwav_container_rf64 || container == drwav_container_aiff) { + drwav_uint8 sizeInBytes[4]; + + if (onRead(pUserData, pHeaderOut->id.fourcc, 4) != 4) { + return DRWAV_AT_END; + } + + if (onRead(pUserData, sizeInBytes, 4) != 4) { + return DRWAV_INVALID_FILE; + } + + pHeaderOut->sizeInBytes = drwav_bytes_to_u32_ex(sizeInBytes, container); + pHeaderOut->paddingSize = drwav__chunk_padding_size_riff(pHeaderOut->sizeInBytes); + + *pRunningBytesReadOut += 8; + } else if (container == drwav_container_w64) { + drwav_uint8 sizeInBytes[8]; + + if (onRead(pUserData, pHeaderOut->id.guid, 16) != 16) { + return DRWAV_AT_END; + } + + if (onRead(pUserData, sizeInBytes, 8) != 8) { + return DRWAV_INVALID_FILE; + } + + pHeaderOut->sizeInBytes = drwav_bytes_to_u64(sizeInBytes); + + /* Subtract 24 from the size because with w64 the reported chunk size includes the size of the header itself. */ + if (pHeaderOut->sizeInBytes >= 24) { + pHeaderOut->sizeInBytes -= 24; + } else { + return DRWAV_INVALID_FILE; + } + + pHeaderOut->paddingSize = drwav__chunk_padding_size_w64(pHeaderOut->sizeInBytes); + *pRunningBytesReadOut += 24; + } else { + return DRWAV_INVALID_FILE; + } + + return DRWAV_SUCCESS; +} + +DRWAV_PRIVATE drwav_bool32 drwav__seek_forward(drwav_seek_proc onSeek, drwav_uint64 offset, void* pUserData) +{ + drwav_uint64 bytesRemainingToSeek = offset; + while (bytesRemainingToSeek > 0) { + if (bytesRemainingToSeek > 0x7FFFFFFF) { + if (!onSeek(pUserData, 0x7FFFFFFF, DRWAV_SEEK_CUR)) { + return DRWAV_FALSE; + } + bytesRemainingToSeek -= 0x7FFFFFFF; + } else { + if (!onSeek(pUserData, (int)bytesRemainingToSeek, DRWAV_SEEK_CUR)) { + return DRWAV_FALSE; + } + bytesRemainingToSeek = 0; + } + } + + return DRWAV_TRUE; +} + +DRWAV_PRIVATE drwav_bool32 drwav__seek_from_start(drwav_seek_proc onSeek, drwav_uint64 offset, void* pUserData) +{ + if (offset <= 0x7FFFFFFF) { + return onSeek(pUserData, (int)offset, DRWAV_SEEK_SET); + } + + /* Larger than 32-bit seek. */ + if (!onSeek(pUserData, 0x7FFFFFFF, DRWAV_SEEK_SET)) { + return DRWAV_FALSE; + } + offset -= 0x7FFFFFFF; + + for (;;) { + if (offset <= 0x7FFFFFFF) { + return onSeek(pUserData, (int)offset, DRWAV_SEEK_CUR); + } + + if (!onSeek(pUserData, 0x7FFFFFFF, DRWAV_SEEK_CUR)) { + return DRWAV_FALSE; + } + offset -= 0x7FFFFFFF; + } + + /* Should never get here. */ + /*return DRWAV_TRUE; */ +} + + + +DRWAV_PRIVATE size_t drwav__on_read(drwav_read_proc onRead, void* pUserData, void* pBufferOut, size_t bytesToRead, drwav_uint64* pCursor) +{ + size_t bytesRead; + + DRWAV_ASSERT(onRead != NULL); + DRWAV_ASSERT(pCursor != NULL); + + bytesRead = onRead(pUserData, pBufferOut, bytesToRead); + *pCursor += bytesRead; + return bytesRead; +} + +#if 0 +DRWAV_PRIVATE drwav_bool32 drwav__on_seek(drwav_seek_proc onSeek, void* pUserData, int offset, drwav_seek_origin origin, drwav_uint64* pCursor) +{ + DRWAV_ASSERT(onSeek != NULL); + DRWAV_ASSERT(pCursor != NULL); + + if (!onSeek(pUserData, offset, origin)) { + return DRWAV_FALSE; + } + + if (origin == DRWAV_SEEK_SET) { + *pCursor = offset; + } else { + *pCursor += offset; + } + + return DRWAV_TRUE; +} +#endif + + +#define DRWAV_SMPL_BYTES 36 +#define DRWAV_SMPL_LOOP_BYTES 24 +#define DRWAV_INST_BYTES 7 +#define DRWAV_ACID_BYTES 24 +#define DRWAV_CUE_BYTES 4 +#define DRWAV_BEXT_BYTES 602 +#define DRWAV_BEXT_DESCRIPTION_BYTES 256 +#define DRWAV_BEXT_ORIGINATOR_NAME_BYTES 32 +#define DRWAV_BEXT_ORIGINATOR_REF_BYTES 32 +#define DRWAV_BEXT_RESERVED_BYTES 180 +#define DRWAV_BEXT_UMID_BYTES 64 +#define DRWAV_CUE_POINT_BYTES 24 +#define DRWAV_LIST_LABEL_OR_NOTE_BYTES 4 +#define DRWAV_LIST_LABELLED_TEXT_BYTES 20 + +#define DRWAV_METADATA_ALIGNMENT 8 + +typedef enum +{ + drwav__metadata_parser_stage_count, + drwav__metadata_parser_stage_read +} drwav__metadata_parser_stage; + +typedef struct +{ + drwav_read_proc onRead; + drwav_seek_proc onSeek; + void *pReadSeekUserData; + drwav__metadata_parser_stage stage; + drwav_metadata *pMetadata; + drwav_uint32 metadataCount; + drwav_uint8 *pData; + drwav_uint8 *pDataCursor; + drwav_uint64 metadataCursor; + drwav_uint64 extraCapacity; +} drwav__metadata_parser; + +DRWAV_PRIVATE size_t drwav__metadata_memory_capacity(drwav__metadata_parser* pParser) +{ + drwav_uint64 cap = sizeof(drwav_metadata) * (drwav_uint64)pParser->metadataCount + pParser->extraCapacity; + if (cap > DRWAV_SIZE_MAX) { + return 0; /* Too big. */ + } + + return (size_t)cap; /* Safe cast thanks to the check above. */ +} + +DRWAV_PRIVATE drwav_uint8* drwav__metadata_get_memory(drwav__metadata_parser* pParser, size_t size, size_t align) +{ + drwav_uint8* pResult; + + if (align) { + drwav_uintptr modulo = (drwav_uintptr)pParser->pDataCursor % align; + if (modulo != 0) { + pParser->pDataCursor += align - modulo; + } + } + + pResult = pParser->pDataCursor; + + /* + Getting to the point where this function is called means there should always be memory + available. Out of memory checks should have been done at an earlier stage. + */ + DRWAV_ASSERT((pResult + size) <= (pParser->pData + drwav__metadata_memory_capacity(pParser))); + + pParser->pDataCursor += size; + return pResult; +} + +DRWAV_PRIVATE void drwav__metadata_request_extra_memory_for_stage_2(drwav__metadata_parser* pParser, size_t bytes, size_t align) +{ + size_t extra = bytes + (align ? (align - 1) : 0); + pParser->extraCapacity += extra; +} + +DRWAV_PRIVATE drwav_result drwav__metadata_alloc(drwav__metadata_parser* pParser, drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pParser->extraCapacity != 0 || pParser->metadataCount != 0) { + pAllocationCallbacks->onFree(pParser->pData, pAllocationCallbacks->pUserData); + + pParser->pData = (drwav_uint8*)pAllocationCallbacks->onMalloc(drwav__metadata_memory_capacity(pParser), pAllocationCallbacks->pUserData); + pParser->pDataCursor = pParser->pData; + + if (pParser->pData == NULL) { + return DRWAV_OUT_OF_MEMORY; + } + + /* + We don't need to worry about specifying an alignment here because malloc always returns something + of suitable alignment. This also means pParser->pMetadata is all that we need to store in order + for us to free when we are done. + */ + pParser->pMetadata = (drwav_metadata*)drwav__metadata_get_memory(pParser, sizeof(drwav_metadata) * pParser->metadataCount, 1); + pParser->metadataCursor = 0; + } + + return DRWAV_SUCCESS; +} + +DRWAV_PRIVATE size_t drwav__metadata_parser_read(drwav__metadata_parser* pParser, void* pBufferOut, size_t bytesToRead, drwav_uint64* pCursor) +{ + if (pCursor != NULL) { + return drwav__on_read(pParser->onRead, pParser->pReadSeekUserData, pBufferOut, bytesToRead, pCursor); + } else { + return pParser->onRead(pParser->pReadSeekUserData, pBufferOut, bytesToRead); + } +} + +DRWAV_PRIVATE drwav_bool32 drwav__metadata_validate_smpl_chunk(const drwav_chunk_header* pChunkHeader, drwav_uint32 loopCount, drwav_uint32 samplerSpecificDataSizeInBytes, drwav_uint64* pTrailingDataSizeInBytes) +{ + drwav_uint64 remainingDataSizeInBytes; + + DRWAV_ASSERT(pChunkHeader != NULL); + + if (pChunkHeader->sizeInBytes < DRWAV_SMPL_BYTES) { + return DRWAV_FALSE; + } + + remainingDataSizeInBytes = pChunkHeader->sizeInBytes - DRWAV_SMPL_BYTES; + if ((drwav_uint64)loopCount > remainingDataSizeInBytes / DRWAV_SMPL_LOOP_BYTES) { + return DRWAV_FALSE; + } + + remainingDataSizeInBytes -= (drwav_uint64)loopCount * DRWAV_SMPL_LOOP_BYTES; + if ((drwav_uint64)samplerSpecificDataSizeInBytes > remainingDataSizeInBytes) { + return DRWAV_FALSE; + } + + if (pTrailingDataSizeInBytes != NULL) { + *pTrailingDataSizeInBytes = remainingDataSizeInBytes - samplerSpecificDataSizeInBytes; + } + + return DRWAV_TRUE; +} + +DRWAV_PRIVATE drwav_uint64 drwav__read_smpl_to_metadata_obj(drwav__metadata_parser* pParser, const drwav_chunk_header* pChunkHeader, drwav_metadata* pMetadata) +{ + drwav_uint8 smplHeaderData[DRWAV_SMPL_BYTES]; + drwav_uint64 totalBytesRead = 0; + size_t bytesJustRead; + + if (pMetadata == NULL) { + return 0; + } + + bytesJustRead = drwav__metadata_parser_read(pParser, smplHeaderData, sizeof(smplHeaderData), &totalBytesRead); + + DRWAV_ASSERT(pParser->stage == drwav__metadata_parser_stage_read); + DRWAV_ASSERT(pChunkHeader != NULL); + + if (pMetadata != NULL && bytesJustRead == sizeof(smplHeaderData)) { + drwav_uint32 iSampleLoop; + drwav_uint32 loopCount; + drwav_uint32 samplerSpecificDataSizeInBytes; + drwav_uint64 trailingDataSizeInBytes; + + /* + When we calculated the amount of memory required for the "smpl" chunk we excluded the chunk entirely + if its loop or sampler-specific data exceeded the chunk size. When this happens, the second stage will + still hit this path but the `pMetadata` will either point at the end of the allocation or at another + chunk. We need to repeat the validation before dereferencing the pMetadata object. + */ + loopCount = drwav_bytes_to_u32(smplHeaderData + 28); + samplerSpecificDataSizeInBytes = drwav_bytes_to_u32(smplHeaderData + 32); + if (!drwav__metadata_validate_smpl_chunk(pChunkHeader, loopCount, samplerSpecificDataSizeInBytes, &trailingDataSizeInBytes)) { + return totalBytesRead; + } + + pMetadata->type = drwav_metadata_type_smpl; + pMetadata->data.smpl.manufacturerId = drwav_bytes_to_u32(smplHeaderData + 0); + pMetadata->data.smpl.productId = drwav_bytes_to_u32(smplHeaderData + 4); + pMetadata->data.smpl.samplePeriodNanoseconds = drwav_bytes_to_u32(smplHeaderData + 8); + pMetadata->data.smpl.midiUnityNote = drwav_bytes_to_u32(smplHeaderData + 12); + pMetadata->data.smpl.midiPitchFraction = drwav_bytes_to_u32(smplHeaderData + 16); + pMetadata->data.smpl.smpteFormat = drwav_bytes_to_u32(smplHeaderData + 20); + pMetadata->data.smpl.smpteOffset = drwav_bytes_to_u32(smplHeaderData + 24); + pMetadata->data.smpl.sampleLoopCount = loopCount; + pMetadata->data.smpl.samplerSpecificDataSizeInBytes = samplerSpecificDataSizeInBytes; + + pMetadata->data.smpl.pLoops = (drwav_smpl_loop*)drwav__metadata_get_memory(pParser, sizeof(drwav_smpl_loop) * pMetadata->data.smpl.sampleLoopCount, DRWAV_METADATA_ALIGNMENT); + + for (iSampleLoop = 0; iSampleLoop < pMetadata->data.smpl.sampleLoopCount; ++iSampleLoop) { + drwav_uint8 smplLoopData[DRWAV_SMPL_LOOP_BYTES]; + bytesJustRead = drwav__metadata_parser_read(pParser, smplLoopData, sizeof(smplLoopData), &totalBytesRead); + + if (bytesJustRead == sizeof(smplLoopData)) { + pMetadata->data.smpl.pLoops[iSampleLoop].cuePointId = drwav_bytes_to_u32(smplLoopData + 0); + pMetadata->data.smpl.pLoops[iSampleLoop].type = drwav_bytes_to_u32(smplLoopData + 4); + pMetadata->data.smpl.pLoops[iSampleLoop].firstSampleOffset = drwav_bytes_to_u32(smplLoopData + 8); + pMetadata->data.smpl.pLoops[iSampleLoop].lastSampleOffset = drwav_bytes_to_u32(smplLoopData + 12); + pMetadata->data.smpl.pLoops[iSampleLoop].sampleFraction = drwav_bytes_to_u32(smplLoopData + 16); + pMetadata->data.smpl.pLoops[iSampleLoop].playCount = drwav_bytes_to_u32(smplLoopData + 20); + } else { + return totalBytesRead; + } + } + + if (pMetadata->data.smpl.samplerSpecificDataSizeInBytes > 0) { + pMetadata->data.smpl.pSamplerSpecificData = drwav__metadata_get_memory(pParser, pMetadata->data.smpl.samplerSpecificDataSizeInBytes, 1); + DRWAV_ASSERT(pMetadata->data.smpl.pSamplerSpecificData != NULL); + + bytesJustRead = drwav__metadata_parser_read(pParser, pMetadata->data.smpl.pSamplerSpecificData, pMetadata->data.smpl.samplerSpecificDataSizeInBytes, &totalBytesRead); + if (bytesJustRead != pMetadata->data.smpl.samplerSpecificDataSizeInBytes) { + return totalBytesRead; + } + } + + if (trailingDataSizeInBytes > 0) { + if (!drwav__seek_forward(pParser->onSeek, trailingDataSizeInBytes, pParser->pReadSeekUserData)) { + return totalBytesRead; + } + totalBytesRead += trailingDataSizeInBytes; + } + } + + return totalBytesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav__read_cue_to_metadata_obj(drwav__metadata_parser* pParser, const drwav_chunk_header* pChunkHeader, drwav_metadata* pMetadata) +{ + drwav_uint8 cueHeaderSectionData[DRWAV_CUE_BYTES]; + drwav_uint64 totalBytesRead = 0; + size_t bytesJustRead; + + if (pMetadata == NULL) { + return 0; + } + + bytesJustRead = drwav__metadata_parser_read(pParser, cueHeaderSectionData, sizeof(cueHeaderSectionData), &totalBytesRead); + + DRWAV_ASSERT(pParser->stage == drwav__metadata_parser_stage_read); + + if (bytesJustRead == sizeof(cueHeaderSectionData)) { + pMetadata->type = drwav_metadata_type_cue; + pMetadata->data.cue.cuePointCount = drwav_bytes_to_u32(cueHeaderSectionData); + + /* + We need to validate the cue point count against the size of the chunk so we don't read + beyond the chunk. + */ + if (pMetadata->data.cue.cuePointCount == (pChunkHeader->sizeInBytes - DRWAV_CUE_BYTES) / DRWAV_CUE_POINT_BYTES) { + pMetadata->data.cue.pCuePoints = (drwav_cue_point*)drwav__metadata_get_memory(pParser, sizeof(drwav_cue_point) * pMetadata->data.cue.cuePointCount, DRWAV_METADATA_ALIGNMENT); + DRWAV_ASSERT(pMetadata->data.cue.pCuePoints != NULL); + + if (pMetadata->data.cue.cuePointCount > 0) { + drwav_uint32 iCuePoint; + + for (iCuePoint = 0; iCuePoint < pMetadata->data.cue.cuePointCount; ++iCuePoint) { + drwav_uint8 cuePointData[DRWAV_CUE_POINT_BYTES]; + bytesJustRead = drwav__metadata_parser_read(pParser, cuePointData, sizeof(cuePointData), &totalBytesRead); + + if (bytesJustRead == sizeof(cuePointData)) { + pMetadata->data.cue.pCuePoints[iCuePoint].id = drwav_bytes_to_u32(cuePointData + 0); + pMetadata->data.cue.pCuePoints[iCuePoint].playOrderPosition = drwav_bytes_to_u32(cuePointData + 4); + pMetadata->data.cue.pCuePoints[iCuePoint].dataChunkId[0] = cuePointData[8]; + pMetadata->data.cue.pCuePoints[iCuePoint].dataChunkId[1] = cuePointData[9]; + pMetadata->data.cue.pCuePoints[iCuePoint].dataChunkId[2] = cuePointData[10]; + pMetadata->data.cue.pCuePoints[iCuePoint].dataChunkId[3] = cuePointData[11]; + pMetadata->data.cue.pCuePoints[iCuePoint].chunkStart = drwav_bytes_to_u32(cuePointData + 12); + pMetadata->data.cue.pCuePoints[iCuePoint].blockStart = drwav_bytes_to_u32(cuePointData + 16); + pMetadata->data.cue.pCuePoints[iCuePoint].sampleOffset = drwav_bytes_to_u32(cuePointData + 20); + } else { + break; + } + } + } + } + } + + return totalBytesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav__read_inst_to_metadata_obj(drwav__metadata_parser* pParser, drwav_metadata* pMetadata) +{ + drwav_uint8 instData[DRWAV_INST_BYTES]; + drwav_uint64 bytesRead; + + if (pMetadata == NULL) { + return 0; + } + + bytesRead = drwav__metadata_parser_read(pParser, instData, sizeof(instData), NULL); + + DRWAV_ASSERT(pParser->stage == drwav__metadata_parser_stage_read); + + if (bytesRead == sizeof(instData)) { + pMetadata->type = drwav_metadata_type_inst; + pMetadata->data.inst.midiUnityNote = (drwav_int8)instData[0]; + pMetadata->data.inst.fineTuneCents = (drwav_int8)instData[1]; + pMetadata->data.inst.gainDecibels = (drwav_int8)instData[2]; + pMetadata->data.inst.lowNote = (drwav_int8)instData[3]; + pMetadata->data.inst.highNote = (drwav_int8)instData[4]; + pMetadata->data.inst.lowVelocity = (drwav_int8)instData[5]; + pMetadata->data.inst.highVelocity = (drwav_int8)instData[6]; + } + + return bytesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav__read_acid_to_metadata_obj(drwav__metadata_parser* pParser, drwav_metadata* pMetadata) +{ + drwav_uint8 acidData[DRWAV_ACID_BYTES]; + drwav_uint64 bytesRead; + + if (pMetadata == NULL) { + return 0; + } + + bytesRead = drwav__metadata_parser_read(pParser, acidData, sizeof(acidData), NULL); + + DRWAV_ASSERT(pParser->stage == drwav__metadata_parser_stage_read); + + if (bytesRead == sizeof(acidData)) { + pMetadata->type = drwav_metadata_type_acid; + pMetadata->data.acid.flags = drwav_bytes_to_u32(acidData + 0); + pMetadata->data.acid.midiUnityNote = drwav_bytes_to_u16(acidData + 4); + pMetadata->data.acid.reserved1 = drwav_bytes_to_u16(acidData + 6); + pMetadata->data.acid.reserved2 = drwav_bytes_to_f32(acidData + 8); + pMetadata->data.acid.numBeats = drwav_bytes_to_u32(acidData + 12); + pMetadata->data.acid.meterDenominator = drwav_bytes_to_u16(acidData + 16); + pMetadata->data.acid.meterNumerator = drwav_bytes_to_u16(acidData + 18); + pMetadata->data.acid.tempo = drwav_bytes_to_f32(acidData + 20); + } + + return bytesRead; +} + +DRWAV_PRIVATE size_t drwav__strlen(const char* str) +{ + size_t result = 0; + + while (*str++) { + result += 1; + } + + return result; +} + +DRWAV_PRIVATE size_t drwav__strlen_clamped(const char* str, size_t maxToRead) +{ + size_t result = 0; + + while (*str++ && result < maxToRead) { + result += 1; + } + + return result; +} + +DRWAV_PRIVATE char* drwav__metadata_copy_string(drwav__metadata_parser* pParser, const char* str, size_t maxToRead) +{ + size_t len = drwav__strlen_clamped(str, maxToRead); + + if (len) { + char* result = (char*)drwav__metadata_get_memory(pParser, len + 1, 1); + DRWAV_ASSERT(result != NULL); + + DRWAV_COPY_MEMORY(result, str, len); + result[len] = '\0'; + + return result; + } else { + return NULL; + } +} + +typedef struct +{ + const void* pBuffer; + size_t sizeInBytes; + size_t cursor; +} drwav_buffer_reader; + +DRWAV_PRIVATE drwav_result drwav_buffer_reader_init(const void* pBuffer, size_t sizeInBytes, drwav_buffer_reader* pReader) +{ + DRWAV_ASSERT(pBuffer != NULL); + DRWAV_ASSERT(pReader != NULL); + + DRWAV_ZERO_OBJECT(pReader); + + pReader->pBuffer = pBuffer; + pReader->sizeInBytes = sizeInBytes; + pReader->cursor = 0; + + return DRWAV_SUCCESS; +} + +DRWAV_PRIVATE const void* drwav_buffer_reader_ptr(const drwav_buffer_reader* pReader) +{ + DRWAV_ASSERT(pReader != NULL); + + return drwav_offset_ptr(pReader->pBuffer, pReader->cursor); +} + +DRWAV_PRIVATE drwav_result drwav_buffer_reader_seek(drwav_buffer_reader* pReader, size_t bytesToSeek) +{ + DRWAV_ASSERT(pReader != NULL); + + if (pReader->cursor + bytesToSeek > pReader->sizeInBytes) { + return DRWAV_BAD_SEEK; /* Seeking too far forward. */ + } + + pReader->cursor += bytesToSeek; + + return DRWAV_SUCCESS; +} + +DRWAV_PRIVATE drwav_result drwav_buffer_reader_read(drwav_buffer_reader* pReader, void* pDst, size_t bytesToRead, size_t* pBytesRead) +{ + drwav_result result = DRWAV_SUCCESS; + size_t bytesRemaining; + + DRWAV_ASSERT(pReader != NULL); + + if (pBytesRead != NULL) { + *pBytesRead = 0; + } + + bytesRemaining = (pReader->sizeInBytes - pReader->cursor); + if (bytesToRead > bytesRemaining) { + bytesToRead = bytesRemaining; + } + + if (pDst == NULL) { + /* Seek. */ + result = drwav_buffer_reader_seek(pReader, bytesToRead); + } else { + /* Read. */ + DRWAV_COPY_MEMORY(pDst, drwav_buffer_reader_ptr(pReader), bytesToRead); + pReader->cursor += bytesToRead; + } + + DRWAV_ASSERT(pReader->cursor <= pReader->sizeInBytes); + + if (result == DRWAV_SUCCESS) { + if (pBytesRead != NULL) { + *pBytesRead = bytesToRead; + } + } + + return DRWAV_SUCCESS; +} + +DRWAV_PRIVATE drwav_result drwav_buffer_reader_read_u16(drwav_buffer_reader* pReader, drwav_uint16* pDst) +{ + drwav_result result; + size_t bytesRead; + drwav_uint8 data[2]; + + DRWAV_ASSERT(pReader != NULL); + DRWAV_ASSERT(pDst != NULL); + + *pDst = 0; /* Safety. */ + + result = drwav_buffer_reader_read(pReader, data, sizeof(*pDst), &bytesRead); + if (result != DRWAV_SUCCESS || bytesRead != sizeof(*pDst)) { + return result; + } + + *pDst = drwav_bytes_to_u16(data); + + return DRWAV_SUCCESS; +} + +DRWAV_PRIVATE drwav_result drwav_buffer_reader_read_u32(drwav_buffer_reader* pReader, drwav_uint32* pDst) +{ + drwav_result result; + size_t bytesRead; + drwav_uint8 data[4]; + + DRWAV_ASSERT(pReader != NULL); + DRWAV_ASSERT(pDst != NULL); + + *pDst = 0; /* Safety. */ + + result = drwav_buffer_reader_read(pReader, data, sizeof(*pDst), &bytesRead); + if (result != DRWAV_SUCCESS || bytesRead != sizeof(*pDst)) { + return result; + } + + *pDst = drwav_bytes_to_u32(data); + + return DRWAV_SUCCESS; +} + + + +DRWAV_PRIVATE drwav_uint64 drwav__read_bext_to_metadata_obj(drwav__metadata_parser* pParser, drwav_metadata* pMetadata, drwav_uint64 chunkSize) +{ + drwav_uint8 bextData[DRWAV_BEXT_BYTES]; + size_t bytesRead = drwav__metadata_parser_read(pParser, bextData, sizeof(bextData), NULL); + + DRWAV_ASSERT(pParser->stage == drwav__metadata_parser_stage_read); + + if (bytesRead == sizeof(bextData)) { + drwav_buffer_reader reader; + drwav_uint32 timeReferenceLow; + drwav_uint32 timeReferenceHigh; + size_t extraBytes; + + pMetadata->type = drwav_metadata_type_bext; + + if (drwav_buffer_reader_init(bextData, bytesRead, &reader) == DRWAV_SUCCESS) { + pMetadata->data.bext.pDescription = drwav__metadata_copy_string(pParser, (const char*)drwav_buffer_reader_ptr(&reader), DRWAV_BEXT_DESCRIPTION_BYTES); + drwav_buffer_reader_seek(&reader, DRWAV_BEXT_DESCRIPTION_BYTES); + + pMetadata->data.bext.pOriginatorName = drwav__metadata_copy_string(pParser, (const char*)drwav_buffer_reader_ptr(&reader), DRWAV_BEXT_ORIGINATOR_NAME_BYTES); + drwav_buffer_reader_seek(&reader, DRWAV_BEXT_ORIGINATOR_NAME_BYTES); + + pMetadata->data.bext.pOriginatorReference = drwav__metadata_copy_string(pParser, (const char*)drwav_buffer_reader_ptr(&reader), DRWAV_BEXT_ORIGINATOR_REF_BYTES); + drwav_buffer_reader_seek(&reader, DRWAV_BEXT_ORIGINATOR_REF_BYTES); + + drwav_buffer_reader_read(&reader, pMetadata->data.bext.pOriginationDate, sizeof(pMetadata->data.bext.pOriginationDate), NULL); + drwav_buffer_reader_read(&reader, pMetadata->data.bext.pOriginationTime, sizeof(pMetadata->data.bext.pOriginationTime), NULL); + + drwav_buffer_reader_read_u32(&reader, &timeReferenceLow); + drwav_buffer_reader_read_u32(&reader, &timeReferenceHigh); + pMetadata->data.bext.timeReference = ((drwav_uint64)timeReferenceHigh << 32) + timeReferenceLow; + + drwav_buffer_reader_read_u16(&reader, &pMetadata->data.bext.version); + + pMetadata->data.bext.pUMID = drwav__metadata_get_memory(pParser, DRWAV_BEXT_UMID_BYTES, 1); + drwav_buffer_reader_read(&reader, pMetadata->data.bext.pUMID, DRWAV_BEXT_UMID_BYTES, NULL); + + drwav_buffer_reader_read_u16(&reader, &pMetadata->data.bext.loudnessValue); + drwav_buffer_reader_read_u16(&reader, &pMetadata->data.bext.loudnessRange); + drwav_buffer_reader_read_u16(&reader, &pMetadata->data.bext.maxTruePeakLevel); + drwav_buffer_reader_read_u16(&reader, &pMetadata->data.bext.maxMomentaryLoudness); + drwav_buffer_reader_read_u16(&reader, &pMetadata->data.bext.maxShortTermLoudness); + + DRWAV_ASSERT((drwav_offset_ptr(drwav_buffer_reader_ptr(&reader), DRWAV_BEXT_RESERVED_BYTES)) == (bextData + DRWAV_BEXT_BYTES)); + + extraBytes = (size_t)(chunkSize - DRWAV_BEXT_BYTES); + if (extraBytes > 0) { + pMetadata->data.bext.pCodingHistory = (char*)drwav__metadata_get_memory(pParser, extraBytes + 1, 1); + DRWAV_ASSERT(pMetadata->data.bext.pCodingHistory != NULL); + + pMetadata->data.bext.codingHistorySize = (drwav_uint32)drwav__metadata_parser_read(pParser, pMetadata->data.bext.pCodingHistory, extraBytes, NULL); + pMetadata->data.bext.pCodingHistory[pMetadata->data.bext.codingHistorySize] = '\0'; /* <-- Explicit null terminator in case of a badly formed file. */ + + bytesRead += pMetadata->data.bext.codingHistorySize; + } else { + pMetadata->data.bext.pCodingHistory = NULL; + pMetadata->data.bext.codingHistorySize = 0; + } + } + } + + return bytesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav__read_list_label_or_note_to_metadata_obj(drwav__metadata_parser* pParser, drwav_metadata* pMetadata, drwav_uint64 chunkSize, drwav_metadata_type type) +{ + drwav_uint8 cueIDBuffer[DRWAV_LIST_LABEL_OR_NOTE_BYTES]; + drwav_uint64 totalBytesRead = 0; + size_t bytesJustRead = drwav__metadata_parser_read(pParser, cueIDBuffer, sizeof(cueIDBuffer), &totalBytesRead); + + DRWAV_ASSERT(pParser->stage == drwav__metadata_parser_stage_read); + + if (bytesJustRead == sizeof(cueIDBuffer)) { + drwav_uint32 sizeIncludingNullTerminator; + + pMetadata->type = type; + pMetadata->data.labelOrNote.cuePointId = drwav_bytes_to_u32(cueIDBuffer); + + sizeIncludingNullTerminator = (drwav_uint32)chunkSize - DRWAV_LIST_LABEL_OR_NOTE_BYTES; + if (sizeIncludingNullTerminator > 0) { + pMetadata->data.labelOrNote.stringLength = sizeIncludingNullTerminator - 1; + pMetadata->data.labelOrNote.pString = (char*)drwav__metadata_get_memory(pParser, sizeIncludingNullTerminator, 1); + DRWAV_ASSERT(pMetadata->data.labelOrNote.pString != NULL); + + drwav__metadata_parser_read(pParser, pMetadata->data.labelOrNote.pString, sizeIncludingNullTerminator, &totalBytesRead); + } else { + pMetadata->data.labelOrNote.stringLength = 0; + pMetadata->data.labelOrNote.pString = NULL; + } + } + + return totalBytesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav__read_list_labelled_cue_region_to_metadata_obj(drwav__metadata_parser* pParser, drwav_metadata* pMetadata, drwav_uint64 chunkSize) +{ + drwav_uint8 buffer[DRWAV_LIST_LABELLED_TEXT_BYTES]; + drwav_uint64 totalBytesRead = 0; + size_t bytesJustRead = drwav__metadata_parser_read(pParser, buffer, sizeof(buffer), &totalBytesRead); + + DRWAV_ASSERT(pParser->stage == drwav__metadata_parser_stage_read); + + if (bytesJustRead == sizeof(buffer)) { + drwav_uint32 sizeIncludingNullTerminator; + + pMetadata->type = drwav_metadata_type_list_labelled_cue_region; + pMetadata->data.labelledCueRegion.cuePointId = drwav_bytes_to_u32(buffer + 0); + pMetadata->data.labelledCueRegion.sampleLength = drwav_bytes_to_u32(buffer + 4); + pMetadata->data.labelledCueRegion.purposeId[0] = buffer[8]; + pMetadata->data.labelledCueRegion.purposeId[1] = buffer[9]; + pMetadata->data.labelledCueRegion.purposeId[2] = buffer[10]; + pMetadata->data.labelledCueRegion.purposeId[3] = buffer[11]; + pMetadata->data.labelledCueRegion.country = drwav_bytes_to_u16(buffer + 12); + pMetadata->data.labelledCueRegion.language = drwav_bytes_to_u16(buffer + 14); + pMetadata->data.labelledCueRegion.dialect = drwav_bytes_to_u16(buffer + 16); + pMetadata->data.labelledCueRegion.codePage = drwav_bytes_to_u16(buffer + 18); + + sizeIncludingNullTerminator = (drwav_uint32)chunkSize - DRWAV_LIST_LABELLED_TEXT_BYTES; + if (sizeIncludingNullTerminator > 0) { + pMetadata->data.labelledCueRegion.stringLength = sizeIncludingNullTerminator - 1; + pMetadata->data.labelledCueRegion.pString = (char*)drwav__metadata_get_memory(pParser, sizeIncludingNullTerminator, 1); + DRWAV_ASSERT(pMetadata->data.labelledCueRegion.pString != NULL); + + drwav__metadata_parser_read(pParser, pMetadata->data.labelledCueRegion.pString, sizeIncludingNullTerminator, &totalBytesRead); + } else { + pMetadata->data.labelledCueRegion.stringLength = 0; + pMetadata->data.labelledCueRegion.pString = NULL; + } + } + + return totalBytesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav__metadata_process_info_text_chunk(drwav__metadata_parser* pParser, drwav_uint64 chunkSize, drwav_metadata_type type) +{ + drwav_uint64 bytesRead = 0; + drwav_uint32 stringSizeWithNullTerminator = (drwav_uint32)chunkSize; + + if (pParser->stage == drwav__metadata_parser_stage_count) { + pParser->metadataCount += 1; + drwav__metadata_request_extra_memory_for_stage_2(pParser, stringSizeWithNullTerminator, 1); + } else { + drwav_metadata* pMetadata = &pParser->pMetadata[pParser->metadataCursor]; + pMetadata->type = type; + if (stringSizeWithNullTerminator > 0) { + pMetadata->data.infoText.stringLength = stringSizeWithNullTerminator - 1; + pMetadata->data.infoText.pString = (char*)drwav__metadata_get_memory(pParser, stringSizeWithNullTerminator, 1); + DRWAV_ASSERT(pMetadata->data.infoText.pString != NULL); + + bytesRead = drwav__metadata_parser_read(pParser, pMetadata->data.infoText.pString, (size_t)stringSizeWithNullTerminator, NULL); + if (bytesRead == chunkSize) { + pParser->metadataCursor += 1; + } else { + /* Failed to parse. */ + } + } else { + pMetadata->data.infoText.stringLength = 0; + pMetadata->data.infoText.pString = NULL; + pParser->metadataCursor += 1; + } + } + + return bytesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav__metadata_process_unknown_chunk(drwav__metadata_parser* pParser, const drwav_uint8* pChunkId, drwav_uint64 chunkSize, drwav_metadata_location location) +{ + drwav_uint64 bytesRead = 0; + + if (location == drwav_metadata_location_invalid) { + return 0; + } + + if (drwav_fourcc_equal(pChunkId, "data") || drwav_fourcc_equal(pChunkId, "fmt ") || drwav_fourcc_equal(pChunkId, "fact")) { + return 0; + } + + if (pParser->stage == drwav__metadata_parser_stage_count) { + pParser->metadataCount += 1; + drwav__metadata_request_extra_memory_for_stage_2(pParser, (size_t)chunkSize, 1); + } else { + drwav_metadata* pMetadata = &pParser->pMetadata[pParser->metadataCursor]; + pMetadata->type = drwav_metadata_type_unknown; + pMetadata->data.unknown.chunkLocation = location; + pMetadata->data.unknown.id[0] = pChunkId[0]; + pMetadata->data.unknown.id[1] = pChunkId[1]; + pMetadata->data.unknown.id[2] = pChunkId[2]; + pMetadata->data.unknown.id[3] = pChunkId[3]; + pMetadata->data.unknown.dataSizeInBytes = (drwav_uint32)chunkSize; + pMetadata->data.unknown.pData = (drwav_uint8 *)drwav__metadata_get_memory(pParser, (size_t)chunkSize, 1); + DRWAV_ASSERT(pMetadata->data.unknown.pData != NULL); + + bytesRead = drwav__metadata_parser_read(pParser, pMetadata->data.unknown.pData, pMetadata->data.unknown.dataSizeInBytes, NULL); + if (bytesRead == pMetadata->data.unknown.dataSizeInBytes) { + pParser->metadataCursor += 1; + } else { + /* Failed to read. */ + } + } + + return bytesRead; +} + +DRWAV_PRIVATE drwav_bool32 drwav__chunk_matches(drwav_metadata_type allowedMetadataTypes, const drwav_uint8* pChunkID, drwav_metadata_type type, const char* pID) +{ + return (allowedMetadataTypes & type) && drwav_fourcc_equal(pChunkID, pID); +} + +DRWAV_PRIVATE drwav_uint64 drwav__metadata_process_chunk(drwav__metadata_parser* pParser, const drwav_chunk_header* pChunkHeader, drwav_metadata_type allowedMetadataTypes) +{ + const drwav_uint8 *pChunkID = pChunkHeader->id.fourcc; + drwav_uint64 bytesRead = 0; + + if (drwav__chunk_matches(allowedMetadataTypes, pChunkID, drwav_metadata_type_smpl, "smpl")) { + if (pChunkHeader->sizeInBytes >= DRWAV_SMPL_BYTES) { + if (pParser->stage == drwav__metadata_parser_stage_count) { + drwav_uint8 buffer[4]; + size_t bytesJustRead; + + if (!pParser->onSeek(pParser->pReadSeekUserData, 28, DRWAV_SEEK_CUR)) { + return bytesRead; + } + bytesRead += 28; + + bytesJustRead = drwav__metadata_parser_read(pParser, buffer, sizeof(buffer), &bytesRead); + if (bytesJustRead == sizeof(buffer)) { + drwav_uint32 loopCount = drwav_bytes_to_u32(buffer); + + bytesJustRead = drwav__metadata_parser_read(pParser, buffer, sizeof(buffer), &bytesRead); + if (bytesJustRead == sizeof(buffer)) { + drwav_uint32 samplerSpecificDataSizeInBytes = drwav_bytes_to_u32(buffer); + + if (drwav__metadata_validate_smpl_chunk(pChunkHeader, loopCount, samplerSpecificDataSizeInBytes, NULL)) { + pParser->metadataCount += 1; + drwav__metadata_request_extra_memory_for_stage_2(pParser, sizeof(drwav_smpl_loop) * loopCount, DRWAV_METADATA_ALIGNMENT); + drwav__metadata_request_extra_memory_for_stage_2(pParser, samplerSpecificDataSizeInBytes, 1); + } else { + /* Incorrectly formed chunk. Loop or sampler-specific data exceeds the size of the chunk. */ + } + } + } + } else { + bytesRead = drwav__read_smpl_to_metadata_obj(pParser, pChunkHeader, &pParser->pMetadata[pParser->metadataCursor]); + if (bytesRead == pChunkHeader->sizeInBytes) { + pParser->metadataCursor += 1; + } else { + /* Failed to parse. */ + } + } + } else { + /* Incorrectly formed chunk. */ + } + } else if (drwav__chunk_matches(allowedMetadataTypes, pChunkID, drwav_metadata_type_inst, "inst")) { + if (pChunkHeader->sizeInBytes == DRWAV_INST_BYTES) { + if (pParser->stage == drwav__metadata_parser_stage_count) { + pParser->metadataCount += 1; + } else { + bytesRead = drwav__read_inst_to_metadata_obj(pParser, &pParser->pMetadata[pParser->metadataCursor]); + if (bytesRead == pChunkHeader->sizeInBytes) { + pParser->metadataCursor += 1; + } else { + /* Failed to parse. */ + } + } + } else { + /* Incorrectly formed chunk. */ + } + } else if (drwav__chunk_matches(allowedMetadataTypes, pChunkID, drwav_metadata_type_acid, "acid")) { + if (pChunkHeader->sizeInBytes == DRWAV_ACID_BYTES) { + if (pParser->stage == drwav__metadata_parser_stage_count) { + pParser->metadataCount += 1; + } else { + bytesRead = drwav__read_acid_to_metadata_obj(pParser, &pParser->pMetadata[pParser->metadataCursor]); + if (bytesRead == pChunkHeader->sizeInBytes) { + pParser->metadataCursor += 1; + } else { + /* Failed to parse. */ + } + } + } else { + /* Incorrectly formed chunk. */ + } + } else if (drwav__chunk_matches(allowedMetadataTypes, pChunkID, drwav_metadata_type_cue, "cue ")) { + if (pChunkHeader->sizeInBytes >= DRWAV_CUE_BYTES) { + if (pParser->stage == drwav__metadata_parser_stage_count) { + size_t cueCount; + + pParser->metadataCount += 1; + cueCount = (size_t)(pChunkHeader->sizeInBytes - DRWAV_CUE_BYTES) / DRWAV_CUE_POINT_BYTES; + drwav__metadata_request_extra_memory_for_stage_2(pParser, sizeof(drwav_cue_point) * cueCount, DRWAV_METADATA_ALIGNMENT); + } else { + bytesRead = drwav__read_cue_to_metadata_obj(pParser, pChunkHeader, &pParser->pMetadata[pParser->metadataCursor]); + if (bytesRead == pChunkHeader->sizeInBytes) { + pParser->metadataCursor += 1; + } else { + /* Failed to parse. */ + } + } + } else { + /* Incorrectly formed chunk. */ + } + } else if (drwav__chunk_matches(allowedMetadataTypes, pChunkID, drwav_metadata_type_bext, "bext")) { + if (pChunkHeader->sizeInBytes >= DRWAV_BEXT_BYTES) { + if (pParser->stage == drwav__metadata_parser_stage_count) { + /* The description field is the largest one in a bext chunk, so that is the max size of this temporary buffer. */ + char buffer[DRWAV_BEXT_DESCRIPTION_BYTES + 1]; + size_t allocSizeNeeded = DRWAV_BEXT_UMID_BYTES; /* We know we will need SMPTE umid size. */ + size_t bytesJustRead; + + buffer[DRWAV_BEXT_DESCRIPTION_BYTES] = '\0'; + bytesJustRead = drwav__metadata_parser_read(pParser, buffer, DRWAV_BEXT_DESCRIPTION_BYTES, &bytesRead); + if (bytesJustRead != DRWAV_BEXT_DESCRIPTION_BYTES) { + return bytesRead; + } + allocSizeNeeded += drwav__strlen(buffer) + 1; + + buffer[DRWAV_BEXT_ORIGINATOR_NAME_BYTES] = '\0'; + bytesJustRead = drwav__metadata_parser_read(pParser, buffer, DRWAV_BEXT_ORIGINATOR_NAME_BYTES, &bytesRead); + if (bytesJustRead != DRWAV_BEXT_ORIGINATOR_NAME_BYTES) { + return bytesRead; + } + allocSizeNeeded += drwav__strlen(buffer) + 1; + + buffer[DRWAV_BEXT_ORIGINATOR_REF_BYTES] = '\0'; + bytesJustRead = drwav__metadata_parser_read(pParser, buffer, DRWAV_BEXT_ORIGINATOR_REF_BYTES, &bytesRead); + if (bytesJustRead != DRWAV_BEXT_ORIGINATOR_REF_BYTES) { + return bytesRead; + } + allocSizeNeeded += drwav__strlen(buffer) + 1; + + /* Coding history. */ + allocSizeNeeded += (size_t)pChunkHeader->sizeInBytes - DRWAV_BEXT_BYTES + 1; + + drwav__metadata_request_extra_memory_for_stage_2(pParser, allocSizeNeeded, 1); + + pParser->metadataCount += 1; + } else { + bytesRead = drwav__read_bext_to_metadata_obj(pParser, &pParser->pMetadata[pParser->metadataCursor], pChunkHeader->sizeInBytes); + if (bytesRead == pChunkHeader->sizeInBytes) { + pParser->metadataCursor += 1; + } else { + /* Failed to parse. */ + } + } + } else { + /* Incorrectly formed chunk. */ + } + } else if (drwav_fourcc_equal(pChunkID, "LIST") || drwav_fourcc_equal(pChunkID, "list")) { + drwav_metadata_location listType = drwav_metadata_location_invalid; + while (bytesRead < pChunkHeader->sizeInBytes) { + drwav_uint8 subchunkId[4]; + drwav_uint8 subchunkSizeBuffer[4]; + drwav_uint64 subchunkDataSize; + drwav_uint64 subchunkBytesRead = 0; + drwav_uint64 bytesJustRead = drwav__metadata_parser_read(pParser, subchunkId, sizeof(subchunkId), &bytesRead); + if (bytesJustRead != sizeof(subchunkId)) { + break; + } + + /* + The first thing in a list chunk should be "adtl" or "INFO". + + - adtl means this list is a Associated Data List Chunk and will contain labels, notes + or labelled cue regions. + - INFO means this list is an Info List Chunk containing info text chunks such as IPRD + which would specifies the album of this wav file. + + No data follows the adtl or INFO id so we just make note of what type this list is and + continue. + */ + if (drwav_fourcc_equal(subchunkId, "adtl")) { + listType = drwav_metadata_location_inside_adtl_list; + continue; + } else if (drwav_fourcc_equal(subchunkId, "INFO")) { + listType = drwav_metadata_location_inside_info_list; + continue; + } + + bytesJustRead = drwav__metadata_parser_read(pParser, subchunkSizeBuffer, sizeof(subchunkSizeBuffer), &bytesRead); + if (bytesJustRead != sizeof(subchunkSizeBuffer)) { + break; + } + subchunkDataSize = drwav_bytes_to_u32(subchunkSizeBuffer); + + if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_label, "labl") || drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_note, "note")) { + if (subchunkDataSize >= DRWAV_LIST_LABEL_OR_NOTE_BYTES) { + drwav_uint64 stringSizeWithNullTerm = subchunkDataSize - DRWAV_LIST_LABEL_OR_NOTE_BYTES; + if (pParser->stage == drwav__metadata_parser_stage_count) { + pParser->metadataCount += 1; + drwav__metadata_request_extra_memory_for_stage_2(pParser, (size_t)stringSizeWithNullTerm, 1); + } else { + subchunkBytesRead = drwav__read_list_label_or_note_to_metadata_obj(pParser, &pParser->pMetadata[pParser->metadataCursor], subchunkDataSize, drwav_fourcc_equal(subchunkId, "labl") ? drwav_metadata_type_list_label : drwav_metadata_type_list_note); + if (subchunkBytesRead == subchunkDataSize) { + pParser->metadataCursor += 1; + } else { + /* Failed to parse. */ + } + } + } else { + /* Incorrectly formed chunk. */ + } + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_labelled_cue_region, "ltxt")) { + if (subchunkDataSize >= DRWAV_LIST_LABELLED_TEXT_BYTES) { + drwav_uint64 stringSizeWithNullTerminator = subchunkDataSize - DRWAV_LIST_LABELLED_TEXT_BYTES; + if (pParser->stage == drwav__metadata_parser_stage_count) { + pParser->metadataCount += 1; + drwav__metadata_request_extra_memory_for_stage_2(pParser, (size_t)stringSizeWithNullTerminator, 1); + } else { + subchunkBytesRead = drwav__read_list_labelled_cue_region_to_metadata_obj(pParser, &pParser->pMetadata[pParser->metadataCursor], subchunkDataSize); + if (subchunkBytesRead == subchunkDataSize) { + pParser->metadataCursor += 1; + } else { + /* Failed to parse. */ + } + } + } else { + /* Incorrectly formed chunk. */ + } + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_software, "ISFT")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_software); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_copyright, "ICOP")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_copyright); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_title, "INAM")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_title); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_artist, "IART")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_artist); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_comment, "ICMT")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_comment); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_date, "ICRD")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_date); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_genre, "IGNR")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_genre); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_album, "IPRD")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_album); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_tracknumber, "ITRK")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_tracknumber); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_location, "IARL")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_location); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_organization, "ICMS")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_organization); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_keywords, "IKEY")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_keywords); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_medium, "IMED")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_medium); + } else if (drwav__chunk_matches(allowedMetadataTypes, subchunkId, drwav_metadata_type_list_info_description, "ISBJ")) { + subchunkBytesRead = drwav__metadata_process_info_text_chunk(pParser, subchunkDataSize, drwav_metadata_type_list_info_description); + } else if ((allowedMetadataTypes & drwav_metadata_type_unknown) != 0) { + subchunkBytesRead = drwav__metadata_process_unknown_chunk(pParser, subchunkId, subchunkDataSize, listType); + } + + bytesRead += subchunkBytesRead; + DRWAV_ASSERT(subchunkBytesRead <= subchunkDataSize); + + if (subchunkBytesRead < subchunkDataSize) { + drwav_uint64 bytesToSeek = subchunkDataSize - subchunkBytesRead; + + if (!pParser->onSeek(pParser->pReadSeekUserData, (int)bytesToSeek, DRWAV_SEEK_CUR)) { + break; + } + bytesRead += bytesToSeek; + } + + if ((subchunkDataSize % 2) == 1) { + if (!pParser->onSeek(pParser->pReadSeekUserData, 1, DRWAV_SEEK_CUR)) { + break; + } + bytesRead += 1; + } + } + } else if ((allowedMetadataTypes & drwav_metadata_type_unknown) != 0) { + bytesRead = drwav__metadata_process_unknown_chunk(pParser, pChunkID, pChunkHeader->sizeInBytes, drwav_metadata_location_top_level); + } + + return bytesRead; +} + + +DRWAV_PRIVATE drwav_uint32 drwav_get_bytes_per_pcm_frame(drwav* pWav) +{ + drwav_uint32 bytesPerFrame; + + /* + The bytes per frame is a bit ambiguous. It can be either be based on the bits per sample, or the block align. The way I'm doing it here + is that if the bits per sample is a multiple of 8, use floor(bitsPerSample*channels/8), otherwise fall back to the block align. + */ + if ((pWav->bitsPerSample & 0x7) == 0) { + /* Bits per sample is a multiple of 8. */ + bytesPerFrame = (pWav->bitsPerSample * pWav->fmt.channels) >> 3; + } else { + bytesPerFrame = pWav->fmt.blockAlign; + } + + /* Validation for known formats. a-law and mu-law should be 1 byte per channel. If it's not, it's not decodable. */ + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ALAW || pWav->translatedFormatTag == DR_WAVE_FORMAT_MULAW) { + if (bytesPerFrame != pWav->fmt.channels) { + return 0; /* Invalid file. */ + } + } + + return bytesPerFrame; +} + +DRWAV_API drwav_uint16 drwav_fmt_get_format(const drwav_fmt* pFMT) +{ + if (pFMT == NULL) { + return 0; + } + + if (pFMT->formatTag != DR_WAVE_FORMAT_EXTENSIBLE) { + return pFMT->formatTag; + } else { + return drwav_bytes_to_u16(pFMT->subFormat); /* Only the first two bytes are required. */ + } +} + +DRWAV_PRIVATE drwav_bool32 drwav_preinit(drwav* pWav, drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pReadSeekTellUserData, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pWav == NULL || onRead == NULL || onSeek == NULL) { /* <-- onTell is optional. */ + return DRWAV_FALSE; + } + + DRWAV_ZERO_MEMORY(pWav, sizeof(*pWav)); + pWav->onRead = onRead; + pWav->onSeek = onSeek; + pWav->onTell = onTell; + pWav->pUserData = pReadSeekTellUserData; + pWav->allocationCallbacks = drwav_copy_allocation_callbacks_or_defaults(pAllocationCallbacks); + + if (pWav->allocationCallbacks.onFree == NULL || (pWav->allocationCallbacks.onMalloc == NULL && pWav->allocationCallbacks.onRealloc == NULL)) { + return DRWAV_FALSE; /* Invalid allocation callbacks. */ + } + + return DRWAV_TRUE; +} + +DRWAV_PRIVATE drwav_bool32 drwav_init__internal(drwav* pWav, drwav_chunk_proc onChunk, void* pChunkUserData, drwav_uint32 flags) +{ + /* This function assumes drwav_preinit() has been called beforehand. */ + drwav_result result; + drwav_uint64 cursor; /* <-- Keeps track of the byte position so we can seek to specific locations. */ + drwav_bool32 sequential; + drwav_uint8 riff[4]; + drwav_fmt fmt; + unsigned short translatedFormatTag; + drwav_uint64 dataChunkSize = 0; /* <-- Important! Don't explicitly set this to 0 anywhere else. Calculation of the size of the data chunk is performed in different paths depending on the container. */ + drwav_uint64 sampleCountFromFactChunk = 0; /* Same as dataChunkSize - make sure this is the only place this is initialized to 0. */ + drwav_uint64 metadataStartPos; + drwav__metadata_parser metadataParser; + drwav_bool8 isProcessingMetadata = DRWAV_FALSE; + drwav_bool8 foundChunk_fmt = DRWAV_FALSE; + drwav_bool8 foundChunk_data = DRWAV_FALSE; + drwav_bool8 isAIFCFormType = DRWAV_FALSE; /* Only used with AIFF. */ + drwav_uint64 aiffFrameCount = 0; + drwav_int64 fileSize = 0; + drwav_bool32 hasKnownFileSize = DRWAV_FALSE; + + cursor = 0; + sequential = (flags & DRWAV_SEQUENTIAL) != 0; + + /* + Whether or not we are processing metadata controls how we load. We can load more efficiently when + metadata is not being processed. Seqential mode cannot support metadata because it involves seeking + backwards. + */ + isProcessingMetadata = !sequential && ((flags & DRWAV_WITH_METADATA) != 0); + + /* + When processing metadata we'll be doing some memory allocations here against untrusted data. We'll do + a basic validation check that they don't exceed the size of the file. + */ + if (isProcessingMetadata && pWav->onTell != NULL && pWav->onSeek != NULL) { + if (pWav->onSeek(pWav->pUserData, 0, DRWAV_SEEK_END)) { + if (pWav->onTell(pWav->pUserData, &fileSize)) { + hasKnownFileSize = DRWAV_TRUE; + } + + pWav->onSeek(pWav->pUserData, 0, DRWAV_SEEK_SET); + } + } + + DRWAV_ZERO_OBJECT(&fmt); + + /* The first 4 bytes should be the RIFF identifier. */ + if (drwav__on_read(pWav->onRead, pWav->pUserData, riff, sizeof(riff), &cursor) != sizeof(riff)) { + return DRWAV_FALSE; + } + + /* + The first 4 bytes can be used to identify the container. For RIFF files it will start with "RIFF" and for + w64 it will start with "riff". + */ + if (drwav_fourcc_equal(riff, "RIFF")) { + pWav->container = drwav_container_riff; + } else if (drwav_fourcc_equal(riff, "RIFX")) { + pWav->container = drwav_container_rifx; + } else if (drwav_fourcc_equal(riff, "riff")) { + int i; + drwav_uint8 riff2[12]; + + pWav->container = drwav_container_w64; + + /* Check the rest of the GUID for validity. */ + if (drwav__on_read(pWav->onRead, pWav->pUserData, riff2, sizeof(riff2), &cursor) != sizeof(riff2)) { + return DRWAV_FALSE; + } + + for (i = 0; i < 12; ++i) { + if (riff2[i] != drwavGUID_W64_RIFF[i+4]) { + return DRWAV_FALSE; + } + } + } else if (drwav_fourcc_equal(riff, "RF64")) { + pWav->container = drwav_container_rf64; + } else if (drwav_fourcc_equal(riff, "FORM")) { + pWav->container = drwav_container_aiff; + } else { + return DRWAV_FALSE; /* Unknown or unsupported container. */ + } + + + if (pWav->container == drwav_container_riff || pWav->container == drwav_container_rifx || pWav->container == drwav_container_rf64) { + drwav_uint8 chunkSizeBytes[4]; + drwav_uint8 wave[4]; + + if (drwav__on_read(pWav->onRead, pWav->pUserData, chunkSizeBytes, sizeof(chunkSizeBytes), &cursor) != sizeof(chunkSizeBytes)) { + return DRWAV_FALSE; + } + + if (pWav->container == drwav_container_riff || pWav->container == drwav_container_rifx) { + if (drwav_bytes_to_u32_ex(chunkSizeBytes, pWav->container) < 36) { + /* + I've had a report of a WAV file failing to load when the size of the WAVE chunk is not encoded + and is instead just set to 0. I'm going to relax the validation here to allow these files to + load. Considering the chunk size isn't actually used this should be safe. With this change my + test suite still passes. + */ + /*return DRWAV_FALSE;*/ /* Chunk size should always be at least 36 bytes. */ + } + } else if (pWav->container == drwav_container_rf64) { + if (drwav_bytes_to_u32_le(chunkSizeBytes) != 0xFFFFFFFF) { + return DRWAV_FALSE; /* Chunk size should always be set to -1/0xFFFFFFFF for RF64. The actual size is retrieved later. */ + } + } else { + return DRWAV_FALSE; /* Should never hit this. */ + } + + if (drwav__on_read(pWav->onRead, pWav->pUserData, wave, sizeof(wave), &cursor) != sizeof(wave)) { + return DRWAV_FALSE; + } + + if (!drwav_fourcc_equal(wave, "WAVE")) { + return DRWAV_FALSE; /* Expecting "WAVE". */ + } + } else if (pWav->container == drwav_container_w64) { + drwav_uint8 chunkSizeBytes[8]; + drwav_uint8 wave[16]; + + if (drwav__on_read(pWav->onRead, pWav->pUserData, chunkSizeBytes, sizeof(chunkSizeBytes), &cursor) != sizeof(chunkSizeBytes)) { + return DRWAV_FALSE; + } + + if (drwav_bytes_to_u64(chunkSizeBytes) < 80) { + return DRWAV_FALSE; + } + + if (drwav__on_read(pWav->onRead, pWav->pUserData, wave, sizeof(wave), &cursor) != sizeof(wave)) { + return DRWAV_FALSE; + } + + if (!drwav_guid_equal(wave, drwavGUID_W64_WAVE)) { + return DRWAV_FALSE; + } + } else if (pWav->container == drwav_container_aiff) { + drwav_uint8 chunkSizeBytes[4]; + drwav_uint8 aiff[4]; + + if (drwav__on_read(pWav->onRead, pWav->pUserData, chunkSizeBytes, sizeof(chunkSizeBytes), &cursor) != sizeof(chunkSizeBytes)) { + return DRWAV_FALSE; + } + + if (drwav_bytes_to_u32_be(chunkSizeBytes) < 18) { + return DRWAV_FALSE; + } + + if (drwav__on_read(pWav->onRead, pWav->pUserData, aiff, sizeof(aiff), &cursor) != sizeof(aiff)) { + return DRWAV_FALSE; + } + + if (drwav_fourcc_equal(aiff, "AIFF")) { + isAIFCFormType = DRWAV_FALSE; + } else if (drwav_fourcc_equal(aiff, "AIFC")) { + isAIFCFormType = DRWAV_TRUE; + } else { + return DRWAV_FALSE; /* Expecting "AIFF" or "AIFC". */ + } + } else { + return DRWAV_FALSE; + } + + + /* For RF64, the "ds64" chunk must come next, before the "fmt " chunk. */ + if (pWav->container == drwav_container_rf64) { + drwav_uint8 sizeBytes[8]; + drwav_uint64 bytesRemainingInChunk; + drwav_chunk_header header; + result = drwav__read_chunk_header(pWav->onRead, pWav->pUserData, pWav->container, &cursor, &header); + if (result != DRWAV_SUCCESS) { + return DRWAV_FALSE; + } + + if (!drwav_fourcc_equal(header.id.fourcc, "ds64")) { + return DRWAV_FALSE; /* Expecting "ds64". */ + } + + bytesRemainingInChunk = header.sizeInBytes + header.paddingSize; + + /* We don't care about the size of the RIFF chunk - skip it. */ + if (!drwav__seek_forward(pWav->onSeek, 8, pWav->pUserData)) { + return DRWAV_FALSE; + } + bytesRemainingInChunk -= 8; + cursor += 8; + + + /* Next 8 bytes is the size of the "data" chunk. */ + if (drwav__on_read(pWav->onRead, pWav->pUserData, sizeBytes, sizeof(sizeBytes), &cursor) != sizeof(sizeBytes)) { + return DRWAV_FALSE; + } + bytesRemainingInChunk -= 8; + dataChunkSize = drwav_bytes_to_u64(sizeBytes); + + + /* Next 8 bytes is the same count which we would usually derived from the FACT chunk if it was available. */ + if (drwav__on_read(pWav->onRead, pWav->pUserData, sizeBytes, sizeof(sizeBytes), &cursor) != sizeof(sizeBytes)) { + return DRWAV_FALSE; + } + bytesRemainingInChunk -= 8; + sampleCountFromFactChunk = drwav_bytes_to_u64(sizeBytes); + + + /* Skip over everything else. */ + if (!drwav__seek_forward(pWav->onSeek, bytesRemainingInChunk, pWav->pUserData)) { + return DRWAV_FALSE; + } + cursor += bytesRemainingInChunk; + } + + + metadataStartPos = cursor; + + /* Don't allow processing of metadata with untested containers. */ + if (pWav->container != drwav_container_riff && pWav->container != drwav_container_rf64) { + isProcessingMetadata = DRWAV_FALSE; + } + + DRWAV_ZERO_MEMORY(&metadataParser, sizeof(metadataParser)); + if (isProcessingMetadata) { + metadataParser.onRead = pWav->onRead; + metadataParser.onSeek = pWav->onSeek; + metadataParser.pReadSeekUserData = pWav->pUserData; + metadataParser.stage = drwav__metadata_parser_stage_count; + } + + + /* + From here on out, chunks might be in any order. In order to robustly handle metadata we'll need + to loop through every chunk and handle them as we find them. In sequential mode we need to get + out of the loop as soon as we find the data chunk because we won't be able to seek back. + */ + for (;;) { /* For each chunk... */ + drwav_chunk_header header; + drwav_uint64 chunkSize; + + result = drwav__read_chunk_header(pWav->onRead, pWav->pUserData, pWav->container, &cursor, &header); + if (result != DRWAV_SUCCESS) { + break; + } + + chunkSize = header.sizeInBytes; + + + /* + Always tell the caller about this chunk. We cannot do this in sequential mode because the + callback is allowed to read from the file, in which case we'll need to rewind. + */ + if (!sequential && onChunk != NULL) { + drwav_uint64 callbackBytesRead = onChunk(pChunkUserData, pWav->onRead, pWav->onSeek, pWav->pUserData, &header, pWav->container, &fmt); + + /* + dr_wav may need to read the contents of the chunk, so we now need to seek back to the position before + we called the callback. + */ + if (callbackBytesRead > 0) { + if (drwav__seek_from_start(pWav->onSeek, cursor, pWav->pUserData) == DRWAV_FALSE) { + return DRWAV_FALSE; + } + } + } + + + /* Explicitly handle known chunks first. */ + + /* "fmt " */ + if (((pWav->container == drwav_container_riff || pWav->container == drwav_container_rifx || pWav->container == drwav_container_rf64) && drwav_fourcc_equal(header.id.fourcc, "fmt ")) || + ((pWav->container == drwav_container_w64) && drwav_guid_equal(header.id.guid, drwavGUID_W64_FMT))) { + drwav_uint8 fmtData[16]; + + if (header.sizeInBytes < sizeof(fmtData)) { + return DRWAV_FALSE; /* Invalid fmt chunk. */ + } + + foundChunk_fmt = DRWAV_TRUE; + + if (pWav->onRead(pWav->pUserData, fmtData, sizeof(fmtData)) != sizeof(fmtData)) { + return DRWAV_FALSE; + } + cursor += sizeof(fmtData); + + fmt.formatTag = drwav_bytes_to_u16_ex(fmtData + 0, pWav->container); + fmt.channels = drwav_bytes_to_u16_ex(fmtData + 2, pWav->container); + fmt.sampleRate = drwav_bytes_to_u32_ex(fmtData + 4, pWav->container); + fmt.avgBytesPerSec = drwav_bytes_to_u32_ex(fmtData + 8, pWav->container); + fmt.blockAlign = drwav_bytes_to_u16_ex(fmtData + 12, pWav->container); + fmt.bitsPerSample = drwav_bytes_to_u16_ex(fmtData + 14, pWav->container); + + fmt.extendedSize = 0; + fmt.validBitsPerSample = 0; + fmt.channelMask = 0; + DRWAV_ZERO_MEMORY(fmt.subFormat, sizeof(fmt.subFormat)); + + if (header.sizeInBytes > 16) { + drwav_uint8 fmt_cbSize[2]; + int bytesReadSoFar = 0; + drwav_uint64 leftoverBytes; + + if (pWav->onRead(pWav->pUserData, fmt_cbSize, sizeof(fmt_cbSize)) != sizeof(fmt_cbSize)) { + return DRWAV_FALSE; /* Expecting more data. */ + } + cursor += sizeof(fmt_cbSize); + + bytesReadSoFar = 18; + + fmt.extendedSize = drwav_bytes_to_u16_ex(fmt_cbSize, pWav->container); + if (fmt.extendedSize > 0) { + /* Simple validation. */ + if (fmt.formatTag == DR_WAVE_FORMAT_EXTENSIBLE) { + if (fmt.extendedSize != 22) { + return DRWAV_FALSE; + } + } + + if (fmt.formatTag == DR_WAVE_FORMAT_EXTENSIBLE) { + drwav_uint8 fmtext[22]; + + if (pWav->onRead(pWav->pUserData, fmtext, fmt.extendedSize) != fmt.extendedSize) { + return DRWAV_FALSE; /* Expecting more data. */ + } + + fmt.validBitsPerSample = drwav_bytes_to_u16_ex(fmtext + 0, pWav->container); + fmt.channelMask = drwav_bytes_to_u32_ex(fmtext + 2, pWav->container); + drwav_bytes_to_guid(fmtext + 6, fmt.subFormat); + } else { + if (pWav->onSeek(pWav->pUserData, fmt.extendedSize, DRWAV_SEEK_CUR) == DRWAV_FALSE) { + return DRWAV_FALSE; + } + } + cursor += fmt.extendedSize; + + bytesReadSoFar += fmt.extendedSize; + } + + /* Seek past any leftover bytes. For w64 the leftover will be defined based on the chunk size. */ + leftoverBytes = header.sizeInBytes - bytesReadSoFar; + if (leftoverBytes > 0x7FFFFFFF) { + return DRWAV_FALSE; + } + + if (pWav->onSeek(pWav->pUserData, (int)leftoverBytes, DRWAV_SEEK_CUR) == DRWAV_FALSE) { + return DRWAV_FALSE; + } + + cursor += leftoverBytes; + } + + if (header.paddingSize > 0) { + if (drwav__seek_forward(pWav->onSeek, header.paddingSize, pWav->pUserData) == DRWAV_FALSE) { + break; + } + cursor += header.paddingSize; + } + + /* Go to the next chunk. Don't include this chunk in metadata. */ + continue; + } + + /* "data" */ + if (((pWav->container == drwav_container_riff || pWav->container == drwav_container_rifx || pWav->container == drwav_container_rf64) && drwav_fourcc_equal(header.id.fourcc, "data")) || + ((pWav->container == drwav_container_w64) && drwav_guid_equal(header.id.guid, drwavGUID_W64_DATA))) { + foundChunk_data = DRWAV_TRUE; + + pWav->dataChunkDataPos = cursor; + + if (pWav->container != drwav_container_rf64) { /* The data chunk size for RF64 will always be set to 0xFFFFFFFF here. It was set to it's true value earlier. */ + dataChunkSize = chunkSize; + } + + /* If we're running in sequential mode, or we're not reading metadata, we have enough now that we can get out of the loop. */ + if (sequential || !isProcessingMetadata) { + break; /* No need to keep reading beyond the data chunk. */ + } else { + chunkSize += header.paddingSize; /* <-- Make sure we seek past the padding. */ + if (drwav__seek_forward(pWav->onSeek, chunkSize, pWav->pUserData) == DRWAV_FALSE) { + break; + } + cursor += chunkSize; + + continue; /* There may be some more metadata to read. */ + } + } + + /* "fact". This is optional. Can use this to get the sample count which is useful for compressed formats. For RF64 we retrieved the sample count from the ds64 chunk earlier. */ + if (((pWav->container == drwav_container_riff || pWav->container == drwav_container_rifx || pWav->container == drwav_container_rf64) && drwav_fourcc_equal(header.id.fourcc, "fact")) || + ((pWav->container == drwav_container_w64) && drwav_guid_equal(header.id.guid, drwavGUID_W64_FACT))) { + if (pWav->container == drwav_container_riff || pWav->container == drwav_container_rifx) { + drwav_uint8 sampleCount[4]; + + if (chunkSize < 4) { + return DRWAV_FALSE; + } + + if (drwav__on_read(pWav->onRead, pWav->pUserData, &sampleCount, 4, &cursor) != 4) { + return DRWAV_FALSE; + } + + chunkSize -= 4; + + sampleCountFromFactChunk = drwav_bytes_to_u32_ex(sampleCount, pWav->container); + } else if (pWav->container == drwav_container_w64) { + if (chunkSize < 8) { + return DRWAV_FALSE; + } + + if (drwav__on_read(pWav->onRead, pWav->pUserData, &sampleCountFromFactChunk, 8, &cursor) != 8) { + return DRWAV_FALSE; + } + + chunkSize -= 8; + } else if (pWav->container == drwav_container_rf64) { + /* We retrieved the sample count from the ds64 chunk earlier so no need to do that here. */ + } + + /* Seek to the next chunk in preparation for the next iteration. */ + chunkSize += header.paddingSize; /* <-- Make sure we seek past the padding. */ + if (drwav__seek_forward(pWav->onSeek, chunkSize, pWav->pUserData) == DRWAV_FALSE) { + break; + } + cursor += chunkSize; + + continue; + } + + + /* "COMM". AIFF/AIFC only. */ + if (pWav->container == drwav_container_aiff && drwav_fourcc_equal(header.id.fourcc, "COMM")) { + drwav_uint8 commData[24]; + drwav_uint32 commDataBytesToRead; + drwav_uint16 channels; + drwav_uint32 frameCount; + drwav_uint16 sampleSizeInBits; + drwav_int64 sampleRate; + drwav_uint16 compressionFormat; + + foundChunk_fmt = DRWAV_TRUE; + + if (isAIFCFormType) { + commDataBytesToRead = 24; + if (header.sizeInBytes < commDataBytesToRead) { + return DRWAV_FALSE; /* Invalid COMM chunk. */ + } + } else { + commDataBytesToRead = 18; + if (header.sizeInBytes != commDataBytesToRead) { + return DRWAV_FALSE; /* INVALID COMM chunk. */ + } + } + + if (drwav__on_read(pWav->onRead, pWav->pUserData, commData, commDataBytesToRead, &cursor) != commDataBytesToRead) { + return DRWAV_FALSE; + } + + + channels = drwav_bytes_to_u16_ex (commData + 0, pWav->container); + frameCount = drwav_bytes_to_u32_ex (commData + 2, pWav->container); + sampleSizeInBits = drwav_bytes_to_u16_ex (commData + 6, pWav->container); + sampleRate = drwav_aiff_extented_to_s64(commData + 8); + + if (sampleRate < 0 || sampleRate > 0xFFFFFFFF) { + return DRWAV_FALSE; /* Invalid sample rate. */ + } + + if (isAIFCFormType) { + const drwav_uint8* type = commData + 18; + + if (drwav_fourcc_equal(type, "NONE")) { + compressionFormat = DR_WAVE_FORMAT_PCM; /* PCM, big-endian. */ + } else if (drwav_fourcc_equal(type, "raw ")) { + compressionFormat = DR_WAVE_FORMAT_PCM; + + /* In my testing, it looks like when the "raw " compression type is used, 8-bit samples should be considered unsigned. */ + if (sampleSizeInBits == 8) { + pWav->aiff.isUnsigned = DRWAV_TRUE; + } + } else if (drwav_fourcc_equal(type, "sowt")) { + compressionFormat = DR_WAVE_FORMAT_PCM; /* PCM, little-endian. */ + pWav->aiff.isLE = DRWAV_TRUE; + } else if (drwav_fourcc_equal(type, "fl32") || drwav_fourcc_equal(type, "fl64") || drwav_fourcc_equal(type, "FL32") || drwav_fourcc_equal(type, "FL64")) { + compressionFormat = DR_WAVE_FORMAT_IEEE_FLOAT; + } else if (drwav_fourcc_equal(type, "alaw") || drwav_fourcc_equal(type, "ALAW")) { + compressionFormat = DR_WAVE_FORMAT_ALAW; + } else if (drwav_fourcc_equal(type, "ulaw") || drwav_fourcc_equal(type, "ULAW")) { + compressionFormat = DR_WAVE_FORMAT_MULAW; + } else if (drwav_fourcc_equal(type, "ima4")) { + compressionFormat = DR_WAVE_FORMAT_DVI_ADPCM; + sampleSizeInBits = 4; + + /* + I haven't been able to figure out how to get correct decoding for IMA ADPCM. Until this is figured out + we'll need to abort when we encounter such an encoding. Advice welcome! + */ + (void)compressionFormat; + (void)sampleSizeInBits; + + return DRWAV_FALSE; + } else { + return DRWAV_FALSE; /* Unknown or unsupported compression format. Need to abort. */ + } + } else { + compressionFormat = DR_WAVE_FORMAT_PCM; /* It's a standard AIFF form which is always compressed. */ + } + + /* With AIFF we want to use the explicitly defined frame count rather than deriving it from the size of the chunk. */ + aiffFrameCount = frameCount; + + /* We should now have enough information to fill out our fmt structure. */ + fmt.formatTag = compressionFormat; + fmt.channels = channels; + fmt.sampleRate = (drwav_uint32)sampleRate; + fmt.bitsPerSample = (sampleSizeInBits + 7) & ~7; /* In AIFF, samples are padded to 8-bit boundaries. We need to round up our bits per sample here. */ + fmt.blockAlign = (drwav_uint16)((drwav_uint32)fmt.channels * fmt.bitsPerSample / 8); + fmt.avgBytesPerSec = fmt.blockAlign * fmt.sampleRate; + + /* + Weird one. I've seen some alaw and ulaw encoded files that for some reason set the bits per sample to 16 when + it should be 8. To get this working I need to explicitly check for this and change it. + */ + if (compressionFormat == DR_WAVE_FORMAT_ALAW || compressionFormat == DR_WAVE_FORMAT_MULAW) { + if (fmt.bitsPerSample > 8) { + fmt.bitsPerSample = 8; + fmt.blockAlign = fmt.channels; + } + } + + /* If the form type is AIFC there will be some additional data in the chunk. We need to seek past it. */ + if (isAIFCFormType) { + if (drwav__seek_forward(pWav->onSeek, (chunkSize - commDataBytesToRead), pWav->pUserData) == DRWAV_FALSE) { + return DRWAV_FALSE; + } + cursor += (chunkSize - commDataBytesToRead); + } + + /* Don't fall through or else we'll end up treating this chunk as metadata which is incorrect. */ + continue; + } + + + /* "SSND". AIFF/AIFC only. This is the AIFF equivalent of the "data" chunk. */ + if (pWav->container == drwav_container_aiff && drwav_fourcc_equal(header.id.fourcc, "SSND")) { + drwav_uint8 offsetAndBlockSizeData[8]; + drwav_uint32 offset; + + foundChunk_data = DRWAV_TRUE; + + if (drwav__on_read(pWav->onRead, pWav->pUserData, offsetAndBlockSizeData, sizeof(offsetAndBlockSizeData), &cursor) != sizeof(offsetAndBlockSizeData)) { + return DRWAV_FALSE; + } + + /* The position of the audio data starts at an offset. */ + offset = drwav_bytes_to_u32_ex(offsetAndBlockSizeData + 0, pWav->container); + pWav->dataChunkDataPos = cursor + offset; + + /* The data chunk size needs to be reduced by the offset or else seeking will break. */ + dataChunkSize = chunkSize; + if (dataChunkSize > offset) { + dataChunkSize -= offset; + } else { + dataChunkSize = 0; + } + + if (sequential) { + if (foundChunk_fmt) { /* <-- Name is misleading, but will be set to true if the COMM chunk has been parsed. */ + /* + Getting here means we're opening in sequential mode and we've found the SSND (data) and COMM (fmt) chunks. We need + to get out of the loop here or else we'll end up going past the data chunk and will have no way of getting back to + it since we're not allowed to seek backwards. + + One subtle detail here is that there is an offset with the SSND chunk. We need to make sure we seek past this offset + so we're left sitting on the first byte of actual audio data. + */ + if (drwav__seek_forward(pWav->onSeek, offset, pWav->pUserData) == DRWAV_FALSE) { + return DRWAV_FALSE; + } + cursor += offset; + + break; + } else { + /* + Getting here means the COMM chunk was not found. In sequential mode, if we haven't yet found the COMM chunk + we'll need to abort because we can't be doing a backwards seek back to the SSND chunk in order to read the + data. For this reason, this configuration of AIFF files are not supported with sequential mode. + */ + return DRWAV_FALSE; + } + } else { + chunkSize += header.paddingSize; /* <-- Make sure we seek past the padding. */ + chunkSize -= sizeof(offsetAndBlockSizeData); /* <-- This was read earlier. */ + + if (drwav__seek_forward(pWav->onSeek, chunkSize, pWav->pUserData) == DRWAV_FALSE) { + break; + } + cursor += chunkSize; + + continue; /* There may be some more metadata to read. */ + } + } + + + /* Getting here means it's not a chunk that we care about internally, but might need to be handled as metadata by the caller. */ + if (isProcessingMetadata) { + if (hasKnownFileSize && header.sizeInBytes > (drwav_uint64)fileSize) { + return DRWAV_FALSE; + } + + drwav__metadata_process_chunk(&metadataParser, &header, drwav_metadata_type_all_including_unknown); + + /* Go back to the start of the chunk so we can normalize the position of the cursor. */ + if (drwav__seek_from_start(pWav->onSeek, cursor, pWav->pUserData) == DRWAV_FALSE) { + break; /* Failed to seek. Can't reliable read the remaining chunks. Get out. */ + } + } + + + /* Make sure we skip past the content of this chunk before we go to the next one. */ + chunkSize += header.paddingSize; /* <-- Make sure we seek past the padding. */ + if (drwav__seek_forward(pWav->onSeek, chunkSize, pWav->pUserData) == DRWAV_FALSE) { + break; + } + cursor += chunkSize; + } + + /* There's some mandatory chunks that must exist. If they were not found in the iteration above we must abort. */ + if (!foundChunk_fmt || !foundChunk_data) { + return DRWAV_FALSE; + } + + /* Basic validation. */ + if ((fmt.sampleRate == 0 || fmt.sampleRate > DRWAV_MAX_SAMPLE_RATE ) || + (fmt.channels == 0 || fmt.channels > DRWAV_MAX_CHANNELS ) || + (fmt.bitsPerSample == 0 || fmt.bitsPerSample > DRWAV_MAX_BITS_PER_SAMPLE) || + fmt.blockAlign == 0) { + return DRWAV_FALSE; /* Probably an invalid WAV file. */ + } + + /* Translate the internal format. */ + translatedFormatTag = fmt.formatTag; + if (translatedFormatTag == DR_WAVE_FORMAT_EXTENSIBLE) { + translatedFormatTag = drwav_bytes_to_u16_ex(fmt.subFormat + 0, pWav->container); + } + + /* Some files have unreliable fact counts. Only use RIFF fact counts for Microsoft ADPCM, after the format is known. */ + if ((pWav->container == drwav_container_riff || pWav->container == drwav_container_rifx) && translatedFormatTag != DR_WAVE_FORMAT_ADPCM) { + sampleCountFromFactChunk = 0; + } + + /* We may have moved passed the data chunk. If so we need to move back. If running in sequential mode we can assume we are already sitting on the data chunk. */ + if (!sequential) { + if (!drwav__seek_from_start(pWav->onSeek, pWav->dataChunkDataPos, pWav->pUserData)) { + return DRWAV_FALSE; + } + cursor = pWav->dataChunkDataPos; + } + + + /* + At this point we should have done the initial parsing of each of our chunks, but we now need to + do a second pass to extract the actual contents of the metadata (the first pass just calculated + the length of the memory allocation). + + We only do this if we've actually got metadata to parse. + */ + if (isProcessingMetadata && metadataParser.metadataCount > 0) { + if (drwav__seek_from_start(pWav->onSeek, metadataStartPos, pWav->pUserData) == DRWAV_FALSE) { + return DRWAV_FALSE; + } + + result = drwav__metadata_alloc(&metadataParser, &pWav->allocationCallbacks); + if (result != DRWAV_SUCCESS) { + return DRWAV_FALSE; + } + + metadataParser.stage = drwav__metadata_parser_stage_read; + + for (;;) { + drwav_chunk_header header; + drwav_uint64 metadataBytesRead; + + result = drwav__read_chunk_header(pWav->onRead, pWav->pUserData, pWav->container, &cursor, &header); + if (result != DRWAV_SUCCESS) { + break; + } + + metadataBytesRead = drwav__metadata_process_chunk(&metadataParser, &header, drwav_metadata_type_all_including_unknown); + + if (metadataParser.metadataCursor == metadataParser.metadataCount) { + break; + } + + /* Move to the end of the chunk so we can keep iterating. */ + if (drwav__seek_forward(pWav->onSeek, (header.sizeInBytes + header.paddingSize) - metadataBytesRead, pWav->pUserData) == DRWAV_FALSE) { + drwav_free(metadataParser.pMetadata, &pWav->allocationCallbacks); + return DRWAV_FALSE; + } + } + + /* Getting here means we're finished parsing the metadata. */ + pWav->pMetadata = metadataParser.pMetadata; + pWav->metadataCount = metadataParser.metadataCount; + } + + /* + It's possible for the size reported in the data chunk to be greater than that of the file. We + need to do a validation check here to make sure we don't exceed the file size. To skip this + check, set the onTell callback to NULL. + */ + if (pWav->onTell != NULL && pWav->onSeek != NULL) { + if (pWav->onSeek(pWav->pUserData, 0, DRWAV_SEEK_END) == DRWAV_TRUE) { + drwav_int64 fileSize; + if (pWav->onTell(pWav->pUserData, &fileSize)) { + if (dataChunkSize + pWav->dataChunkDataPos > (drwav_uint64)fileSize) { + dataChunkSize = (drwav_uint64)fileSize - pWav->dataChunkDataPos; + } + } + } else { + /* + Failed to seek to the end of the file. It might not be supported by the backend so in + this case we cannot perform the validation check. + */ + } + } + + /* + I've seen a WAV file in the wild where a RIFF-ecapsulated file has the size of it's "RIFF" and + "data" chunks set to 0xFFFFFFFF when the file is definitely not that big. In this case we're + going to have to calculate the size by reading and discarding bytes, and then seeking back. We + cannot do this in sequential mode. We just assume that the rest of the file is audio data. + */ + if (dataChunkSize == 0xFFFFFFFF && (pWav->container == drwav_container_riff || pWav->container == drwav_container_rifx) && pWav->isSequentialWrite == DRWAV_FALSE) { + dataChunkSize = 0; + + for (;;) { + drwav_uint8 temp[4096]; + size_t bytesRead = pWav->onRead(pWav->pUserData, temp, sizeof(temp)); + dataChunkSize += bytesRead; + + if (bytesRead < sizeof(temp)) { + break; + } + } + } + + /* At this point we want to be sitting on the first byte of the raw audio data. */ + if (drwav__seek_from_start(pWav->onSeek, pWav->dataChunkDataPos, pWav->pUserData) == DRWAV_FALSE) { + drwav_free(pWav->pMetadata, &pWav->allocationCallbacks); + return DRWAV_FALSE; + } + + + pWav->fmt = fmt; + pWav->sampleRate = fmt.sampleRate; + pWav->channels = fmt.channels; + pWav->bitsPerSample = fmt.bitsPerSample; + pWav->translatedFormatTag = translatedFormatTag; + + /* + I've had a report where files would start glitching after seeking. The reason for this is the data + chunk is not a clean multiple of the PCM frame size in bytes. Where this becomes a problem is when + seeking, because the number of bytes remaining in the data chunk is used to calculate the current + byte position. If this byte position is not aligned to the number of bytes in a PCM frame, it will + result in the seek not being cleanly positioned at the start of the PCM frame thereby resulting in + all decoded frames after that being corrupted. + + To address this, we need to round the data chunk size down to the nearest multiple of the frame size. + */ + if (!drwav__is_compressed_format_tag(translatedFormatTag)) { + drwav_uint32 bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame > 0) { + dataChunkSize -= (dataChunkSize % bytesPerFrame); + } + } + + pWav->bytesRemaining = dataChunkSize; + pWav->dataChunkDataSize = dataChunkSize; + + if (sampleCountFromFactChunk != 0) { + pWav->totalPCMFrameCount = sampleCountFromFactChunk; + } else if (aiffFrameCount != 0) { + pWav->totalPCMFrameCount = aiffFrameCount; + } else { + drwav_uint32 bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + drwav_free(pWav->pMetadata, &pWav->allocationCallbacks); + return DRWAV_FALSE; /* Invalid file. */ + } + + pWav->totalPCMFrameCount = dataChunkSize / bytesPerFrame; + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ADPCM) { + drwav_uint64 totalBlockHeaderSizeInBytes; + drwav_uint64 blockCount = dataChunkSize / fmt.blockAlign; + + /* Make sure any trailing partial block is accounted for. */ + if ((blockCount * fmt.blockAlign) < dataChunkSize) { + blockCount += 1; + } + + /* We decode two samples per byte. There will be blockCount headers in the data chunk. This is enough to know how to calculate the total PCM frame count. */ + totalBlockHeaderSizeInBytes = blockCount * (6*fmt.channels); + if (totalBlockHeaderSizeInBytes >= dataChunkSize) { /* <-- We'll be subtracting totalBlockHeaderSizeInBytes from dataChunkSize next so it must be validated. */ + drwav_free(pWav->pMetadata, &pWav->allocationCallbacks); + return DRWAV_FALSE; /* Invalid file. */ + } + + pWav->totalPCMFrameCount = ((dataChunkSize - totalBlockHeaderSizeInBytes) * 2) / fmt.channels; + } + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_DVI_ADPCM) { + drwav_uint64 totalBlockHeaderSizeInBytes; + drwav_uint64 blockCount = dataChunkSize / fmt.blockAlign; + + /* Make sure any trailing partial block is accounted for. */ + if ((blockCount * fmt.blockAlign) < dataChunkSize) { + blockCount += 1; + } + + /* We decode two samples per byte. There will be blockCount headers in the data chunk. This is enough to know how to calculate the total PCM frame count. */ + totalBlockHeaderSizeInBytes = blockCount * (4*fmt.channels); + if (totalBlockHeaderSizeInBytes >= dataChunkSize) { /* <-- We'll be subtracting totalBlockHeaderSizeInBytes from dataChunkSize next so it must be validated. */ + drwav_free(pWav->pMetadata, &pWav->allocationCallbacks); + return DRWAV_FALSE; /* Invalid file. */ + } + + pWav->totalPCMFrameCount = ((dataChunkSize - totalBlockHeaderSizeInBytes) * 2) / fmt.channels; + + /* The header includes a decoded sample for each channel which acts as the initial predictor sample. */ + pWav->totalPCMFrameCount += blockCount; + } + } + + /* Some formats only support a certain number of channels. */ + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ADPCM || pWav->translatedFormatTag == DR_WAVE_FORMAT_DVI_ADPCM) { + if (pWav->channels > 2) { + drwav_free(pWav->pMetadata, &pWav->allocationCallbacks); + return DRWAV_FALSE; + } + } + + /* The number of bytes per frame must be known. If not, it's an invalid file and not decodable. */ + if (drwav_get_bytes_per_pcm_frame(pWav) == 0) { + drwav_free(pWav->pMetadata, &pWav->allocationCallbacks); + return DRWAV_FALSE; + } + + return DRWAV_TRUE; +} + +DRWAV_API drwav_bool32 drwav_init(drwav* pWav, drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_ex(pWav, onRead, onSeek, onTell, NULL, pUserData, NULL, 0, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_ex(drwav* pWav, drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, drwav_chunk_proc onChunk, void* pReadSeekTellUserData, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (!drwav_preinit(pWav, onRead, onSeek, onTell, pReadSeekTellUserData, pAllocationCallbacks)) { + return DRWAV_FALSE; + } + + return drwav_init__internal(pWav, onChunk, pChunkUserData, flags); +} + +DRWAV_API drwav_bool32 drwav_init_with_metadata(drwav* pWav, drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (!drwav_preinit(pWav, onRead, onSeek, onTell, pUserData, pAllocationCallbacks)) { + return DRWAV_FALSE; + } + + return drwav_init__internal(pWav, NULL, NULL, flags | DRWAV_WITH_METADATA); +} + +DRWAV_API drwav_metadata* drwav_take_ownership_of_metadata(drwav* pWav) +{ + drwav_metadata *result = pWav->pMetadata; + + pWav->pMetadata = NULL; + pWav->metadataCount = 0; + + return result; +} + + +DRWAV_PRIVATE size_t drwav__write(drwav* pWav, const void* pData, size_t dataSize) +{ + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pWav->onWrite != NULL); + + /* Generic write. Assumes no byte reordering required. */ + return pWav->onWrite(pWav->pUserData, pData, dataSize); +} + +DRWAV_PRIVATE size_t drwav__write_byte(drwav* pWav, drwav_uint8 byte) +{ + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pWav->onWrite != NULL); + + return pWav->onWrite(pWav->pUserData, &byte, 1); +} + +DRWAV_PRIVATE size_t drwav__write_u16ne_to_le(drwav* pWav, drwav_uint16 value) +{ + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pWav->onWrite != NULL); + + if (!drwav__is_little_endian()) { + value = drwav__bswap16(value); + } + + return drwav__write(pWav, &value, 2); +} + +DRWAV_PRIVATE size_t drwav__write_u32ne_to_le(drwav* pWav, drwav_uint32 value) +{ + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pWav->onWrite != NULL); + + if (!drwav__is_little_endian()) { + value = drwav__bswap32(value); + } + + return drwav__write(pWav, &value, 4); +} + +DRWAV_PRIVATE size_t drwav__write_u64ne_to_le(drwav* pWav, drwav_uint64 value) +{ + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pWav->onWrite != NULL); + + if (!drwav__is_little_endian()) { + value = drwav__bswap64(value); + } + + return drwav__write(pWav, &value, 8); +} + +DRWAV_PRIVATE size_t drwav__write_f32ne_to_le(drwav* pWav, float value) +{ + union { + drwav_uint32 u32; + float f32; + } u; + + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pWav->onWrite != NULL); + + u.f32 = value; + + if (!drwav__is_little_endian()) { + u.u32 = drwav__bswap32(u.u32); + } + + return drwav__write(pWav, &u.u32, 4); +} + +DRWAV_PRIVATE size_t drwav__write_or_count(drwav* pWav, const void* pData, size_t dataSize) +{ + if (pWav == NULL) { + return dataSize; + } + + return drwav__write(pWav, pData, dataSize); +} + +DRWAV_PRIVATE size_t drwav__write_or_count_byte(drwav* pWav, drwav_uint8 byte) +{ + if (pWav == NULL) { + return 1; + } + + return drwav__write_byte(pWav, byte); +} + +DRWAV_PRIVATE size_t drwav__write_or_count_u16ne_to_le(drwav* pWav, drwav_uint16 value) +{ + if (pWav == NULL) { + return 2; + } + + return drwav__write_u16ne_to_le(pWav, value); +} + +DRWAV_PRIVATE size_t drwav__write_or_count_u32ne_to_le(drwav* pWav, drwav_uint32 value) +{ + if (pWav == NULL) { + return 4; + } + + return drwav__write_u32ne_to_le(pWav, value); +} + +#if 0 /* Unused for now. */ +DRWAV_PRIVATE size_t drwav__write_or_count_u64ne_to_le(drwav* pWav, drwav_uint64 value) +{ + if (pWav == NULL) { + return 8; + } + + return drwav__write_u64ne_to_le(pWav, value); +} +#endif + +DRWAV_PRIVATE size_t drwav__write_or_count_f32ne_to_le(drwav* pWav, float value) +{ + if (pWav == NULL) { + return 4; + } + + return drwav__write_f32ne_to_le(pWav, value); +} + +DRWAV_PRIVATE size_t drwav__write_or_count_string_to_fixed_size_buf(drwav* pWav, char* str, size_t bufFixedSize) +{ + size_t len; + + if (pWav == NULL) { + return bufFixedSize; + } + + len = drwav__strlen_clamped(str, bufFixedSize); + drwav__write_or_count(pWav, str, len); + + if (len < bufFixedSize) { + size_t i; + for (i = 0; i < bufFixedSize - len; ++i) { + drwav__write_byte(pWav, 0); + } + } + + return bufFixedSize; +} + + +/* pWav can be NULL meaning just count the bytes that would be written. */ +DRWAV_PRIVATE size_t drwav__write_or_count_metadata(drwav* pWav, drwav_metadata* pMetadatas, drwav_uint32 metadataCount) +{ + size_t bytesWritten = 0; + drwav_bool32 hasListAdtl = DRWAV_FALSE; + drwav_bool32 hasListInfo = DRWAV_FALSE; + drwav_uint32 iMetadata; + + if (pMetadatas == NULL || metadataCount == 0) { + return 0; + } + + for (iMetadata = 0; iMetadata < metadataCount; ++iMetadata) { + drwav_metadata* pMetadata = &pMetadatas[iMetadata]; + drwav_uint32 chunkSize = 0; + + if ((pMetadata->type & drwav_metadata_type_list_all_info_strings) || (pMetadata->type == drwav_metadata_type_unknown && pMetadata->data.unknown.chunkLocation == drwav_metadata_location_inside_info_list)) { + hasListInfo = DRWAV_TRUE; + } + + if ((pMetadata->type & drwav_metadata_type_list_all_adtl) || (pMetadata->type == drwav_metadata_type_unknown && pMetadata->data.unknown.chunkLocation == drwav_metadata_location_inside_adtl_list)) { + hasListAdtl = DRWAV_TRUE; + } + + switch (pMetadata->type) { + case drwav_metadata_type_smpl: + { + drwav_uint32 iLoop; + + chunkSize = DRWAV_SMPL_BYTES + DRWAV_SMPL_LOOP_BYTES * pMetadata->data.smpl.sampleLoopCount + pMetadata->data.smpl.samplerSpecificDataSizeInBytes; + + bytesWritten += drwav__write_or_count(pWav, "smpl", 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, chunkSize); + + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.manufacturerId); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.productId); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.samplePeriodNanoseconds); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.midiUnityNote); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.midiPitchFraction); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.smpteFormat); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.smpteOffset); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.sampleLoopCount); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.samplerSpecificDataSizeInBytes); + + for (iLoop = 0; iLoop < pMetadata->data.smpl.sampleLoopCount; ++iLoop) { + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.pLoops[iLoop].cuePointId); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.pLoops[iLoop].type); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.pLoops[iLoop].firstSampleOffset); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.pLoops[iLoop].lastSampleOffset); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.pLoops[iLoop].sampleFraction); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.smpl.pLoops[iLoop].playCount); + } + + if (pMetadata->data.smpl.samplerSpecificDataSizeInBytes > 0) { + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.smpl.pSamplerSpecificData, pMetadata->data.smpl.samplerSpecificDataSizeInBytes); + } + } break; + + case drwav_metadata_type_inst: + { + chunkSize = DRWAV_INST_BYTES; + + bytesWritten += drwav__write_or_count(pWav, "inst", 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, chunkSize); + bytesWritten += drwav__write_or_count(pWav, &pMetadata->data.inst.midiUnityNote, 1); + bytesWritten += drwav__write_or_count(pWav, &pMetadata->data.inst.fineTuneCents, 1); + bytesWritten += drwav__write_or_count(pWav, &pMetadata->data.inst.gainDecibels, 1); + bytesWritten += drwav__write_or_count(pWav, &pMetadata->data.inst.lowNote, 1); + bytesWritten += drwav__write_or_count(pWav, &pMetadata->data.inst.highNote, 1); + bytesWritten += drwav__write_or_count(pWav, &pMetadata->data.inst.lowVelocity, 1); + bytesWritten += drwav__write_or_count(pWav, &pMetadata->data.inst.highVelocity, 1); + } break; + + case drwav_metadata_type_cue: + { + drwav_uint32 iCuePoint; + + chunkSize = DRWAV_CUE_BYTES + DRWAV_CUE_POINT_BYTES * pMetadata->data.cue.cuePointCount; + + bytesWritten += drwav__write_or_count(pWav, "cue ", 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, chunkSize); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.cue.cuePointCount); + for (iCuePoint = 0; iCuePoint < pMetadata->data.cue.cuePointCount; ++iCuePoint) { + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.cue.pCuePoints[iCuePoint].id); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.cue.pCuePoints[iCuePoint].playOrderPosition); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.cue.pCuePoints[iCuePoint].dataChunkId, 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.cue.pCuePoints[iCuePoint].chunkStart); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.cue.pCuePoints[iCuePoint].blockStart); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.cue.pCuePoints[iCuePoint].sampleOffset); + } + } break; + + case drwav_metadata_type_acid: + { + chunkSize = DRWAV_ACID_BYTES; + + bytesWritten += drwav__write_or_count(pWav, "acid", 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, chunkSize); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.acid.flags); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.acid.midiUnityNote); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.acid.reserved1); + bytesWritten += drwav__write_or_count_f32ne_to_le(pWav, pMetadata->data.acid.reserved2); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.acid.numBeats); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.acid.meterDenominator); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.acid.meterNumerator); + bytesWritten += drwav__write_or_count_f32ne_to_le(pWav, pMetadata->data.acid.tempo); + } break; + + case drwav_metadata_type_bext: + { + char reservedBuf[DRWAV_BEXT_RESERVED_BYTES]; + drwav_uint32 timeReferenceLow; + drwav_uint32 timeReferenceHigh; + + chunkSize = DRWAV_BEXT_BYTES + pMetadata->data.bext.codingHistorySize; + + bytesWritten += drwav__write_or_count(pWav, "bext", 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, chunkSize); + + bytesWritten += drwav__write_or_count_string_to_fixed_size_buf(pWav, pMetadata->data.bext.pDescription, DRWAV_BEXT_DESCRIPTION_BYTES); + bytesWritten += drwav__write_or_count_string_to_fixed_size_buf(pWav, pMetadata->data.bext.pOriginatorName, DRWAV_BEXT_ORIGINATOR_NAME_BYTES); + bytesWritten += drwav__write_or_count_string_to_fixed_size_buf(pWav, pMetadata->data.bext.pOriginatorReference, DRWAV_BEXT_ORIGINATOR_REF_BYTES); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.bext.pOriginationDate, sizeof(pMetadata->data.bext.pOriginationDate)); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.bext.pOriginationTime, sizeof(pMetadata->data.bext.pOriginationTime)); + + timeReferenceLow = (drwav_uint32)(pMetadata->data.bext.timeReference & 0xFFFFFFFF); + timeReferenceHigh = (drwav_uint32)(pMetadata->data.bext.timeReference >> 32); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, timeReferenceLow); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, timeReferenceHigh); + + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.bext.version); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.bext.pUMID, DRWAV_BEXT_UMID_BYTES); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.bext.loudnessValue); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.bext.loudnessRange); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.bext.maxTruePeakLevel); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.bext.maxMomentaryLoudness); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.bext.maxShortTermLoudness); + + DRWAV_ZERO_MEMORY(reservedBuf, sizeof(reservedBuf)); + bytesWritten += drwav__write_or_count(pWav, reservedBuf, sizeof(reservedBuf)); + + if (pMetadata->data.bext.codingHistorySize > 0) { + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.bext.pCodingHistory, pMetadata->data.bext.codingHistorySize); + } + } break; + + case drwav_metadata_type_unknown: + { + if (pMetadata->data.unknown.chunkLocation == drwav_metadata_location_top_level) { + chunkSize = pMetadata->data.unknown.dataSizeInBytes; + + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.unknown.id, 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, chunkSize); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.unknown.pData, pMetadata->data.unknown.dataSizeInBytes); + } + } break; + + default: break; + } + if ((chunkSize % 2) != 0) { + bytesWritten += drwav__write_or_count_byte(pWav, 0); + } + } + + if (hasListInfo) { + drwav_uint32 chunkSize = 4; /* Start with 4 bytes for "INFO". */ + for (iMetadata = 0; iMetadata < metadataCount; ++iMetadata) { + drwav_metadata* pMetadata = &pMetadatas[iMetadata]; + + if ((pMetadata->type & drwav_metadata_type_list_all_info_strings)) { + chunkSize += 8; /* For id and string size. */ + chunkSize += pMetadata->data.infoText.stringLength + 1; /* Include null terminator. */ + } else if (pMetadata->type == drwav_metadata_type_unknown && pMetadata->data.unknown.chunkLocation == drwav_metadata_location_inside_info_list) { + chunkSize += 8; /* For id string size. */ + chunkSize += pMetadata->data.unknown.dataSizeInBytes; + } + + if ((chunkSize % 2) != 0) { + chunkSize += 1; + } + } + + bytesWritten += drwav__write_or_count(pWav, "LIST", 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, chunkSize); + bytesWritten += drwav__write_or_count(pWav, "INFO", 4); + + for (iMetadata = 0; iMetadata < metadataCount; ++iMetadata) { + drwav_metadata* pMetadata = &pMetadatas[iMetadata]; + drwav_uint32 subchunkSize = 0; + + if (pMetadata->type & drwav_metadata_type_list_all_info_strings) { + const char* pID = NULL; + + switch (pMetadata->type) { + case drwav_metadata_type_list_info_software: pID = "ISFT"; break; + case drwav_metadata_type_list_info_copyright: pID = "ICOP"; break; + case drwav_metadata_type_list_info_title: pID = "INAM"; break; + case drwav_metadata_type_list_info_artist: pID = "IART"; break; + case drwav_metadata_type_list_info_comment: pID = "ICMT"; break; + case drwav_metadata_type_list_info_date: pID = "ICRD"; break; + case drwav_metadata_type_list_info_genre: pID = "IGNR"; break; + case drwav_metadata_type_list_info_album: pID = "IPRD"; break; + case drwav_metadata_type_list_info_tracknumber: pID = "ITRK"; break; + case drwav_metadata_type_list_info_location: pID = "IARL"; break; + case drwav_metadata_type_list_info_organization: pID = "ICMS"; break; + case drwav_metadata_type_list_info_keywords: pID = "IKEY"; break; + case drwav_metadata_type_list_info_medium: pID = "IMED"; break; + case drwav_metadata_type_list_info_description: pID = "ISBJ"; break; + default: break; + } + + DRWAV_ASSERT(pID != NULL); + + if (pMetadata->data.infoText.stringLength) { + subchunkSize = pMetadata->data.infoText.stringLength + 1; + bytesWritten += drwav__write_or_count(pWav, pID, 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, subchunkSize); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.infoText.pString, pMetadata->data.infoText.stringLength); + bytesWritten += drwav__write_or_count_byte(pWav, '\0'); + } + } else if (pMetadata->type == drwav_metadata_type_unknown && pMetadata->data.unknown.chunkLocation == drwav_metadata_location_inside_info_list) { + if (pMetadata->data.unknown.dataSizeInBytes) { + subchunkSize = pMetadata->data.unknown.dataSizeInBytes; + + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.unknown.id, 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.unknown.dataSizeInBytes); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.unknown.pData, subchunkSize); + } + } + + if ((subchunkSize % 2) != 0) { + bytesWritten += drwav__write_or_count_byte(pWav, 0); + } + } + } + + if (hasListAdtl) { + drwav_uint32 chunkSize = 4; /* start with 4 bytes for "adtl" */ + + for (iMetadata = 0; iMetadata < metadataCount; ++iMetadata) { + drwav_metadata* pMetadata = &pMetadatas[iMetadata]; + + switch (pMetadata->type) + { + case drwav_metadata_type_list_label: + case drwav_metadata_type_list_note: + { + chunkSize += 8; /* for id and chunk size */ + chunkSize += DRWAV_LIST_LABEL_OR_NOTE_BYTES; + + if (pMetadata->data.labelOrNote.stringLength > 0) { + chunkSize += pMetadata->data.labelOrNote.stringLength + 1; + } + } break; + + case drwav_metadata_type_list_labelled_cue_region: + { + chunkSize += 8; /* for id and chunk size */ + chunkSize += DRWAV_LIST_LABELLED_TEXT_BYTES; + + if (pMetadata->data.labelledCueRegion.stringLength > 0) { + chunkSize += pMetadata->data.labelledCueRegion.stringLength + 1; + } + } break; + + case drwav_metadata_type_unknown: + { + if (pMetadata->data.unknown.chunkLocation == drwav_metadata_location_inside_adtl_list) { + chunkSize += 8; /* for id and chunk size */ + chunkSize += pMetadata->data.unknown.dataSizeInBytes; + } + } break; + + default: break; + } + + if ((chunkSize % 2) != 0) { + chunkSize += 1; + } + } + + bytesWritten += drwav__write_or_count(pWav, "LIST", 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, chunkSize); + bytesWritten += drwav__write_or_count(pWav, "adtl", 4); + + for (iMetadata = 0; iMetadata < metadataCount; ++iMetadata) { + drwav_metadata* pMetadata = &pMetadatas[iMetadata]; + drwav_uint32 subchunkSize = 0; + + switch (pMetadata->type) + { + case drwav_metadata_type_list_label: + case drwav_metadata_type_list_note: + { + if (pMetadata->data.labelOrNote.stringLength > 0) { + const char *pID = NULL; + + if (pMetadata->type == drwav_metadata_type_list_label) { + pID = "labl"; + } + else if (pMetadata->type == drwav_metadata_type_list_note) { + pID = "note"; + } + + DRWAV_ASSERT(pID != NULL); + DRWAV_ASSERT(pMetadata->data.labelOrNote.pString != NULL); + + subchunkSize = DRWAV_LIST_LABEL_OR_NOTE_BYTES; + + bytesWritten += drwav__write_or_count(pWav, pID, 4); + subchunkSize += pMetadata->data.labelOrNote.stringLength + 1; + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, subchunkSize); + + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.labelOrNote.cuePointId); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.labelOrNote.pString, pMetadata->data.labelOrNote.stringLength); + bytesWritten += drwav__write_or_count_byte(pWav, '\0'); + } + } break; + + case drwav_metadata_type_list_labelled_cue_region: + { + subchunkSize = DRWAV_LIST_LABELLED_TEXT_BYTES; + + bytesWritten += drwav__write_or_count(pWav, "ltxt", 4); + if (pMetadata->data.labelledCueRegion.stringLength > 0) { + subchunkSize += pMetadata->data.labelledCueRegion.stringLength + 1; + } + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, subchunkSize); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.labelledCueRegion.cuePointId); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, pMetadata->data.labelledCueRegion.sampleLength); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.labelledCueRegion.purposeId, 4); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.labelledCueRegion.country); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.labelledCueRegion.language); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.labelledCueRegion.dialect); + bytesWritten += drwav__write_or_count_u16ne_to_le(pWav, pMetadata->data.labelledCueRegion.codePage); + + if (pMetadata->data.labelledCueRegion.stringLength > 0) { + DRWAV_ASSERT(pMetadata->data.labelledCueRegion.pString != NULL); + + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.labelledCueRegion.pString, pMetadata->data.labelledCueRegion.stringLength); + bytesWritten += drwav__write_or_count_byte(pWav, '\0'); + } + } break; + + case drwav_metadata_type_unknown: + { + if (pMetadata->data.unknown.chunkLocation == drwav_metadata_location_inside_adtl_list) { + subchunkSize = pMetadata->data.unknown.dataSizeInBytes; + + DRWAV_ASSERT(pMetadata->data.unknown.pData != NULL); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.unknown.id, 4); + bytesWritten += drwav__write_or_count_u32ne_to_le(pWav, subchunkSize); + bytesWritten += drwav__write_or_count(pWav, pMetadata->data.unknown.pData, subchunkSize); + } + } break; + + default: break; + } + + if ((subchunkSize % 2) != 0) { + bytesWritten += drwav__write_or_count_byte(pWav, 0); + } + } + } + + DRWAV_ASSERT((bytesWritten % 2) == 0); + + return bytesWritten; +} + +DRWAV_PRIVATE drwav_uint32 drwav__riff_chunk_size_riff(drwav_uint64 dataChunkSize, drwav_metadata* pMetadata, drwav_uint32 metadataCount) +{ + drwav_uint64 chunkSize = 4 + 24 + (drwav_uint64)drwav__write_or_count_metadata(NULL, pMetadata, metadataCount) + 8 + dataChunkSize + drwav__chunk_padding_size_riff(dataChunkSize); /* 4 = "WAVE". 24 = "fmt " chunk. 8 = "data" + u32 data size. */ + if (chunkSize > 0xFFFFFFFFUL) { + chunkSize = 0xFFFFFFFFUL; + } + + return (drwav_uint32)chunkSize; /* Safe cast due to the clamp above. */ +} + +DRWAV_PRIVATE drwav_uint32 drwav__data_chunk_size_riff(drwav_uint64 dataChunkSize) +{ + if (dataChunkSize <= 0xFFFFFFFFUL) { + return (drwav_uint32)dataChunkSize; + } else { + return 0xFFFFFFFFUL; + } +} + +DRWAV_PRIVATE drwav_uint64 drwav__riff_chunk_size_w64(drwav_uint64 dataChunkSize) +{ + drwav_uint64 dataSubchunkPaddingSize = drwav__chunk_padding_size_w64(dataChunkSize); + + return 80 + 24 + dataChunkSize + dataSubchunkPaddingSize; /* +24 because W64 includes the size of the GUID and size fields. */ +} + +DRWAV_PRIVATE drwav_uint64 drwav__data_chunk_size_w64(drwav_uint64 dataChunkSize) +{ + return 24 + dataChunkSize; /* +24 because W64 includes the size of the GUID and size fields. */ +} + +DRWAV_PRIVATE drwav_uint64 drwav__riff_chunk_size_rf64(drwav_uint64 dataChunkSize, drwav_metadata *metadata, drwav_uint32 numMetadata) +{ + drwav_uint64 chunkSize = 4 + 36 + 24 + (drwav_uint64)drwav__write_or_count_metadata(NULL, metadata, numMetadata) + 8 + dataChunkSize + drwav__chunk_padding_size_riff(dataChunkSize); /* 4 = "WAVE". 36 = "ds64" chunk. 24 = "fmt " chunk. 8 = "data" + u32 data size. */ + if (chunkSize > 0xFFFFFFFFUL) { + chunkSize = 0xFFFFFFFFUL; + } + + return chunkSize; +} + +DRWAV_PRIVATE drwav_uint64 drwav__data_chunk_size_rf64(drwav_uint64 dataChunkSize) +{ + return dataChunkSize; +} + + + +DRWAV_PRIVATE drwav_bool32 drwav_preinit_write(drwav* pWav, const drwav_data_format* pFormat, drwav_bool32 isSequential, drwav_write_proc onWrite, drwav_seek_proc onSeek, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pWav == NULL || onWrite == NULL) { + return DRWAV_FALSE; + } + + if (!isSequential && onSeek == NULL) { + return DRWAV_FALSE; /* <-- onSeek is required when in non-sequential mode. */ + } + + /* Not currently supporting compressed formats. Will need to add support for the "fact" chunk before we enable this. */ + if (pFormat->format == DR_WAVE_FORMAT_EXTENSIBLE) { + return DRWAV_FALSE; + } + if (pFormat->format == DR_WAVE_FORMAT_ADPCM || pFormat->format == DR_WAVE_FORMAT_DVI_ADPCM) { + return DRWAV_FALSE; + } + + DRWAV_ZERO_MEMORY(pWav, sizeof(*pWav)); + pWav->onWrite = onWrite; + pWav->onSeek = onSeek; + pWav->pUserData = pUserData; + pWav->allocationCallbacks = drwav_copy_allocation_callbacks_or_defaults(pAllocationCallbacks); + + if (pWav->allocationCallbacks.onFree == NULL || (pWav->allocationCallbacks.onMalloc == NULL && pWav->allocationCallbacks.onRealloc == NULL)) { + return DRWAV_FALSE; /* Invalid allocation callbacks. */ + } + + pWav->fmt.formatTag = (drwav_uint16)pFormat->format; + pWav->fmt.channels = (drwav_uint16)pFormat->channels; + pWav->fmt.sampleRate = pFormat->sampleRate; + pWav->fmt.avgBytesPerSec = (drwav_uint32)((pFormat->bitsPerSample * pFormat->sampleRate * pFormat->channels) / 8); + pWav->fmt.blockAlign = (drwav_uint16)((pFormat->channels * pFormat->bitsPerSample) / 8); + pWav->fmt.bitsPerSample = (drwav_uint16)pFormat->bitsPerSample; + pWav->fmt.extendedSize = 0; + pWav->isSequentialWrite = isSequential; + + return DRWAV_TRUE; +} + + +DRWAV_PRIVATE drwav_bool32 drwav_init_write__internal(drwav* pWav, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount) +{ + /* The function assumes drwav_preinit_write() was called beforehand. */ + + size_t runningPos = 0; + drwav_uint64 initialDataChunkSize = 0; + drwav_uint64 chunkSizeFMT; + + /* + The initial values for the "RIFF" and "data" chunks depends on whether or not we are initializing in sequential mode or not. In + sequential mode we set this to its final values straight away since they can be calculated from the total sample count. In non- + sequential mode we initialize it all to zero and fill it out in drwav_uninit() using a backwards seek. + */ + if (pWav->isSequentialWrite) { + initialDataChunkSize = (totalSampleCount * pWav->fmt.bitsPerSample) / 8; + + /* + The RIFF container has a limit on the number of samples. drwav is not allowing this. There's no practical limits for Wave64 + so for the sake of simplicity I'm not doing any validation for that. + */ + if (pFormat->container == drwav_container_riff) { + if (initialDataChunkSize > (0xFFFFFFFFUL - 36)) { + return DRWAV_FALSE; /* Not enough room to store every sample. */ + } + } + } + + pWav->dataChunkDataSizeTargetWrite = initialDataChunkSize; + + + /* "RIFF" chunk. */ + if (pFormat->container == drwav_container_riff) { + drwav_uint32 chunkSizeRIFF = 36 + (drwav_uint32)initialDataChunkSize; /* +36 = "WAVE" + [sizeof "fmt " chunk] + [data chunk header] */ + runningPos += drwav__write(pWav, "RIFF", 4); + runningPos += drwav__write_u32ne_to_le(pWav, chunkSizeRIFF); + runningPos += drwav__write(pWav, "WAVE", 4); + } else if (pFormat->container == drwav_container_w64) { + drwav_uint64 chunkSizeRIFF = 80 + 24 + initialDataChunkSize; /* +24 because W64 includes the size of the GUID and size fields. */ + runningPos += drwav__write(pWav, drwavGUID_W64_RIFF, 16); + runningPos += drwav__write_u64ne_to_le(pWav, chunkSizeRIFF); + runningPos += drwav__write(pWav, drwavGUID_W64_WAVE, 16); + } else if (pFormat->container == drwav_container_rf64) { + runningPos += drwav__write(pWav, "RF64", 4); + runningPos += drwav__write_u32ne_to_le(pWav, 0xFFFFFFFF); /* Always 0xFFFFFFFF for RF64. Set to a proper value in the "ds64" chunk. */ + runningPos += drwav__write(pWav, "WAVE", 4); + } else { + return DRWAV_FALSE; /* Container not supported for writing. */ + } + + + /* "ds64" chunk (RF64 only). */ + if (pFormat->container == drwav_container_rf64) { + drwav_uint32 initialds64ChunkSize = 28; /* 28 = [Size of RIFF (8 bytes)] + [Size of DATA (8 bytes)] + [Sample Count (8 bytes)] + [Table Length (4 bytes)]. Table length always set to 0. */ + drwav_uint64 initialRiffChunkSize = 8 + initialds64ChunkSize + initialDataChunkSize; /* +8 for the ds64 header. */ + + runningPos += drwav__write(pWav, "ds64", 4); + runningPos += drwav__write_u32ne_to_le(pWav, initialds64ChunkSize); /* Size of ds64. */ + runningPos += drwav__write_u64ne_to_le(pWav, initialRiffChunkSize); /* Size of RIFF. Set to true value at the end. */ + runningPos += drwav__write_u64ne_to_le(pWav, initialDataChunkSize); /* Size of DATA. Set to true value at the end. */ + runningPos += drwav__write_u64ne_to_le(pWav, totalSampleCount); /* Sample count. */ + runningPos += drwav__write_u32ne_to_le(pWav, 0); /* Table length. Always set to zero in our case since we're not doing any other chunks than "DATA". */ + } + + + /* "fmt " chunk. */ + if (pFormat->container == drwav_container_riff || pFormat->container == drwav_container_rf64) { + chunkSizeFMT = 16; + runningPos += drwav__write(pWav, "fmt ", 4); + runningPos += drwav__write_u32ne_to_le(pWav, (drwav_uint32)chunkSizeFMT); + } else if (pFormat->container == drwav_container_w64) { + chunkSizeFMT = 40; + runningPos += drwav__write(pWav, drwavGUID_W64_FMT, 16); + runningPos += drwav__write_u64ne_to_le(pWav, chunkSizeFMT); + } + + runningPos += drwav__write_u16ne_to_le(pWav, pWav->fmt.formatTag); + runningPos += drwav__write_u16ne_to_le(pWav, pWav->fmt.channels); + runningPos += drwav__write_u32ne_to_le(pWav, pWav->fmt.sampleRate); + runningPos += drwav__write_u32ne_to_le(pWav, pWav->fmt.avgBytesPerSec); + runningPos += drwav__write_u16ne_to_le(pWav, pWav->fmt.blockAlign); + runningPos += drwav__write_u16ne_to_le(pWav, pWav->fmt.bitsPerSample); + + /* TODO: is a 'fact' chunk required for DR_WAVE_FORMAT_IEEE_FLOAT? */ + + if (!pWav->isSequentialWrite && pWav->pMetadata != NULL && pWav->metadataCount > 0 && (pFormat->container == drwav_container_riff || pFormat->container == drwav_container_rf64)) { + runningPos += drwav__write_or_count_metadata(pWav, pWav->pMetadata, pWav->metadataCount); + } + + pWav->dataChunkDataPos = runningPos; + + /* "data" chunk. */ + if (pFormat->container == drwav_container_riff) { + drwav_uint32 chunkSizeDATA = (drwav_uint32)initialDataChunkSize; + runningPos += drwav__write(pWav, "data", 4); + runningPos += drwav__write_u32ne_to_le(pWav, chunkSizeDATA); + } else if (pFormat->container == drwav_container_w64) { + drwav_uint64 chunkSizeDATA = 24 + initialDataChunkSize; /* +24 because W64 includes the size of the GUID and size fields. */ + runningPos += drwav__write(pWav, drwavGUID_W64_DATA, 16); + runningPos += drwav__write_u64ne_to_le(pWav, chunkSizeDATA); + } else if (pFormat->container == drwav_container_rf64) { + runningPos += drwav__write(pWav, "data", 4); + runningPos += drwav__write_u32ne_to_le(pWav, 0xFFFFFFFF); /* Always set to 0xFFFFFFFF for RF64. The true size of the data chunk is specified in the ds64 chunk. */ + } + + /* Set some properties for the client's convenience. */ + pWav->container = pFormat->container; + pWav->channels = (drwav_uint16)pFormat->channels; + pWav->sampleRate = pFormat->sampleRate; + pWav->bitsPerSample = (drwav_uint16)pFormat->bitsPerSample; + pWav->translatedFormatTag = (drwav_uint16)pFormat->format; + pWav->dataChunkDataPos = runningPos; + + return DRWAV_TRUE; +} + + +DRWAV_API drwav_bool32 drwav_init_write(drwav* pWav, const drwav_data_format* pFormat, drwav_write_proc onWrite, drwav_seek_proc onSeek, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (!drwav_preinit_write(pWav, pFormat, DRWAV_FALSE, onWrite, onSeek, pUserData, pAllocationCallbacks)) { + return DRWAV_FALSE; + } + + return drwav_init_write__internal(pWav, pFormat, 0); /* DRWAV_FALSE = Not Sequential */ +} + +DRWAV_API drwav_bool32 drwav_init_write_sequential(drwav* pWav, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, drwav_write_proc onWrite, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (!drwav_preinit_write(pWav, pFormat, DRWAV_TRUE, onWrite, NULL, pUserData, pAllocationCallbacks)) { + return DRWAV_FALSE; + } + + return drwav_init_write__internal(pWav, pFormat, totalSampleCount); /* DRWAV_TRUE = Sequential */ +} + +DRWAV_API drwav_bool32 drwav_init_write_sequential_pcm_frames(drwav* pWav, const drwav_data_format* pFormat, drwav_uint64 totalPCMFrameCount, drwav_write_proc onWrite, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pFormat == NULL) { + return DRWAV_FALSE; + } + + return drwav_init_write_sequential(pWav, pFormat, totalPCMFrameCount*pFormat->channels, onWrite, pUserData, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_write_with_metadata(drwav* pWav, const drwav_data_format* pFormat, drwav_write_proc onWrite, drwav_seek_proc onSeek, void* pUserData, const drwav_allocation_callbacks* pAllocationCallbacks, drwav_metadata* pMetadata, drwav_uint32 metadataCount) +{ + if (!drwav_preinit_write(pWav, pFormat, DRWAV_FALSE, onWrite, onSeek, pUserData, pAllocationCallbacks)) { + return DRWAV_FALSE; + } + + pWav->pMetadata = pMetadata; + pWav->metadataCount = metadataCount; + + return drwav_init_write__internal(pWav, pFormat, 0); +} + + +DRWAV_API drwav_uint64 drwav_target_write_size_bytes(const drwav_data_format* pFormat, drwav_uint64 totalFrameCount, drwav_metadata* pMetadata, drwav_uint32 metadataCount) +{ + /* Casting totalFrameCount to drwav_int64 for VC6 compatibility. No issues in practice because nobody is going to exhaust the whole 63 bits. */ + drwav_uint64 targetDataSizeBytes = (drwav_uint64)((drwav_int64)totalFrameCount * pFormat->channels * pFormat->bitsPerSample/8.0); + drwav_uint64 riffChunkSizeBytes; + drwav_uint64 fileSizeBytes = 0; + + if (pFormat->container == drwav_container_riff) { + riffChunkSizeBytes = drwav__riff_chunk_size_riff(targetDataSizeBytes, pMetadata, metadataCount); + fileSizeBytes = (8 + riffChunkSizeBytes); /* +8 because WAV doesn't include the size of the ChunkID and ChunkSize fields. */ + } else if (pFormat->container == drwav_container_w64) { + riffChunkSizeBytes = drwav__riff_chunk_size_w64(targetDataSizeBytes); + fileSizeBytes = riffChunkSizeBytes; + } else if (pFormat->container == drwav_container_rf64) { + riffChunkSizeBytes = drwav__riff_chunk_size_rf64(targetDataSizeBytes, pMetadata, metadataCount); + fileSizeBytes = (8 + riffChunkSizeBytes); /* +8 because WAV doesn't include the size of the ChunkID and ChunkSize fields. */ + } + + return fileSizeBytes; +} + + +#ifndef DR_WAV_NO_STDIO + +/* Errno */ +/* drwav_result_from_errno() is only used for fopen() and wfopen() so putting it inside DR_WAV_NO_STDIO for now. If something else needs this later we can move it out. */ +#include <errno.h> +DRWAV_PRIVATE drwav_result drwav_result_from_errno(int e) +{ + switch (e) + { + case 0: return DRWAV_SUCCESS; + #ifdef EPERM + case EPERM: return DRWAV_INVALID_OPERATION; + #endif + #ifdef ENOENT + case ENOENT: return DRWAV_DOES_NOT_EXIST; + #endif + #ifdef ESRCH + case ESRCH: return DRWAV_DOES_NOT_EXIST; + #endif + #ifdef EINTR + case EINTR: return DRWAV_INTERRUPT; + #endif + #ifdef EIO + case EIO: return DRWAV_IO_ERROR; + #endif + #ifdef ENXIO + case ENXIO: return DRWAV_DOES_NOT_EXIST; + #endif + #ifdef E2BIG + case E2BIG: return DRWAV_INVALID_ARGS; + #endif + #ifdef ENOEXEC + case ENOEXEC: return DRWAV_INVALID_FILE; + #endif + #ifdef EBADF + case EBADF: return DRWAV_INVALID_FILE; + #endif + #ifdef ECHILD + case ECHILD: return DRWAV_ERROR; + #endif + #ifdef EAGAIN + case EAGAIN: return DRWAV_UNAVAILABLE; + #endif + #ifdef ENOMEM + case ENOMEM: return DRWAV_OUT_OF_MEMORY; + #endif + #ifdef EACCES + case EACCES: return DRWAV_ACCESS_DENIED; + #endif + #ifdef EFAULT + case EFAULT: return DRWAV_BAD_ADDRESS; + #endif + #ifdef ENOTBLK + case ENOTBLK: return DRWAV_ERROR; + #endif + #ifdef EBUSY + case EBUSY: return DRWAV_BUSY; + #endif + #ifdef EEXIST + case EEXIST: return DRWAV_ALREADY_EXISTS; + #endif + #ifdef EXDEV + case EXDEV: return DRWAV_ERROR; + #endif + #ifdef ENODEV + case ENODEV: return DRWAV_DOES_NOT_EXIST; + #endif + #ifdef ENOTDIR + case ENOTDIR: return DRWAV_NOT_DIRECTORY; + #endif + #ifdef EISDIR + case EISDIR: return DRWAV_IS_DIRECTORY; + #endif + #ifdef EINVAL + case EINVAL: return DRWAV_INVALID_ARGS; + #endif + #ifdef ENFILE + case ENFILE: return DRWAV_TOO_MANY_OPEN_FILES; + #endif + #ifdef EMFILE + case EMFILE: return DRWAV_TOO_MANY_OPEN_FILES; + #endif + #ifdef ENOTTY + case ENOTTY: return DRWAV_INVALID_OPERATION; + #endif + #ifdef ETXTBSY + case ETXTBSY: return DRWAV_BUSY; + #endif + #ifdef EFBIG + case EFBIG: return DRWAV_TOO_BIG; + #endif + #ifdef ENOSPC + case ENOSPC: return DRWAV_NO_SPACE; + #endif + #ifdef ESPIPE + case ESPIPE: return DRWAV_BAD_SEEK; + #endif + #ifdef EROFS + case EROFS: return DRWAV_ACCESS_DENIED; + #endif + #ifdef EMLINK + case EMLINK: return DRWAV_TOO_MANY_LINKS; + #endif + #ifdef EPIPE + case EPIPE: return DRWAV_BAD_PIPE; + #endif + #ifdef EDOM + case EDOM: return DRWAV_OUT_OF_RANGE; + #endif + #ifdef ERANGE + case ERANGE: return DRWAV_OUT_OF_RANGE; + #endif + #ifdef EDEADLK + case EDEADLK: return DRWAV_DEADLOCK; + #endif + #ifdef ENAMETOOLONG + case ENAMETOOLONG: return DRWAV_PATH_TOO_LONG; + #endif + #ifdef ENOLCK + case ENOLCK: return DRWAV_ERROR; + #endif + #ifdef ENOSYS + case ENOSYS: return DRWAV_NOT_IMPLEMENTED; + #endif + #if defined(ENOTEMPTY) && ENOTEMPTY != EEXIST /* In AIX, ENOTEMPTY and EEXIST use the same value. */ + case ENOTEMPTY: return DRWAV_DIRECTORY_NOT_EMPTY; + #endif + #ifdef ELOOP + case ELOOP: return DRWAV_TOO_MANY_LINKS; + #endif + #ifdef ENOMSG + case ENOMSG: return DRWAV_NO_MESSAGE; + #endif + #ifdef EIDRM + case EIDRM: return DRWAV_ERROR; + #endif + #ifdef ECHRNG + case ECHRNG: return DRWAV_ERROR; + #endif + #ifdef EL2NSYNC + case EL2NSYNC: return DRWAV_ERROR; + #endif + #ifdef EL3HLT + case EL3HLT: return DRWAV_ERROR; + #endif + #ifdef EL3RST + case EL3RST: return DRWAV_ERROR; + #endif + #ifdef ELNRNG + case ELNRNG: return DRWAV_OUT_OF_RANGE; + #endif + #ifdef EUNATCH + case EUNATCH: return DRWAV_ERROR; + #endif + #ifdef ENOCSI + case ENOCSI: return DRWAV_ERROR; + #endif + #ifdef EL2HLT + case EL2HLT: return DRWAV_ERROR; + #endif + #ifdef EBADE + case EBADE: return DRWAV_ERROR; + #endif + #ifdef EBADR + case EBADR: return DRWAV_ERROR; + #endif + #ifdef EXFULL + case EXFULL: return DRWAV_ERROR; + #endif + #ifdef ENOANO + case ENOANO: return DRWAV_ERROR; + #endif + #ifdef EBADRQC + case EBADRQC: return DRWAV_ERROR; + #endif + #ifdef EBADSLT + case EBADSLT: return DRWAV_ERROR; + #endif + #ifdef EBFONT + case EBFONT: return DRWAV_INVALID_FILE; + #endif + #ifdef ENOSTR + case ENOSTR: return DRWAV_ERROR; + #endif + #ifdef ENODATA + case ENODATA: return DRWAV_NO_DATA_AVAILABLE; + #endif + #ifdef ETIME + case ETIME: return DRWAV_TIMEOUT; + #endif + #ifdef ENOSR + case ENOSR: return DRWAV_NO_DATA_AVAILABLE; + #endif + #ifdef ENONET + case ENONET: return DRWAV_NO_NETWORK; + #endif + #ifdef ENOPKG + case ENOPKG: return DRWAV_ERROR; + #endif + #ifdef EREMOTE + case EREMOTE: return DRWAV_ERROR; + #endif + #ifdef ENOLINK + case ENOLINK: return DRWAV_ERROR; + #endif + #ifdef EADV + case EADV: return DRWAV_ERROR; + #endif + #ifdef ESRMNT + case ESRMNT: return DRWAV_ERROR; + #endif + #ifdef ECOMM + case ECOMM: return DRWAV_ERROR; + #endif + #ifdef EPROTO + case EPROTO: return DRWAV_ERROR; + #endif + #ifdef EMULTIHOP + case EMULTIHOP: return DRWAV_ERROR; + #endif + #ifdef EDOTDOT + case EDOTDOT: return DRWAV_ERROR; + #endif + #ifdef EBADMSG + case EBADMSG: return DRWAV_BAD_MESSAGE; + #endif + #ifdef EOVERFLOW + case EOVERFLOW: return DRWAV_TOO_BIG; + #endif + #ifdef ENOTUNIQ + case ENOTUNIQ: return DRWAV_NOT_UNIQUE; + #endif + #ifdef EBADFD + case EBADFD: return DRWAV_ERROR; + #endif + #ifdef EREMCHG + case EREMCHG: return DRWAV_ERROR; + #endif + #ifdef ELIBACC + case ELIBACC: return DRWAV_ACCESS_DENIED; + #endif + #ifdef ELIBBAD + case ELIBBAD: return DRWAV_INVALID_FILE; + #endif + #ifdef ELIBSCN + case ELIBSCN: return DRWAV_INVALID_FILE; + #endif + #ifdef ELIBMAX + case ELIBMAX: return DRWAV_ERROR; + #endif + #ifdef ELIBEXEC + case ELIBEXEC: return DRWAV_ERROR; + #endif + #ifdef EILSEQ + case EILSEQ: return DRWAV_INVALID_DATA; + #endif + #ifdef ERESTART + case ERESTART: return DRWAV_ERROR; + #endif + #ifdef ESTRPIPE + case ESTRPIPE: return DRWAV_ERROR; + #endif + #ifdef EUSERS + case EUSERS: return DRWAV_ERROR; + #endif + #ifdef ENOTSOCK + case ENOTSOCK: return DRWAV_NOT_SOCKET; + #endif + #ifdef EDESTADDRREQ + case EDESTADDRREQ: return DRWAV_NO_ADDRESS; + #endif + #ifdef EMSGSIZE + case EMSGSIZE: return DRWAV_TOO_BIG; + #endif + #ifdef EPROTOTYPE + case EPROTOTYPE: return DRWAV_BAD_PROTOCOL; + #endif + #ifdef ENOPROTOOPT + case ENOPROTOOPT: return DRWAV_PROTOCOL_UNAVAILABLE; + #endif + #ifdef EPROTONOSUPPORT + case EPROTONOSUPPORT: return DRWAV_PROTOCOL_NOT_SUPPORTED; + #endif + #ifdef ESOCKTNOSUPPORT + case ESOCKTNOSUPPORT: return DRWAV_SOCKET_NOT_SUPPORTED; + #endif + #ifdef EOPNOTSUPP + case EOPNOTSUPP: return DRWAV_INVALID_OPERATION; + #endif + #ifdef EPFNOSUPPORT + case EPFNOSUPPORT: return DRWAV_PROTOCOL_FAMILY_NOT_SUPPORTED; + #endif + #ifdef EAFNOSUPPORT + case EAFNOSUPPORT: return DRWAV_ADDRESS_FAMILY_NOT_SUPPORTED; + #endif + #ifdef EADDRINUSE + case EADDRINUSE: return DRWAV_ALREADY_IN_USE; + #endif + #ifdef EADDRNOTAVAIL + case EADDRNOTAVAIL: return DRWAV_ERROR; + #endif + #ifdef ENETDOWN + case ENETDOWN: return DRWAV_NO_NETWORK; + #endif + #ifdef ENETUNREACH + case ENETUNREACH: return DRWAV_NO_NETWORK; + #endif + #ifdef ENETRESET + case ENETRESET: return DRWAV_NO_NETWORK; + #endif + #ifdef ECONNABORTED + case ECONNABORTED: return DRWAV_NO_NETWORK; + #endif + #ifdef ECONNRESET + case ECONNRESET: return DRWAV_CONNECTION_RESET; + #endif + #ifdef ENOBUFS + case ENOBUFS: return DRWAV_NO_SPACE; + #endif + #ifdef EISCONN + case EISCONN: return DRWAV_ALREADY_CONNECTED; + #endif + #ifdef ENOTCONN + case ENOTCONN: return DRWAV_NOT_CONNECTED; + #endif + #ifdef ESHUTDOWN + case ESHUTDOWN: return DRWAV_ERROR; + #endif + #ifdef ETOOMANYREFS + case ETOOMANYREFS: return DRWAV_ERROR; + #endif + #ifdef ETIMEDOUT + case ETIMEDOUT: return DRWAV_TIMEOUT; + #endif + #ifdef ECONNREFUSED + case ECONNREFUSED: return DRWAV_CONNECTION_REFUSED; + #endif + #ifdef EHOSTDOWN + case EHOSTDOWN: return DRWAV_NO_HOST; + #endif + #ifdef EHOSTUNREACH + case EHOSTUNREACH: return DRWAV_NO_HOST; + #endif + #ifdef EALREADY + case EALREADY: return DRWAV_IN_PROGRESS; + #endif + #ifdef EINPROGRESS + case EINPROGRESS: return DRWAV_IN_PROGRESS; + #endif + #ifdef ESTALE + case ESTALE: return DRWAV_INVALID_FILE; + #endif + #ifdef EUCLEAN + case EUCLEAN: return DRWAV_ERROR; + #endif + #ifdef ENOTNAM + case ENOTNAM: return DRWAV_ERROR; + #endif + #ifdef ENAVAIL + case ENAVAIL: return DRWAV_ERROR; + #endif + #ifdef EISNAM + case EISNAM: return DRWAV_ERROR; + #endif + #ifdef EREMOTEIO + case EREMOTEIO: return DRWAV_IO_ERROR; + #endif + #ifdef EDQUOT + case EDQUOT: return DRWAV_NO_SPACE; + #endif + #ifdef ENOMEDIUM + case ENOMEDIUM: return DRWAV_DOES_NOT_EXIST; + #endif + #ifdef EMEDIUMTYPE + case EMEDIUMTYPE: return DRWAV_ERROR; + #endif + #ifdef ECANCELED + case ECANCELED: return DRWAV_CANCELLED; + #endif + #ifdef ENOKEY + case ENOKEY: return DRWAV_ERROR; + #endif + #ifdef EKEYEXPIRED + case EKEYEXPIRED: return DRWAV_ERROR; + #endif + #ifdef EKEYREVOKED + case EKEYREVOKED: return DRWAV_ERROR; + #endif + #ifdef EKEYREJECTED + case EKEYREJECTED: return DRWAV_ERROR; + #endif + #ifdef EOWNERDEAD + case EOWNERDEAD: return DRWAV_ERROR; + #endif + #ifdef ENOTRECOVERABLE + case ENOTRECOVERABLE: return DRWAV_ERROR; + #endif + #ifdef ERFKILL + case ERFKILL: return DRWAV_ERROR; + #endif + #ifdef EHWPOISON + case EHWPOISON: return DRWAV_ERROR; + #endif + default: return DRWAV_ERROR; + } +} +/* End Errno */ + +/* fopen */ +DRWAV_PRIVATE drwav_result drwav_fopen(FILE** ppFile, const char* pFilePath, const char* pOpenMode) +{ +#if defined(_MSC_VER) && _MSC_VER >= 1400 + errno_t err; +#endif + + if (ppFile != NULL) { + *ppFile = NULL; /* Safety. */ + } + + if (pFilePath == NULL || pOpenMode == NULL || ppFile == NULL) { + return DRWAV_INVALID_ARGS; + } + +#if defined(_MSC_VER) && _MSC_VER >= 1400 + err = fopen_s(ppFile, pFilePath, pOpenMode); + if (err != 0) { + return drwav_result_from_errno(err); + } +#else +#if defined(_WIN32) || defined(__APPLE__) + *ppFile = fopen(pFilePath, pOpenMode); +#else + #if defined(_FILE_OFFSET_BITS) && _FILE_OFFSET_BITS == 64 && defined(_LARGEFILE64_SOURCE) + *ppFile = fopen64(pFilePath, pOpenMode); + #else + *ppFile = fopen(pFilePath, pOpenMode); + #endif +#endif + if (*ppFile == NULL) { + drwav_result result = drwav_result_from_errno(errno); + if (result == DRWAV_SUCCESS) { + result = DRWAV_ERROR; /* Just a safety check to make sure we never ever return success when pFile == NULL. */ + } + + return result; + } +#endif + + return DRWAV_SUCCESS; +} + +/* +_wfopen() isn't always available in all compilation environments. + + * Windows only. + * MSVC seems to support it universally as far back as VC6 from what I can tell (haven't checked further back). + * MinGW-64 (both 32- and 64-bit) seems to support it. + * MinGW wraps it in !defined(__STRICT_ANSI__). + * OpenWatcom wraps it in !defined(_NO_EXT_KEYS). + +This can be reviewed as compatibility issues arise. The preference is to use _wfopen_s() and _wfopen() as opposed to the wcsrtombs() +fallback, so if you notice your compiler not detecting this properly I'm happy to look at adding support. +*/ +#if defined(_WIN32) + #if defined(_MSC_VER) || defined(__MINGW64__) || (!defined(__STRICT_ANSI__) && !defined(_NO_EXT_KEYS)) + #define DRWAV_HAS_WFOPEN + #endif +#endif + +#ifndef DR_WAV_NO_WCHAR +DRWAV_PRIVATE drwav_result drwav_wfopen(FILE** ppFile, const wchar_t* pFilePath, const wchar_t* pOpenMode, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (ppFile != NULL) { + *ppFile = NULL; /* Safety. */ + } + + if (pFilePath == NULL || pOpenMode == NULL || ppFile == NULL) { + return DRWAV_INVALID_ARGS; + } + +#if defined(DRWAV_HAS_WFOPEN) + { + /* Use _wfopen() on Windows. */ + #if defined(_MSC_VER) && _MSC_VER >= 1400 + errno_t err = _wfopen_s(ppFile, pFilePath, pOpenMode); + if (err != 0) { + return drwav_result_from_errno(err); + } + #else + *ppFile = _wfopen(pFilePath, pOpenMode); + if (*ppFile == NULL) { + return drwav_result_from_errno(errno); + } + #endif + (void)pAllocationCallbacks; + } +#else + /* + Use fopen() on anything other than Windows. Requires a conversion. This is annoying because + fopen() is locale specific. The only real way I can think of to do this is with wcsrtombs(). Note + that wcstombs() is apparently not thread-safe because it uses a static global mbstate_t object for + maintaining state. I've checked this with -std=c89 and it works, but if somebody get's a compiler + error I'll look into improving compatibility. + */ + + /* + Some compilers don't support wchar_t or wcsrtombs() which we're using below. In this case we just + need to abort with an error. If you encounter a compiler lacking such support, add it to this list + and submit a bug report and it'll be added to the library upstream. + */ + #if defined(__DJGPP__) + { + /* Nothing to do here. This will fall through to the error check below. */ + } + #else + { + mbstate_t mbs; + size_t lenMB; + const wchar_t* pFilePathTemp = pFilePath; + char* pFilePathMB = NULL; + char pOpenModeMB[32] = {0}; + + /* Get the length first. */ + DRWAV_ZERO_OBJECT(&mbs); + lenMB = wcsrtombs(NULL, &pFilePathTemp, 0, &mbs); + if (lenMB == (size_t)-1) { + return drwav_result_from_errno(errno); + } + + pFilePathMB = (char*)drwav__malloc_from_callbacks(lenMB + 1, pAllocationCallbacks); + if (pFilePathMB == NULL) { + return DRWAV_OUT_OF_MEMORY; + } + + pFilePathTemp = pFilePath; + DRWAV_ZERO_OBJECT(&mbs); + wcsrtombs(pFilePathMB, &pFilePathTemp, lenMB + 1, &mbs); + + /* The open mode should always consist of ASCII characters so we should be able to do a trivial conversion. */ + { + size_t i = 0; + for (;;) { + if (pOpenMode[i] == 0) { + pOpenModeMB[i] = '\0'; + break; + } + + pOpenModeMB[i] = (char)pOpenMode[i]; + i += 1; + } + } + + *ppFile = fopen(pFilePathMB, pOpenModeMB); + + drwav__free_from_callbacks(pFilePathMB, pAllocationCallbacks); + } + #endif + + if (*ppFile == NULL) { + return DRWAV_ERROR; + } +#endif + + return DRWAV_SUCCESS; +} +#endif +/* End fopen */ + + +DRWAV_PRIVATE size_t drwav__on_read_stdio(void* pUserData, void* pBufferOut, size_t bytesToRead) +{ + return fread(pBufferOut, 1, bytesToRead, (FILE*)pUserData); +} + +DRWAV_PRIVATE size_t drwav__on_write_stdio(void* pUserData, const void* pData, size_t bytesToWrite) +{ + return fwrite(pData, 1, bytesToWrite, (FILE*)pUserData); +} + +DRWAV_PRIVATE drwav_bool32 drwav__on_seek_stdio(void* pUserData, int offset, drwav_seek_origin origin) +{ + int whence = SEEK_SET; + if (origin == DRWAV_SEEK_CUR) { + whence = SEEK_CUR; + } else if (origin == DRWAV_SEEK_END) { + whence = SEEK_END; + } + + return fseek((FILE*)pUserData, offset, whence) == 0; +} + +DRWAV_PRIVATE drwav_bool32 drwav__on_tell_stdio(void* pUserData, drwav_int64* pCursor) +{ + FILE* pFileStdio = (FILE*)pUserData; + drwav_int64 result; + + /* These were all validated at a higher level. */ + DRWAV_ASSERT(pFileStdio != NULL); + DRWAV_ASSERT(pCursor != NULL); + +#if defined(_WIN32) && !defined(NXDK) + #if defined(_MSC_VER) && _MSC_VER > 1200 + result = _ftelli64(pFileStdio); + #else + result = ftell(pFileStdio); + #endif +#else + result = ftell(pFileStdio); +#endif + + *pCursor = result; + + return DRWAV_TRUE; +} + +DRWAV_API drwav_bool32 drwav_init_file(drwav* pWav, const char* filename, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_file_ex(pWav, filename, NULL, NULL, 0, pAllocationCallbacks); +} + + +DRWAV_PRIVATE drwav_bool32 drwav_init_file__internal_FILE(drwav* pWav, FILE* pFile, drwav_chunk_proc onChunk, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav_bool32 result; + + result = drwav_preinit(pWav, drwav__on_read_stdio, drwav__on_seek_stdio, drwav__on_tell_stdio, (void*)pFile, pAllocationCallbacks); + if (result != DRWAV_TRUE) { + fclose(pFile); + return result; + } + + result = drwav_init__internal(pWav, onChunk, pChunkUserData, flags); + if (result != DRWAV_TRUE) { + fclose(pFile); + return result; + } + + return DRWAV_TRUE; +} + +DRWAV_API drwav_bool32 drwav_init_file_ex(drwav* pWav, const char* filename, drwav_chunk_proc onChunk, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + FILE* pFile; + if (drwav_fopen(&pFile, filename, "rb") != DRWAV_SUCCESS) { + return DRWAV_FALSE; + } + + /* This takes ownership of the FILE* object. */ + return drwav_init_file__internal_FILE(pWav, pFile, onChunk, pChunkUserData, flags, pAllocationCallbacks); +} + +#ifndef DR_WAV_NO_WCHAR +DRWAV_API drwav_bool32 drwav_init_file_w(drwav* pWav, const wchar_t* filename, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_file_ex_w(pWav, filename, NULL, NULL, 0, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_file_ex_w(drwav* pWav, const wchar_t* filename, drwav_chunk_proc onChunk, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + FILE* pFile; + if (drwav_wfopen(&pFile, filename, L"rb", pAllocationCallbacks) != DRWAV_SUCCESS) { + return DRWAV_FALSE; + } + + /* This takes ownership of the FILE* object. */ + return drwav_init_file__internal_FILE(pWav, pFile, onChunk, pChunkUserData, flags, pAllocationCallbacks); +} +#endif + +DRWAV_API drwav_bool32 drwav_init_file_with_metadata(drwav* pWav, const char* filename, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + FILE* pFile; + if (drwav_fopen(&pFile, filename, "rb") != DRWAV_SUCCESS) { + return DRWAV_FALSE; + } + + /* This takes ownership of the FILE* object. */ + return drwav_init_file__internal_FILE(pWav, pFile, NULL, NULL, flags | DRWAV_WITH_METADATA, pAllocationCallbacks); +} + +#ifndef DR_WAV_NO_WCHAR +DRWAV_API drwav_bool32 drwav_init_file_with_metadata_w(drwav* pWav, const wchar_t* filename, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + FILE* pFile; + if (drwav_wfopen(&pFile, filename, L"rb", pAllocationCallbacks) != DRWAV_SUCCESS) { + return DRWAV_FALSE; + } + + /* This takes ownership of the FILE* object. */ + return drwav_init_file__internal_FILE(pWav, pFile, NULL, NULL, flags | DRWAV_WITH_METADATA, pAllocationCallbacks); +} +#endif + + +DRWAV_PRIVATE drwav_bool32 drwav_init_file_write__internal_FILE(drwav* pWav, FILE* pFile, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, drwav_bool32 isSequential, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav_bool32 result; + + result = drwav_preinit_write(pWav, pFormat, isSequential, drwav__on_write_stdio, drwav__on_seek_stdio, (void*)pFile, pAllocationCallbacks); + if (result != DRWAV_TRUE) { + fclose(pFile); + return result; + } + + result = drwav_init_write__internal(pWav, pFormat, totalSampleCount); + if (result != DRWAV_TRUE) { + fclose(pFile); + return result; + } + + return DRWAV_TRUE; +} + +DRWAV_PRIVATE drwav_bool32 drwav_init_file_write__internal(drwav* pWav, const char* filename, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, drwav_bool32 isSequential, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + FILE* pFile; + if (drwav_fopen(&pFile, filename, "wb") != DRWAV_SUCCESS) { + return DRWAV_FALSE; + } + + /* This takes ownership of the FILE* object. */ + return drwav_init_file_write__internal_FILE(pWav, pFile, pFormat, totalSampleCount, isSequential, pAllocationCallbacks); +} + +#ifndef DR_WAV_NO_WCHAR +DRWAV_PRIVATE drwav_bool32 drwav_init_file_write_w__internal(drwav* pWav, const wchar_t* filename, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, drwav_bool32 isSequential, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + FILE* pFile; + if (drwav_wfopen(&pFile, filename, L"wb", pAllocationCallbacks) != DRWAV_SUCCESS) { + return DRWAV_FALSE; + } + + /* This takes ownership of the FILE* object. */ + return drwav_init_file_write__internal_FILE(pWav, pFile, pFormat, totalSampleCount, isSequential, pAllocationCallbacks); +} +#endif + +DRWAV_API drwav_bool32 drwav_init_file_write(drwav* pWav, const char* filename, const drwav_data_format* pFormat, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_file_write__internal(pWav, filename, pFormat, 0, DRWAV_FALSE, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_file_write_sequential(drwav* pWav, const char* filename, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_file_write__internal(pWav, filename, pFormat, totalSampleCount, DRWAV_TRUE, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_file_write_sequential_pcm_frames(drwav* pWav, const char* filename, const drwav_data_format* pFormat, drwav_uint64 totalPCMFrameCount, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pFormat == NULL) { + return DRWAV_FALSE; + } + + return drwav_init_file_write_sequential(pWav, filename, pFormat, totalPCMFrameCount*pFormat->channels, pAllocationCallbacks); +} + +#ifndef DR_WAV_NO_WCHAR +DRWAV_API drwav_bool32 drwav_init_file_write_w(drwav* pWav, const wchar_t* filename, const drwav_data_format* pFormat, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_file_write_w__internal(pWav, filename, pFormat, 0, DRWAV_FALSE, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_file_write_sequential_w(drwav* pWav, const wchar_t* filename, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_file_write_w__internal(pWav, filename, pFormat, totalSampleCount, DRWAV_TRUE, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_file_write_sequential_pcm_frames_w(drwav* pWav, const wchar_t* filename, const drwav_data_format* pFormat, drwav_uint64 totalPCMFrameCount, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pFormat == NULL) { + return DRWAV_FALSE; + } + + return drwav_init_file_write_sequential_w(pWav, filename, pFormat, totalPCMFrameCount*pFormat->channels, pAllocationCallbacks); +} +#endif +#endif /* DR_WAV_NO_STDIO */ + + +DRWAV_PRIVATE size_t drwav__on_read_memory(void* pUserData, void* pBufferOut, size_t bytesToRead) +{ + drwav* pWav = (drwav*)pUserData; + size_t bytesRemaining; + + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pWav->memoryStream.dataSize >= pWav->memoryStream.currentReadPos); + + bytesRemaining = pWav->memoryStream.dataSize - pWav->memoryStream.currentReadPos; + if (bytesToRead > bytesRemaining) { + bytesToRead = bytesRemaining; + } + + if (bytesToRead > 0) { + DRWAV_COPY_MEMORY(pBufferOut, pWav->memoryStream.data + pWav->memoryStream.currentReadPos, bytesToRead); + pWav->memoryStream.currentReadPos += bytesToRead; + } + + return bytesToRead; +} + +DRWAV_PRIVATE drwav_bool32 drwav__on_seek_memory(void* pUserData, int offset, drwav_seek_origin origin) +{ + drwav* pWav = (drwav*)pUserData; + drwav_int64 newCursor; + + DRWAV_ASSERT(pWav != NULL); + + if (origin == DRWAV_SEEK_SET) { + newCursor = 0; + } else if (origin == DRWAV_SEEK_CUR) { + newCursor = (drwav_int64)pWav->memoryStream.currentReadPos; + } else if (origin == DRWAV_SEEK_END) { + newCursor = (drwav_int64)pWav->memoryStream.dataSize; + } else { + DRWAV_ASSERT(!"Invalid seek origin"); + return DRWAV_FALSE; + } + + newCursor += offset; + + if (newCursor < 0) { + return DRWAV_FALSE; /* Trying to seek prior to the start of the buffer. */ + } + if ((size_t)newCursor > pWav->memoryStream.dataSize) { + return DRWAV_FALSE; /* Trying to seek beyond the end of the buffer. */ + } + + pWav->memoryStream.currentReadPos = (size_t)newCursor; + + return DRWAV_TRUE; +} + +DRWAV_PRIVATE size_t drwav__on_write_memory(void* pUserData, const void* pDataIn, size_t bytesToWrite) +{ + drwav* pWav = (drwav*)pUserData; + size_t bytesRemaining; + + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pWav->memoryStreamWrite.dataCapacity >= pWav->memoryStreamWrite.currentWritePos); + + bytesRemaining = pWav->memoryStreamWrite.dataCapacity - pWav->memoryStreamWrite.currentWritePos; + if (bytesRemaining < bytesToWrite) { + /* Need to reallocate. */ + void* pNewData; + size_t newDataCapacity = (pWav->memoryStreamWrite.dataCapacity == 0) ? 256 : pWav->memoryStreamWrite.dataCapacity * 2; + + /* If doubling wasn't enough, just make it the minimum required size to write the data. */ + if ((newDataCapacity - pWav->memoryStreamWrite.currentWritePos) < bytesToWrite) { + newDataCapacity = pWav->memoryStreamWrite.currentWritePos + bytesToWrite; + } + + pNewData = drwav__realloc_from_callbacks(*pWav->memoryStreamWrite.ppData, newDataCapacity, pWav->memoryStreamWrite.dataCapacity, &pWav->allocationCallbacks); + if (pNewData == NULL) { + return 0; + } + + *pWav->memoryStreamWrite.ppData = pNewData; + pWav->memoryStreamWrite.dataCapacity = newDataCapacity; + } + + DRWAV_COPY_MEMORY(((drwav_uint8*)(*pWav->memoryStreamWrite.ppData)) + pWav->memoryStreamWrite.currentWritePos, pDataIn, bytesToWrite); + + pWav->memoryStreamWrite.currentWritePos += bytesToWrite; + if (pWav->memoryStreamWrite.dataSize < pWav->memoryStreamWrite.currentWritePos) { + pWav->memoryStreamWrite.dataSize = pWav->memoryStreamWrite.currentWritePos; + } + + *pWav->memoryStreamWrite.pDataSize = pWav->memoryStreamWrite.dataSize; + + return bytesToWrite; +} + +DRWAV_PRIVATE drwav_bool32 drwav__on_seek_memory_write(void* pUserData, int offset, drwav_seek_origin origin) +{ + drwav* pWav = (drwav*)pUserData; + drwav_int64 newCursor; + + DRWAV_ASSERT(pWav != NULL); + + if (origin == DRWAV_SEEK_SET) { + newCursor = 0; + } else if (origin == DRWAV_SEEK_CUR) { + newCursor = (drwav_int64)pWav->memoryStreamWrite.currentWritePos; + } else if (origin == DRWAV_SEEK_END) { + newCursor = (drwav_int64)pWav->memoryStreamWrite.dataSize; + } else { + DRWAV_ASSERT(!"Invalid seek origin"); + return DRWAV_FALSE; + } + + newCursor += offset; + + if (newCursor < 0) { + return DRWAV_FALSE; /* Trying to seek prior to the start of the buffer. */ + } + if ((size_t)newCursor > pWav->memoryStreamWrite.dataSize) { + return DRWAV_FALSE; /* Trying to seek beyond the end of the buffer. */ + } + + pWav->memoryStreamWrite.currentWritePos = (size_t)newCursor; + + return DRWAV_TRUE; +} + +DRWAV_PRIVATE drwav_bool32 drwav__on_tell_memory(void* pUserData, drwav_int64* pCursor) +{ + drwav* pWav = (drwav*)pUserData; + + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(pCursor != NULL); + + *pCursor = (drwav_int64)pWav->memoryStream.currentReadPos; + return DRWAV_TRUE; +} + +DRWAV_API drwav_bool32 drwav_init_memory(drwav* pWav, const void* data, size_t dataSize, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_memory_ex(pWav, data, dataSize, NULL, NULL, 0, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_memory_ex(drwav* pWav, const void* data, size_t dataSize, drwav_chunk_proc onChunk, void* pChunkUserData, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (data == NULL || dataSize == 0) { + return DRWAV_FALSE; + } + + if (!drwav_preinit(pWav, drwav__on_read_memory, drwav__on_seek_memory, drwav__on_tell_memory, pWav, pAllocationCallbacks)) { + return DRWAV_FALSE; + } + + pWav->memoryStream.data = (const drwav_uint8*)data; + pWav->memoryStream.dataSize = dataSize; + pWav->memoryStream.currentReadPos = 0; + + return drwav_init__internal(pWav, onChunk, pChunkUserData, flags); +} + +DRWAV_API drwav_bool32 drwav_init_memory_with_metadata(drwav* pWav, const void* data, size_t dataSize, drwav_uint32 flags, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (data == NULL || dataSize == 0) { + return DRWAV_FALSE; + } + + if (!drwav_preinit(pWav, drwav__on_read_memory, drwav__on_seek_memory, drwav__on_tell_memory, pWav, pAllocationCallbacks)) { + return DRWAV_FALSE; + } + + pWav->memoryStream.data = (const drwav_uint8*)data; + pWav->memoryStream.dataSize = dataSize; + pWav->memoryStream.currentReadPos = 0; + + return drwav_init__internal(pWav, NULL, NULL, flags | DRWAV_WITH_METADATA); +} + + +DRWAV_PRIVATE drwav_bool32 drwav_init_memory_write__internal(drwav* pWav, void** ppData, size_t* pDataSize, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, drwav_bool32 isSequential, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (ppData == NULL || pDataSize == NULL) { + return DRWAV_FALSE; + } + + *ppData = NULL; /* Important because we're using realloc()! */ + *pDataSize = 0; + + if (!drwav_preinit_write(pWav, pFormat, isSequential, drwav__on_write_memory, drwav__on_seek_memory_write, pWav, pAllocationCallbacks)) { + return DRWAV_FALSE; + } + + pWav->memoryStreamWrite.ppData = ppData; + pWav->memoryStreamWrite.pDataSize = pDataSize; + pWav->memoryStreamWrite.dataSize = 0; + pWav->memoryStreamWrite.dataCapacity = 0; + pWav->memoryStreamWrite.currentWritePos = 0; + + return drwav_init_write__internal(pWav, pFormat, totalSampleCount); +} + +DRWAV_API drwav_bool32 drwav_init_memory_write(drwav* pWav, void** ppData, size_t* pDataSize, const drwav_data_format* pFormat, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_memory_write__internal(pWav, ppData, pDataSize, pFormat, 0, DRWAV_FALSE, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_memory_write_sequential(drwav* pWav, void** ppData, size_t* pDataSize, const drwav_data_format* pFormat, drwav_uint64 totalSampleCount, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + return drwav_init_memory_write__internal(pWav, ppData, pDataSize, pFormat, totalSampleCount, DRWAV_TRUE, pAllocationCallbacks); +} + +DRWAV_API drwav_bool32 drwav_init_memory_write_sequential_pcm_frames(drwav* pWav, void** ppData, size_t* pDataSize, const drwav_data_format* pFormat, drwav_uint64 totalPCMFrameCount, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pFormat == NULL) { + return DRWAV_FALSE; + } + + return drwav_init_memory_write_sequential(pWav, ppData, pDataSize, pFormat, totalPCMFrameCount*pFormat->channels, pAllocationCallbacks); +} + +DRWAV_PRIVATE drwav_uint32 drwav_write_padding(drwav* pWav) +{ + /* Do not adjust pWav->dataChunkDataSize - this should not include the padding. */ + drwav_uint32 paddingSize = drwav_calculate_padding_size(pWav->container, pWav->dataChunkDataSize); + + if (paddingSize > 0) { + drwav_uint64 paddingData = 0; + drwav__write(pWav, &paddingData, paddingSize); /* Byte order does not matter for this. */ + } + + return paddingSize; +} + +DRWAV_PRIVATE void drwav_write_chunk_sizes(drwav* pWav) +{ + /* + When using sequential mode, these will have been filled in at initialization time. We only need + to do this when using non-sequential mode. + */ + if (pWav->onSeek != NULL && !pWav->isSequentialWrite) { + if (pWav->container == drwav_container_riff) { + /* The "RIFF" chunk size. */ + if (pWav->onSeek(pWav->pUserData, 4, DRWAV_SEEK_SET)) { + drwav_uint32 riffChunkSize = drwav__riff_chunk_size_riff(pWav->dataChunkDataSize, pWav->pMetadata, pWav->metadataCount); + drwav__write_u32ne_to_le(pWav, riffChunkSize); + } + + /* The "data" chunk size. */ + if (pWav->onSeek(pWav->pUserData, (int)pWav->dataChunkDataPos - 4, DRWAV_SEEK_SET)) { + drwav_uint32 dataChunkSize = drwav__data_chunk_size_riff(pWav->dataChunkDataSize); + drwav__write_u32ne_to_le(pWav, dataChunkSize); + } + } else if (pWav->container == drwav_container_w64) { + /* The "RIFF" chunk size. */ + if (pWav->onSeek(pWav->pUserData, 16, DRWAV_SEEK_SET)) { + drwav_uint64 riffChunkSize = drwav__riff_chunk_size_w64(pWav->dataChunkDataSize); + drwav__write_u64ne_to_le(pWav, riffChunkSize); + } + + /* The "data" chunk size. */ + if (pWav->onSeek(pWav->pUserData, (int)pWav->dataChunkDataPos - 8, DRWAV_SEEK_SET)) { + drwav_uint64 dataChunkSize = drwav__data_chunk_size_w64(pWav->dataChunkDataSize); + drwav__write_u64ne_to_le(pWav, dataChunkSize); + } + } else if (pWav->container == drwav_container_rf64) { + /* We only need to update the ds64 chunk. The "RIFF" and "data" chunks always have their sizes set to 0xFFFFFFFF for RF64. */ + int ds64BodyPos = 12 + 8; + + /* The "RIFF" chunk size. */ + if (pWav->onSeek(pWav->pUserData, ds64BodyPos + 0, DRWAV_SEEK_SET)) { + drwav_uint64 riffChunkSize = drwav__riff_chunk_size_rf64(pWav->dataChunkDataSize, pWav->pMetadata, pWav->metadataCount); + drwav__write_u64ne_to_le(pWav, riffChunkSize); + } + + /* The "data" chunk size. */ + if (pWav->onSeek(pWav->pUserData, ds64BodyPos + 8, DRWAV_SEEK_SET)) { + drwav_uint64 dataChunkSize = drwav__data_chunk_size_rf64(pWav->dataChunkDataSize); + drwav__write_u64ne_to_le(pWav, dataChunkSize); + } + } + } +} + +DRWAV_API drwav_result drwav_uninit(drwav* pWav) +{ + drwav_result result = DRWAV_SUCCESS; + + if (pWav == NULL) { + return DRWAV_INVALID_ARGS; + } + + if (pWav->onWrite != NULL) { + if (pWav->isSequentialWrite) { + /* + Padding will not have been written in `drwav_write_*()` in sequential mode so we'll want to + do it explicitly here. + */ + drwav_write_padding(pWav); + + /* Validation for sequential mode. */ + if (pWav->dataChunkDataSize != pWav->dataChunkDataSizeTargetWrite) { + result = DRWAV_INVALID_FILE; + } + } + } else { + drwav_free(pWav->pMetadata, &pWav->allocationCallbacks); + } + +#ifndef DR_WAV_NO_STDIO + /* + If we opened the file with drwav_open_file() we will want to close the file handle. We can know whether or not drwav_open_file() + was used by looking at the onRead and onSeek callbacks. + */ + if (pWav->onRead == drwav__on_read_stdio || pWav->onWrite == drwav__on_write_stdio) { + fclose((FILE*)pWav->pUserData); + } +#endif + + return result; +} + + + +DRWAV_API size_t drwav_read_raw(drwav* pWav, size_t bytesToRead, void* pBufferOut) +{ + size_t bytesRead; + drwav_uint32 bytesPerFrame; + + if (pWav == NULL || bytesToRead == 0) { + return 0; /* Invalid args. */ + } + + if (bytesToRead > pWav->bytesRemaining) { + bytesToRead = (size_t)pWav->bytesRemaining; + } + + if (bytesToRead == 0) { + return 0; /* At end. */ + } + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; /* Could not determine the bytes per frame. */ + } + + if (pBufferOut != NULL) { + bytesRead = pWav->onRead(pWav->pUserData, pBufferOut, bytesToRead); + } else { + /* We need to seek. If we fail, we need to read-and-discard to make sure we get a good byte count. */ + bytesRead = 0; + while (bytesRead < bytesToRead) { + size_t bytesToSeek = (bytesToRead - bytesRead); + if (bytesToSeek > 0x7FFFFFFF) { + bytesToSeek = 0x7FFFFFFF; + } + + if (pWav->onSeek(pWav->pUserData, (int)bytesToSeek, DRWAV_SEEK_CUR) == DRWAV_FALSE) { + break; + } + + bytesRead += bytesToSeek; + } + + /* When we get here we may need to read-and-discard some data. */ + while (bytesRead < bytesToRead) { + drwav_uint8 buffer[4096]; + size_t bytesSeeked; + size_t bytesToSeek = (bytesToRead - bytesRead); + if (bytesToSeek > sizeof(buffer)) { + bytesToSeek = sizeof(buffer); + } + + bytesSeeked = pWav->onRead(pWav->pUserData, buffer, bytesToSeek); + bytesRead += bytesSeeked; + + if (bytesSeeked < bytesToSeek) { + break; /* Reached the end. */ + } + } + } + + pWav->bytesRemaining -= bytesRead; + pWav->readCursorInPCMFrames = (pWav->dataChunkDataSize - pWav->bytesRemaining) / bytesPerFrame; + + return bytesRead; +} + + + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_le(drwav* pWav, drwav_uint64 framesToRead, void* pBufferOut) +{ + drwav_uint32 bytesPerFrame; + drwav_uint64 bytesToRead; /* Intentionally uint64 instead of size_t so we can do a check that we're not reading too much on 32-bit builds. */ + drwav_uint64 framesRemainingInFile; + + if (pWav == NULL || framesToRead == 0) { + return 0; + } + + /* Cannot use this function for compressed formats. */ + if (drwav__is_compressed_format_tag(pWav->translatedFormatTag)) { + return 0; + } + + framesRemainingInFile = pWav->totalPCMFrameCount - pWav->readCursorInPCMFrames; + if (framesToRead > framesRemainingInFile) { + framesToRead = framesRemainingInFile; + } + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + /* Don't try to read more samples than can potentially fit in the output buffer. */ + bytesToRead = framesToRead * bytesPerFrame; + if (bytesToRead > DRWAV_SIZE_MAX) { + bytesToRead = (DRWAV_SIZE_MAX / bytesPerFrame) * bytesPerFrame; /* Round the number of bytes to read to a clean frame boundary. */ + } + + /* + Doing an explicit check here just to make it clear that we don't want to be attempt to read anything if there's no bytes to read. There + *could* be a time where it evaluates to 0 due to overflowing. + */ + if (bytesToRead == 0) { + return 0; + } + + return drwav_read_raw(pWav, (size_t)bytesToRead, pBufferOut) / bytesPerFrame; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_be(drwav* pWav, drwav_uint64 framesToRead, void* pBufferOut) +{ + drwav_uint64 framesRead = drwav_read_pcm_frames_le(pWav, framesToRead, pBufferOut); + + if (pBufferOut != NULL) { + drwav_uint32 bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; /* Could not get the bytes per frame which means bytes per sample cannot be determined and we don't know how to byte swap. */ + } + + drwav__bswap_samples(pBufferOut, framesRead*pWav->channels, bytesPerFrame/pWav->channels); + } + + return framesRead; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames(drwav* pWav, drwav_uint64 framesToRead, void* pBufferOut) +{ + drwav_uint64 framesRead = 0; + + if (drwav_is_container_be(pWav->container)) { + /* + Special case for AIFF. AIFF is a big-endian encoded format, but it supports a format that is + PCM in little-endian encoding. In this case, we fall through this branch and treate it as + little-endian. + */ + if (pWav->container != drwav_container_aiff || pWav->aiff.isLE == DRWAV_FALSE) { + if (drwav__is_little_endian()) { + framesRead = drwav_read_pcm_frames_be(pWav, framesToRead, pBufferOut); + } else { + framesRead = drwav_read_pcm_frames_le(pWav, framesToRead, pBufferOut); + } + + goto post_process; + } + } + + /* Getting here means the data should be considered little-endian. */ + if (drwav__is_little_endian()) { + framesRead = drwav_read_pcm_frames_le(pWav, framesToRead, pBufferOut); + } else { + framesRead = drwav_read_pcm_frames_be(pWav, framesToRead, pBufferOut); + } + + /* + Here is where we check if we need to do a signed/unsigned conversion for AIFF. The reason we need to do this + is because dr_wav always assumes an 8-bit sample is unsigned, whereas AIFF can have signed 8-bit formats. + */ + post_process: + { + if (pWav->container == drwav_container_aiff && pWav->translatedFormatTag == DR_WAVE_FORMAT_PCM && pWav->bitsPerSample == 8 && pWav->aiff.isUnsigned == DRWAV_FALSE) { + if (pBufferOut != NULL) { + drwav_uint64 iSample; + + for (iSample = 0; iSample < framesRead * pWav->channels; iSample += 1) { + ((drwav_uint8*)pBufferOut)[iSample] += 128; + } + } + } + } + + return framesRead; +} + + + +DRWAV_PRIVATE drwav_bool32 drwav_seek_to_first_pcm_frame(drwav* pWav) +{ + if (pWav->onWrite != NULL) { + return DRWAV_FALSE; /* No seeking in write mode. */ + } + + if (!pWav->onSeek(pWav->pUserData, (int)pWav->dataChunkDataPos, DRWAV_SEEK_SET)) { + return DRWAV_FALSE; + } + + if (drwav__is_compressed_format_tag(pWav->translatedFormatTag)) { + /* Cached data needs to be cleared for compressed formats. */ + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ADPCM) { + DRWAV_ZERO_OBJECT(&pWav->msadpcm); + } else if (pWav->translatedFormatTag == DR_WAVE_FORMAT_DVI_ADPCM) { + DRWAV_ZERO_OBJECT(&pWav->ima); + } else { + DRWAV_ASSERT(DRWAV_FALSE); /* If this assertion is triggered it means I've implemented a new compressed format but forgot to add a branch for it here. */ + } + } + + pWav->readCursorInPCMFrames = 0; + pWav->bytesRemaining = pWav->dataChunkDataSize; + + return DRWAV_TRUE; +} + +DRWAV_API drwav_bool32 drwav_seek_to_pcm_frame(drwav* pWav, drwav_uint64 targetFrameIndex) +{ + /* Seeking should be compatible with wave files > 2GB. */ + + if (pWav == NULL || pWav->onSeek == NULL) { + return DRWAV_FALSE; + } + + /* No seeking in write mode. */ + if (pWav->onWrite != NULL) { + return DRWAV_FALSE; + } + + /* If there are no samples, just return DRWAV_TRUE without doing anything. */ + if (pWav->totalPCMFrameCount == 0) { + return DRWAV_TRUE; + } + + /* Make sure the sample is clamped. */ + if (targetFrameIndex > pWav->totalPCMFrameCount) { + targetFrameIndex = pWav->totalPCMFrameCount; + } + + /* + For compressed formats we just use a slow generic seek. If we are seeking forward we just seek forward. If we are going backwards we need + to seek back to the start. + */ + if (drwav__is_compressed_format_tag(pWav->translatedFormatTag)) { + /* TODO: This can be optimized. */ + + /* + If we're seeking forward it's simple - just keep reading samples until we hit the sample we're requesting. If we're seeking backwards, + we first need to seek back to the start and then just do the same thing as a forward seek. + */ + if (targetFrameIndex < pWav->readCursorInPCMFrames) { + if (!drwav_seek_to_first_pcm_frame(pWav)) { + return DRWAV_FALSE; + } + } + + if (targetFrameIndex > pWav->readCursorInPCMFrames) { + drwav_uint64 offsetInFrames = targetFrameIndex - pWav->readCursorInPCMFrames; + + drwav_int16 devnull[2048]; + while (offsetInFrames > 0) { + drwav_uint64 framesRead = 0; + drwav_uint64 framesToRead = offsetInFrames; + if (framesToRead > drwav_countof(devnull)/pWav->channels) { + framesToRead = drwav_countof(devnull)/pWav->channels; + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ADPCM) { + framesRead = drwav_read_pcm_frames_s16__msadpcm(pWav, framesToRead, devnull); + } else if (pWav->translatedFormatTag == DR_WAVE_FORMAT_DVI_ADPCM) { + framesRead = drwav_read_pcm_frames_s16__ima(pWav, framesToRead, devnull); + } else { + DRWAV_ASSERT(DRWAV_FALSE); /* If this assertion is triggered it means I've implemented a new compressed format but forgot to add a branch for it here. */ + } + + if (framesRead != framesToRead) { + return DRWAV_FALSE; + } + + offsetInFrames -= framesRead; + } + } + } else { + drwav_uint64 currentBytePos; + drwav_uint64 targetBytePos; + drwav_uint64 offset; + drwav_uint32 bytesPerFrame; + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return DRWAV_FALSE; /* Not able to calculate offset. */ + } + + currentBytePos = pWav->dataChunkDataSize - pWav->bytesRemaining; + targetBytePos = targetFrameIndex * bytesPerFrame; + + pWav->readCursorInPCMFrames = currentBytePos / bytesPerFrame; + + if (currentBytePos < targetBytePos) { + /* Offset forwards. */ + offset = (targetBytePos - currentBytePos); + } else { + /* Offset backwards. */ + if (!drwav_seek_to_first_pcm_frame(pWav)) { + return DRWAV_FALSE; + } + offset = targetBytePos; + } + + while (offset > 0) { + int offset32 = ((offset > INT_MAX) ? INT_MAX : (int)offset); + if (!pWav->onSeek(pWav->pUserData, offset32, DRWAV_SEEK_CUR)) { + return DRWAV_FALSE; + } + + pWav->bytesRemaining -= offset32; + pWav->readCursorInPCMFrames = (pWav->dataChunkDataSize - pWav->bytesRemaining) / bytesPerFrame; + offset -= offset32; + } + } + + return DRWAV_TRUE; +} + +DRWAV_API drwav_result drwav_get_cursor_in_pcm_frames(drwav* pWav, drwav_uint64* pCursor) +{ + if (pCursor == NULL) { + return DRWAV_INVALID_ARGS; + } + + *pCursor = 0; /* Safety. */ + + if (pWav == NULL) { + return DRWAV_INVALID_ARGS; + } + + *pCursor = pWav->readCursorInPCMFrames; + + return DRWAV_SUCCESS; +} + +DRWAV_API drwav_result drwav_get_length_in_pcm_frames(drwav* pWav, drwav_uint64* pLength) +{ + if (pLength == NULL) { + return DRWAV_INVALID_ARGS; + } + + *pLength = 0; /* Safety. */ + + if (pWav == NULL) { + return DRWAV_INVALID_ARGS; + } + + *pLength = pWav->totalPCMFrameCount; + + return DRWAV_SUCCESS; +} + + +DRWAV_API size_t drwav_write_raw(drwav* pWav, size_t bytesToWrite, const void* pData) +{ + size_t bytesWritten; + + if (pWav == NULL || bytesToWrite == 0 || pData == NULL) { + return 0; + } + + bytesWritten = pWav->onWrite(pWav->pUserData, pData, bytesToWrite); + pWav->dataChunkDataSize += bytesWritten; + + if (!pWav->isSequentialWrite) { + drwav_uint32 padding; + + /* Padding. */ + padding = drwav_write_padding(pWav); + + /* Chunk sizes. */ + drwav_write_chunk_sizes(pWav); + + /* Now seek back to just before the padding in preparation for the next writes. */ + if (pWav->onSeek != NULL) { + pWav->onSeek(pWav->pUserData, -(int)padding, DRWAV_SEEK_END); /* Safe cast. */ + } + } + + return bytesWritten; +} + +DRWAV_API drwav_uint64 drwav_write_pcm_frames_le(drwav* pWav, drwav_uint64 framesToWrite, const void* pData) +{ + drwav_uint64 bytesToWrite; + drwav_uint64 bytesWritten; + const drwav_uint8* pRunningData; + + if (pWav == NULL || framesToWrite == 0 || pData == NULL) { + return 0; + } + + bytesToWrite = ((framesToWrite * pWav->channels * pWav->bitsPerSample) / 8); + if (bytesToWrite > DRWAV_SIZE_MAX) { + return 0; + } + + bytesWritten = 0; + pRunningData = (const drwav_uint8*)pData; + + while (bytesToWrite > 0) { + size_t bytesJustWritten; + drwav_uint64 bytesToWriteThisIteration; + + bytesToWriteThisIteration = bytesToWrite; + DRWAV_ASSERT(bytesToWriteThisIteration <= DRWAV_SIZE_MAX); /* <-- This is checked above. */ + + bytesJustWritten = drwav_write_raw(pWav, (size_t)bytesToWriteThisIteration, pRunningData); + if (bytesJustWritten == 0) { + break; + } + + bytesToWrite -= bytesJustWritten; + bytesWritten += bytesJustWritten; + pRunningData += bytesJustWritten; + } + + return (bytesWritten * 8) / pWav->bitsPerSample / pWav->channels; +} + +DRWAV_API drwav_uint64 drwav_write_pcm_frames_be(drwav* pWav, drwav_uint64 framesToWrite, const void* pData) +{ + drwav_uint64 bytesToWrite; + drwav_uint64 bytesWritten; + drwav_uint32 bytesPerSample; + const drwav_uint8* pRunningData; + + if (pWav == NULL || framesToWrite == 0 || pData == NULL) { + return 0; + } + + bytesToWrite = ((framesToWrite * pWav->channels * pWav->bitsPerSample) / 8); + if (bytesToWrite > DRWAV_SIZE_MAX) { + return 0; + } + + bytesWritten = 0; + pRunningData = (const drwav_uint8*)pData; + + bytesPerSample = drwav_get_bytes_per_pcm_frame(pWav) / pWav->channels; + if (bytesPerSample == 0) { + return 0; /* Cannot determine bytes per sample, or bytes per sample is less than one byte. */ + } + + while (bytesToWrite > 0) { + drwav_uint8 temp[4096]; + drwav_uint32 sampleCount; + size_t bytesJustWritten; + drwav_uint64 bytesToWriteThisIteration; + + bytesToWriteThisIteration = bytesToWrite; + DRWAV_ASSERT(bytesToWriteThisIteration <= DRWAV_SIZE_MAX); /* <-- This is checked above. */ + + /* + WAV files are always little-endian. We need to byte swap on big-endian architectures. Since our input buffer is read-only we need + to use an intermediary buffer for the conversion. + */ + sampleCount = sizeof(temp)/bytesPerSample; + + if (bytesToWriteThisIteration > ((drwav_uint64)sampleCount)*bytesPerSample) { + bytesToWriteThisIteration = ((drwav_uint64)sampleCount)*bytesPerSample; + } + + DRWAV_COPY_MEMORY(temp, pRunningData, (size_t)bytesToWriteThisIteration); + drwav__bswap_samples(temp, sampleCount, bytesPerSample); + + bytesJustWritten = drwav_write_raw(pWav, (size_t)bytesToWriteThisIteration, temp); + if (bytesJustWritten == 0) { + break; + } + + bytesToWrite -= bytesJustWritten; + bytesWritten += bytesJustWritten; + pRunningData += bytesJustWritten; + } + + return (bytesWritten * 8) / pWav->bitsPerSample / pWav->channels; +} + +DRWAV_API drwav_uint64 drwav_write_pcm_frames(drwav* pWav, drwav_uint64 framesToWrite, const void* pData) +{ + if (drwav__is_little_endian()) { + return drwav_write_pcm_frames_le(pWav, framesToWrite, pData); + } else { + return drwav_write_pcm_frames_be(pWav, framesToWrite, pData); + } +} + + +DRWAV_PRIVATE DRWAV_INLINE drwav_int32 drwav__msadpcm_calc_sample(drwav_int32 prevSample0, drwav_int32 prevSample1, drwav_int32 coeff1, drwav_int32 coeff2, drwav_int32 nibble, drwav_int32 delta) +{ + drwav_int32 prediction; + + prediction = ((prevSample1 * coeff1) + (prevSample0 * coeff2)) >> 8; + + /* The prediction magnitude cannot exceed 98304. Use 64-bit arithmetic only when the addition could overflow which in my testing is just a tiny bit faster. */ + if (delta > (0x7FFFFFFF - 98304) / 8) { + return (drwav_int32)drwav_clamp(prediction + ((drwav_int64)nibble * delta), -32768, 32767); + } else { + return drwav_clamp(prediction + (nibble * delta), -32768, 32767); + } +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s16__msadpcm(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + drwav_uint64 totalFramesRead = 0; + + static const drwav_int32 adaptationTable[] = { + 230, 230, 230, 230, 307, 409, 512, 614, + 768, 614, 512, 409, 307, 230, 230, 230 + }; + static const drwav_int32 coeff1Table[] = { 256, 512, 0, 192, 240, 460, 392 }; + static const drwav_int32 coeff2Table[] = { 0, -256, 0, 64, 0, -208, -232 }; + + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(framesToRead > 0); + + /* TODO: Lots of room for optimization here. */ + + while (pWav->readCursorInPCMFrames < pWav->totalPCMFrameCount) { + DRWAV_ASSERT(framesToRead > 0); /* This loop iteration will never get hit with framesToRead == 0 because it's asserted at the top, and we check for 0 inside the loop just below. */ + + /* If there are no cached frames we need to load a new block. */ + if (pWav->msadpcm.cachedFrameCount == 0 && pWav->msadpcm.bytesRemainingInBlock == 0) { + if (pWav->channels == 1) { + /* Mono. */ + drwav_uint8 header[7]; + if (pWav->onRead(pWav->pUserData, header, sizeof(header)) != sizeof(header)) { + return totalFramesRead; + } + pWav->msadpcm.bytesRemainingInBlock = pWav->fmt.blockAlign - sizeof(header); + + pWav->msadpcm.predictor[0] = header[0]; + pWav->msadpcm.delta[0] = drwav_bytes_to_s16(header + 1); + pWav->msadpcm.prevFrames[0][1] = (drwav_int32)drwav_bytes_to_s16(header + 3); + pWav->msadpcm.prevFrames[0][0] = (drwav_int32)drwav_bytes_to_s16(header + 5); + pWav->msadpcm.cachedFrames[2] = pWav->msadpcm.prevFrames[0][0]; + pWav->msadpcm.cachedFrames[3] = pWav->msadpcm.prevFrames[0][1]; + pWav->msadpcm.cachedFrameCount = 2; + + /* The predictor is used as an index into coeff1Table so we'll need to validate to ensure it never overflows. */ + if (pWav->msadpcm.predictor[0] >= drwav_countof(coeff1Table) || pWav->msadpcm.predictor[0] >= drwav_countof(coeff2Table)) { + return totalFramesRead; /* Invalid file. */ + } + } else { + /* Stereo. */ + drwav_uint8 header[14]; + if (pWav->onRead(pWav->pUserData, header, sizeof(header)) != sizeof(header)) { + return totalFramesRead; + } + pWav->msadpcm.bytesRemainingInBlock = pWav->fmt.blockAlign - sizeof(header); + + pWav->msadpcm.predictor[0] = header[0]; + pWav->msadpcm.predictor[1] = header[1]; + pWav->msadpcm.delta[0] = drwav_bytes_to_s16(header + 2); + pWav->msadpcm.delta[1] = drwav_bytes_to_s16(header + 4); + pWav->msadpcm.prevFrames[0][1] = (drwav_int32)drwav_bytes_to_s16(header + 6); + pWav->msadpcm.prevFrames[1][1] = (drwav_int32)drwav_bytes_to_s16(header + 8); + pWav->msadpcm.prevFrames[0][0] = (drwav_int32)drwav_bytes_to_s16(header + 10); + pWav->msadpcm.prevFrames[1][0] = (drwav_int32)drwav_bytes_to_s16(header + 12); + + pWav->msadpcm.cachedFrames[0] = pWav->msadpcm.prevFrames[0][0]; + pWav->msadpcm.cachedFrames[1] = pWav->msadpcm.prevFrames[1][0]; + pWav->msadpcm.cachedFrames[2] = pWav->msadpcm.prevFrames[0][1]; + pWav->msadpcm.cachedFrames[3] = pWav->msadpcm.prevFrames[1][1]; + pWav->msadpcm.cachedFrameCount = 2; + + /* The predictor is used as an index into coeff1Table so we'll need to validate to ensure it never overflows. */ + if (pWav->msadpcm.predictor[0] >= drwav_countof(coeff1Table) || pWav->msadpcm.predictor[0] >= drwav_countof(coeff2Table) || + pWav->msadpcm.predictor[1] >= drwav_countof(coeff1Table) || pWav->msadpcm.predictor[1] >= drwav_countof(coeff2Table)) { + return totalFramesRead; /* Invalid file. */ + } + } + } + + /* Output anything that's cached. */ + while (framesToRead > 0 && pWav->msadpcm.cachedFrameCount > 0 && pWav->readCursorInPCMFrames < pWav->totalPCMFrameCount) { + if (pBufferOut != NULL) { + drwav_uint32 iSample = 0; + for (iSample = 0; iSample < pWav->channels; iSample += 1) { + pBufferOut[iSample] = (drwav_int16)pWav->msadpcm.cachedFrames[(drwav_countof(pWav->msadpcm.cachedFrames) - (pWav->msadpcm.cachedFrameCount*pWav->channels)) + iSample]; + } + + pBufferOut += pWav->channels; + } + + framesToRead -= 1; + totalFramesRead += 1; + pWav->readCursorInPCMFrames += 1; + pWav->msadpcm.cachedFrameCount -= 1; + } + + if (framesToRead == 0) { + break; + } + + + /* + If there's nothing left in the cache, just go ahead and load more. If there's nothing left to load in the current block we just continue to the next + loop iteration which will trigger the loading of a new block. + */ + if (pWav->msadpcm.cachedFrameCount == 0) { + if (pWav->msadpcm.bytesRemainingInBlock == 0) { + continue; + } else { + drwav_uint8 nibbles; + drwav_int32 nibble0; + drwav_int32 nibble1; + + if (pWav->onRead(pWav->pUserData, &nibbles, 1) != 1) { + return totalFramesRead; + } + pWav->msadpcm.bytesRemainingInBlock -= 1; + + /* TODO: Optimize away these if statements. */ + nibble0 = ((nibbles & 0xF0) >> 4); if ((nibbles & 0x80)) { nibble0 |= 0xFFFFFFF0UL; } + nibble1 = ((nibbles & 0x0F) >> 0); if ((nibbles & 0x08)) { nibble1 |= 0xFFFFFFF0UL; } + + if (pWav->channels == 1) { + /* Mono. */ + drwav_int32 newSample0; + drwav_int32 newSample1; + + /* The predictor is read from the file and then indexed into a table. Check that it's in bounds. */ + if (pWav->msadpcm.predictor[0] >= drwav_countof(coeff1Table) || pWav->msadpcm.predictor[0] >= drwav_countof(coeff2Table)) { + return totalFramesRead; + } + + newSample0 = drwav__msadpcm_calc_sample(pWav->msadpcm.prevFrames[0][0], pWav->msadpcm.prevFrames[0][1], coeff1Table[pWav->msadpcm.predictor[0]], coeff2Table[pWav->msadpcm.predictor[0]], nibble0, pWav->msadpcm.delta[0]); + + pWav->msadpcm.delta[0] = (drwav_int32)drwav_clamp(((drwav_int64)adaptationTable[((nibbles & 0xF0) >> 4)] * pWav->msadpcm.delta[0]) >> 8, 16, 0x7FFFFFFF); + + pWav->msadpcm.prevFrames[0][0] = pWav->msadpcm.prevFrames[0][1]; + pWav->msadpcm.prevFrames[0][1] = newSample0; + + + newSample1 = drwav__msadpcm_calc_sample(pWav->msadpcm.prevFrames[0][0], pWav->msadpcm.prevFrames[0][1], coeff1Table[pWav->msadpcm.predictor[0]], coeff2Table[pWav->msadpcm.predictor[0]], nibble1, pWav->msadpcm.delta[0]); + + pWav->msadpcm.delta[0] = (drwav_int32)drwav_clamp(((drwav_int64)adaptationTable[((nibbles & 0x0F) >> 0)] * pWav->msadpcm.delta[0]) >> 8, 16, 0x7FFFFFFF); + + pWav->msadpcm.prevFrames[0][0] = pWav->msadpcm.prevFrames[0][1]; + pWav->msadpcm.prevFrames[0][1] = newSample1; + + pWav->msadpcm.cachedFrames[2] = newSample0; + pWav->msadpcm.cachedFrames[3] = newSample1; + pWav->msadpcm.cachedFrameCount = 2; + } else { + /* Stereo. */ + drwav_int32 newSample0; + drwav_int32 newSample1; + + /* Left. */ + if (pWav->msadpcm.predictor[0] >= drwav_countof(coeff1Table) || pWav->msadpcm.predictor[0] >= drwav_countof(coeff2Table)) { + return totalFramesRead; /* Out of bounds. Invalid file. */ + } + + newSample0 = drwav__msadpcm_calc_sample(pWav->msadpcm.prevFrames[0][0], pWav->msadpcm.prevFrames[0][1], coeff1Table[pWav->msadpcm.predictor[0]], coeff2Table[pWav->msadpcm.predictor[0]], nibble0, pWav->msadpcm.delta[0]); + + pWav->msadpcm.delta[0] = (drwav_int32)drwav_clamp(((drwav_int64)adaptationTable[((nibbles & 0xF0) >> 4)] * pWav->msadpcm.delta[0]) >> 8, 16, 0x7FFFFFFF); + + pWav->msadpcm.prevFrames[0][0] = pWav->msadpcm.prevFrames[0][1]; + pWav->msadpcm.prevFrames[0][1] = newSample0; + + + /* Right. */ + if (pWav->msadpcm.predictor[1] >= drwav_countof(coeff1Table) || pWav->msadpcm.predictor[1] >= drwav_countof(coeff2Table)) { + return totalFramesRead; /* Out of bounds. Invalid file. */ + } + + newSample1 = drwav__msadpcm_calc_sample(pWav->msadpcm.prevFrames[1][0], pWav->msadpcm.prevFrames[1][1], coeff1Table[pWav->msadpcm.predictor[1]], coeff2Table[pWav->msadpcm.predictor[1]], nibble1, pWav->msadpcm.delta[1]); + + pWav->msadpcm.delta[1] = (drwav_int32)drwav_clamp(((drwav_int64)adaptationTable[((nibbles & 0x0F) >> 0)] * pWav->msadpcm.delta[1]) >> 8, 16, 0x7FFFFFFF); + + pWav->msadpcm.prevFrames[1][0] = pWav->msadpcm.prevFrames[1][1]; + pWav->msadpcm.prevFrames[1][1] = newSample1; + + pWav->msadpcm.cachedFrames[2] = newSample0; + pWav->msadpcm.cachedFrames[3] = newSample1; + pWav->msadpcm.cachedFrameCount = 1; + } + } + } + } + + return totalFramesRead; +} + + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s16__ima(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + drwav_uint64 totalFramesRead = 0; + drwav_uint32 iChannel; + + static const drwav_int32 indexTable[16] = { + -1, -1, -1, -1, 2, 4, 6, 8, + -1, -1, -1, -1, 2, 4, 6, 8 + }; + + static const drwav_int32 stepTable[89] = { + 7, 8, 9, 10, 11, 12, 13, 14, 16, 17, + 19, 21, 23, 25, 28, 31, 34, 37, 41, 45, + 50, 55, 60, 66, 73, 80, 88, 97, 107, 118, + 130, 143, 157, 173, 190, 209, 230, 253, 279, 307, + 337, 371, 408, 449, 494, 544, 598, 658, 724, 796, + 876, 963, 1060, 1166, 1282, 1411, 1552, 1707, 1878, 2066, + 2272, 2499, 2749, 3024, 3327, 3660, 4026, 4428, 4871, 5358, + 5894, 6484, 7132, 7845, 8630, 9493, 10442, 11487, 12635, 13899, + 15289, 16818, 18500, 20350, 22385, 24623, 27086, 29794, 32767 + }; + + DRWAV_ASSERT(pWav != NULL); + DRWAV_ASSERT(framesToRead > 0); + + /* TODO: Lots of room for optimization here. */ + + while (pWav->readCursorInPCMFrames < pWav->totalPCMFrameCount) { + DRWAV_ASSERT(framesToRead > 0); /* This loop iteration will never get hit with framesToRead == 0 because it's asserted at the top, and we check for 0 inside the loop just below. */ + + /* If there are no cached samples we need to load a new block. */ + if (pWav->ima.cachedFrameCount == 0 && pWav->ima.bytesRemainingInBlock == 0) { + if (pWav->channels == 1) { + /* Mono. */ + drwav_uint8 header[4]; + if (pWav->onRead(pWav->pUserData, header, sizeof(header)) != sizeof(header)) { + return totalFramesRead; + } + pWav->ima.bytesRemainingInBlock = pWav->fmt.blockAlign - sizeof(header); + + if (header[2] >= drwav_countof(stepTable)) { + pWav->onSeek(pWav->pUserData, pWav->ima.bytesRemainingInBlock, DRWAV_SEEK_CUR); + pWav->ima.bytesRemainingInBlock = 0; + return totalFramesRead; /* Invalid data. */ + } + + pWav->ima.predictor[0] = (drwav_int16)drwav_bytes_to_u16(header + 0); + pWav->ima.stepIndex[0] = drwav_clamp(header[2], 0, (drwav_int32)drwav_countof(stepTable)-1); /* Clamp not necessary because we checked above, but adding here to silence a static analysis warning. */ + pWav->ima.cachedFrames[drwav_countof(pWav->ima.cachedFrames) - 1] = pWav->ima.predictor[0]; + pWav->ima.cachedFrameCount = 1; + } else { + /* Stereo. */ + drwav_uint8 header[8]; + if (pWav->onRead(pWav->pUserData, header, sizeof(header)) != sizeof(header)) { + return totalFramesRead; + } + pWav->ima.bytesRemainingInBlock = pWav->fmt.blockAlign - sizeof(header); + + if (header[2] >= drwav_countof(stepTable) || header[6] >= drwav_countof(stepTable)) { + pWav->onSeek(pWav->pUserData, pWav->ima.bytesRemainingInBlock, DRWAV_SEEK_CUR); + pWav->ima.bytesRemainingInBlock = 0; + return totalFramesRead; /* Invalid data. */ + } + + pWav->ima.predictor[0] = drwav_bytes_to_s16(header + 0); + pWav->ima.stepIndex[0] = drwav_clamp(header[2], 0, (drwav_int32)drwav_countof(stepTable)-1); /* Clamp not necessary because we checked above, but adding here to silence a static analysis warning. */ + pWav->ima.predictor[1] = drwav_bytes_to_s16(header + 4); + pWav->ima.stepIndex[1] = drwav_clamp(header[6], 0, (drwav_int32)drwav_countof(stepTable)-1); /* Clamp not necessary because we checked above, but adding here to silence a static analysis warning. */ + + pWav->ima.cachedFrames[drwav_countof(pWav->ima.cachedFrames) - 2] = pWav->ima.predictor[0]; + pWav->ima.cachedFrames[drwav_countof(pWav->ima.cachedFrames) - 1] = pWav->ima.predictor[1]; + pWav->ima.cachedFrameCount = 1; + } + } + + /* Output anything that's cached. */ + while (framesToRead > 0 && pWav->ima.cachedFrameCount > 0 && pWav->readCursorInPCMFrames < pWav->totalPCMFrameCount) { + if (pBufferOut != NULL) { + drwav_uint32 iSample; + for (iSample = 0; iSample < pWav->channels; iSample += 1) { + pBufferOut[iSample] = (drwav_int16)pWav->ima.cachedFrames[(drwav_countof(pWav->ima.cachedFrames) - (pWav->ima.cachedFrameCount*pWav->channels)) + iSample]; + } + pBufferOut += pWav->channels; + } + + framesToRead -= 1; + totalFramesRead += 1; + pWav->readCursorInPCMFrames += 1; + pWav->ima.cachedFrameCount -= 1; + } + + if (framesToRead == 0) { + break; + } + + /* + If there's nothing left in the cache, just go ahead and load more. If there's nothing left to load in the current block we just continue to the next + loop iteration which will trigger the loading of a new block. + */ + if (pWav->ima.cachedFrameCount == 0) { + if (pWav->ima.bytesRemainingInBlock == 0) { + continue; + } else { + /* + From what I can tell with stereo streams, it looks like every 4 bytes (8 samples) is for one channel. So it goes 4 bytes for the + left channel, 4 bytes for the right channel. + */ + pWav->ima.cachedFrameCount = 8; + for (iChannel = 0; iChannel < pWav->channels; ++iChannel) { + drwav_uint32 iByte; + drwav_uint8 nibbles[4]; + if (pWav->onRead(pWav->pUserData, &nibbles, 4) != 4) { + pWav->ima.cachedFrameCount = 0; + return totalFramesRead; + } + pWav->ima.bytesRemainingInBlock -= 4; + + for (iByte = 0; iByte < 4; ++iByte) { + drwav_uint8 nibble0 = ((nibbles[iByte] & 0x0F) >> 0); + drwav_uint8 nibble1 = ((nibbles[iByte] & 0xF0) >> 4); + + drwav_int32 step = stepTable[pWav->ima.stepIndex[iChannel]]; + drwav_int32 predictor = pWav->ima.predictor[iChannel]; + + drwav_int32 diff = step >> 3; + if (nibble0 & 1) diff += step >> 2; + if (nibble0 & 2) diff += step >> 1; + if (nibble0 & 4) diff += step; + if (nibble0 & 8) diff = -diff; + + predictor = drwav_clamp(predictor + diff, -32768, 32767); + pWav->ima.predictor[iChannel] = predictor; + pWav->ima.stepIndex[iChannel] = drwav_clamp(pWav->ima.stepIndex[iChannel] + indexTable[nibble0], 0, (drwav_int32)drwav_countof(stepTable)-1); + pWav->ima.cachedFrames[(drwav_countof(pWav->ima.cachedFrames) - (pWav->ima.cachedFrameCount*pWav->channels)) + (iByte*2+0)*pWav->channels + iChannel] = predictor; + + + step = stepTable[pWav->ima.stepIndex[iChannel]]; + predictor = pWav->ima.predictor[iChannel]; + + diff = step >> 3; + if (nibble1 & 1) diff += step >> 2; + if (nibble1 & 2) diff += step >> 1; + if (nibble1 & 4) diff += step; + if (nibble1 & 8) diff = -diff; + + predictor = drwav_clamp(predictor + diff, -32768, 32767); + pWav->ima.predictor[iChannel] = predictor; + pWav->ima.stepIndex[iChannel] = drwav_clamp(pWav->ima.stepIndex[iChannel] + indexTable[nibble1], 0, (drwav_int32)drwav_countof(stepTable)-1); + pWav->ima.cachedFrames[(drwav_countof(pWav->ima.cachedFrames) - (pWav->ima.cachedFrameCount*pWav->channels)) + (iByte*2+1)*pWav->channels + iChannel] = predictor; + } + } + } + } + } + + return totalFramesRead; +} + + +#ifndef DR_WAV_NO_CONVERSION_API +static const unsigned short g_drwavAlawTable[256] = { + 0xEA80, 0xEB80, 0xE880, 0xE980, 0xEE80, 0xEF80, 0xEC80, 0xED80, 0xE280, 0xE380, 0xE080, 0xE180, 0xE680, 0xE780, 0xE480, 0xE580, + 0xF540, 0xF5C0, 0xF440, 0xF4C0, 0xF740, 0xF7C0, 0xF640, 0xF6C0, 0xF140, 0xF1C0, 0xF040, 0xF0C0, 0xF340, 0xF3C0, 0xF240, 0xF2C0, + 0xAA00, 0xAE00, 0xA200, 0xA600, 0xBA00, 0xBE00, 0xB200, 0xB600, 0x8A00, 0x8E00, 0x8200, 0x8600, 0x9A00, 0x9E00, 0x9200, 0x9600, + 0xD500, 0xD700, 0xD100, 0xD300, 0xDD00, 0xDF00, 0xD900, 0xDB00, 0xC500, 0xC700, 0xC100, 0xC300, 0xCD00, 0xCF00, 0xC900, 0xCB00, + 0xFEA8, 0xFEB8, 0xFE88, 0xFE98, 0xFEE8, 0xFEF8, 0xFEC8, 0xFED8, 0xFE28, 0xFE38, 0xFE08, 0xFE18, 0xFE68, 0xFE78, 0xFE48, 0xFE58, + 0xFFA8, 0xFFB8, 0xFF88, 0xFF98, 0xFFE8, 0xFFF8, 0xFFC8, 0xFFD8, 0xFF28, 0xFF38, 0xFF08, 0xFF18, 0xFF68, 0xFF78, 0xFF48, 0xFF58, + 0xFAA0, 0xFAE0, 0xFA20, 0xFA60, 0xFBA0, 0xFBE0, 0xFB20, 0xFB60, 0xF8A0, 0xF8E0, 0xF820, 0xF860, 0xF9A0, 0xF9E0, 0xF920, 0xF960, + 0xFD50, 0xFD70, 0xFD10, 0xFD30, 0xFDD0, 0xFDF0, 0xFD90, 0xFDB0, 0xFC50, 0xFC70, 0xFC10, 0xFC30, 0xFCD0, 0xFCF0, 0xFC90, 0xFCB0, + 0x1580, 0x1480, 0x1780, 0x1680, 0x1180, 0x1080, 0x1380, 0x1280, 0x1D80, 0x1C80, 0x1F80, 0x1E80, 0x1980, 0x1880, 0x1B80, 0x1A80, + 0x0AC0, 0x0A40, 0x0BC0, 0x0B40, 0x08C0, 0x0840, 0x09C0, 0x0940, 0x0EC0, 0x0E40, 0x0FC0, 0x0F40, 0x0CC0, 0x0C40, 0x0DC0, 0x0D40, + 0x5600, 0x5200, 0x5E00, 0x5A00, 0x4600, 0x4200, 0x4E00, 0x4A00, 0x7600, 0x7200, 0x7E00, 0x7A00, 0x6600, 0x6200, 0x6E00, 0x6A00, + 0x2B00, 0x2900, 0x2F00, 0x2D00, 0x2300, 0x2100, 0x2700, 0x2500, 0x3B00, 0x3900, 0x3F00, 0x3D00, 0x3300, 0x3100, 0x3700, 0x3500, + 0x0158, 0x0148, 0x0178, 0x0168, 0x0118, 0x0108, 0x0138, 0x0128, 0x01D8, 0x01C8, 0x01F8, 0x01E8, 0x0198, 0x0188, 0x01B8, 0x01A8, + 0x0058, 0x0048, 0x0078, 0x0068, 0x0018, 0x0008, 0x0038, 0x0028, 0x00D8, 0x00C8, 0x00F8, 0x00E8, 0x0098, 0x0088, 0x00B8, 0x00A8, + 0x0560, 0x0520, 0x05E0, 0x05A0, 0x0460, 0x0420, 0x04E0, 0x04A0, 0x0760, 0x0720, 0x07E0, 0x07A0, 0x0660, 0x0620, 0x06E0, 0x06A0, + 0x02B0, 0x0290, 0x02F0, 0x02D0, 0x0230, 0x0210, 0x0270, 0x0250, 0x03B0, 0x0390, 0x03F0, 0x03D0, 0x0330, 0x0310, 0x0370, 0x0350 +}; + +static const unsigned short g_drwavMulawTable[256] = { + 0x8284, 0x8684, 0x8A84, 0x8E84, 0x9284, 0x9684, 0x9A84, 0x9E84, 0xA284, 0xA684, 0xAA84, 0xAE84, 0xB284, 0xB684, 0xBA84, 0xBE84, + 0xC184, 0xC384, 0xC584, 0xC784, 0xC984, 0xCB84, 0xCD84, 0xCF84, 0xD184, 0xD384, 0xD584, 0xD784, 0xD984, 0xDB84, 0xDD84, 0xDF84, + 0xE104, 0xE204, 0xE304, 0xE404, 0xE504, 0xE604, 0xE704, 0xE804, 0xE904, 0xEA04, 0xEB04, 0xEC04, 0xED04, 0xEE04, 0xEF04, 0xF004, + 0xF0C4, 0xF144, 0xF1C4, 0xF244, 0xF2C4, 0xF344, 0xF3C4, 0xF444, 0xF4C4, 0xF544, 0xF5C4, 0xF644, 0xF6C4, 0xF744, 0xF7C4, 0xF844, + 0xF8A4, 0xF8E4, 0xF924, 0xF964, 0xF9A4, 0xF9E4, 0xFA24, 0xFA64, 0xFAA4, 0xFAE4, 0xFB24, 0xFB64, 0xFBA4, 0xFBE4, 0xFC24, 0xFC64, + 0xFC94, 0xFCB4, 0xFCD4, 0xFCF4, 0xFD14, 0xFD34, 0xFD54, 0xFD74, 0xFD94, 0xFDB4, 0xFDD4, 0xFDF4, 0xFE14, 0xFE34, 0xFE54, 0xFE74, + 0xFE8C, 0xFE9C, 0xFEAC, 0xFEBC, 0xFECC, 0xFEDC, 0xFEEC, 0xFEFC, 0xFF0C, 0xFF1C, 0xFF2C, 0xFF3C, 0xFF4C, 0xFF5C, 0xFF6C, 0xFF7C, + 0xFF88, 0xFF90, 0xFF98, 0xFFA0, 0xFFA8, 0xFFB0, 0xFFB8, 0xFFC0, 0xFFC8, 0xFFD0, 0xFFD8, 0xFFE0, 0xFFE8, 0xFFF0, 0xFFF8, 0x0000, + 0x7D7C, 0x797C, 0x757C, 0x717C, 0x6D7C, 0x697C, 0x657C, 0x617C, 0x5D7C, 0x597C, 0x557C, 0x517C, 0x4D7C, 0x497C, 0x457C, 0x417C, + 0x3E7C, 0x3C7C, 0x3A7C, 0x387C, 0x367C, 0x347C, 0x327C, 0x307C, 0x2E7C, 0x2C7C, 0x2A7C, 0x287C, 0x267C, 0x247C, 0x227C, 0x207C, + 0x1EFC, 0x1DFC, 0x1CFC, 0x1BFC, 0x1AFC, 0x19FC, 0x18FC, 0x17FC, 0x16FC, 0x15FC, 0x14FC, 0x13FC, 0x12FC, 0x11FC, 0x10FC, 0x0FFC, + 0x0F3C, 0x0EBC, 0x0E3C, 0x0DBC, 0x0D3C, 0x0CBC, 0x0C3C, 0x0BBC, 0x0B3C, 0x0ABC, 0x0A3C, 0x09BC, 0x093C, 0x08BC, 0x083C, 0x07BC, + 0x075C, 0x071C, 0x06DC, 0x069C, 0x065C, 0x061C, 0x05DC, 0x059C, 0x055C, 0x051C, 0x04DC, 0x049C, 0x045C, 0x041C, 0x03DC, 0x039C, + 0x036C, 0x034C, 0x032C, 0x030C, 0x02EC, 0x02CC, 0x02AC, 0x028C, 0x026C, 0x024C, 0x022C, 0x020C, 0x01EC, 0x01CC, 0x01AC, 0x018C, + 0x0174, 0x0164, 0x0154, 0x0144, 0x0134, 0x0124, 0x0114, 0x0104, 0x00F4, 0x00E4, 0x00D4, 0x00C4, 0x00B4, 0x00A4, 0x0094, 0x0084, + 0x0078, 0x0070, 0x0068, 0x0060, 0x0058, 0x0050, 0x0048, 0x0040, 0x0038, 0x0030, 0x0028, 0x0020, 0x0018, 0x0010, 0x0008, 0x0000 +}; + +static DRWAV_INLINE drwav_int16 drwav__alaw_to_s16(drwav_uint8 sampleIn) +{ + return (short)g_drwavAlawTable[sampleIn]; +} + +static DRWAV_INLINE drwav_int16 drwav__mulaw_to_s16(drwav_uint8 sampleIn) +{ + return (short)g_drwavMulawTable[sampleIn]; +} + + + +DRWAV_PRIVATE void drwav__pcm_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t totalSampleCount, unsigned int bytesPerSample) +{ + size_t i; + + /* Special case for 8-bit sample data because it's treated as unsigned. */ + if (bytesPerSample == 1) { + drwav_u8_to_s16(pOut, pIn, totalSampleCount); + return; + } + + + /* Slightly more optimal implementation for common formats. */ + if (bytesPerSample == 2) { + for (i = 0; i < totalSampleCount; ++i) { + *pOut++ = ((const drwav_int16*)pIn)[i]; + } + return; + } + if (bytesPerSample == 3) { + drwav_s24_to_s16(pOut, pIn, totalSampleCount); + return; + } + if (bytesPerSample == 4) { + drwav_s32_to_s16(pOut, (const drwav_int32*)pIn, totalSampleCount); + return; + } + + + /* Anything more than 64 bits per sample is not supported. */ + if (bytesPerSample > 8) { + DRWAV_ZERO_MEMORY(pOut, totalSampleCount * sizeof(*pOut)); + return; + } + + + /* Generic, slow converter. */ + for (i = 0; i < totalSampleCount; ++i) { + drwav_uint64 sample = 0; + unsigned int shift = (8 - bytesPerSample) * 8; + + unsigned int j; + for (j = 0; j < bytesPerSample; j += 1) { + DRWAV_ASSERT(j < 8); + sample |= (drwav_uint64)(pIn[j]) << shift; + shift += 8; + } + + if (!drwav__is_little_endian()) { + sample = drwav__bswap64(sample); + } + + pIn += j; + *pOut++ = (drwav_int16)((drwav_int64)sample >> 48); + } +} + +DRWAV_PRIVATE void drwav__ieee_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t totalSampleCount, unsigned int bytesPerSample) +{ + if (bytesPerSample == 4) { + drwav_f32_to_s16(pOut, (const float*)pIn, totalSampleCount); + return; + } else if (bytesPerSample == 8) { + drwav_f64_to_s16(pOut, (const double*)pIn, totalSampleCount); + return; + } else { + /* Only supporting 32- and 64-bit float. Output silence in all other cases. Contributions welcome for 16-bit float. */ + DRWAV_ZERO_MEMORY(pOut, totalSampleCount * sizeof(*pOut)); + return; + } +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s16__pcm(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + /* Fast path. */ + if ((pWav->translatedFormatTag == DR_WAVE_FORMAT_PCM && pWav->bitsPerSample == 16) || pBufferOut == NULL) { + return drwav_read_pcm_frames(pWav, framesToRead, pBufferOut); + } + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav__pcm_to_s16(pBufferOut, sampleData, (size_t)samplesRead, bytesPerSample); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s16__ieee(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + if (pBufferOut == NULL) { + return drwav_read_pcm_frames(pWav, framesToRead, NULL); + } + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav__ieee_to_s16(pBufferOut, sampleData, (size_t)samplesRead, bytesPerSample); /* Safe cast. */ + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s16__alaw(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + if (pBufferOut == NULL) { + return drwav_read_pcm_frames(pWav, framesToRead, NULL); + } + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav_alaw_to_s16(pBufferOut, sampleData, (size_t)samplesRead); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s16__mulaw(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + if (pBufferOut == NULL) { + return drwav_read_pcm_frames(pWav, framesToRead, NULL); + } + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav_mulaw_to_s16(pBufferOut, sampleData, (size_t)samplesRead); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s16(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + if (pWav == NULL || framesToRead == 0) { + return 0; + } + + if (pBufferOut == NULL) { + return drwav_read_pcm_frames(pWav, framesToRead, NULL); + } + + /* Don't try to read more samples than can potentially fit in the output buffer. */ + if (framesToRead * pWav->channels * sizeof(drwav_int16) > DRWAV_SIZE_MAX) { + framesToRead = DRWAV_SIZE_MAX / sizeof(drwav_int16) / pWav->channels; + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_PCM) { + return drwav_read_pcm_frames_s16__pcm(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_IEEE_FLOAT) { + return drwav_read_pcm_frames_s16__ieee(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ALAW) { + return drwav_read_pcm_frames_s16__alaw(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_MULAW) { + return drwav_read_pcm_frames_s16__mulaw(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ADPCM) { + return drwav_read_pcm_frames_s16__msadpcm(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_DVI_ADPCM) { + return drwav_read_pcm_frames_s16__ima(pWav, framesToRead, pBufferOut); + } + + return 0; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s16le(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + drwav_uint64 framesRead = drwav_read_pcm_frames_s16(pWav, framesToRead, pBufferOut); + if (pBufferOut != NULL && drwav__is_little_endian() == DRWAV_FALSE) { + drwav__bswap_samples_s16(pBufferOut, framesRead*pWav->channels); + } + + return framesRead; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s16be(drwav* pWav, drwav_uint64 framesToRead, drwav_int16* pBufferOut) +{ + drwav_uint64 framesRead = drwav_read_pcm_frames_s16(pWav, framesToRead, pBufferOut); + if (pBufferOut != NULL && drwav__is_little_endian() == DRWAV_TRUE) { + drwav__bswap_samples_s16(pBufferOut, framesRead*pWav->channels); + } + + return framesRead; +} + + +DRWAV_API void drwav_u8_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + int r; + size_t i; + for (i = 0; i < sampleCount; ++i) { + int x = pIn[i]; + r = x << 8; + r = r - 32768; + pOut[i] = (short)r; + } +} + +DRWAV_API void drwav_s24_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + int r; + size_t i; + for (i = 0; i < sampleCount; ++i) { + int x = ((int)(((unsigned int)(((const drwav_uint8*)pIn)[i*3+0]) << 8) | ((unsigned int)(((const drwav_uint8*)pIn)[i*3+1]) << 16) | ((unsigned int)(((const drwav_uint8*)pIn)[i*3+2])) << 24)) >> 8; + r = x >> 8; + pOut[i] = (short)r; + } +} + +DRWAV_API void drwav_s32_to_s16(drwav_int16* pOut, const drwav_int32* pIn, size_t sampleCount) +{ + int r; + size_t i; + for (i = 0; i < sampleCount; ++i) { + int x = pIn[i]; + r = x >> 16; + pOut[i] = (short)r; + } +} + +DRWAV_API void drwav_f32_to_s16(drwav_int16* pOut, const float* pIn, size_t sampleCount) +{ + size_t i; + for (i = 0; i < sampleCount; ++i) { + float x = pIn[i]; + if (x != x) { + pOut[i] = 0; /* NaN */ + } else if (x <= -1) { + pOut[i] = (-32767 - 1); + } else if (x >= 1) { + pOut[i] = 32767; + } else { + pOut[i] = (drwav_int16)(x * 32768.0f); + } + } +} + +DRWAV_API void drwav_f64_to_s16(drwav_int16* pOut, const double* pIn, size_t sampleCount) +{ + size_t i; + for (i = 0; i < sampleCount; ++i) { + double x = pIn[i]; + if (x != x) { + pOut[i] = 0; /* NaN */ + } else if (x <= -1) { + pOut[i] = (-32767 - 1); + } else if (x >= 1) { + pOut[i] = 32767; + } else { + pOut[i] = (drwav_int16)(x * 32768.0); + } + } +} + +DRWAV_API void drwav_alaw_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + for (i = 0; i < sampleCount; ++i) { + pOut[i] = drwav__alaw_to_s16(pIn[i]); + } +} + +DRWAV_API void drwav_mulaw_to_s16(drwav_int16* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + for (i = 0; i < sampleCount; ++i) { + pOut[i] = drwav__mulaw_to_s16(pIn[i]); + } +} + + +DRWAV_PRIVATE void drwav__pcm_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount, unsigned int bytesPerSample) +{ + unsigned int i; + + /* Special case for 8-bit sample data because it's treated as unsigned. */ + if (bytesPerSample == 1) { + drwav_u8_to_f32(pOut, pIn, sampleCount); + return; + } + + /* Slightly more optimal implementation for common formats. */ + if (bytesPerSample == 2) { + drwav_s16_to_f32(pOut, (const drwav_int16*)pIn, sampleCount); + return; + } + if (bytesPerSample == 3) { + drwav_s24_to_f32(pOut, pIn, sampleCount); + return; + } + if (bytesPerSample == 4) { + drwav_s32_to_f32(pOut, (const drwav_int32*)pIn, sampleCount); + return; + } + + + /* Anything more than 64 bits per sample is not supported. */ + if (bytesPerSample > 8) { + DRWAV_ZERO_MEMORY(pOut, sampleCount * sizeof(*pOut)); + return; + } + + + /* Generic, slow converter. */ + for (i = 0; i < sampleCount; ++i) { + drwav_uint64 sample = 0; + unsigned int shift = (8 - bytesPerSample) * 8; + + unsigned int j; + for (j = 0; j < bytesPerSample; j += 1) { + DRWAV_ASSERT(j < 8); + sample |= (drwav_uint64)(pIn[j]) << shift; + shift += 8; + } + + if (!drwav__is_little_endian()) { + sample = drwav__bswap64(sample); + } + + pIn += j; + *pOut++ = (float)((drwav_int64)sample / 9223372036854775807.0); + } +} + +DRWAV_PRIVATE void drwav__ieee_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount, unsigned int bytesPerSample) +{ + if (bytesPerSample == 4) { + unsigned int i; + for (i = 0; i < sampleCount; ++i) { + *pOut++ = ((const float*)pIn)[i]; + } + return; + } else if (bytesPerSample == 8) { + drwav_f64_to_f32(pOut, (const double*)pIn, sampleCount); + return; + } else { + /* Only supporting 32- and 64-bit float. Output silence in all other cases. Contributions welcome for 16-bit float. */ + DRWAV_ZERO_MEMORY(pOut, sampleCount * sizeof(*pOut)); + return; + } +} + + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_f32__pcm(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav__pcm_to_f32(pBufferOut, sampleData, (size_t)samplesRead, bytesPerSample); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_f32__msadpcm_ima(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut) +{ + /* + We're just going to borrow the implementation from the drwav_read_s16() since ADPCM is a little bit more complicated than other formats and I don't + want to duplicate that code. + */ + drwav_uint64 totalFramesRead; + drwav_int16 samples16[2048]; + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, drwav_countof(samples16)/pWav->channels); + drwav_uint64 framesRead = drwav_read_pcm_frames_s16(pWav, framesToReadThisIteration, samples16); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + drwav_s16_to_f32(pBufferOut, samples16, (size_t)(framesRead*pWav->channels)); /* <-- Safe cast because we're clamping to 2048. */ + + pBufferOut += framesRead*pWav->channels; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_f32__ieee(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + /* Fast path. */ + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_IEEE_FLOAT && pWav->bitsPerSample == 32) { + return drwav_read_pcm_frames(pWav, framesToRead, pBufferOut); + } + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav__ieee_to_f32(pBufferOut, sampleData, (size_t)samplesRead, bytesPerSample); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_f32__alaw(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav_alaw_to_f32(pBufferOut, sampleData, (size_t)samplesRead); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_f32__mulaw(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav_mulaw_to_f32(pBufferOut, sampleData, (size_t)samplesRead); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_f32(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut) +{ + if (pWav == NULL || framesToRead == 0) { + return 0; + } + + if (pBufferOut == NULL) { + return drwav_read_pcm_frames(pWav, framesToRead, NULL); + } + + /* Don't try to read more samples than can potentially fit in the output buffer. */ + if (framesToRead * pWav->channels * sizeof(float) > DRWAV_SIZE_MAX) { + framesToRead = DRWAV_SIZE_MAX / sizeof(float) / pWav->channels; + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_PCM) { + return drwav_read_pcm_frames_f32__pcm(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ADPCM || pWav->translatedFormatTag == DR_WAVE_FORMAT_DVI_ADPCM) { + return drwav_read_pcm_frames_f32__msadpcm_ima(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_IEEE_FLOAT) { + return drwav_read_pcm_frames_f32__ieee(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ALAW) { + return drwav_read_pcm_frames_f32__alaw(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_MULAW) { + return drwav_read_pcm_frames_f32__mulaw(pWav, framesToRead, pBufferOut); + } + + return 0; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_f32le(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut) +{ + drwav_uint64 framesRead = drwav_read_pcm_frames_f32(pWav, framesToRead, pBufferOut); + if (pBufferOut != NULL && drwav__is_little_endian() == DRWAV_FALSE) { + drwav__bswap_samples_f32(pBufferOut, framesRead*pWav->channels); + } + + return framesRead; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_f32be(drwav* pWav, drwav_uint64 framesToRead, float* pBufferOut) +{ + drwav_uint64 framesRead = drwav_read_pcm_frames_f32(pWav, framesToRead, pBufferOut); + if (pBufferOut != NULL && drwav__is_little_endian() == DRWAV_TRUE) { + drwav__bswap_samples_f32(pBufferOut, framesRead*pWav->channels); + } + + return framesRead; +} + + +DRWAV_API void drwav_u8_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = ((int)pIn[i] - 128) * 0.0078125f; + } +} + +DRWAV_API void drwav_s16_to_f32(float* pOut, const drwav_int16* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = pIn[i] * 0.000030517578125f; + } +} + +DRWAV_API void drwav_s24_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + double x; + drwav_uint32 a = ((drwav_uint32)(pIn[i*3+0]) << 8); + drwav_uint32 b = ((drwav_uint32)(pIn[i*3+1]) << 16); + drwav_uint32 c = ((drwav_uint32)(pIn[i*3+2]) << 24); + + x = (double)((drwav_int32)(a | b | c) >> 8); + *pOut++ = (float)(x * 0.00000011920928955078125); + } +} + +DRWAV_API void drwav_s32_to_f32(float* pOut, const drwav_int32* pIn, size_t sampleCount) +{ + size_t i; + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = (float)(pIn[i] / 2147483648.0); + } +} + +DRWAV_API void drwav_f64_to_f32(float* pOut, const double* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = (float)pIn[i]; + } +} + +DRWAV_API void drwav_alaw_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = drwav__alaw_to_s16(pIn[i]) / 32768.0f; + } +} + +DRWAV_API void drwav_mulaw_to_f32(float* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = drwav__mulaw_to_s16(pIn[i]) / 32768.0f; + } +} + + + +DRWAV_PRIVATE void drwav__pcm_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t totalSampleCount, unsigned int bytesPerSample) +{ + unsigned int i; + + /* Special case for 8-bit sample data because it's treated as unsigned. */ + if (bytesPerSample == 1) { + drwav_u8_to_s32(pOut, pIn, totalSampleCount); + return; + } + + /* Slightly more optimal implementation for common formats. */ + if (bytesPerSample == 2) { + drwav_s16_to_s32(pOut, (const drwav_int16*)pIn, totalSampleCount); + return; + } + if (bytesPerSample == 3) { + drwav_s24_to_s32(pOut, pIn, totalSampleCount); + return; + } + if (bytesPerSample == 4) { + for (i = 0; i < totalSampleCount; ++i) { + *pOut++ = ((const drwav_int32*)pIn)[i]; + } + return; + } + + + /* Anything more than 64 bits per sample is not supported. */ + if (bytesPerSample > 8) { + DRWAV_ZERO_MEMORY(pOut, totalSampleCount * sizeof(*pOut)); + return; + } + + + /* Generic, slow converter. */ + for (i = 0; i < totalSampleCount; ++i) { + drwav_uint64 sample = 0; + unsigned int shift = (8 - bytesPerSample) * 8; + + unsigned int j; + for (j = 0; j < bytesPerSample; j += 1) { + DRWAV_ASSERT(j < 8); + sample |= (drwav_uint64)(pIn[j]) << shift; + shift += 8; + } + + if (!drwav__is_little_endian()) { + sample = drwav__bswap64(sample); + } + + pIn += j; + *pOut++ = (drwav_int32)((drwav_int64)sample >> 32); + } +} + +DRWAV_PRIVATE void drwav__ieee_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t totalSampleCount, unsigned int bytesPerSample) +{ + if (bytesPerSample == 4) { + drwav_f32_to_s32(pOut, (const float*)pIn, totalSampleCount); + return; + } else if (bytesPerSample == 8) { + drwav_f64_to_s32(pOut, (const double*)pIn, totalSampleCount); + return; + } else { + /* Only supporting 32- and 64-bit float. Output silence in all other cases. Contributions welcome for 16-bit float. */ + DRWAV_ZERO_MEMORY(pOut, totalSampleCount * sizeof(*pOut)); + return; + } +} + + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s32__pcm(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + /* Fast path. */ + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_PCM && pWav->bitsPerSample == 32) { + return drwav_read_pcm_frames(pWav, framesToRead, pBufferOut); + } + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav__pcm_to_s32(pBufferOut, sampleData, (size_t)samplesRead, bytesPerSample); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s32__msadpcm_ima(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut) +{ + /* + We're just going to borrow the implementation from the drwav_read_s16() since ADPCM is a little bit more complicated than other formats and I don't + want to duplicate that code. + */ + drwav_uint64 totalFramesRead = 0; + drwav_int16 samples16[2048]; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, drwav_countof(samples16)/pWav->channels); + drwav_uint64 framesRead = drwav_read_pcm_frames_s16(pWav, framesToReadThisIteration, samples16); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + drwav_s16_to_s32(pBufferOut, samples16, (size_t)(framesRead*pWav->channels)); /* <-- Safe cast because we're clamping to 2048. */ + + pBufferOut += framesRead*pWav->channels; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s32__ieee(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav__ieee_to_s32(pBufferOut, sampleData, (size_t)samplesRead, bytesPerSample); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s32__alaw(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav_alaw_to_s32(pBufferOut, sampleData, (size_t)samplesRead); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_PRIVATE drwav_uint64 drwav_read_pcm_frames_s32__mulaw(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut) +{ + drwav_uint64 totalFramesRead; + drwav_uint8 sampleData[4096] = {0}; + drwav_uint32 bytesPerFrame; + drwav_uint32 bytesPerSample; + drwav_uint64 samplesRead; + + bytesPerFrame = drwav_get_bytes_per_pcm_frame(pWav); + if (bytesPerFrame == 0) { + return 0; + } + + bytesPerSample = bytesPerFrame / pWav->channels; + if (bytesPerSample == 0 || (bytesPerFrame % pWav->channels) != 0) { + return 0; /* Only byte-aligned formats are supported. */ + } + + totalFramesRead = 0; + + while (framesToRead > 0) { + drwav_uint64 framesToReadThisIteration = drwav_min(framesToRead, sizeof(sampleData)/bytesPerFrame); + drwav_uint64 framesRead = drwav_read_pcm_frames(pWav, framesToReadThisIteration, sampleData); + if (framesRead == 0) { + break; + } + + DRWAV_ASSERT(framesRead <= framesToReadThisIteration); /* If this fails it means there's a bug in drwav_read_pcm_frames(). */ + + /* Validation to ensure we don't read too much from out intermediary buffer. This is to protect from invalid files. */ + samplesRead = framesRead * pWav->channels; + if ((samplesRead * bytesPerSample) > sizeof(sampleData)) { + DRWAV_ASSERT(DRWAV_FALSE); /* This should never happen with a valid file. */ + break; + } + + drwav_mulaw_to_s32(pBufferOut, sampleData, (size_t)samplesRead); + + pBufferOut += samplesRead; + framesToRead -= framesRead; + totalFramesRead += framesRead; + } + + return totalFramesRead; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s32(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut) +{ + if (pWav == NULL || framesToRead == 0) { + return 0; + } + + if (pBufferOut == NULL) { + return drwav_read_pcm_frames(pWav, framesToRead, NULL); + } + + /* Don't try to read more samples than can potentially fit in the output buffer. */ + if (framesToRead * pWav->channels * sizeof(drwav_int32) > DRWAV_SIZE_MAX) { + framesToRead = DRWAV_SIZE_MAX / sizeof(drwav_int32) / pWav->channels; + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_PCM) { + return drwav_read_pcm_frames_s32__pcm(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ADPCM || pWav->translatedFormatTag == DR_WAVE_FORMAT_DVI_ADPCM) { + return drwav_read_pcm_frames_s32__msadpcm_ima(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_IEEE_FLOAT) { + return drwav_read_pcm_frames_s32__ieee(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_ALAW) { + return drwav_read_pcm_frames_s32__alaw(pWav, framesToRead, pBufferOut); + } + + if (pWav->translatedFormatTag == DR_WAVE_FORMAT_MULAW) { + return drwav_read_pcm_frames_s32__mulaw(pWav, framesToRead, pBufferOut); + } + + return 0; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s32le(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut) +{ + drwav_uint64 framesRead = drwav_read_pcm_frames_s32(pWav, framesToRead, pBufferOut); + if (pBufferOut != NULL && drwav__is_little_endian() == DRWAV_FALSE) { + drwav__bswap_samples_s32(pBufferOut, framesRead*pWav->channels); + } + + return framesRead; +} + +DRWAV_API drwav_uint64 drwav_read_pcm_frames_s32be(drwav* pWav, drwav_uint64 framesToRead, drwav_int32* pBufferOut) +{ + drwav_uint64 framesRead = drwav_read_pcm_frames_s32(pWav, framesToRead, pBufferOut); + if (pBufferOut != NULL && drwav__is_little_endian() == DRWAV_TRUE) { + drwav__bswap_samples_s32(pBufferOut, framesRead*pWav->channels); + } + + return framesRead; +} + + +DRWAV_API void drwav_u8_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = ((int)pIn[i] - 128) * 16777216; + } +} + +DRWAV_API void drwav_s16_to_s32(drwav_int32* pOut, const drwav_int16* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = (drwav_int32)pIn[i] * 65536; + } +} + +DRWAV_API void drwav_s24_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + unsigned int s0 = pIn[i*3 + 0]; + unsigned int s1 = pIn[i*3 + 1]; + unsigned int s2 = pIn[i*3 + 2]; + + drwav_int32 sample32 = (drwav_int32)((s0 << 8) | (s1 << 16) | (s2 << 24)); + *pOut++ = sample32; + } +} + +DRWAV_API void drwav_f32_to_s32(drwav_int32* pOut, const float* pIn, size_t sampleCount) +{ + size_t i; + for (i = 0; i < sampleCount; ++i) { + float x = pIn[i]; + if (x != x) { + pOut[i] = 0; /* NaN */ + } else if (x <= -1) { + pOut[i] = (-2147483647 - 1); + } else if (x >= 1) { + pOut[i] = 2147483647; + } else { + pOut[i] = (drwav_int32)(x * 2147483648.0f); + } + } +} + +DRWAV_API void drwav_f64_to_s32(drwav_int32* pOut, const double* pIn, size_t sampleCount) +{ + size_t i; + for (i = 0; i < sampleCount; ++i) { + double x = pIn[i]; + if (x != x) { + pOut[i] = 0; /* NaN */ + } else if (x <= -1) { + pOut[i] = (-2147483647 - 1); + } else if (x >= 1) { + pOut[i] = 2147483647; + } else { + pOut[i] = (drwav_int32)(x * 2147483648.0); + } + } +} + +DRWAV_API void drwav_alaw_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i = 0; i < sampleCount; ++i) { + *pOut++ = (drwav_int32)drwav__alaw_to_s16(pIn[i]) * 65536; + } +} + +DRWAV_API void drwav_mulaw_to_s32(drwav_int32* pOut, const drwav_uint8* pIn, size_t sampleCount) +{ + size_t i; + + if (pOut == NULL || pIn == NULL) { + return; + } + + for (i= 0; i < sampleCount; ++i) { + *pOut++ = (drwav_int32)drwav__mulaw_to_s16(pIn[i]) * 65536; + } +} + + + +DRWAV_PRIVATE drwav_int16* drwav__read_pcm_frames_and_close_s16(drwav* pWav, unsigned int* channels, unsigned int* sampleRate, drwav_uint64* totalFrameCount) +{ + drwav_uint64 sampleDataSize; + drwav_int16* pSampleData; + drwav_uint64 framesRead; + + DRWAV_ASSERT(pWav != NULL); + + /* Check for overflow before multiplication. */ + if (pWav->channels == 0 || pWav->totalPCMFrameCount > DRWAV_SIZE_MAX / pWav->channels / sizeof(drwav_int16)) { + drwav_uninit(pWav); + return NULL; /* Overflow or invalid channels. */ + } + + sampleDataSize = pWav->totalPCMFrameCount * pWav->channels * sizeof(drwav_int16); + if (sampleDataSize > DRWAV_SIZE_MAX) { + drwav_uninit(pWav); + return NULL; /* File's too big. */ + } + + pSampleData = (drwav_int16*)drwav__malloc_from_callbacks((size_t)sampleDataSize, &pWav->allocationCallbacks); /* <-- Safe cast due to the check above. */ + if (pSampleData == NULL) { + drwav_uninit(pWav); + return NULL; /* Failed to allocate memory. */ + } + + framesRead = drwav_read_pcm_frames_s16(pWav, (size_t)pWav->totalPCMFrameCount, pSampleData); + if (framesRead != pWav->totalPCMFrameCount) { + drwav__free_from_callbacks(pSampleData, &pWav->allocationCallbacks); + drwav_uninit(pWav); + return NULL; /* There was an error reading the samples. */ + } + + drwav_uninit(pWav); + + if (sampleRate) { + *sampleRate = pWav->sampleRate; + } + if (channels) { + *channels = pWav->channels; + } + if (totalFrameCount) { + *totalFrameCount = pWav->totalPCMFrameCount; + } + + return pSampleData; +} + +DRWAV_PRIVATE float* drwav__read_pcm_frames_and_close_f32(drwav* pWav, unsigned int* channels, unsigned int* sampleRate, drwav_uint64* totalFrameCount) +{ + drwav_uint64 sampleDataSize; + float* pSampleData; + drwav_uint64 framesRead; + + DRWAV_ASSERT(pWav != NULL); + + /* Check for overflow before multiplication. */ + if (pWav->channels == 0 || pWav->totalPCMFrameCount > DRWAV_SIZE_MAX / pWav->channels / sizeof(float)) { + drwav_uninit(pWav); + return NULL; /* Overflow or invalid channels. */ + } + + sampleDataSize = pWav->totalPCMFrameCount * pWav->channels * sizeof(float); + if (sampleDataSize > DRWAV_SIZE_MAX) { + drwav_uninit(pWav); + return NULL; /* File's too big. */ + } + + pSampleData = (float*)drwav__malloc_from_callbacks((size_t)sampleDataSize, &pWav->allocationCallbacks); /* <-- Safe cast due to the check above. */ + if (pSampleData == NULL) { + drwav_uninit(pWav); + return NULL; /* Failed to allocate memory. */ + } + + framesRead = drwav_read_pcm_frames_f32(pWav, (size_t)pWav->totalPCMFrameCount, pSampleData); + if (framesRead != pWav->totalPCMFrameCount) { + drwav__free_from_callbacks(pSampleData, &pWav->allocationCallbacks); + drwav_uninit(pWav); + return NULL; /* There was an error reading the samples. */ + } + + drwav_uninit(pWav); + + if (sampleRate) { + *sampleRate = pWav->sampleRate; + } + if (channels) { + *channels = pWav->channels; + } + if (totalFrameCount) { + *totalFrameCount = pWav->totalPCMFrameCount; + } + + return pSampleData; +} + +DRWAV_PRIVATE drwav_int32* drwav__read_pcm_frames_and_close_s32(drwav* pWav, unsigned int* channels, unsigned int* sampleRate, drwav_uint64* totalFrameCount) +{ + drwav_uint64 sampleDataSize; + drwav_int32* pSampleData; + drwav_uint64 framesRead; + + DRWAV_ASSERT(pWav != NULL); + + /* Check for overflow before multiplication. */ + if (pWav->channels == 0 || pWav->totalPCMFrameCount > DRWAV_SIZE_MAX / pWav->channels / sizeof(drwav_int32)) { + drwav_uninit(pWav); + return NULL; /* Overflow or invalid channels. */ + } + + sampleDataSize = pWav->totalPCMFrameCount * pWav->channels * sizeof(drwav_int32); + if (sampleDataSize > DRWAV_SIZE_MAX) { + drwav_uninit(pWav); + return NULL; /* File's too big. */ + } + + pSampleData = (drwav_int32*)drwav__malloc_from_callbacks((size_t)sampleDataSize, &pWav->allocationCallbacks); /* <-- Safe cast due to the check above. */ + if (pSampleData == NULL) { + drwav_uninit(pWav); + return NULL; /* Failed to allocate memory. */ + } + + framesRead = drwav_read_pcm_frames_s32(pWav, (size_t)pWav->totalPCMFrameCount, pSampleData); + if (framesRead != pWav->totalPCMFrameCount) { + drwav__free_from_callbacks(pSampleData, &pWav->allocationCallbacks); + drwav_uninit(pWav); + return NULL; /* There was an error reading the samples. */ + } + + drwav_uninit(pWav); + + if (sampleRate) { + *sampleRate = pWav->sampleRate; + } + if (channels) { + *channels = pWav->channels; + } + if (totalFrameCount) { + *totalFrameCount = pWav->totalPCMFrameCount; + } + + return pSampleData; +} + + + +DRWAV_API drwav_int16* drwav_open_and_read_pcm_frames_s16(drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init(&wav, onRead, onSeek, onTell, pUserData, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_s16(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +DRWAV_API float* drwav_open_and_read_pcm_frames_f32(drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init(&wav, onRead, onSeek, onTell, pUserData, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_f32(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +DRWAV_API drwav_int32* drwav_open_and_read_pcm_frames_s32(drwav_read_proc onRead, drwav_seek_proc onSeek, drwav_tell_proc onTell, void* pUserData, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init(&wav, onRead, onSeek, onTell, pUserData, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_s32(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +#ifndef DR_WAV_NO_STDIO +DRWAV_API drwav_int16* drwav_open_file_and_read_pcm_frames_s16(const char* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_file(&wav, filename, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_s16(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +DRWAV_API float* drwav_open_file_and_read_pcm_frames_f32(const char* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_file(&wav, filename, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_f32(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +DRWAV_API drwav_int32* drwav_open_file_and_read_pcm_frames_s32(const char* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_file(&wav, filename, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_s32(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + + +#ifndef DR_WAV_NO_WCHAR +DRWAV_API drwav_int16* drwav_open_file_and_read_pcm_frames_s16_w(const wchar_t* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (channelsOut) { + *channelsOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_file_w(&wav, filename, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_s16(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +DRWAV_API float* drwav_open_file_and_read_pcm_frames_f32_w(const wchar_t* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (channelsOut) { + *channelsOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_file_w(&wav, filename, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_f32(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +DRWAV_API drwav_int32* drwav_open_file_and_read_pcm_frames_s32_w(const wchar_t* filename, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (channelsOut) { + *channelsOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_file_w(&wav, filename, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_s32(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} +#endif /* DR_WAV_NO_WCHAR */ +#endif /* DR_WAV_NO_STDIO */ + +DRWAV_API drwav_int16* drwav_open_memory_and_read_pcm_frames_s16(const void* data, size_t dataSize, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_memory(&wav, data, dataSize, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_s16(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +DRWAV_API float* drwav_open_memory_and_read_pcm_frames_f32(const void* data, size_t dataSize, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_memory(&wav, data, dataSize, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_f32(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} + +DRWAV_API drwav_int32* drwav_open_memory_and_read_pcm_frames_s32(const void* data, size_t dataSize, unsigned int* channelsOut, unsigned int* sampleRateOut, drwav_uint64* totalFrameCountOut, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + drwav wav; + + if (channelsOut) { + *channelsOut = 0; + } + if (sampleRateOut) { + *sampleRateOut = 0; + } + if (totalFrameCountOut) { + *totalFrameCountOut = 0; + } + + if (!drwav_init_memory(&wav, data, dataSize, pAllocationCallbacks)) { + return NULL; + } + + return drwav__read_pcm_frames_and_close_s32(&wav, channelsOut, sampleRateOut, totalFrameCountOut); +} +#endif /* DR_WAV_NO_CONVERSION_API */ + + +DRWAV_API void drwav_free(void* p, const drwav_allocation_callbacks* pAllocationCallbacks) +{ + if (pAllocationCallbacks != NULL) { + drwav__free_from_callbacks(p, pAllocationCallbacks); + } else { + drwav__free_default(p, NULL); + } +} + +DRWAV_API drwav_uint16 drwav_bytes_to_u16(const drwav_uint8* data) +{ + return ((drwav_uint16)data[0] << 0) | ((drwav_uint16)data[1] << 8); +} + +DRWAV_API drwav_int16 drwav_bytes_to_s16(const drwav_uint8* data) +{ + return (drwav_int16)drwav_bytes_to_u16(data); +} + +DRWAV_API drwav_uint32 drwav_bytes_to_u32(const drwav_uint8* data) +{ + return drwav_bytes_to_u32_le(data); +} + +DRWAV_API float drwav_bytes_to_f32(const drwav_uint8* data) +{ + union { + drwav_uint32 u32; + float f32; + } value; + + value.u32 = drwav_bytes_to_u32(data); + return value.f32; +} + +DRWAV_API drwav_int32 drwav_bytes_to_s32(const drwav_uint8* data) +{ + return (drwav_int32)drwav_bytes_to_u32(data); +} + +DRWAV_API drwav_uint64 drwav_bytes_to_u64(const drwav_uint8* data) +{ + return + ((drwav_uint64)data[0] << 0) | ((drwav_uint64)data[1] << 8) | ((drwav_uint64)data[2] << 16) | ((drwav_uint64)data[3] << 24) | + ((drwav_uint64)data[4] << 32) | ((drwav_uint64)data[5] << 40) | ((drwav_uint64)data[6] << 48) | ((drwav_uint64)data[7] << 56); +} + +DRWAV_API drwav_int64 drwav_bytes_to_s64(const drwav_uint8* data) +{ + return (drwav_int64)drwav_bytes_to_u64(data); +} + + +DRWAV_API drwav_bool32 drwav_guid_equal(const drwav_uint8 a[16], const drwav_uint8 b[16]) +{ + int i; + for (i = 0; i < 16; i += 1) { + if (a[i] != b[i]) { + return DRWAV_FALSE; + } + } + + return DRWAV_TRUE; +} + +DRWAV_API drwav_bool32 drwav_fourcc_equal(const drwav_uint8* a, const char* b) +{ + return + a[0] == b[0] && + a[1] == b[1] && + a[2] == b[2] && + a[3] == b[3]; +} + +#ifdef __MRC__ +/* Undo the pragma at the beginning of this file. */ +#pragma options opt reset +#endif + +#endif /* dr_wav_c */ +#endif /* DR_WAV_IMPLEMENTATION */ + +/* +This software is available as a choice of the following licenses. Choose +whichever you prefer. + +=============================================================================== +ALTERNATIVE 1 - Public Domain (www.unlicense.org) +=============================================================================== +This is free and unencumbered software released into the public domain. + +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. + +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +For more information, please refer to <http://unlicense.org/> + +=============================================================================== +ALTERNATIVE 2 - MIT No Attribution +=============================================================================== +Copyright 2023 David Reid + +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +*/ diff --git a/vendor/fonts/Go-Mono.ttf b/vendor/fonts/Go-Mono.ttf Binary files differnew file mode 100644 index 0000000..853d473 --- /dev/null +++ b/vendor/fonts/Go-Mono.ttf diff --git a/vendor/fonts/LICENSE b/vendor/fonts/LICENSE new file mode 100644 index 0000000..7043c36 --- /dev/null +++ b/vendor/fonts/LICENSE @@ -0,0 +1,36 @@ +These fonts were created by the Bigelow & Holmes foundry specifically for the +Go project. See https://blog.golang.org/go-fonts for details. + +They are licensed under the same open source license as the rest of the Go +project's software: + +Copyright (c) 2016 Bigelow & Holmes Inc.. All rights reserved. + +Distribution of this font is governed by the following license. If you do not +agree to this license, including the disclaimer, do not distribute or modify +this font. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the documentation + and/or other materials provided with the distribution. + + * Neither the name of Google Inc. nor the names of its contributors may be + used to endorse or promote products derived from this software without + specific prior written permission. + +DISCLAIMER: THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, +THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE +FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/vendor/fonts/README.md b/vendor/fonts/README.md new file mode 100644 index 0000000..f86058e --- /dev/null +++ b/vendor/fonts/README.md @@ -0,0 +1,52 @@ +# Native text + +Linux Sketchbook text uses bundled **Go Mono** through PangoCairo: HarfBuzz +shaping, Fontconfig fallback and FreeType/Cairo grayscale antialiasing with light +hinting. This follows the rendering quality choices in Pardes `src/gui/font.c`; +it does not copy Pardes's terminal-cell shaping, bold synthesis or glyph atlas. + +`Go-Mono.ttf` is from the Go project's official `golang/image` repository: +https://github.com/golang/image/tree/master/font/gofont/ttfs +SHA-256: `8bc66a0154bbf69cd24e5bde41a12ec9495c8a242c5def2255e4a164900f1ed7`. +The complete Bigelow & Holmes / Go font license is in `LICENSE`. + +The font is embedded in the binary with C23 `#embed`. It is registered in a +private Fontconfig config through a Linux memfd, so moving the executable does +not require a font installation or the source checkout. Installed outline fonts +provide Unicode fallback, in the order listed in `vendor/text.c`; actual glyph +coverage and emoji color depend on those fonts. Color font paint is preserved. +No system Fontconfig settings are modified. + +The `f.text`/`f.print` API is unchanged. Their x/y positions mean line-box top. +Scale 1 uses a nominal 13px font: Go Mono measures 8px ASCII advance and 16px +line height on the tested stack. At other sizes advances may round to a physical +pixel. Applications should allow a 16×scale line box and avoid slicing UTF-8 at +arbitrary byte boundaries. Pango shapes each supplied string independently; +applications still perform wrapping and clipping. Explicit newlines are not a +layout API. Other platforms retain SDL's debug-font fallback for now. + +Text is rasterized at the renderer's physical pixel density (including logical +presentation scale), and texture origins snap to physical pixels. Whole-run +textures retain antialiasing and shaping; no scaled 8×8 bitmap text is involved. +The renderer-owned cache retains at most 512 strings / 16 MiB of pixel data. +Only prior-frame textures can be evicted; exceptionally dense frames flush and +release temporary textures rather than growing the cache. Renderer destruction +releases the font map, descriptors and keys after SDL releases its textures. + +Native Linux builds need `pkg-config`, `pangocairo`, and `pangoft2` development +packages, in addition to SDL. The resulting executable uses the system Pango, +Cairo, HarfBuzz, Fontconfig and FreeType libraries. + +Run the independent regression and specimen (with system SDL3): + +```sh +cc -std=c23 -Wall -Wextra -Werror -O2 test/fonts.c -o /tmp/sketchbook-fonts \ + $(pkg-config --cflags --libs sdl3 pangocairo pangoft2) -lm +/tmp/sketchbook-fonts /tmp/sketchbook-fonts.png +``` + +It checks the primary font's advance/line box, Unicode fallback, zero-advance +combining marks, Arabic lam-alef shaping, 2× rendering density, clipping, cache +eviction and renderer cleanup. The specimen includes Latin, Greek, Cyrillic, +Japanese, Arabic, Hebrew and symbols/emoji. Missing fallback fonts fail the +coverage assertion rather than silently claiming complete Unicode support. diff --git a/vendor/stb_vorbis.c b/vendor/stb_vorbis.c new file mode 100644 index 0000000..3e5c250 --- /dev/null +++ b/vendor/stb_vorbis.c @@ -0,0 +1,5584 @@ +// Ogg Vorbis audio decoder - v1.22 - public domain +// http://nothings.org/stb_vorbis/ +// +// Original version written by Sean Barrett in 2007. +// +// Originally sponsored by RAD Game Tools. Seeking implementation +// sponsored by Phillip Bennefall, Marc Andersen, Aaron Baker, +// Elias Software, Aras Pranckevicius, and Sean Barrett. +// +// LICENSE +// +// See end of file for license information. +// +// Limitations: +// +// - floor 0 not supported (used in old ogg vorbis files pre-2004) +// - lossless sample-truncation at beginning ignored +// - cannot concatenate multiple vorbis streams +// - sample positions are 32-bit, limiting seekable 192Khz +// files to around 6 hours (Ogg supports 64-bit) +// +// Feature contributors: +// Dougall Johnson (sample-exact seeking) +// +// Bugfix/warning contributors: +// Terje Mathisen Niklas Frykholm Andy Hill +// Casey Muratori John Bolton Gargaj +// Laurent Gomila Marc LeBlanc Ronny Chevalier +// Bernhard Wodo Evan Balster github:alxprd +// Tom Beaumont Ingo Leitgeb Nicolas Guillemot +// Phillip Bennefall Rohit Thiago Goulart +// github:manxorist Saga Musix github:infatum +// Timur Gagiev Maxwell Koo Peter Waller +// github:audinowho Dougall Johnson David Reid +// github:Clownacy Pedro J. Estebanez Remi Verschelde +// AnthoFoxo github:morlat Gabriel Ravier +// +// Partial history: +// 1.22 - 2021-07-11 - various small fixes +// 1.21 - 2021-07-02 - fix bug for files with no comments +// 1.20 - 2020-07-11 - several small fixes +// 1.19 - 2020-02-05 - warnings +// 1.18 - 2020-02-02 - fix seek bugs; parse header comments; misc warnings etc. +// 1.17 - 2019-07-08 - fix CVE-2019-13217..CVE-2019-13223 (by ForAllSecure) +// 1.16 - 2019-03-04 - fix warnings +// 1.15 - 2019-02-07 - explicit failure if Ogg Skeleton data is found +// 1.14 - 2018-02-11 - delete bogus dealloca usage +// 1.13 - 2018-01-29 - fix truncation of last frame (hopefully) +// 1.12 - 2017-11-21 - limit residue begin/end to blocksize/2 to avoid large temp allocs in bad/corrupt files +// 1.11 - 2017-07-23 - fix MinGW compilation +// 1.10 - 2017-03-03 - more robust seeking; fix negative ilog(); clear error in open_memory +// 1.09 - 2016-04-04 - back out 'truncation of last frame' fix from previous version +// 1.08 - 2016-04-02 - warnings; setup memory leaks; truncation of last frame +// 1.07 - 2015-01-16 - fixes for crashes on invalid files; warning fixes; const +// 1.06 - 2015-08-31 - full, correct support for seeking API (Dougall Johnson) +// some crash fixes when out of memory or with corrupt files +// fix some inappropriately signed shifts +// 1.05 - 2015-04-19 - don't define __forceinline if it's redundant +// 1.04 - 2014-08-27 - fix missing const-correct case in API +// 1.03 - 2014-08-07 - warning fixes +// 1.02 - 2014-07-09 - declare qsort comparison as explicitly _cdecl in Windows +// 1.01 - 2014-06-18 - fix stb_vorbis_get_samples_float (interleaved was correct) +// 1.0 - 2014-05-26 - fix memory leaks; fix warnings; fix bugs in >2-channel; +// (API change) report sample rate for decode-full-file funcs +// +// See end of file for full version history. + + +////////////////////////////////////////////////////////////////////////////// +// +// HEADER BEGINS HERE +// + +#ifndef STB_VORBIS_INCLUDE_STB_VORBIS_H +#define STB_VORBIS_INCLUDE_STB_VORBIS_H + +#if defined(STB_VORBIS_NO_CRT) && !defined(STB_VORBIS_NO_STDIO) +#define STB_VORBIS_NO_STDIO 1 +#endif + +#ifndef STB_VORBIS_NO_STDIO +#include <stdio.h> +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +/////////// THREAD SAFETY + +// Individual stb_vorbis* handles are not thread-safe; you cannot decode from +// them from multiple threads at the same time. However, you can have multiple +// stb_vorbis* handles and decode from them independently in multiple thrads. + + +/////////// MEMORY ALLOCATION + +// normally stb_vorbis uses malloc() to allocate memory at startup, +// and alloca() to allocate temporary memory during a frame on the +// stack. (Memory consumption will depend on the amount of setup +// data in the file and how you set the compile flags for speed +// vs. size. In my test files the maximal-size usage is ~150KB.) +// +// You can modify the wrapper functions in the source (setup_malloc, +// setup_temp_malloc, temp_malloc) to change this behavior, or you +// can use a simpler allocation model: you pass in a buffer from +// which stb_vorbis will allocate _all_ its memory (including the +// temp memory). "open" may fail with a VORBIS_outofmem if you +// do not pass in enough data; there is no way to determine how +// much you do need except to succeed (at which point you can +// query get_info to find the exact amount required. yes I know +// this is lame). +// +// If you pass in a non-NULL buffer of the type below, allocation +// will occur from it as described above. Otherwise just pass NULL +// to use malloc()/alloca() + +typedef struct +{ + char *alloc_buffer; + int alloc_buffer_length_in_bytes; +} stb_vorbis_alloc; + + +/////////// FUNCTIONS USEABLE WITH ALL INPUT MODES + +typedef struct stb_vorbis stb_vorbis; + +typedef struct +{ + unsigned int sample_rate; + int channels; + + unsigned int setup_memory_required; + unsigned int setup_temp_memory_required; + unsigned int temp_memory_required; + + int max_frame_size; +} stb_vorbis_info; + +typedef struct +{ + char *vendor; + + int comment_list_length; + char **comment_list; +} stb_vorbis_comment; + +// get general information about the file +extern stb_vorbis_info stb_vorbis_get_info(stb_vorbis *f); + +// get ogg comments +extern stb_vorbis_comment stb_vorbis_get_comment(stb_vorbis *f); + +// get the last error detected (clears it, too) +extern int stb_vorbis_get_error(stb_vorbis *f); + +// close an ogg vorbis file and free all memory in use +extern void stb_vorbis_close(stb_vorbis *f); + +// this function returns the offset (in samples) from the beginning of the +// file that will be returned by the next decode, if it is known, or -1 +// otherwise. after a flush_pushdata() call, this may take a while before +// it becomes valid again. +// NOT WORKING YET after a seek with PULLDATA API +extern int stb_vorbis_get_sample_offset(stb_vorbis *f); + +// returns the current seek point within the file, or offset from the beginning +// of the memory buffer. In pushdata mode it returns 0. +extern unsigned int stb_vorbis_get_file_offset(stb_vorbis *f); + +/////////// PUSHDATA API + +#ifndef STB_VORBIS_NO_PUSHDATA_API + +// this API allows you to get blocks of data from any source and hand +// them to stb_vorbis. you have to buffer them; stb_vorbis will tell +// you how much it used, and you have to give it the rest next time; +// and stb_vorbis may not have enough data to work with and you will +// need to give it the same data again PLUS more. Note that the Vorbis +// specification does not bound the size of an individual frame. + +extern stb_vorbis *stb_vorbis_open_pushdata( + const unsigned char * datablock, int datablock_length_in_bytes, + int *datablock_memory_consumed_in_bytes, + int *error, + const stb_vorbis_alloc *alloc_buffer); +// create a vorbis decoder by passing in the initial data block containing +// the ogg&vorbis headers (you don't need to do parse them, just provide +// the first N bytes of the file--you're told if it's not enough, see below) +// on success, returns an stb_vorbis *, does not set error, returns the amount of +// data parsed/consumed on this call in *datablock_memory_consumed_in_bytes; +// on failure, returns NULL on error and sets *error, does not change *datablock_memory_consumed +// if returns NULL and *error is VORBIS_need_more_data, then the input block was +// incomplete and you need to pass in a larger block from the start of the file + +extern int stb_vorbis_decode_frame_pushdata( + stb_vorbis *f, + const unsigned char *datablock, int datablock_length_in_bytes, + int *channels, // place to write number of float * buffers + float ***output, // place to write float ** array of float * buffers + int *samples // place to write number of output samples + ); +// decode a frame of audio sample data if possible from the passed-in data block +// +// return value: number of bytes we used from datablock +// +// possible cases: +// 0 bytes used, 0 samples output (need more data) +// N bytes used, 0 samples output (resynching the stream, keep going) +// N bytes used, M samples output (one frame of data) +// note that after opening a file, you will ALWAYS get one N-bytes,0-sample +// frame, because Vorbis always "discards" the first frame. +// +// Note that on resynch, stb_vorbis will rarely consume all of the buffer, +// instead only datablock_length_in_bytes-3 or less. This is because it wants +// to avoid missing parts of a page header if they cross a datablock boundary, +// without writing state-machiney code to record a partial detection. +// +// The number of channels returned are stored in *channels (which can be +// NULL--it is always the same as the number of channels reported by +// get_info). *output will contain an array of float* buffers, one per +// channel. In other words, (*output)[0][0] contains the first sample from +// the first channel, and (*output)[1][0] contains the first sample from +// the second channel. +// +// *output points into stb_vorbis's internal output buffer storage; these +// buffers are owned by stb_vorbis and application code should not free +// them or modify their contents. They are transient and will be overwritten +// once you ask for more data to get decoded, so be sure to grab any data +// you need before then. + +extern void stb_vorbis_flush_pushdata(stb_vorbis *f); +// inform stb_vorbis that your next datablock will not be contiguous with +// previous ones (e.g. you've seeked in the data); future attempts to decode +// frames will cause stb_vorbis to resynchronize (as noted above), and +// once it sees a valid Ogg page (typically 4-8KB, as large as 64KB), it +// will begin decoding the _next_ frame. +// +// if you want to seek using pushdata, you need to seek in your file, then +// call stb_vorbis_flush_pushdata(), then start calling decoding, then once +// decoding is returning you data, call stb_vorbis_get_sample_offset, and +// if you don't like the result, seek your file again and repeat. +#endif + + +////////// PULLING INPUT API + +#ifndef STB_VORBIS_NO_PULLDATA_API +// This API assumes stb_vorbis is allowed to pull data from a source-- +// either a block of memory containing the _entire_ vorbis stream, or a +// FILE * that you or it create, or possibly some other reading mechanism +// if you go modify the source to replace the FILE * case with some kind +// of callback to your code. (But if you don't support seeking, you may +// just want to go ahead and use pushdata.) + +#if !defined(STB_VORBIS_NO_STDIO) && !defined(STB_VORBIS_NO_INTEGER_CONVERSION) +extern int stb_vorbis_decode_filename(const char *filename, int *channels, int *sample_rate, short **output); +#endif +#if !defined(STB_VORBIS_NO_INTEGER_CONVERSION) +extern int stb_vorbis_decode_memory(const unsigned char *mem, int len, int *channels, int *sample_rate, short **output); +#endif +// decode an entire file and output the data interleaved into a malloc()ed +// buffer stored in *output. The return value is the number of samples +// decoded, or -1 if the file could not be opened or was not an ogg vorbis file. +// When you're done with it, just free() the pointer returned in *output. + +extern stb_vorbis * stb_vorbis_open_memory(const unsigned char *data, int len, + int *error, const stb_vorbis_alloc *alloc_buffer); +// create an ogg vorbis decoder from an ogg vorbis stream in memory (note +// this must be the entire stream!). on failure, returns NULL and sets *error + +#ifndef STB_VORBIS_NO_STDIO +extern stb_vorbis * stb_vorbis_open_filename(const char *filename, + int *error, const stb_vorbis_alloc *alloc_buffer); +// create an ogg vorbis decoder from a filename via fopen(). on failure, +// returns NULL and sets *error (possibly to VORBIS_file_open_failure). + +extern stb_vorbis * stb_vorbis_open_file(FILE *f, int close_handle_on_close, + int *error, const stb_vorbis_alloc *alloc_buffer); +// create an ogg vorbis decoder from an open FILE *, looking for a stream at +// the _current_ seek point (ftell). on failure, returns NULL and sets *error. +// note that stb_vorbis must "own" this stream; if you seek it in between +// calls to stb_vorbis, it will become confused. Moreover, if you attempt to +// perform stb_vorbis_seek_*() operations on this file, it will assume it +// owns the _entire_ rest of the file after the start point. Use the next +// function, stb_vorbis_open_file_section(), to limit it. + +extern stb_vorbis * stb_vorbis_open_file_section(FILE *f, int close_handle_on_close, + int *error, const stb_vorbis_alloc *alloc_buffer, unsigned int len); +// create an ogg vorbis decoder from an open FILE *, looking for a stream at +// the _current_ seek point (ftell); the stream will be of length 'len' bytes. +// on failure, returns NULL and sets *error. note that stb_vorbis must "own" +// this stream; if you seek it in between calls to stb_vorbis, it will become +// confused. +#endif + +extern int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number); +extern int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number); +// these functions seek in the Vorbis file to (approximately) 'sample_number'. +// after calling seek_frame(), the next call to get_frame_*() will include +// the specified sample. after calling stb_vorbis_seek(), the next call to +// stb_vorbis_get_samples_* will start with the specified sample. If you +// do not need to seek to EXACTLY the target sample when using get_samples_*, +// you can also use seek_frame(). + +extern int stb_vorbis_seek_start(stb_vorbis *f); +// this function is equivalent to stb_vorbis_seek(f,0) + +extern unsigned int stb_vorbis_stream_length_in_samples(stb_vorbis *f); +extern float stb_vorbis_stream_length_in_seconds(stb_vorbis *f); +// these functions return the total length of the vorbis stream + +extern int stb_vorbis_get_frame_float(stb_vorbis *f, int *channels, float ***output); +// decode the next frame and return the number of samples. the number of +// channels returned are stored in *channels (which can be NULL--it is always +// the same as the number of channels reported by get_info). *output will +// contain an array of float* buffers, one per channel. These outputs will +// be overwritten on the next call to stb_vorbis_get_frame_*. +// +// You generally should not intermix calls to stb_vorbis_get_frame_*() +// and stb_vorbis_get_samples_*(), since the latter calls the former. + +#ifndef STB_VORBIS_NO_INTEGER_CONVERSION +extern int stb_vorbis_get_frame_short_interleaved(stb_vorbis *f, int num_c, short *buffer, int num_shorts); +extern int stb_vorbis_get_frame_short (stb_vorbis *f, int num_c, short **buffer, int num_samples); +#endif +// decode the next frame and return the number of *samples* per channel. +// Note that for interleaved data, you pass in the number of shorts (the +// size of your array), but the return value is the number of samples per +// channel, not the total number of samples. +// +// The data is coerced to the number of channels you request according to the +// channel coercion rules (see below). You must pass in the size of your +// buffer(s) so that stb_vorbis will not overwrite the end of the buffer. +// The maximum buffer size needed can be gotten from get_info(); however, +// the Vorbis I specification implies an absolute maximum of 4096 samples +// per channel. + +// Channel coercion rules: +// Let M be the number of channels requested, and N the number of channels present, +// and Cn be the nth channel; let stereo L be the sum of all L and center channels, +// and stereo R be the sum of all R and center channels (channel assignment from the +// vorbis spec). +// M N output +// 1 k sum(Ck) for all k +// 2 * stereo L, stereo R +// k l k > l, the first l channels, then 0s +// k l k <= l, the first k channels +// Note that this is not _good_ surround etc. mixing at all! It's just so +// you get something useful. + +extern int stb_vorbis_get_samples_float_interleaved(stb_vorbis *f, int channels, float *buffer, int num_floats); +extern int stb_vorbis_get_samples_float(stb_vorbis *f, int channels, float **buffer, int num_samples); +// gets num_samples samples, not necessarily on a frame boundary--this requires +// buffering so you have to supply the buffers. DOES NOT APPLY THE COERCION RULES. +// Returns the number of samples stored per channel; it may be less than requested +// at the end of the file. If there are no more samples in the file, returns 0. + +#ifndef STB_VORBIS_NO_INTEGER_CONVERSION +extern int stb_vorbis_get_samples_short_interleaved(stb_vorbis *f, int channels, short *buffer, int num_shorts); +extern int stb_vorbis_get_samples_short(stb_vorbis *f, int channels, short **buffer, int num_samples); +#endif +// gets num_samples samples, not necessarily on a frame boundary--this requires +// buffering so you have to supply the buffers. Applies the coercion rules above +// to produce 'channels' channels. Returns the number of samples stored per channel; +// it may be less than requested at the end of the file. If there are no more +// samples in the file, returns 0. + +#endif + +//////// ERROR CODES + +enum STBVorbisError +{ + VORBIS__no_error, + + VORBIS_need_more_data=1, // not a real error + + VORBIS_invalid_api_mixing, // can't mix API modes + VORBIS_outofmem, // not enough memory + VORBIS_feature_not_supported, // uses floor 0 + VORBIS_too_many_channels, // STB_VORBIS_MAX_CHANNELS is too small + VORBIS_file_open_failure, // fopen() failed + VORBIS_seek_without_length, // can't seek in unknown-length file + + VORBIS_unexpected_eof=10, // file is truncated? + VORBIS_seek_invalid, // seek past EOF + + // decoding errors (corrupt/invalid stream) -- you probably + // don't care about the exact details of these + + // vorbis errors: + VORBIS_invalid_setup=20, + VORBIS_invalid_stream, + + // ogg errors: + VORBIS_missing_capture_pattern=30, + VORBIS_invalid_stream_structure_version, + VORBIS_continued_packet_flag_invalid, + VORBIS_incorrect_stream_serial_number, + VORBIS_invalid_first_page, + VORBIS_bad_packet_type, + VORBIS_cant_find_last_page, + VORBIS_seek_failed, + VORBIS_ogg_skeleton_not_supported +}; + + +#ifdef __cplusplus +} +#endif + +#endif // STB_VORBIS_INCLUDE_STB_VORBIS_H +// +// HEADER ENDS HERE +// +////////////////////////////////////////////////////////////////////////////// + +#ifndef STB_VORBIS_HEADER_ONLY + +// global configuration settings (e.g. set these in the project/makefile), +// or just set them in this file at the top (although ideally the first few +// should be visible when the header file is compiled too, although it's not +// crucial) + +// STB_VORBIS_NO_PUSHDATA_API +// does not compile the code for the various stb_vorbis_*_pushdata() +// functions +// #define STB_VORBIS_NO_PUSHDATA_API + +// STB_VORBIS_NO_PULLDATA_API +// does not compile the code for the non-pushdata APIs +// #define STB_VORBIS_NO_PULLDATA_API + +// STB_VORBIS_NO_STDIO +// does not compile the code for the APIs that use FILE *s internally +// or externally (implied by STB_VORBIS_NO_PULLDATA_API) +// #define STB_VORBIS_NO_STDIO + +// STB_VORBIS_NO_INTEGER_CONVERSION +// does not compile the code for converting audio sample data from +// float to integer (implied by STB_VORBIS_NO_PULLDATA_API) +// #define STB_VORBIS_NO_INTEGER_CONVERSION + +// STB_VORBIS_NO_FAST_SCALED_FLOAT +// does not use a fast float-to-int trick to accelerate float-to-int on +// most platforms which requires endianness be defined correctly. +//#define STB_VORBIS_NO_FAST_SCALED_FLOAT + + +// STB_VORBIS_MAX_CHANNELS [number] +// globally define this to the maximum number of channels you need. +// The spec does not put a restriction on channels except that +// the count is stored in a byte, so 255 is the hard limit. +// Reducing this saves about 16 bytes per value, so using 16 saves +// (255-16)*16 or around 4KB. Plus anything other memory usage +// I forgot to account for. Can probably go as low as 8 (7.1 audio), +// 6 (5.1 audio), or 2 (stereo only). +#ifndef STB_VORBIS_MAX_CHANNELS +#define STB_VORBIS_MAX_CHANNELS 16 // enough for anyone? +#endif + +// STB_VORBIS_PUSHDATA_CRC_COUNT [number] +// after a flush_pushdata(), stb_vorbis begins scanning for the +// next valid page, without backtracking. when it finds something +// that looks like a page, it streams through it and verifies its +// CRC32. Should that validation fail, it keeps scanning. But it's +// possible that _while_ streaming through to check the CRC32 of +// one candidate page, it sees another candidate page. This #define +// determines how many "overlapping" candidate pages it can search +// at once. Note that "real" pages are typically ~4KB to ~8KB, whereas +// garbage pages could be as big as 64KB, but probably average ~16KB. +// So don't hose ourselves by scanning an apparent 64KB page and +// missing a ton of real ones in the interim; so minimum of 2 +#ifndef STB_VORBIS_PUSHDATA_CRC_COUNT +#define STB_VORBIS_PUSHDATA_CRC_COUNT 4 +#endif + +// STB_VORBIS_FAST_HUFFMAN_LENGTH [number] +// sets the log size of the huffman-acceleration table. Maximum +// supported value is 24. with larger numbers, more decodings are O(1), +// but the table size is larger so worse cache missing, so you'll have +// to probe (and try multiple ogg vorbis files) to find the sweet spot. +#ifndef STB_VORBIS_FAST_HUFFMAN_LENGTH +#define STB_VORBIS_FAST_HUFFMAN_LENGTH 10 +#endif + +// STB_VORBIS_FAST_BINARY_LENGTH [number] +// sets the log size of the binary-search acceleration table. this +// is used in similar fashion to the fast-huffman size to set initial +// parameters for the binary search + +// STB_VORBIS_FAST_HUFFMAN_INT +// The fast huffman tables are much more efficient if they can be +// stored as 16-bit results instead of 32-bit results. This restricts +// the codebooks to having only 65535 possible outcomes, though. +// (At least, accelerated by the huffman table.) +#ifndef STB_VORBIS_FAST_HUFFMAN_INT +#define STB_VORBIS_FAST_HUFFMAN_SHORT +#endif + +// STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH +// If the 'fast huffman' search doesn't succeed, then stb_vorbis falls +// back on binary searching for the correct one. This requires storing +// extra tables with the huffman codes in sorted order. Defining this +// symbol trades off space for speed by forcing a linear search in the +// non-fast case, except for "sparse" codebooks. +// #define STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH + +// STB_VORBIS_DIVIDES_IN_RESIDUE +// stb_vorbis precomputes the result of the scalar residue decoding +// that would otherwise require a divide per chunk. you can trade off +// space for time by defining this symbol. +// #define STB_VORBIS_DIVIDES_IN_RESIDUE + +// STB_VORBIS_DIVIDES_IN_CODEBOOK +// vorbis VQ codebooks can be encoded two ways: with every case explicitly +// stored, or with all elements being chosen from a small range of values, +// and all values possible in all elements. By default, stb_vorbis expands +// this latter kind out to look like the former kind for ease of decoding, +// because otherwise an integer divide-per-vector-element is required to +// unpack the index. If you define STB_VORBIS_DIVIDES_IN_CODEBOOK, you can +// trade off storage for speed. +//#define STB_VORBIS_DIVIDES_IN_CODEBOOK + +#ifdef STB_VORBIS_CODEBOOK_SHORTS +#error "STB_VORBIS_CODEBOOK_SHORTS is no longer supported as it produced incorrect results for some input formats" +#endif + +// STB_VORBIS_DIVIDE_TABLE +// this replaces small integer divides in the floor decode loop with +// table lookups. made less than 1% difference, so disabled by default. + +// STB_VORBIS_NO_INLINE_DECODE +// disables the inlining of the scalar codebook fast-huffman decode. +// might save a little codespace; useful for debugging +// #define STB_VORBIS_NO_INLINE_DECODE + +// STB_VORBIS_NO_DEFER_FLOOR +// Normally we only decode the floor without synthesizing the actual +// full curve. We can instead synthesize the curve immediately. This +// requires more memory and is very likely slower, so I don't think +// you'd ever want to do it except for debugging. +// #define STB_VORBIS_NO_DEFER_FLOOR + + + + +////////////////////////////////////////////////////////////////////////////// + +#ifdef STB_VORBIS_NO_PULLDATA_API + #define STB_VORBIS_NO_INTEGER_CONVERSION + #define STB_VORBIS_NO_STDIO +#endif + +#if defined(STB_VORBIS_NO_CRT) && !defined(STB_VORBIS_NO_STDIO) + #define STB_VORBIS_NO_STDIO 1 +#endif + +#ifndef STB_VORBIS_NO_INTEGER_CONVERSION +#ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT + + // only need endianness for fast-float-to-int, which we don't + // use for pushdata + + #ifndef STB_VORBIS_BIG_ENDIAN + #define STB_VORBIS_ENDIAN 0 + #else + #define STB_VORBIS_ENDIAN 1 + #endif + +#endif +#endif + + +#ifndef STB_VORBIS_NO_STDIO +#include <stdio.h> +#endif + +#ifndef STB_VORBIS_NO_CRT + #include <stdlib.h> + #include <string.h> + #include <assert.h> + #include <math.h> + + // find definition of alloca if it's not in stdlib.h: + #if defined(_MSC_VER) || defined(__MINGW32__) + #include <malloc.h> + #endif + #if defined(__linux__) || defined(__linux) || defined(__sun__) || defined(__EMSCRIPTEN__) || defined(__NEWLIB__) + #include <alloca.h> + #endif +#else // STB_VORBIS_NO_CRT + #define NULL 0 + #define malloc(s) 0 + #define free(s) ((void) 0) + #define realloc(s) 0 +#endif // STB_VORBIS_NO_CRT + +#include <limits.h> + +#ifdef __MINGW32__ + // eff you mingw: + // "fixed": + // http://sourceforge.net/p/mingw-w64/mailman/message/32882927/ + // "no that broke the build, reverted, who cares about C": + // http://sourceforge.net/p/mingw-w64/mailman/message/32890381/ + #ifdef __forceinline + #undef __forceinline + #endif + #define __forceinline + #ifndef alloca + #define alloca __builtin_alloca + #endif +#elif !defined(_MSC_VER) + #if __GNUC__ + #define __forceinline inline + #else + #define __forceinline + #endif +#endif + +#if STB_VORBIS_MAX_CHANNELS > 256 +#error "Value of STB_VORBIS_MAX_CHANNELS outside of allowed range" +#endif + +#if STB_VORBIS_FAST_HUFFMAN_LENGTH > 24 +#error "Value of STB_VORBIS_FAST_HUFFMAN_LENGTH outside of allowed range" +#endif + + +#if 0 +#include <crtdbg.h> +#define CHECK(f) _CrtIsValidHeapPointer(f->channel_buffers[1]) +#else +#define CHECK(f) ((void) 0) +#endif + +#define MAX_BLOCKSIZE_LOG 13 // from specification +#define MAX_BLOCKSIZE (1 << MAX_BLOCKSIZE_LOG) + + +typedef unsigned char uint8; +typedef signed char int8; +typedef unsigned short uint16; +typedef signed short int16; +typedef unsigned int uint32; +typedef signed int int32; + +#ifndef TRUE +#define TRUE 1 +#define FALSE 0 +#endif + +typedef float codetype; + +#ifdef _MSC_VER +#define STBV_NOTUSED(v) (void)(v) +#else +#define STBV_NOTUSED(v) (void)sizeof(v) +#endif + +// @NOTE +// +// Some arrays below are tagged "//varies", which means it's actually +// a variable-sized piece of data, but rather than malloc I assume it's +// small enough it's better to just allocate it all together with the +// main thing +// +// Most of the variables are specified with the smallest size I could pack +// them into. It might give better performance to make them all full-sized +// integers. It should be safe to freely rearrange the structures or change +// the sizes larger--nothing relies on silently truncating etc., nor the +// order of variables. + +#define FAST_HUFFMAN_TABLE_SIZE (1 << STB_VORBIS_FAST_HUFFMAN_LENGTH) +#define FAST_HUFFMAN_TABLE_MASK (FAST_HUFFMAN_TABLE_SIZE - 1) + +typedef struct +{ + int dimensions, entries; + uint8 *codeword_lengths; + float minimum_value; + float delta_value; + uint8 value_bits; + uint8 lookup_type; + uint8 sequence_p; + uint8 sparse; + uint32 lookup_values; + codetype *multiplicands; + uint32 *codewords; + #ifdef STB_VORBIS_FAST_HUFFMAN_SHORT + int16 fast_huffman[FAST_HUFFMAN_TABLE_SIZE]; + #else + int32 fast_huffman[FAST_HUFFMAN_TABLE_SIZE]; + #endif + uint32 *sorted_codewords; + int *sorted_values; + int sorted_entries; +} Codebook; + +typedef struct +{ + uint8 order; + uint16 rate; + uint16 bark_map_size; + uint8 amplitude_bits; + uint8 amplitude_offset; + uint8 number_of_books; + uint8 book_list[16]; // varies +} Floor0; + +typedef struct +{ + uint8 partitions; + uint8 partition_class_list[32]; // varies + uint8 class_dimensions[16]; // varies + uint8 class_subclasses[16]; // varies + uint8 class_masterbooks[16]; // varies + int16 subclass_books[16][8]; // varies + uint16 Xlist[31*8+2]; // varies + uint8 sorted_order[31*8+2]; + uint8 neighbors[31*8+2][2]; + uint8 floor1_multiplier; + uint8 rangebits; + int values; +} Floor1; + +typedef union +{ + Floor0 floor0; + Floor1 floor1; +} Floor; + +typedef struct +{ + uint32 begin, end; + uint32 part_size; + uint8 classifications; + uint8 classbook; + uint8 **classdata; + int16 (*residue_books)[8]; +} Residue; + +typedef struct +{ + uint8 magnitude; + uint8 angle; + uint8 mux; +} MappingChannel; + +typedef struct +{ + uint16 coupling_steps; + MappingChannel *chan; + uint8 submaps; + uint8 submap_floor[15]; // varies + uint8 submap_residue[15]; // varies +} Mapping; + +typedef struct +{ + uint8 blockflag; + uint8 mapping; + uint16 windowtype; + uint16 transformtype; +} Mode; + +typedef struct +{ + uint32 goal_crc; // expected crc if match + int bytes_left; // bytes left in packet + uint32 crc_so_far; // running crc + int bytes_done; // bytes processed in _current_ chunk + uint32 sample_loc; // granule pos encoded in page +} CRCscan; + +typedef struct +{ + uint32 page_start, page_end; + uint32 last_decoded_sample; +} ProbedPage; + +struct stb_vorbis +{ + // user-accessible info + unsigned int sample_rate; + int channels; + + unsigned int setup_memory_required; + unsigned int temp_memory_required; + unsigned int setup_temp_memory_required; + + char *vendor; + int comment_list_length; + char **comment_list; + + // input config +#ifndef STB_VORBIS_NO_STDIO + FILE *f; + uint32 f_start; + int close_on_free; +#endif + + uint8 *stream; + uint8 *stream_start; + uint8 *stream_end; + + uint32 stream_len; + + uint8 push_mode; + + // the page to seek to when seeking to start, may be zero + uint32 first_audio_page_offset; + + // p_first is the page on which the first audio packet ends + // (but not necessarily the page on which it starts) + ProbedPage p_first, p_last; + + // memory management + stb_vorbis_alloc alloc; + int setup_offset; + int temp_offset; + + // run-time results + int eof; + enum STBVorbisError error; + + // user-useful data + + // header info + int blocksize[2]; + int blocksize_0, blocksize_1; + int codebook_count; + Codebook *codebooks; + int floor_count; + uint16 floor_types[64]; // varies + Floor *floor_config; + int residue_count; + uint16 residue_types[64]; // varies + Residue *residue_config; + int mapping_count; + Mapping *mapping; + int mode_count; + Mode mode_config[64]; // varies + + uint32 total_samples; + + // decode buffer + float *channel_buffers[STB_VORBIS_MAX_CHANNELS]; + float *outputs [STB_VORBIS_MAX_CHANNELS]; + + float *previous_window[STB_VORBIS_MAX_CHANNELS]; + int previous_length; + + #ifndef STB_VORBIS_NO_DEFER_FLOOR + int16 *finalY[STB_VORBIS_MAX_CHANNELS]; + #else + float *floor_buffers[STB_VORBIS_MAX_CHANNELS]; + #endif + + uint32 current_loc; // sample location of next frame to decode + int current_loc_valid; + + // per-blocksize precomputed data + + // twiddle factors + float *A[2],*B[2],*C[2]; + float *window[2]; + uint16 *bit_reverse[2]; + + // current page/packet/segment streaming info + uint32 serial; // stream serial number for verification + int last_page; + int segment_count; + uint8 segments[255]; + uint8 page_flag; + uint8 bytes_in_seg; + uint8 first_decode; + int next_seg; + int last_seg; // flag that we're on the last segment + int last_seg_which; // what was the segment number of the last seg? + uint32 acc; + int valid_bits; + int packet_bytes; + int end_seg_with_known_loc; + uint32 known_loc_for_packet; + int discard_samples_deferred; + uint32 samples_output; + + // push mode scanning + int page_crc_tests; // only in push_mode: number of tests active; -1 if not searching +#ifndef STB_VORBIS_NO_PUSHDATA_API + CRCscan scan[STB_VORBIS_PUSHDATA_CRC_COUNT]; +#endif + + // sample-access + int channel_buffer_start; + int channel_buffer_end; +}; + +#if defined(STB_VORBIS_NO_PUSHDATA_API) + #define IS_PUSH_MODE(f) FALSE +#elif defined(STB_VORBIS_NO_PULLDATA_API) + #define IS_PUSH_MODE(f) TRUE +#else + #define IS_PUSH_MODE(f) ((f)->push_mode) +#endif + +typedef struct stb_vorbis vorb; + +static int error(vorb *f, enum STBVorbisError e) +{ + f->error = e; + if (!f->eof && e != VORBIS_need_more_data) { + f->error=e; // breakpoint for debugging + } + return 0; +} + + +// these functions are used for allocating temporary memory +// while decoding. if you can afford the stack space, use +// alloca(); otherwise, provide a temp buffer and it will +// allocate out of those. + +#define array_size_required(count,size) (count*(sizeof(void *)+(size))) + +#define temp_alloc(f,size) (f->alloc.alloc_buffer ? setup_temp_malloc(f,size) : alloca(size)) +#define temp_free(f,p) (void)0 +#define temp_alloc_save(f) ((f)->temp_offset) +#define temp_alloc_restore(f,p) ((f)->temp_offset = (p)) + +#define temp_block_array(f,count,size) make_block_array(temp_alloc(f,array_size_required(count,size)), count, size) + +// given a sufficiently large block of memory, make an array of pointers to subblocks of it +static void *make_block_array(void *mem, int count, int size) +{ + int i; + void ** p = (void **) mem; + char *q = (char *) (p + count); + for (i=0; i < count; ++i) { + p[i] = q; + q += size; + } + return p; +} + +static void *setup_malloc(vorb *f, int sz) +{ + sz = (sz+7) & ~7; // round up to nearest 8 for alignment of future allocs. + f->setup_memory_required += sz; + if (f->alloc.alloc_buffer) { + void *p = (char *) f->alloc.alloc_buffer + f->setup_offset; + if (f->setup_offset + sz > f->temp_offset) return NULL; + f->setup_offset += sz; + return p; + } + return sz ? malloc(sz) : NULL; +} + +static void setup_free(vorb *f, void *p) +{ + if (f->alloc.alloc_buffer) return; // do nothing; setup mem is a stack + free(p); +} + +static void *setup_temp_malloc(vorb *f, int sz) +{ + sz = (sz+7) & ~7; // round up to nearest 8 for alignment of future allocs. + if (f->alloc.alloc_buffer) { + if (f->temp_offset - sz < f->setup_offset) return NULL; + f->temp_offset -= sz; + return (char *) f->alloc.alloc_buffer + f->temp_offset; + } + return malloc(sz); +} + +static void setup_temp_free(vorb *f, void *p, int sz) +{ + if (f->alloc.alloc_buffer) { + f->temp_offset += (sz+7)&~7; + return; + } + free(p); +} + +#define CRC32_POLY 0x04c11db7 // from spec + +static uint32 crc_table[256]; +static void crc32_init(void) +{ + int i,j; + uint32 s; + for(i=0; i < 256; i++) { + for (s=(uint32) i << 24, j=0; j < 8; ++j) + s = (s << 1) ^ (s >= (1U<<31) ? CRC32_POLY : 0); + crc_table[i] = s; + } +} + +static __forceinline uint32 crc32_update(uint32 crc, uint8 byte) +{ + return (crc << 8) ^ crc_table[byte ^ (crc >> 24)]; +} + + +// used in setup, and for huffman that doesn't go fast path +static unsigned int bit_reverse(unsigned int n) +{ + n = ((n & 0xAAAAAAAA) >> 1) | ((n & 0x55555555) << 1); + n = ((n & 0xCCCCCCCC) >> 2) | ((n & 0x33333333) << 2); + n = ((n & 0xF0F0F0F0) >> 4) | ((n & 0x0F0F0F0F) << 4); + n = ((n & 0xFF00FF00) >> 8) | ((n & 0x00FF00FF) << 8); + return (n >> 16) | (n << 16); +} + +static float square(float x) +{ + return x*x; +} + +// this is a weird definition of log2() for which log2(1) = 1, log2(2) = 2, log2(4) = 3 +// as required by the specification. fast(?) implementation from stb.h +// @OPTIMIZE: called multiple times per-packet with "constants"; move to setup +static int ilog(int32 n) +{ + static signed char log2_4[16] = { 0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4 }; + + if (n < 0) return 0; // signed n returns 0 + + // 2 compares if n < 16, 3 compares otherwise (4 if signed or n > 1<<29) + if (n < (1 << 14)) + if (n < (1 << 4)) return 0 + log2_4[n ]; + else if (n < (1 << 9)) return 5 + log2_4[n >> 5]; + else return 10 + log2_4[n >> 10]; + else if (n < (1 << 24)) + if (n < (1 << 19)) return 15 + log2_4[n >> 15]; + else return 20 + log2_4[n >> 20]; + else if (n < (1 << 29)) return 25 + log2_4[n >> 25]; + else return 30 + log2_4[n >> 30]; +} + +#ifndef M_PI + #define M_PI 3.14159265358979323846264f // from CRC +#endif + +// code length assigned to a value with no huffman encoding +#define NO_CODE 255 + +/////////////////////// LEAF SETUP FUNCTIONS ////////////////////////// +// +// these functions are only called at setup, and only a few times +// per file + +static float float32_unpack(uint32 x) +{ + // from the specification + uint32 mantissa = x & 0x1fffff; + uint32 sign = x & 0x80000000; + uint32 exp = (x & 0x7fe00000) >> 21; + double res = sign ? -(double)mantissa : (double)mantissa; + return (float) ldexp((float)res, (int)exp-788); +} + + +// zlib & jpeg huffman tables assume that the output symbols +// can either be arbitrarily arranged, or have monotonically +// increasing frequencies--they rely on the lengths being sorted; +// this makes for a very simple generation algorithm. +// vorbis allows a huffman table with non-sorted lengths. This +// requires a more sophisticated construction, since symbols in +// order do not map to huffman codes "in order". +static void add_entry(Codebook *c, uint32 huff_code, int symbol, int count, int len, uint32 *values) +{ + if (!c->sparse) { + c->codewords [symbol] = huff_code; + } else { + c->codewords [count] = huff_code; + c->codeword_lengths[count] = len; + values [count] = symbol; + } +} + +static int compute_codewords(Codebook *c, uint8 *len, int n, uint32 *values) +{ + int i,k,m=0; + uint32 available[32]; + + memset(available, 0, sizeof(available)); + // find the first entry + for (k=0; k < n; ++k) if (len[k] < NO_CODE) break; + if (k == n) { assert(c->sorted_entries == 0); return TRUE; } + assert(len[k] < 32); // no error return required, code reading lens checks this + // add to the list + add_entry(c, 0, k, m++, len[k], values); + // add all available leaves + for (i=1; i <= len[k]; ++i) + available[i] = 1U << (32-i); + // note that the above code treats the first case specially, + // but it's really the same as the following code, so they + // could probably be combined (except the initial code is 0, + // and I use 0 in available[] to mean 'empty') + for (i=k+1; i < n; ++i) { + uint32 res; + int z = len[i], y; + if (z == NO_CODE) continue; + assert(z < 32); // no error return required, code reading lens checks this + // find lowest available leaf (should always be earliest, + // which is what the specification calls for) + // note that this property, and the fact we can never have + // more than one free leaf at a given level, isn't totally + // trivial to prove, but it seems true and the assert never + // fires, so! + while (z > 0 && !available[z]) --z; + if (z == 0) { return FALSE; } + res = available[z]; + available[z] = 0; + add_entry(c, bit_reverse(res), i, m++, len[i], values); + // propagate availability up the tree + if (z != len[i]) { + for (y=len[i]; y > z; --y) { + assert(available[y] == 0); + available[y] = res + (1 << (32-y)); + } + } + } + return TRUE; +} + +// accelerated huffman table allows fast O(1) match of all symbols +// of length <= STB_VORBIS_FAST_HUFFMAN_LENGTH +static void compute_accelerated_huffman(Codebook *c) +{ + int i, len; + for (i=0; i < FAST_HUFFMAN_TABLE_SIZE; ++i) + c->fast_huffman[i] = -1; + + len = c->sparse ? c->sorted_entries : c->entries; + #ifdef STB_VORBIS_FAST_HUFFMAN_SHORT + if (len > 32767) len = 32767; // largest possible value we can encode! + #endif + for (i=0; i < len; ++i) { + if (c->codeword_lengths[i] <= STB_VORBIS_FAST_HUFFMAN_LENGTH) { + uint32 z = c->sparse ? bit_reverse(c->sorted_codewords[i]) : c->codewords[i]; + // set table entries for all bit combinations in the higher bits + while (z < FAST_HUFFMAN_TABLE_SIZE) { + c->fast_huffman[z] = i; + z += 1 << c->codeword_lengths[i]; + } + } + } +} + +#ifdef _MSC_VER +#define STBV_CDECL __cdecl +#else +#define STBV_CDECL +#endif + +static int STBV_CDECL uint32_compare(const void *p, const void *q) +{ + uint32 x = * (uint32 *) p; + uint32 y = * (uint32 *) q; + return x < y ? -1 : x > y; +} + +static int include_in_sort(Codebook *c, uint8 len) +{ + if (c->sparse) { assert(len != NO_CODE); return TRUE; } + if (len == NO_CODE) return FALSE; + if (len > STB_VORBIS_FAST_HUFFMAN_LENGTH) return TRUE; + return FALSE; +} + +// if the fast table above doesn't work, we want to binary +// search them... need to reverse the bits +static void compute_sorted_huffman(Codebook *c, uint8 *lengths, uint32 *values) +{ + int i, len; + // build a list of all the entries + // OPTIMIZATION: don't include the short ones, since they'll be caught by FAST_HUFFMAN. + // this is kind of a frivolous optimization--I don't see any performance improvement, + // but it's like 4 extra lines of code, so. + if (!c->sparse) { + int k = 0; + for (i=0; i < c->entries; ++i) + if (include_in_sort(c, lengths[i])) + c->sorted_codewords[k++] = bit_reverse(c->codewords[i]); + assert(k == c->sorted_entries); + } else { + for (i=0; i < c->sorted_entries; ++i) + c->sorted_codewords[i] = bit_reverse(c->codewords[i]); + } + + qsort(c->sorted_codewords, c->sorted_entries, sizeof(c->sorted_codewords[0]), uint32_compare); + c->sorted_codewords[c->sorted_entries] = 0xffffffff; + + len = c->sparse ? c->sorted_entries : c->entries; + // now we need to indicate how they correspond; we could either + // #1: sort a different data structure that says who they correspond to + // #2: for each sorted entry, search the original list to find who corresponds + // #3: for each original entry, find the sorted entry + // #1 requires extra storage, #2 is slow, #3 can use binary search! + for (i=0; i < len; ++i) { + int huff_len = c->sparse ? lengths[values[i]] : lengths[i]; + if (include_in_sort(c,huff_len)) { + uint32 code = bit_reverse(c->codewords[i]); + int x=0, n=c->sorted_entries; + while (n > 1) { + // invariant: sc[x] <= code < sc[x+n] + int m = x + (n >> 1); + if (c->sorted_codewords[m] <= code) { + x = m; + n -= (n>>1); + } else { + n >>= 1; + } + } + assert(c->sorted_codewords[x] == code); + if (c->sparse) { + c->sorted_values[x] = values[i]; + c->codeword_lengths[x] = huff_len; + } else { + c->sorted_values[x] = i; + } + } + } +} + +// only run while parsing the header (3 times) +static int vorbis_validate(uint8 *data) +{ + static uint8 vorbis[6] = { 'v', 'o', 'r', 'b', 'i', 's' }; + return memcmp(data, vorbis, 6) == 0; +} + +// called from setup only, once per code book +// (formula implied by specification) +static int lookup1_values(int entries, int dim) +{ + int r = (int) floor(exp((float) log((float) entries) / dim)); + if ((int) floor(pow((float) r+1, dim)) <= entries) // (int) cast for MinGW warning; + ++r; // floor() to avoid _ftol() when non-CRT + if (pow((float) r+1, dim) <= entries) + return -1; + if ((int) floor(pow((float) r, dim)) > entries) + return -1; + return r; +} + +// called twice per file +static void compute_twiddle_factors(int n, float *A, float *B, float *C) +{ + int n4 = n >> 2, n8 = n >> 3; + int k,k2; + + for (k=k2=0; k < n4; ++k,k2+=2) { + A[k2 ] = (float) cos(4*k*M_PI/n); + A[k2+1] = (float) -sin(4*k*M_PI/n); + B[k2 ] = (float) cos((k2+1)*M_PI/n/2) * 0.5f; + B[k2+1] = (float) sin((k2+1)*M_PI/n/2) * 0.5f; + } + for (k=k2=0; k < n8; ++k,k2+=2) { + C[k2 ] = (float) cos(2*(k2+1)*M_PI/n); + C[k2+1] = (float) -sin(2*(k2+1)*M_PI/n); + } +} + +static void compute_window(int n, float *window) +{ + int n2 = n >> 1, i; + for (i=0; i < n2; ++i) + window[i] = (float) sin(0.5 * M_PI * square((float) sin((i - 0 + 0.5) / n2 * 0.5 * M_PI))); +} + +static void compute_bitreverse(int n, uint16 *rev) +{ + int ld = ilog(n) - 1; // ilog is off-by-one from normal definitions + int i, n8 = n >> 3; + for (i=0; i < n8; ++i) + rev[i] = (bit_reverse(i) >> (32-ld+3)) << 2; +} + +static int init_blocksize(vorb *f, int b, int n) +{ + int n2 = n >> 1, n4 = n >> 2, n8 = n >> 3; + f->A[b] = (float *) setup_malloc(f, sizeof(float) * n2); + f->B[b] = (float *) setup_malloc(f, sizeof(float) * n2); + f->C[b] = (float *) setup_malloc(f, sizeof(float) * n4); + if (!f->A[b] || !f->B[b] || !f->C[b]) return error(f, VORBIS_outofmem); + compute_twiddle_factors(n, f->A[b], f->B[b], f->C[b]); + f->window[b] = (float *) setup_malloc(f, sizeof(float) * n2); + if (!f->window[b]) return error(f, VORBIS_outofmem); + compute_window(n, f->window[b]); + f->bit_reverse[b] = (uint16 *) setup_malloc(f, sizeof(uint16) * n8); + if (!f->bit_reverse[b]) return error(f, VORBIS_outofmem); + compute_bitreverse(n, f->bit_reverse[b]); + return TRUE; +} + +static void neighbors(uint16 *x, int n, int *plow, int *phigh) +{ + int low = -1; + int high = 65536; + int i; + for (i=0; i < n; ++i) { + if (x[i] > low && x[i] < x[n]) { *plow = i; low = x[i]; } + if (x[i] < high && x[i] > x[n]) { *phigh = i; high = x[i]; } + } +} + +// this has been repurposed so y is now the original index instead of y +typedef struct +{ + uint16 x,id; +} stbv__floor_ordering; + +static int STBV_CDECL point_compare(const void *p, const void *q) +{ + stbv__floor_ordering *a = (stbv__floor_ordering *) p; + stbv__floor_ordering *b = (stbv__floor_ordering *) q; + return a->x < b->x ? -1 : a->x > b->x; +} + +// +/////////////////////// END LEAF SETUP FUNCTIONS ////////////////////////// + + +#if defined(STB_VORBIS_NO_STDIO) + #define USE_MEMORY(z) TRUE +#else + #define USE_MEMORY(z) ((z)->stream) +#endif + +static uint8 get8(vorb *z) +{ + if (USE_MEMORY(z)) { + if (z->stream >= z->stream_end) { z->eof = TRUE; return 0; } + return *z->stream++; + } + + #ifndef STB_VORBIS_NO_STDIO + { + int c = fgetc(z->f); + if (c == EOF) { z->eof = TRUE; return 0; } + return c; + } + #endif +} + +static uint32 get32(vorb *f) +{ + uint32 x; + x = get8(f); + x += get8(f) << 8; + x += get8(f) << 16; + x += (uint32) get8(f) << 24; + return x; +} + +static int getn(vorb *z, uint8 *data, int n) +{ + if (USE_MEMORY(z)) { + if (z->stream+n > z->stream_end) { z->eof = 1; return 0; } + memcpy(data, z->stream, n); + z->stream += n; + return 1; + } + + #ifndef STB_VORBIS_NO_STDIO + if (fread(data, n, 1, z->f) == 1) + return 1; + else { + z->eof = 1; + return 0; + } + #endif +} + +static void skip(vorb *z, int n) +{ + if (USE_MEMORY(z)) { + z->stream += n; + if (z->stream >= z->stream_end) z->eof = 1; + return; + } + #ifndef STB_VORBIS_NO_STDIO + { + long x = ftell(z->f); + fseek(z->f, x+n, SEEK_SET); + } + #endif +} + +static int set_file_offset(stb_vorbis *f, unsigned int loc) +{ + #ifndef STB_VORBIS_NO_PUSHDATA_API + if (f->push_mode) return 0; + #endif + f->eof = 0; + if (USE_MEMORY(f)) { + if (f->stream_start + loc >= f->stream_end || f->stream_start + loc < f->stream_start) { + f->stream = f->stream_end; + f->eof = 1; + return 0; + } else { + f->stream = f->stream_start + loc; + return 1; + } + } + #ifndef STB_VORBIS_NO_STDIO + if (loc + f->f_start < loc || loc >= 0x80000000) { + loc = 0x7fffffff; + f->eof = 1; + } else { + loc += f->f_start; + } + if (!fseek(f->f, loc, SEEK_SET)) + return 1; + f->eof = 1; + fseek(f->f, f->f_start, SEEK_END); + return 0; + #endif +} + + +static uint8 ogg_page_header[4] = { 0x4f, 0x67, 0x67, 0x53 }; + +static int capture_pattern(vorb *f) +{ + if (0x4f != get8(f)) return FALSE; + if (0x67 != get8(f)) return FALSE; + if (0x67 != get8(f)) return FALSE; + if (0x53 != get8(f)) return FALSE; + return TRUE; +} + +#define PAGEFLAG_continued_packet 1 +#define PAGEFLAG_first_page 2 +#define PAGEFLAG_last_page 4 + +static int start_page_no_capturepattern(vorb *f) +{ + uint32 loc0,loc1,n; + if (f->first_decode && !IS_PUSH_MODE(f)) { + f->p_first.page_start = stb_vorbis_get_file_offset(f) - 4; + } + // stream structure version + if (0 != get8(f)) return error(f, VORBIS_invalid_stream_structure_version); + // header flag + f->page_flag = get8(f); + // absolute granule position + loc0 = get32(f); + loc1 = get32(f); + // @TODO: validate loc0,loc1 as valid positions? + // stream serial number -- vorbis doesn't interleave, so discard + get32(f); + //if (f->serial != get32(f)) return error(f, VORBIS_incorrect_stream_serial_number); + // page sequence number + n = get32(f); + f->last_page = n; + // CRC32 + get32(f); + // page_segments + f->segment_count = get8(f); + if (!getn(f, f->segments, f->segment_count)) + return error(f, VORBIS_unexpected_eof); + // assume we _don't_ know any the sample position of any segments + f->end_seg_with_known_loc = -2; + if (loc0 != ~0U || loc1 != ~0U) { + int i; + // determine which packet is the last one that will complete + for (i=f->segment_count-1; i >= 0; --i) + if (f->segments[i] < 255) + break; + // 'i' is now the index of the _last_ segment of a packet that ends + if (i >= 0) { + f->end_seg_with_known_loc = i; + f->known_loc_for_packet = loc0; + } + } + if (f->first_decode) { + int i,len; + len = 0; + for (i=0; i < f->segment_count; ++i) + len += f->segments[i]; + len += 27 + f->segment_count; + f->p_first.page_end = f->p_first.page_start + len; + f->p_first.last_decoded_sample = loc0; + } + f->next_seg = 0; + return TRUE; +} + +static int start_page(vorb *f) +{ + if (!capture_pattern(f)) return error(f, VORBIS_missing_capture_pattern); + return start_page_no_capturepattern(f); +} + +static int start_packet(vorb *f) +{ + while (f->next_seg == -1) { + if (!start_page(f)) return FALSE; + if (f->page_flag & PAGEFLAG_continued_packet) + return error(f, VORBIS_continued_packet_flag_invalid); + } + f->last_seg = FALSE; + f->valid_bits = 0; + f->packet_bytes = 0; + f->bytes_in_seg = 0; + // f->next_seg is now valid + return TRUE; +} + +static int maybe_start_packet(vorb *f) +{ + if (f->next_seg == -1) { + int x = get8(f); + if (f->eof) return FALSE; // EOF at page boundary is not an error! + if (0x4f != x ) return error(f, VORBIS_missing_capture_pattern); + if (0x67 != get8(f)) return error(f, VORBIS_missing_capture_pattern); + if (0x67 != get8(f)) return error(f, VORBIS_missing_capture_pattern); + if (0x53 != get8(f)) return error(f, VORBIS_missing_capture_pattern); + if (!start_page_no_capturepattern(f)) return FALSE; + if (f->page_flag & PAGEFLAG_continued_packet) { + // set up enough state that we can read this packet if we want, + // e.g. during recovery + f->last_seg = FALSE; + f->bytes_in_seg = 0; + return error(f, VORBIS_continued_packet_flag_invalid); + } + } + return start_packet(f); +} + +static int next_segment(vorb *f) +{ + int len; + if (f->last_seg) return 0; + if (f->next_seg == -1) { + f->last_seg_which = f->segment_count-1; // in case start_page fails + if (!start_page(f)) { f->last_seg = 1; return 0; } + if (!(f->page_flag & PAGEFLAG_continued_packet)) return error(f, VORBIS_continued_packet_flag_invalid); + } + len = f->segments[f->next_seg++]; + if (len < 255) { + f->last_seg = TRUE; + f->last_seg_which = f->next_seg-1; + } + if (f->next_seg >= f->segment_count) + f->next_seg = -1; + assert(f->bytes_in_seg == 0); + f->bytes_in_seg = len; + return len; +} + +#define EOP (-1) +#define INVALID_BITS (-1) + +static int get8_packet_raw(vorb *f) +{ + if (!f->bytes_in_seg) { // CLANG! + if (f->last_seg) return EOP; + else if (!next_segment(f)) return EOP; + } + assert(f->bytes_in_seg > 0); + --f->bytes_in_seg; + ++f->packet_bytes; + return get8(f); +} + +static int get8_packet(vorb *f) +{ + int x = get8_packet_raw(f); + f->valid_bits = 0; + return x; +} + +static int get32_packet(vorb *f) +{ + uint32 x; + x = get8_packet(f); + x += get8_packet(f) << 8; + x += get8_packet(f) << 16; + x += (uint32) get8_packet(f) << 24; + return x; +} + +static void flush_packet(vorb *f) +{ + while (get8_packet_raw(f) != EOP); +} + +// @OPTIMIZE: this is the secondary bit decoder, so it's probably not as important +// as the huffman decoder? +static uint32 get_bits(vorb *f, int n) +{ + uint32 z; + + if (f->valid_bits < 0) return 0; + if (f->valid_bits < n) { + if (n > 24) { + // the accumulator technique below would not work correctly in this case + z = get_bits(f, 24); + z += get_bits(f, n-24) << 24; + return z; + } + if (f->valid_bits == 0) f->acc = 0; + while (f->valid_bits < n) { + int z = get8_packet_raw(f); + if (z == EOP) { + f->valid_bits = INVALID_BITS; + return 0; + } + f->acc += z << f->valid_bits; + f->valid_bits += 8; + } + } + + assert(f->valid_bits >= n); + z = f->acc & ((1 << n)-1); + f->acc >>= n; + f->valid_bits -= n; + return z; +} + +// @OPTIMIZE: primary accumulator for huffman +// expand the buffer to as many bits as possible without reading off end of packet +// it might be nice to allow f->valid_bits and f->acc to be stored in registers, +// e.g. cache them locally and decode locally +static __forceinline void prep_huffman(vorb *f) +{ + if (f->valid_bits <= 24) { + if (f->valid_bits == 0) f->acc = 0; + do { + int z; + if (f->last_seg && !f->bytes_in_seg) return; + z = get8_packet_raw(f); + if (z == EOP) return; + f->acc += (unsigned) z << f->valid_bits; + f->valid_bits += 8; + } while (f->valid_bits <= 24); + } +} + +enum +{ + VORBIS_packet_id = 1, + VORBIS_packet_comment = 3, + VORBIS_packet_setup = 5 +}; + +static int codebook_decode_scalar_raw(vorb *f, Codebook *c) +{ + int i; + prep_huffman(f); + + if (c->codewords == NULL && c->sorted_codewords == NULL) + return -1; + + // cases to use binary search: sorted_codewords && !c->codewords + // sorted_codewords && c->entries > 8 + if (c->entries > 8 ? c->sorted_codewords!=NULL : !c->codewords) { + // binary search + uint32 code = bit_reverse(f->acc); + int x=0, n=c->sorted_entries, len; + + while (n > 1) { + // invariant: sc[x] <= code < sc[x+n] + int m = x + (n >> 1); + if (c->sorted_codewords[m] <= code) { + x = m; + n -= (n>>1); + } else { + n >>= 1; + } + } + // x is now the sorted index + if (!c->sparse) x = c->sorted_values[x]; + // x is now sorted index if sparse, or symbol otherwise + len = c->codeword_lengths[x]; + if (f->valid_bits >= len) { + f->acc >>= len; + f->valid_bits -= len; + return x; + } + + f->valid_bits = 0; + return -1; + } + + // if small, linear search + assert(!c->sparse); + for (i=0; i < c->entries; ++i) { + if (c->codeword_lengths[i] == NO_CODE) continue; + if (c->codewords[i] == (f->acc & ((1 << c->codeword_lengths[i])-1))) { + if (f->valid_bits >= c->codeword_lengths[i]) { + f->acc >>= c->codeword_lengths[i]; + f->valid_bits -= c->codeword_lengths[i]; + return i; + } + f->valid_bits = 0; + return -1; + } + } + + error(f, VORBIS_invalid_stream); + f->valid_bits = 0; + return -1; +} + +#ifndef STB_VORBIS_NO_INLINE_DECODE + +#define DECODE_RAW(var, f,c) \ + if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH) \ + prep_huffman(f); \ + var = f->acc & FAST_HUFFMAN_TABLE_MASK; \ + var = c->fast_huffman[var]; \ + if (var >= 0) { \ + int n = c->codeword_lengths[var]; \ + f->acc >>= n; \ + f->valid_bits -= n; \ + if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } \ + } else { \ + var = codebook_decode_scalar_raw(f,c); \ + } + +#else + +static int codebook_decode_scalar(vorb *f, Codebook *c) +{ + int i; + if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH) + prep_huffman(f); + // fast huffman table lookup + i = f->acc & FAST_HUFFMAN_TABLE_MASK; + i = c->fast_huffman[i]; + if (i >= 0) { + f->acc >>= c->codeword_lengths[i]; + f->valid_bits -= c->codeword_lengths[i]; + if (f->valid_bits < 0) { f->valid_bits = 0; return -1; } + return i; + } + return codebook_decode_scalar_raw(f,c); +} + +#define DECODE_RAW(var,f,c) var = codebook_decode_scalar(f,c); + +#endif + +#define DECODE(var,f,c) \ + DECODE_RAW(var,f,c) \ + if (c->sparse) var = c->sorted_values[var]; + +#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK + #define DECODE_VQ(var,f,c) DECODE_RAW(var,f,c) +#else + #define DECODE_VQ(var,f,c) DECODE(var,f,c) +#endif + + + + + + +// CODEBOOK_ELEMENT_FAST is an optimization for the CODEBOOK_FLOATS case +// where we avoid one addition +#define CODEBOOK_ELEMENT(c,off) (c->multiplicands[off]) +#define CODEBOOK_ELEMENT_FAST(c,off) (c->multiplicands[off]) +#define CODEBOOK_ELEMENT_BASE(c) (0) + +static int codebook_decode_start(vorb *f, Codebook *c) +{ + int z = -1; + + // type 0 is only legal in a scalar context + if (c->lookup_type == 0) + error(f, VORBIS_invalid_stream); + else { + DECODE_VQ(z,f,c); + if (c->sparse) assert(z < c->sorted_entries); + if (z < 0) { // check for EOP + if (!f->bytes_in_seg) + if (f->last_seg) + return z; + error(f, VORBIS_invalid_stream); + } + } + return z; +} + +static int codebook_decode(vorb *f, Codebook *c, float *output, int len) +{ + int i,z = codebook_decode_start(f,c); + if (z < 0) return FALSE; + if (len > c->dimensions) len = c->dimensions; + +#ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK + if (c->lookup_type == 1) { + float last = CODEBOOK_ELEMENT_BASE(c); + int div = 1; + for (i=0; i < len; ++i) { + int off = (z / div) % c->lookup_values; + float val = CODEBOOK_ELEMENT_FAST(c,off) + last; + output[i] += val; + if (c->sequence_p) last = val + c->minimum_value; + div *= c->lookup_values; + } + return TRUE; + } +#endif + + z *= c->dimensions; + if (c->sequence_p) { + float last = CODEBOOK_ELEMENT_BASE(c); + for (i=0; i < len; ++i) { + float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last; + output[i] += val; + last = val + c->minimum_value; + } + } else { + float last = CODEBOOK_ELEMENT_BASE(c); + for (i=0; i < len; ++i) { + output[i] += CODEBOOK_ELEMENT_FAST(c,z+i) + last; + } + } + + return TRUE; +} + +static int codebook_decode_step(vorb *f, Codebook *c, float *output, int len, int step) +{ + int i,z = codebook_decode_start(f,c); + float last = CODEBOOK_ELEMENT_BASE(c); + if (z < 0) return FALSE; + if (len > c->dimensions) len = c->dimensions; + +#ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK + if (c->lookup_type == 1) { + int div = 1; + for (i=0; i < len; ++i) { + int off = (z / div) % c->lookup_values; + float val = CODEBOOK_ELEMENT_FAST(c,off) + last; + output[i*step] += val; + if (c->sequence_p) last = val; + div *= c->lookup_values; + } + return TRUE; + } +#endif + + z *= c->dimensions; + for (i=0; i < len; ++i) { + float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last; + output[i*step] += val; + if (c->sequence_p) last = val; + } + + return TRUE; +} + +static int codebook_decode_deinterleave_repeat(vorb *f, Codebook *c, float **outputs, int ch, int *c_inter_p, int *p_inter_p, int len, int total_decode) +{ + int c_inter = *c_inter_p; + int p_inter = *p_inter_p; + int i,z, effective = c->dimensions; + + // type 0 is only legal in a scalar context + if (c->lookup_type == 0) return error(f, VORBIS_invalid_stream); + + while (total_decode > 0) { + float last = CODEBOOK_ELEMENT_BASE(c); + DECODE_VQ(z,f,c); + #ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK + assert(!c->sparse || z < c->sorted_entries); + #endif + if (z < 0) { + if (!f->bytes_in_seg) + if (f->last_seg) return FALSE; + return error(f, VORBIS_invalid_stream); + } + + // if this will take us off the end of the buffers, stop short! + // we check by computing the length of the virtual interleaved + // buffer (len*ch), our current offset within it (p_inter*ch)+(c_inter), + // and the length we'll be using (effective) + if (c_inter + p_inter*ch + effective > len * ch) { + effective = len*ch - (p_inter*ch - c_inter); + } + + #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK + if (c->lookup_type == 1) { + int div = 1; + for (i=0; i < effective; ++i) { + int off = (z / div) % c->lookup_values; + float val = CODEBOOK_ELEMENT_FAST(c,off) + last; + if (outputs[c_inter]) + outputs[c_inter][p_inter] += val; + if (++c_inter == ch) { c_inter = 0; ++p_inter; } + if (c->sequence_p) last = val; + div *= c->lookup_values; + } + } else + #endif + { + z *= c->dimensions; + if (c->sequence_p) { + for (i=0; i < effective; ++i) { + float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last; + if (outputs[c_inter]) + outputs[c_inter][p_inter] += val; + if (++c_inter == ch) { c_inter = 0; ++p_inter; } + last = val; + } + } else { + for (i=0; i < effective; ++i) { + float val = CODEBOOK_ELEMENT_FAST(c,z+i) + last; + if (outputs[c_inter]) + outputs[c_inter][p_inter] += val; + if (++c_inter == ch) { c_inter = 0; ++p_inter; } + } + } + } + + total_decode -= effective; + } + *c_inter_p = c_inter; + *p_inter_p = p_inter; + return TRUE; +} + +static int predict_point(int x, int x0, int x1, int y0, int y1) +{ + int dy = y1 - y0; + int adx = x1 - x0; + // @OPTIMIZE: force int division to round in the right direction... is this necessary on x86? + int err = abs(dy) * (x - x0); + int off = err / adx; + return dy < 0 ? y0 - off : y0 + off; +} + +// the following table is block-copied from the specification +static float inverse_db_table[256] = +{ + 1.0649863e-07f, 1.1341951e-07f, 1.2079015e-07f, 1.2863978e-07f, + 1.3699951e-07f, 1.4590251e-07f, 1.5538408e-07f, 1.6548181e-07f, + 1.7623575e-07f, 1.8768855e-07f, 1.9988561e-07f, 2.1287530e-07f, + 2.2670913e-07f, 2.4144197e-07f, 2.5713223e-07f, 2.7384213e-07f, + 2.9163793e-07f, 3.1059021e-07f, 3.3077411e-07f, 3.5226968e-07f, + 3.7516214e-07f, 3.9954229e-07f, 4.2550680e-07f, 4.5315863e-07f, + 4.8260743e-07f, 5.1396998e-07f, 5.4737065e-07f, 5.8294187e-07f, + 6.2082472e-07f, 6.6116941e-07f, 7.0413592e-07f, 7.4989464e-07f, + 7.9862701e-07f, 8.5052630e-07f, 9.0579828e-07f, 9.6466216e-07f, + 1.0273513e-06f, 1.0941144e-06f, 1.1652161e-06f, 1.2409384e-06f, + 1.3215816e-06f, 1.4074654e-06f, 1.4989305e-06f, 1.5963394e-06f, + 1.7000785e-06f, 1.8105592e-06f, 1.9282195e-06f, 2.0535261e-06f, + 2.1869758e-06f, 2.3290978e-06f, 2.4804557e-06f, 2.6416497e-06f, + 2.8133190e-06f, 2.9961443e-06f, 3.1908506e-06f, 3.3982101e-06f, + 3.6190449e-06f, 3.8542308e-06f, 4.1047004e-06f, 4.3714470e-06f, + 4.6555282e-06f, 4.9580707e-06f, 5.2802740e-06f, 5.6234160e-06f, + 5.9888572e-06f, 6.3780469e-06f, 6.7925283e-06f, 7.2339451e-06f, + 7.7040476e-06f, 8.2047000e-06f, 8.7378876e-06f, 9.3057248e-06f, + 9.9104632e-06f, 1.0554501e-05f, 1.1240392e-05f, 1.1970856e-05f, + 1.2748789e-05f, 1.3577278e-05f, 1.4459606e-05f, 1.5399272e-05f, + 1.6400004e-05f, 1.7465768e-05f, 1.8600792e-05f, 1.9809576e-05f, + 2.1096914e-05f, 2.2467911e-05f, 2.3928002e-05f, 2.5482978e-05f, + 2.7139006e-05f, 2.8902651e-05f, 3.0780908e-05f, 3.2781225e-05f, + 3.4911534e-05f, 3.7180282e-05f, 3.9596466e-05f, 4.2169667e-05f, + 4.4910090e-05f, 4.7828601e-05f, 5.0936773e-05f, 5.4246931e-05f, + 5.7772202e-05f, 6.1526565e-05f, 6.5524908e-05f, 6.9783085e-05f, + 7.4317983e-05f, 7.9147585e-05f, 8.4291040e-05f, 8.9768747e-05f, + 9.5602426e-05f, 0.00010181521f, 0.00010843174f, 0.00011547824f, + 0.00012298267f, 0.00013097477f, 0.00013948625f, 0.00014855085f, + 0.00015820453f, 0.00016848555f, 0.00017943469f, 0.00019109536f, + 0.00020351382f, 0.00021673929f, 0.00023082423f, 0.00024582449f, + 0.00026179955f, 0.00027881276f, 0.00029693158f, 0.00031622787f, + 0.00033677814f, 0.00035866388f, 0.00038197188f, 0.00040679456f, + 0.00043323036f, 0.00046138411f, 0.00049136745f, 0.00052329927f, + 0.00055730621f, 0.00059352311f, 0.00063209358f, 0.00067317058f, + 0.00071691700f, 0.00076350630f, 0.00081312324f, 0.00086596457f, + 0.00092223983f, 0.00098217216f, 0.0010459992f, 0.0011139742f, + 0.0011863665f, 0.0012634633f, 0.0013455702f, 0.0014330129f, + 0.0015261382f, 0.0016253153f, 0.0017309374f, 0.0018434235f, + 0.0019632195f, 0.0020908006f, 0.0022266726f, 0.0023713743f, + 0.0025254795f, 0.0026895994f, 0.0028643847f, 0.0030505286f, + 0.0032487691f, 0.0034598925f, 0.0036847358f, 0.0039241906f, + 0.0041792066f, 0.0044507950f, 0.0047400328f, 0.0050480668f, + 0.0053761186f, 0.0057254891f, 0.0060975636f, 0.0064938176f, + 0.0069158225f, 0.0073652516f, 0.0078438871f, 0.0083536271f, + 0.0088964928f, 0.009474637f, 0.010090352f, 0.010746080f, + 0.011444421f, 0.012188144f, 0.012980198f, 0.013823725f, + 0.014722068f, 0.015678791f, 0.016697687f, 0.017782797f, + 0.018938423f, 0.020169149f, 0.021479854f, 0.022875735f, + 0.024362330f, 0.025945531f, 0.027631618f, 0.029427276f, + 0.031339626f, 0.033376252f, 0.035545228f, 0.037855157f, + 0.040315199f, 0.042935108f, 0.045725273f, 0.048696758f, + 0.051861348f, 0.055231591f, 0.058820850f, 0.062643361f, + 0.066714279f, 0.071049749f, 0.075666962f, 0.080584227f, + 0.085821044f, 0.091398179f, 0.097337747f, 0.10366330f, + 0.11039993f, 0.11757434f, 0.12521498f, 0.13335215f, + 0.14201813f, 0.15124727f, 0.16107617f, 0.17154380f, + 0.18269168f, 0.19456402f, 0.20720788f, 0.22067342f, + 0.23501402f, 0.25028656f, 0.26655159f, 0.28387361f, + 0.30232132f, 0.32196786f, 0.34289114f, 0.36517414f, + 0.38890521f, 0.41417847f, 0.44109412f, 0.46975890f, + 0.50028648f, 0.53279791f, 0.56742212f, 0.60429640f, + 0.64356699f, 0.68538959f, 0.72993007f, 0.77736504f, + 0.82788260f, 0.88168307f, 0.9389798f, 1.0f +}; + + +// @OPTIMIZE: if you want to replace this bresenham line-drawing routine, +// note that you must produce bit-identical output to decode correctly; +// this specific sequence of operations is specified in the spec (it's +// drawing integer-quantized frequency-space lines that the encoder +// expects to be exactly the same) +// ... also, isn't the whole point of Bresenham's algorithm to NOT +// have to divide in the setup? sigh. +#ifndef STB_VORBIS_NO_DEFER_FLOOR +#define LINE_OP(a,b) a *= b +#else +#define LINE_OP(a,b) a = b +#endif + +#ifdef STB_VORBIS_DIVIDE_TABLE +#define DIVTAB_NUMER 32 +#define DIVTAB_DENOM 64 +int8 integer_divide_table[DIVTAB_NUMER][DIVTAB_DENOM]; // 2KB +#endif + +static __forceinline void draw_line(float *output, int x0, int y0, int x1, int y1, int n) +{ + int dy = y1 - y0; + int adx = x1 - x0; + int ady = abs(dy); + int base; + int x=x0,y=y0; + int err = 0; + int sy; + +#ifdef STB_VORBIS_DIVIDE_TABLE + if (adx < DIVTAB_DENOM && ady < DIVTAB_NUMER) { + if (dy < 0) { + base = -integer_divide_table[ady][adx]; + sy = base-1; + } else { + base = integer_divide_table[ady][adx]; + sy = base+1; + } + } else { + base = dy / adx; + if (dy < 0) + sy = base - 1; + else + sy = base+1; + } +#else + base = dy / adx; + if (dy < 0) + sy = base - 1; + else + sy = base+1; +#endif + ady -= abs(base) * adx; + if (x1 > n) x1 = n; + if (x < x1) { + LINE_OP(output[x], inverse_db_table[y&255]); + for (++x; x < x1; ++x) { + err += ady; + if (err >= adx) { + err -= adx; + y += sy; + } else + y += base; + LINE_OP(output[x], inverse_db_table[y&255]); + } + } +} + +static int residue_decode(vorb *f, Codebook *book, float *target, int offset, int n, int rtype) +{ + int k; + if (rtype == 0) { + int step = n / book->dimensions; + for (k=0; k < step; ++k) + if (!codebook_decode_step(f, book, target+offset+k, n-offset-k, step)) + return FALSE; + } else { + for (k=0; k < n; ) { + if (!codebook_decode(f, book, target+offset, n-k)) + return FALSE; + k += book->dimensions; + offset += book->dimensions; + } + } + return TRUE; +} + +// n is 1/2 of the blocksize -- +// specification: "Correct per-vector decode length is [n]/2" +static void decode_residue(vorb *f, float *residue_buffers[], int ch, int n, int rn, uint8 *do_not_decode) +{ + int i,j,pass; + Residue *r = f->residue_config + rn; + int rtype = f->residue_types[rn]; + int c = r->classbook; + int classwords = f->codebooks[c].dimensions; + unsigned int actual_size = rtype == 2 ? n*2 : n; + unsigned int limit_r_begin = (r->begin < actual_size ? r->begin : actual_size); + unsigned int limit_r_end = (r->end < actual_size ? r->end : actual_size); + int n_read = limit_r_end - limit_r_begin; + int part_read = n_read / r->part_size; + int temp_alloc_point = temp_alloc_save(f); + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + uint8 ***part_classdata = (uint8 ***) temp_block_array(f,f->channels, part_read * sizeof(**part_classdata)); + #else + int **classifications = (int **) temp_block_array(f,f->channels, part_read * sizeof(**classifications)); + #endif + + CHECK(f); + + for (i=0; i < ch; ++i) + if (!do_not_decode[i]) + memset(residue_buffers[i], 0, sizeof(float) * n); + + if (rtype == 2 && ch != 1) { + for (j=0; j < ch; ++j) + if (!do_not_decode[j]) + break; + if (j == ch) + goto done; + + for (pass=0; pass < 8; ++pass) { + int pcount = 0, class_set = 0; + if (ch == 2) { + while (pcount < part_read) { + int z = r->begin + pcount*r->part_size; + int c_inter = (z & 1), p_inter = z>>1; + if (pass == 0) { + Codebook *c = f->codebooks+r->classbook; + int q; + DECODE(q,f,c); + if (q == EOP) goto done; + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + part_classdata[0][class_set] = r->classdata[q]; + #else + for (i=classwords-1; i >= 0; --i) { + classifications[0][i+pcount] = q % r->classifications; + q /= r->classifications; + } + #endif + } + for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) { + int z = r->begin + pcount*r->part_size; + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + int c = part_classdata[0][class_set][i]; + #else + int c = classifications[0][pcount]; + #endif + int b = r->residue_books[c][pass]; + if (b >= 0) { + Codebook *book = f->codebooks + b; + #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK + if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size)) + goto done; + #else + // saves 1% + if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size)) + goto done; + #endif + } else { + z += r->part_size; + c_inter = z & 1; + p_inter = z >> 1; + } + } + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + ++class_set; + #endif + } + } else if (ch > 2) { + while (pcount < part_read) { + int z = r->begin + pcount*r->part_size; + int c_inter = z % ch, p_inter = z/ch; + if (pass == 0) { + Codebook *c = f->codebooks+r->classbook; + int q; + DECODE(q,f,c); + if (q == EOP) goto done; + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + part_classdata[0][class_set] = r->classdata[q]; + #else + for (i=classwords-1; i >= 0; --i) { + classifications[0][i+pcount] = q % r->classifications; + q /= r->classifications; + } + #endif + } + for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) { + int z = r->begin + pcount*r->part_size; + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + int c = part_classdata[0][class_set][i]; + #else + int c = classifications[0][pcount]; + #endif + int b = r->residue_books[c][pass]; + if (b >= 0) { + Codebook *book = f->codebooks + b; + if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size)) + goto done; + } else { + z += r->part_size; + c_inter = z % ch; + p_inter = z / ch; + } + } + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + ++class_set; + #endif + } + } + } + goto done; + } + CHECK(f); + + for (pass=0; pass < 8; ++pass) { + int pcount = 0, class_set=0; + while (pcount < part_read) { + if (pass == 0) { + for (j=0; j < ch; ++j) { + if (!do_not_decode[j]) { + Codebook *c = f->codebooks+r->classbook; + int temp; + DECODE(temp,f,c); + if (temp == EOP) goto done; + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + part_classdata[j][class_set] = r->classdata[temp]; + #else + for (i=classwords-1; i >= 0; --i) { + classifications[j][i+pcount] = temp % r->classifications; + temp /= r->classifications; + } + #endif + } + } + } + for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) { + for (j=0; j < ch; ++j) { + if (!do_not_decode[j]) { + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + int c = part_classdata[j][class_set][i]; + #else + int c = classifications[j][pcount]; + #endif + int b = r->residue_books[c][pass]; + if (b >= 0) { + float *target = residue_buffers[j]; + int offset = r->begin + pcount * r->part_size; + int n = r->part_size; + Codebook *book = f->codebooks + b; + if (!residue_decode(f, book, target, offset, n, rtype)) + goto done; + } + } + } + } + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + ++class_set; + #endif + } + } + done: + CHECK(f); + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + temp_free(f,part_classdata); + #else + temp_free(f,classifications); + #endif + temp_alloc_restore(f,temp_alloc_point); +} + + +#if 0 +// slow way for debugging +void inverse_mdct_slow(float *buffer, int n) +{ + int i,j; + int n2 = n >> 1; + float *x = (float *) malloc(sizeof(*x) * n2); + memcpy(x, buffer, sizeof(*x) * n2); + for (i=0; i < n; ++i) { + float acc = 0; + for (j=0; j < n2; ++j) + // formula from paper: + //acc += n/4.0f * x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1)); + // formula from wikipedia + //acc += 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5)); + // these are equivalent, except the formula from the paper inverts the multiplier! + // however, what actually works is NO MULTIPLIER!?! + //acc += 64 * 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5)); + acc += x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1)); + buffer[i] = acc; + } + free(x); +} +#elif 0 +// same as above, but just barely able to run in real time on modern machines +void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype) +{ + float mcos[16384]; + int i,j; + int n2 = n >> 1, nmask = (n << 2) -1; + float *x = (float *) malloc(sizeof(*x) * n2); + memcpy(x, buffer, sizeof(*x) * n2); + for (i=0; i < 4*n; ++i) + mcos[i] = (float) cos(M_PI / 2 * i / n); + + for (i=0; i < n; ++i) { + float acc = 0; + for (j=0; j < n2; ++j) + acc += x[j] * mcos[(2 * i + 1 + n2)*(2*j+1) & nmask]; + buffer[i] = acc; + } + free(x); +} +#elif 0 +// transform to use a slow dct-iv; this is STILL basically trivial, +// but only requires half as many ops +void dct_iv_slow(float *buffer, int n) +{ + float mcos[16384]; + float x[2048]; + int i,j; + int n2 = n >> 1, nmask = (n << 3) - 1; + memcpy(x, buffer, sizeof(*x) * n); + for (i=0; i < 8*n; ++i) + mcos[i] = (float) cos(M_PI / 4 * i / n); + for (i=0; i < n; ++i) { + float acc = 0; + for (j=0; j < n; ++j) + acc += x[j] * mcos[((2 * i + 1)*(2*j+1)) & nmask]; + buffer[i] = acc; + } +} + +void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype) +{ + int i, n4 = n >> 2, n2 = n >> 1, n3_4 = n - n4; + float temp[4096]; + + memcpy(temp, buffer, n2 * sizeof(float)); + dct_iv_slow(temp, n2); // returns -c'-d, a-b' + + for (i=0; i < n4 ; ++i) buffer[i] = temp[i+n4]; // a-b' + for ( ; i < n3_4; ++i) buffer[i] = -temp[n3_4 - i - 1]; // b-a', c+d' + for ( ; i < n ; ++i) buffer[i] = -temp[i - n3_4]; // c'+d +} +#endif + +#ifndef LIBVORBIS_MDCT +#define LIBVORBIS_MDCT 0 +#endif + +#if LIBVORBIS_MDCT +// directly call the vorbis MDCT using an interface documented +// by Jeff Roberts... useful for performance comparison +typedef struct +{ + int n; + int log2n; + + float *trig; + int *bitrev; + + float scale; +} mdct_lookup; + +extern void mdct_init(mdct_lookup *lookup, int n); +extern void mdct_clear(mdct_lookup *l); +extern void mdct_backward(mdct_lookup *init, float *in, float *out); + +mdct_lookup M1,M2; + +void inverse_mdct(float *buffer, int n, vorb *f, int blocktype) +{ + mdct_lookup *M; + if (M1.n == n) M = &M1; + else if (M2.n == n) M = &M2; + else if (M1.n == 0) { mdct_init(&M1, n); M = &M1; } + else { + if (M2.n) __asm int 3; + mdct_init(&M2, n); + M = &M2; + } + + mdct_backward(M, buffer, buffer); +} +#endif + + +// the following were split out into separate functions while optimizing; +// they could be pushed back up but eh. __forceinline showed no change; +// they're probably already being inlined. +static void imdct_step3_iter0_loop(int n, float *e, int i_off, int k_off, float *A) +{ + float *ee0 = e + i_off; + float *ee2 = ee0 + k_off; + int i; + + assert((n & 3) == 0); + for (i=(n>>2); i > 0; --i) { + float k00_20, k01_21; + k00_20 = ee0[ 0] - ee2[ 0]; + k01_21 = ee0[-1] - ee2[-1]; + ee0[ 0] += ee2[ 0];//ee0[ 0] = ee0[ 0] + ee2[ 0]; + ee0[-1] += ee2[-1];//ee0[-1] = ee0[-1] + ee2[-1]; + ee2[ 0] = k00_20 * A[0] - k01_21 * A[1]; + ee2[-1] = k01_21 * A[0] + k00_20 * A[1]; + A += 8; + + k00_20 = ee0[-2] - ee2[-2]; + k01_21 = ee0[-3] - ee2[-3]; + ee0[-2] += ee2[-2];//ee0[-2] = ee0[-2] + ee2[-2]; + ee0[-3] += ee2[-3];//ee0[-3] = ee0[-3] + ee2[-3]; + ee2[-2] = k00_20 * A[0] - k01_21 * A[1]; + ee2[-3] = k01_21 * A[0] + k00_20 * A[1]; + A += 8; + + k00_20 = ee0[-4] - ee2[-4]; + k01_21 = ee0[-5] - ee2[-5]; + ee0[-4] += ee2[-4];//ee0[-4] = ee0[-4] + ee2[-4]; + ee0[-5] += ee2[-5];//ee0[-5] = ee0[-5] + ee2[-5]; + ee2[-4] = k00_20 * A[0] - k01_21 * A[1]; + ee2[-5] = k01_21 * A[0] + k00_20 * A[1]; + A += 8; + + k00_20 = ee0[-6] - ee2[-6]; + k01_21 = ee0[-7] - ee2[-7]; + ee0[-6] += ee2[-6];//ee0[-6] = ee0[-6] + ee2[-6]; + ee0[-7] += ee2[-7];//ee0[-7] = ee0[-7] + ee2[-7]; + ee2[-6] = k00_20 * A[0] - k01_21 * A[1]; + ee2[-7] = k01_21 * A[0] + k00_20 * A[1]; + A += 8; + ee0 -= 8; + ee2 -= 8; + } +} + +static void imdct_step3_inner_r_loop(int lim, float *e, int d0, int k_off, float *A, int k1) +{ + int i; + float k00_20, k01_21; + + float *e0 = e + d0; + float *e2 = e0 + k_off; + + for (i=lim >> 2; i > 0; --i) { + k00_20 = e0[-0] - e2[-0]; + k01_21 = e0[-1] - e2[-1]; + e0[-0] += e2[-0];//e0[-0] = e0[-0] + e2[-0]; + e0[-1] += e2[-1];//e0[-1] = e0[-1] + e2[-1]; + e2[-0] = (k00_20)*A[0] - (k01_21) * A[1]; + e2[-1] = (k01_21)*A[0] + (k00_20) * A[1]; + + A += k1; + + k00_20 = e0[-2] - e2[-2]; + k01_21 = e0[-3] - e2[-3]; + e0[-2] += e2[-2];//e0[-2] = e0[-2] + e2[-2]; + e0[-3] += e2[-3];//e0[-3] = e0[-3] + e2[-3]; + e2[-2] = (k00_20)*A[0] - (k01_21) * A[1]; + e2[-3] = (k01_21)*A[0] + (k00_20) * A[1]; + + A += k1; + + k00_20 = e0[-4] - e2[-4]; + k01_21 = e0[-5] - e2[-5]; + e0[-4] += e2[-4];//e0[-4] = e0[-4] + e2[-4]; + e0[-5] += e2[-5];//e0[-5] = e0[-5] + e2[-5]; + e2[-4] = (k00_20)*A[0] - (k01_21) * A[1]; + e2[-5] = (k01_21)*A[0] + (k00_20) * A[1]; + + A += k1; + + k00_20 = e0[-6] - e2[-6]; + k01_21 = e0[-7] - e2[-7]; + e0[-6] += e2[-6];//e0[-6] = e0[-6] + e2[-6]; + e0[-7] += e2[-7];//e0[-7] = e0[-7] + e2[-7]; + e2[-6] = (k00_20)*A[0] - (k01_21) * A[1]; + e2[-7] = (k01_21)*A[0] + (k00_20) * A[1]; + + e0 -= 8; + e2 -= 8; + + A += k1; + } +} + +static void imdct_step3_inner_s_loop(int n, float *e, int i_off, int k_off, float *A, int a_off, int k0) +{ + int i; + float A0 = A[0]; + float A1 = A[0+1]; + float A2 = A[0+a_off]; + float A3 = A[0+a_off+1]; + float A4 = A[0+a_off*2+0]; + float A5 = A[0+a_off*2+1]; + float A6 = A[0+a_off*3+0]; + float A7 = A[0+a_off*3+1]; + + float k00,k11; + + float *ee0 = e +i_off; + float *ee2 = ee0+k_off; + + for (i=n; i > 0; --i) { + k00 = ee0[ 0] - ee2[ 0]; + k11 = ee0[-1] - ee2[-1]; + ee0[ 0] = ee0[ 0] + ee2[ 0]; + ee0[-1] = ee0[-1] + ee2[-1]; + ee2[ 0] = (k00) * A0 - (k11) * A1; + ee2[-1] = (k11) * A0 + (k00) * A1; + + k00 = ee0[-2] - ee2[-2]; + k11 = ee0[-3] - ee2[-3]; + ee0[-2] = ee0[-2] + ee2[-2]; + ee0[-3] = ee0[-3] + ee2[-3]; + ee2[-2] = (k00) * A2 - (k11) * A3; + ee2[-3] = (k11) * A2 + (k00) * A3; + + k00 = ee0[-4] - ee2[-4]; + k11 = ee0[-5] - ee2[-5]; + ee0[-4] = ee0[-4] + ee2[-4]; + ee0[-5] = ee0[-5] + ee2[-5]; + ee2[-4] = (k00) * A4 - (k11) * A5; + ee2[-5] = (k11) * A4 + (k00) * A5; + + k00 = ee0[-6] - ee2[-6]; + k11 = ee0[-7] - ee2[-7]; + ee0[-6] = ee0[-6] + ee2[-6]; + ee0[-7] = ee0[-7] + ee2[-7]; + ee2[-6] = (k00) * A6 - (k11) * A7; + ee2[-7] = (k11) * A6 + (k00) * A7; + + ee0 -= k0; + ee2 -= k0; + } +} + +static __forceinline void iter_54(float *z) +{ + float k00,k11,k22,k33; + float y0,y1,y2,y3; + + k00 = z[ 0] - z[-4]; + y0 = z[ 0] + z[-4]; + y2 = z[-2] + z[-6]; + k22 = z[-2] - z[-6]; + + z[-0] = y0 + y2; // z0 + z4 + z2 + z6 + z[-2] = y0 - y2; // z0 + z4 - z2 - z6 + + // done with y0,y2 + + k33 = z[-3] - z[-7]; + + z[-4] = k00 + k33; // z0 - z4 + z3 - z7 + z[-6] = k00 - k33; // z0 - z4 - z3 + z7 + + // done with k33 + + k11 = z[-1] - z[-5]; + y1 = z[-1] + z[-5]; + y3 = z[-3] + z[-7]; + + z[-1] = y1 + y3; // z1 + z5 + z3 + z7 + z[-3] = y1 - y3; // z1 + z5 - z3 - z7 + z[-5] = k11 - k22; // z1 - z5 + z2 - z6 + z[-7] = k11 + k22; // z1 - z5 - z2 + z6 +} + +static void imdct_step3_inner_s_loop_ld654(int n, float *e, int i_off, float *A, int base_n) +{ + int a_off = base_n >> 3; + float A2 = A[0+a_off]; + float *z = e + i_off; + float *base = z - 16 * n; + + while (z > base) { + float k00,k11; + float l00,l11; + + k00 = z[-0] - z[ -8]; + k11 = z[-1] - z[ -9]; + l00 = z[-2] - z[-10]; + l11 = z[-3] - z[-11]; + z[ -0] = z[-0] + z[ -8]; + z[ -1] = z[-1] + z[ -9]; + z[ -2] = z[-2] + z[-10]; + z[ -3] = z[-3] + z[-11]; + z[ -8] = k00; + z[ -9] = k11; + z[-10] = (l00+l11) * A2; + z[-11] = (l11-l00) * A2; + + k00 = z[ -4] - z[-12]; + k11 = z[ -5] - z[-13]; + l00 = z[ -6] - z[-14]; + l11 = z[ -7] - z[-15]; + z[ -4] = z[ -4] + z[-12]; + z[ -5] = z[ -5] + z[-13]; + z[ -6] = z[ -6] + z[-14]; + z[ -7] = z[ -7] + z[-15]; + z[-12] = k11; + z[-13] = -k00; + z[-14] = (l11-l00) * A2; + z[-15] = (l00+l11) * -A2; + + iter_54(z); + iter_54(z-8); + z -= 16; + } +} + +static void inverse_mdct(float *buffer, int n, vorb *f, int blocktype) +{ + int n2 = n >> 1, n4 = n >> 2, n8 = n >> 3, l; + int ld; + // @OPTIMIZE: reduce register pressure by using fewer variables? + int save_point = temp_alloc_save(f); + float *buf2 = (float *) temp_alloc(f, n2 * sizeof(*buf2)); + float *u=NULL,*v=NULL; + // twiddle factors + float *A = f->A[blocktype]; + + // IMDCT algorithm from "The use of multirate filter banks for coding of high quality digital audio" + // See notes about bugs in that paper in less-optimal implementation 'inverse_mdct_old' after this function. + + // kernel from paper + + + // merged: + // copy and reflect spectral data + // step 0 + + // note that it turns out that the items added together during + // this step are, in fact, being added to themselves (as reflected + // by step 0). inexplicable inefficiency! this became obvious + // once I combined the passes. + + // so there's a missing 'times 2' here (for adding X to itself). + // this propagates through linearly to the end, where the numbers + // are 1/2 too small, and need to be compensated for. + + { + float *d,*e, *AA, *e_stop; + d = &buf2[n2-2]; + AA = A; + e = &buffer[0]; + e_stop = &buffer[n2]; + while (e != e_stop) { + d[1] = (e[0] * AA[0] - e[2]*AA[1]); + d[0] = (e[0] * AA[1] + e[2]*AA[0]); + d -= 2; + AA += 2; + e += 4; + } + + e = &buffer[n2-3]; + while (d >= buf2) { + d[1] = (-e[2] * AA[0] - -e[0]*AA[1]); + d[0] = (-e[2] * AA[1] + -e[0]*AA[0]); + d -= 2; + AA += 2; + e -= 4; + } + } + + // now we use symbolic names for these, so that we can + // possibly swap their meaning as we change which operations + // are in place + + u = buffer; + v = buf2; + + // step 2 (paper output is w, now u) + // this could be in place, but the data ends up in the wrong + // place... _somebody_'s got to swap it, so this is nominated + { + float *AA = &A[n2-8]; + float *d0,*d1, *e0, *e1; + + e0 = &v[n4]; + e1 = &v[0]; + + d0 = &u[n4]; + d1 = &u[0]; + + while (AA >= A) { + float v40_20, v41_21; + + v41_21 = e0[1] - e1[1]; + v40_20 = e0[0] - e1[0]; + d0[1] = e0[1] + e1[1]; + d0[0] = e0[0] + e1[0]; + d1[1] = v41_21*AA[4] - v40_20*AA[5]; + d1[0] = v40_20*AA[4] + v41_21*AA[5]; + + v41_21 = e0[3] - e1[3]; + v40_20 = e0[2] - e1[2]; + d0[3] = e0[3] + e1[3]; + d0[2] = e0[2] + e1[2]; + d1[3] = v41_21*AA[0] - v40_20*AA[1]; + d1[2] = v40_20*AA[0] + v41_21*AA[1]; + + AA -= 8; + + d0 += 4; + d1 += 4; + e0 += 4; + e1 += 4; + } + } + + // step 3 + ld = ilog(n) - 1; // ilog is off-by-one from normal definitions + + // optimized step 3: + + // the original step3 loop can be nested r inside s or s inside r; + // it's written originally as s inside r, but this is dumb when r + // iterates many times, and s few. So I have two copies of it and + // switch between them halfway. + + // this is iteration 0 of step 3 + imdct_step3_iter0_loop(n >> 4, u, n2-1-n4*0, -(n >> 3), A); + imdct_step3_iter0_loop(n >> 4, u, n2-1-n4*1, -(n >> 3), A); + + // this is iteration 1 of step 3 + imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*0, -(n >> 4), A, 16); + imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*1, -(n >> 4), A, 16); + imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*2, -(n >> 4), A, 16); + imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*3, -(n >> 4), A, 16); + + l=2; + for (; l < (ld-3)>>1; ++l) { + int k0 = n >> (l+2), k0_2 = k0>>1; + int lim = 1 << (l+1); + int i; + for (i=0; i < lim; ++i) + imdct_step3_inner_r_loop(n >> (l+4), u, n2-1 - k0*i, -k0_2, A, 1 << (l+3)); + } + + for (; l < ld-6; ++l) { + int k0 = n >> (l+2), k1 = 1 << (l+3), k0_2 = k0>>1; + int rlim = n >> (l+6), r; + int lim = 1 << (l+1); + int i_off; + float *A0 = A; + i_off = n2-1; + for (r=rlim; r > 0; --r) { + imdct_step3_inner_s_loop(lim, u, i_off, -k0_2, A0, k1, k0); + A0 += k1*4; + i_off -= 8; + } + } + + // iterations with count: + // ld-6,-5,-4 all interleaved together + // the big win comes from getting rid of needless flops + // due to the constants on pass 5 & 4 being all 1 and 0; + // combining them to be simultaneous to improve cache made little difference + imdct_step3_inner_s_loop_ld654(n >> 5, u, n2-1, A, n); + + // output is u + + // step 4, 5, and 6 + // cannot be in-place because of step 5 + { + uint16 *bitrev = f->bit_reverse[blocktype]; + // weirdly, I'd have thought reading sequentially and writing + // erratically would have been better than vice-versa, but in + // fact that's not what my testing showed. (That is, with + // j = bitreverse(i), do you read i and write j, or read j and write i.) + + float *d0 = &v[n4-4]; + float *d1 = &v[n2-4]; + while (d0 >= v) { + int k4; + + k4 = bitrev[0]; + d1[3] = u[k4+0]; + d1[2] = u[k4+1]; + d0[3] = u[k4+2]; + d0[2] = u[k4+3]; + + k4 = bitrev[1]; + d1[1] = u[k4+0]; + d1[0] = u[k4+1]; + d0[1] = u[k4+2]; + d0[0] = u[k4+3]; + + d0 -= 4; + d1 -= 4; + bitrev += 2; + } + } + // (paper output is u, now v) + + + // data must be in buf2 + assert(v == buf2); + + // step 7 (paper output is v, now v) + // this is now in place + { + float *C = f->C[blocktype]; + float *d, *e; + + d = v; + e = v + n2 - 4; + + while (d < e) { + float a02,a11,b0,b1,b2,b3; + + a02 = d[0] - e[2]; + a11 = d[1] + e[3]; + + b0 = C[1]*a02 + C[0]*a11; + b1 = C[1]*a11 - C[0]*a02; + + b2 = d[0] + e[ 2]; + b3 = d[1] - e[ 3]; + + d[0] = b2 + b0; + d[1] = b3 + b1; + e[2] = b2 - b0; + e[3] = b1 - b3; + + a02 = d[2] - e[0]; + a11 = d[3] + e[1]; + + b0 = C[3]*a02 + C[2]*a11; + b1 = C[3]*a11 - C[2]*a02; + + b2 = d[2] + e[ 0]; + b3 = d[3] - e[ 1]; + + d[2] = b2 + b0; + d[3] = b3 + b1; + e[0] = b2 - b0; + e[1] = b1 - b3; + + C += 4; + d += 4; + e -= 4; + } + } + + // data must be in buf2 + + + // step 8+decode (paper output is X, now buffer) + // this generates pairs of data a la 8 and pushes them directly through + // the decode kernel (pushing rather than pulling) to avoid having + // to make another pass later + + // this cannot POSSIBLY be in place, so we refer to the buffers directly + + { + float *d0,*d1,*d2,*d3; + + float *B = f->B[blocktype] + n2 - 8; + float *e = buf2 + n2 - 8; + d0 = &buffer[0]; + d1 = &buffer[n2-4]; + d2 = &buffer[n2]; + d3 = &buffer[n-4]; + while (e >= v) { + float p0,p1,p2,p3; + + p3 = e[6]*B[7] - e[7]*B[6]; + p2 = -e[6]*B[6] - e[7]*B[7]; + + d0[0] = p3; + d1[3] = - p3; + d2[0] = p2; + d3[3] = p2; + + p1 = e[4]*B[5] - e[5]*B[4]; + p0 = -e[4]*B[4] - e[5]*B[5]; + + d0[1] = p1; + d1[2] = - p1; + d2[1] = p0; + d3[2] = p0; + + p3 = e[2]*B[3] - e[3]*B[2]; + p2 = -e[2]*B[2] - e[3]*B[3]; + + d0[2] = p3; + d1[1] = - p3; + d2[2] = p2; + d3[1] = p2; + + p1 = e[0]*B[1] - e[1]*B[0]; + p0 = -e[0]*B[0] - e[1]*B[1]; + + d0[3] = p1; + d1[0] = - p1; + d2[3] = p0; + d3[0] = p0; + + B -= 8; + e -= 8; + d0 += 4; + d2 += 4; + d1 -= 4; + d3 -= 4; + } + } + + temp_free(f,buf2); + temp_alloc_restore(f,save_point); +} + +#if 0 +// this is the original version of the above code, if you want to optimize it from scratch +void inverse_mdct_naive(float *buffer, int n) +{ + float s; + float A[1 << 12], B[1 << 12], C[1 << 11]; + int i,k,k2,k4, n2 = n >> 1, n4 = n >> 2, n8 = n >> 3, l; + int n3_4 = n - n4, ld; + // how can they claim this only uses N words?! + // oh, because they're only used sparsely, whoops + float u[1 << 13], X[1 << 13], v[1 << 13], w[1 << 13]; + // set up twiddle factors + + for (k=k2=0; k < n4; ++k,k2+=2) { + A[k2 ] = (float) cos(4*k*M_PI/n); + A[k2+1] = (float) -sin(4*k*M_PI/n); + B[k2 ] = (float) cos((k2+1)*M_PI/n/2); + B[k2+1] = (float) sin((k2+1)*M_PI/n/2); + } + for (k=k2=0; k < n8; ++k,k2+=2) { + C[k2 ] = (float) cos(2*(k2+1)*M_PI/n); + C[k2+1] = (float) -sin(2*(k2+1)*M_PI/n); + } + + // IMDCT algorithm from "The use of multirate filter banks for coding of high quality digital audio" + // Note there are bugs in that pseudocode, presumably due to them attempting + // to rename the arrays nicely rather than representing the way their actual + // implementation bounces buffers back and forth. As a result, even in the + // "some formulars corrected" version, a direct implementation fails. These + // are noted below as "paper bug". + + // copy and reflect spectral data + for (k=0; k < n2; ++k) u[k] = buffer[k]; + for ( ; k < n ; ++k) u[k] = -buffer[n - k - 1]; + // kernel from paper + // step 1 + for (k=k2=k4=0; k < n4; k+=1, k2+=2, k4+=4) { + v[n-k4-1] = (u[k4] - u[n-k4-1]) * A[k2] - (u[k4+2] - u[n-k4-3])*A[k2+1]; + v[n-k4-3] = (u[k4] - u[n-k4-1]) * A[k2+1] + (u[k4+2] - u[n-k4-3])*A[k2]; + } + // step 2 + for (k=k4=0; k < n8; k+=1, k4+=4) { + w[n2+3+k4] = v[n2+3+k4] + v[k4+3]; + w[n2+1+k4] = v[n2+1+k4] + v[k4+1]; + w[k4+3] = (v[n2+3+k4] - v[k4+3])*A[n2-4-k4] - (v[n2+1+k4]-v[k4+1])*A[n2-3-k4]; + w[k4+1] = (v[n2+1+k4] - v[k4+1])*A[n2-4-k4] + (v[n2+3+k4]-v[k4+3])*A[n2-3-k4]; + } + // step 3 + ld = ilog(n) - 1; // ilog is off-by-one from normal definitions + for (l=0; l < ld-3; ++l) { + int k0 = n >> (l+2), k1 = 1 << (l+3); + int rlim = n >> (l+4), r4, r; + int s2lim = 1 << (l+2), s2; + for (r=r4=0; r < rlim; r4+=4,++r) { + for (s2=0; s2 < s2lim; s2+=2) { + u[n-1-k0*s2-r4] = w[n-1-k0*s2-r4] + w[n-1-k0*(s2+1)-r4]; + u[n-3-k0*s2-r4] = w[n-3-k0*s2-r4] + w[n-3-k0*(s2+1)-r4]; + u[n-1-k0*(s2+1)-r4] = (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1] + - (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1+1]; + u[n-3-k0*(s2+1)-r4] = (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1] + + (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1+1]; + } + } + if (l+1 < ld-3) { + // paper bug: ping-ponging of u&w here is omitted + memcpy(w, u, sizeof(u)); + } + } + + // step 4 + for (i=0; i < n8; ++i) { + int j = bit_reverse(i) >> (32-ld+3); + assert(j < n8); + if (i == j) { + // paper bug: original code probably swapped in place; if copying, + // need to directly copy in this case + int i8 = i << 3; + v[i8+1] = u[i8+1]; + v[i8+3] = u[i8+3]; + v[i8+5] = u[i8+5]; + v[i8+7] = u[i8+7]; + } else if (i < j) { + int i8 = i << 3, j8 = j << 3; + v[j8+1] = u[i8+1], v[i8+1] = u[j8 + 1]; + v[j8+3] = u[i8+3], v[i8+3] = u[j8 + 3]; + v[j8+5] = u[i8+5], v[i8+5] = u[j8 + 5]; + v[j8+7] = u[i8+7], v[i8+7] = u[j8 + 7]; + } + } + // step 5 + for (k=0; k < n2; ++k) { + w[k] = v[k*2+1]; + } + // step 6 + for (k=k2=k4=0; k < n8; ++k, k2 += 2, k4 += 4) { + u[n-1-k2] = w[k4]; + u[n-2-k2] = w[k4+1]; + u[n3_4 - 1 - k2] = w[k4+2]; + u[n3_4 - 2 - k2] = w[k4+3]; + } + // step 7 + for (k=k2=0; k < n8; ++k, k2 += 2) { + v[n2 + k2 ] = ( u[n2 + k2] + u[n-2-k2] + C[k2+1]*(u[n2+k2]-u[n-2-k2]) + C[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2; + v[n-2 - k2] = ( u[n2 + k2] + u[n-2-k2] - C[k2+1]*(u[n2+k2]-u[n-2-k2]) - C[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2; + v[n2+1+ k2] = ( u[n2+1+k2] - u[n-1-k2] + C[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C[k2]*(u[n2+k2]-u[n-2-k2]))/2; + v[n-1 - k2] = (-u[n2+1+k2] + u[n-1-k2] + C[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C[k2]*(u[n2+k2]-u[n-2-k2]))/2; + } + // step 8 + for (k=k2=0; k < n4; ++k,k2 += 2) { + X[k] = v[k2+n2]*B[k2 ] + v[k2+1+n2]*B[k2+1]; + X[n2-1-k] = v[k2+n2]*B[k2+1] - v[k2+1+n2]*B[k2 ]; + } + + // decode kernel to output + // determined the following value experimentally + // (by first figuring out what made inverse_mdct_slow work); then matching that here + // (probably vorbis encoder premultiplies by n or n/2, to save it on the decoder?) + s = 0.5; // theoretically would be n4 + + // [[[ note! the s value of 0.5 is compensated for by the B[] in the current code, + // so it needs to use the "old" B values to behave correctly, or else + // set s to 1.0 ]]] + for (i=0; i < n4 ; ++i) buffer[i] = s * X[i+n4]; + for ( ; i < n3_4; ++i) buffer[i] = -s * X[n3_4 - i - 1]; + for ( ; i < n ; ++i) buffer[i] = -s * X[i - n3_4]; +} +#endif + +static float *get_window(vorb *f, int len) +{ + len <<= 1; + if (len == f->blocksize_0) return f->window[0]; + if (len == f->blocksize_1) return f->window[1]; + return NULL; +} + +#ifndef STB_VORBIS_NO_DEFER_FLOOR +typedef int16 YTYPE; +#else +typedef int YTYPE; +#endif +static int do_floor(vorb *f, Mapping *map, int i, int n, float *target, YTYPE *finalY, uint8 *step2_flag) +{ + int n2 = n >> 1; + int s = map->chan[i].mux, floor; + floor = map->submap_floor[s]; + if (f->floor_types[floor] == 0) { + return error(f, VORBIS_invalid_stream); + } else { + Floor1 *g = &f->floor_config[floor].floor1; + int j,q; + int lx = 0, ly = finalY[0] * g->floor1_multiplier; + for (q=1; q < g->values; ++q) { + j = g->sorted_order[q]; + #ifndef STB_VORBIS_NO_DEFER_FLOOR + STBV_NOTUSED(step2_flag); + if (finalY[j] >= 0) + #else + if (step2_flag[j]) + #endif + { + int hy = finalY[j] * g->floor1_multiplier; + int hx = g->Xlist[j]; + if (lx != hx) + draw_line(target, lx,ly, hx,hy, n2); + CHECK(f); + lx = hx, ly = hy; + } + } + if (lx < n2) { + // optimization of: draw_line(target, lx,ly, n,ly, n2); + for (j=lx; j < n2; ++j) + LINE_OP(target[j], inverse_db_table[ly]); + CHECK(f); + } + } + return TRUE; +} + +// The meaning of "left" and "right" +// +// For a given frame: +// we compute samples from 0..n +// window_center is n/2 +// we'll window and mix the samples from left_start to left_end with data from the previous frame +// all of the samples from left_end to right_start can be output without mixing; however, +// this interval is 0-length except when transitioning between short and long frames +// all of the samples from right_start to right_end need to be mixed with the next frame, +// which we don't have, so those get saved in a buffer +// frame N's right_end-right_start, the number of samples to mix with the next frame, +// has to be the same as frame N+1's left_end-left_start (which they are by +// construction) + +static int vorbis_decode_initial(vorb *f, int *p_left_start, int *p_left_end, int *p_right_start, int *p_right_end, int *mode) +{ + Mode *m; + int i, n, prev, next, window_center; + f->channel_buffer_start = f->channel_buffer_end = 0; + + retry: + if (f->eof) return FALSE; + if (!maybe_start_packet(f)) + return FALSE; + // check packet type + if (get_bits(f,1) != 0) { + if (IS_PUSH_MODE(f)) + return error(f,VORBIS_bad_packet_type); + while (EOP != get8_packet(f)); + goto retry; + } + + if (f->alloc.alloc_buffer) + assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset); + + i = get_bits(f, ilog(f->mode_count-1)); + if (i == EOP) return FALSE; + if (i >= f->mode_count) return FALSE; + *mode = i; + m = f->mode_config + i; + if (m->blockflag) { + n = f->blocksize_1; + prev = get_bits(f,1); + next = get_bits(f,1); + } else { + prev = next = 0; + n = f->blocksize_0; + } + +// WINDOWING + + window_center = n >> 1; + if (m->blockflag && !prev) { + *p_left_start = (n - f->blocksize_0) >> 2; + *p_left_end = (n + f->blocksize_0) >> 2; + } else { + *p_left_start = 0; + *p_left_end = window_center; + } + if (m->blockflag && !next) { + *p_right_start = (n*3 - f->blocksize_0) >> 2; + *p_right_end = (n*3 + f->blocksize_0) >> 2; + } else { + *p_right_start = window_center; + *p_right_end = n; + } + + return TRUE; +} + +static int vorbis_decode_packet_rest(vorb *f, int *len, Mode *m, int left_start, int left_end, int right_start, int right_end, int *p_left) +{ + Mapping *map; + int i,j,k,n,n2; + int zero_channel[256]; + int really_zero_channel[256]; + +// WINDOWING + + STBV_NOTUSED(left_end); + n = f->blocksize[m->blockflag]; + map = &f->mapping[m->mapping]; + +// FLOORS + n2 = n >> 1; + + CHECK(f); + + for (i=0; i < f->channels; ++i) { + int s = map->chan[i].mux, floor; + zero_channel[i] = FALSE; + floor = map->submap_floor[s]; + if (f->floor_types[floor] == 0) { + return error(f, VORBIS_invalid_stream); + } else { + Floor1 *g = &f->floor_config[floor].floor1; + if (get_bits(f, 1)) { + short *finalY; + uint8 step2_flag[256]; + static int range_list[4] = { 256, 128, 86, 64 }; + int range = range_list[g->floor1_multiplier-1]; + int offset = 2; + finalY = f->finalY[i]; + finalY[0] = get_bits(f, ilog(range)-1); + finalY[1] = get_bits(f, ilog(range)-1); + for (j=0; j < g->partitions; ++j) { + int pclass = g->partition_class_list[j]; + int cdim = g->class_dimensions[pclass]; + int cbits = g->class_subclasses[pclass]; + int csub = (1 << cbits)-1; + int cval = 0; + if (cbits) { + Codebook *c = f->codebooks + g->class_masterbooks[pclass]; + DECODE(cval,f,c); + } + for (k=0; k < cdim; ++k) { + int book = g->subclass_books[pclass][cval & csub]; + cval = cval >> cbits; + if (book >= 0) { + int temp; + Codebook *c = f->codebooks + book; + DECODE(temp,f,c); + finalY[offset++] = temp; + } else + finalY[offset++] = 0; + } + } + if (f->valid_bits == INVALID_BITS) goto error; // behavior according to spec + step2_flag[0] = step2_flag[1] = 1; + for (j=2; j < g->values; ++j) { + int low, high, pred, highroom, lowroom, room, val; + low = g->neighbors[j][0]; + high = g->neighbors[j][1]; + //neighbors(g->Xlist, j, &low, &high); + pred = predict_point(g->Xlist[j], g->Xlist[low], g->Xlist[high], finalY[low], finalY[high]); + val = finalY[j]; + highroom = range - pred; + lowroom = pred; + if (highroom < lowroom) + room = highroom * 2; + else + room = lowroom * 2; + if (val) { + step2_flag[low] = step2_flag[high] = 1; + step2_flag[j] = 1; + if (val >= room) + if (highroom > lowroom) + finalY[j] = val - lowroom + pred; + else + finalY[j] = pred - val + highroom - 1; + else + if (val & 1) + finalY[j] = pred - ((val+1)>>1); + else + finalY[j] = pred + (val>>1); + } else { + step2_flag[j] = 0; + finalY[j] = pred; + } + } + +#ifdef STB_VORBIS_NO_DEFER_FLOOR + do_floor(f, map, i, n, f->floor_buffers[i], finalY, step2_flag); +#else + // defer final floor computation until _after_ residue + for (j=0; j < g->values; ++j) { + if (!step2_flag[j]) + finalY[j] = -1; + } +#endif + } else { + error: + zero_channel[i] = TRUE; + } + // So we just defer everything else to later + + // at this point we've decoded the floor into buffer + } + } + CHECK(f); + // at this point we've decoded all floors + + if (f->alloc.alloc_buffer) + assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset); + + // re-enable coupled channels if necessary + memcpy(really_zero_channel, zero_channel, sizeof(really_zero_channel[0]) * f->channels); + for (i=0; i < map->coupling_steps; ++i) + if (!zero_channel[map->chan[i].magnitude] || !zero_channel[map->chan[i].angle]) { + zero_channel[map->chan[i].magnitude] = zero_channel[map->chan[i].angle] = FALSE; + } + + CHECK(f); +// RESIDUE DECODE + for (i=0; i < map->submaps; ++i) { + float *residue_buffers[STB_VORBIS_MAX_CHANNELS]; + int r; + uint8 do_not_decode[256]; + int ch = 0; + for (j=0; j < f->channels; ++j) { + if (map->chan[j].mux == i) { + if (zero_channel[j]) { + do_not_decode[ch] = TRUE; + residue_buffers[ch] = NULL; + } else { + do_not_decode[ch] = FALSE; + residue_buffers[ch] = f->channel_buffers[j]; + } + ++ch; + } + } + r = map->submap_residue[i]; + decode_residue(f, residue_buffers, ch, n2, r, do_not_decode); + } + + if (f->alloc.alloc_buffer) + assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset); + CHECK(f); + +// INVERSE COUPLING + for (i = map->coupling_steps-1; i >= 0; --i) { + int n2 = n >> 1; + float *m = f->channel_buffers[map->chan[i].magnitude]; + float *a = f->channel_buffers[map->chan[i].angle ]; + for (j=0; j < n2; ++j) { + float a2,m2; + if (m[j] > 0) + if (a[j] > 0) + m2 = m[j], a2 = m[j] - a[j]; + else + a2 = m[j], m2 = m[j] + a[j]; + else + if (a[j] > 0) + m2 = m[j], a2 = m[j] + a[j]; + else + a2 = m[j], m2 = m[j] - a[j]; + m[j] = m2; + a[j] = a2; + } + } + CHECK(f); + + // finish decoding the floors +#ifndef STB_VORBIS_NO_DEFER_FLOOR + for (i=0; i < f->channels; ++i) { + if (really_zero_channel[i]) { + memset(f->channel_buffers[i], 0, sizeof(*f->channel_buffers[i]) * n2); + } else { + do_floor(f, map, i, n, f->channel_buffers[i], f->finalY[i], NULL); + } + } +#else + for (i=0; i < f->channels; ++i) { + if (really_zero_channel[i]) { + memset(f->channel_buffers[i], 0, sizeof(*f->channel_buffers[i]) * n2); + } else { + for (j=0; j < n2; ++j) + f->channel_buffers[i][j] *= f->floor_buffers[i][j]; + } + } +#endif + +// INVERSE MDCT + CHECK(f); + for (i=0; i < f->channels; ++i) + inverse_mdct(f->channel_buffers[i], n, f, m->blockflag); + CHECK(f); + + // this shouldn't be necessary, unless we exited on an error + // and want to flush to get to the next packet + flush_packet(f); + + if (f->first_decode) { + // assume we start so first non-discarded sample is sample 0 + // this isn't to spec, but spec would require us to read ahead + // and decode the size of all current frames--could be done, + // but presumably it's not a commonly used feature + f->current_loc = 0u - n2; // start of first frame is positioned for discard (NB this is an intentional unsigned overflow/wrap-around) + // we might have to discard samples "from" the next frame too, + // if we're lapping a large block then a small at the start? + f->discard_samples_deferred = n - right_end; + f->current_loc_valid = TRUE; + f->first_decode = FALSE; + } else if (f->discard_samples_deferred) { + if (f->discard_samples_deferred >= right_start - left_start) { + f->discard_samples_deferred -= (right_start - left_start); + left_start = right_start; + *p_left = left_start; + } else { + left_start += f->discard_samples_deferred; + *p_left = left_start; + f->discard_samples_deferred = 0; + } + } else if (f->previous_length == 0 && f->current_loc_valid) { + // we're recovering from a seek... that means we're going to discard + // the samples from this packet even though we know our position from + // the last page header, so we need to update the position based on + // the discarded samples here + // but wait, the code below is going to add this in itself even + // on a discard, so we don't need to do it here... + } + + // check if we have ogg information about the sample # for this packet + if (f->last_seg_which == f->end_seg_with_known_loc) { + // if we have a valid current loc, and this is final: + if (f->current_loc_valid && (f->page_flag & PAGEFLAG_last_page)) { + uint32 current_end = f->known_loc_for_packet; + // then let's infer the size of the (probably) short final frame + if (current_end < f->current_loc + (right_end-left_start)) { + if (current_end < f->current_loc) { + // negative truncation, that's impossible! + *len = 0; + } else { + *len = current_end - f->current_loc; + } + *len += left_start; // this doesn't seem right, but has no ill effect on my test files + if (*len > right_end) *len = right_end; // this should never happen + f->current_loc += *len; + return TRUE; + } + } + // otherwise, just set our sample loc + // guess that the ogg granule pos refers to the _middle_ of the + // last frame? + // set f->current_loc to the position of left_start + f->current_loc = f->known_loc_for_packet - (n2-left_start); + f->current_loc_valid = TRUE; + } + if (f->current_loc_valid) + f->current_loc += (right_start - left_start); + + if (f->alloc.alloc_buffer) + assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset); + *len = right_end; // ignore samples after the window goes to 0 + CHECK(f); + + return TRUE; +} + +static int vorbis_decode_packet(vorb *f, int *len, int *p_left, int *p_right) +{ + int mode, left_end, right_end; + if (!vorbis_decode_initial(f, p_left, &left_end, p_right, &right_end, &mode)) return 0; + return vorbis_decode_packet_rest(f, len, f->mode_config + mode, *p_left, left_end, *p_right, right_end, p_left); +} + +static int vorbis_finish_frame(stb_vorbis *f, int len, int left, int right) +{ + int prev,i,j; + // we use right&left (the start of the right- and left-window sin()-regions) + // to determine how much to return, rather than inferring from the rules + // (same result, clearer code); 'left' indicates where our sin() window + // starts, therefore where the previous window's right edge starts, and + // therefore where to start mixing from the previous buffer. 'right' + // indicates where our sin() ending-window starts, therefore that's where + // we start saving, and where our returned-data ends. + + // mixin from previous window + if (f->previous_length) { + int i,j, n = f->previous_length; + float *w = get_window(f, n); + if (w == NULL) return 0; + for (i=0; i < f->channels; ++i) { + for (j=0; j < n; ++j) + f->channel_buffers[i][left+j] = + f->channel_buffers[i][left+j]*w[ j] + + f->previous_window[i][ j]*w[n-1-j]; + } + } + + prev = f->previous_length; + + // last half of this data becomes previous window + f->previous_length = len - right; + + // @OPTIMIZE: could avoid this copy by double-buffering the + // output (flipping previous_window with channel_buffers), but + // then previous_window would have to be 2x as large, and + // channel_buffers couldn't be temp mem (although they're NOT + // currently temp mem, they could be (unless we want to level + // performance by spreading out the computation)) + for (i=0; i < f->channels; ++i) + for (j=0; right+j < len; ++j) + f->previous_window[i][j] = f->channel_buffers[i][right+j]; + + if (!prev) + // there was no previous packet, so this data isn't valid... + // this isn't entirely true, only the would-have-overlapped data + // isn't valid, but this seems to be what the spec requires + return 0; + + // truncate a short frame + if (len < right) right = len; + + f->samples_output += right-left; + + return right - left; +} + +static int vorbis_pump_first_frame(stb_vorbis *f) +{ + int len, right, left, res; + res = vorbis_decode_packet(f, &len, &left, &right); + if (res) + vorbis_finish_frame(f, len, left, right); + return res; +} + +#ifndef STB_VORBIS_NO_PUSHDATA_API +static int is_whole_packet_present(stb_vorbis *f) +{ + // make sure that we have the packet available before continuing... + // this requires a full ogg parse, but we know we can fetch from f->stream + + // instead of coding this out explicitly, we could save the current read state, + // read the next packet with get8() until end-of-packet, check f->eof, then + // reset the state? but that would be slower, esp. since we'd have over 256 bytes + // of state to restore (primarily the page segment table) + + int s = f->next_seg, first = TRUE; + uint8 *p = f->stream; + + if (s != -1) { // if we're not starting the packet with a 'continue on next page' flag + for (; s < f->segment_count; ++s) { + p += f->segments[s]; + if (f->segments[s] < 255) // stop at first short segment + break; + } + // either this continues, or it ends it... + if (s == f->segment_count) + s = -1; // set 'crosses page' flag + if (p > f->stream_end) return error(f, VORBIS_need_more_data); + first = FALSE; + } + for (; s == -1;) { + uint8 *q; + int n; + + // check that we have the page header ready + if (p + 26 >= f->stream_end) return error(f, VORBIS_need_more_data); + // validate the page + if (memcmp(p, ogg_page_header, 4)) return error(f, VORBIS_invalid_stream); + if (p[4] != 0) return error(f, VORBIS_invalid_stream); + if (first) { // the first segment must NOT have 'continued_packet', later ones MUST + if (f->previous_length) + if ((p[5] & PAGEFLAG_continued_packet)) return error(f, VORBIS_invalid_stream); + // if no previous length, we're resynching, so we can come in on a continued-packet, + // which we'll just drop + } else { + if (!(p[5] & PAGEFLAG_continued_packet)) return error(f, VORBIS_invalid_stream); + } + n = p[26]; // segment counts + q = p+27; // q points to segment table + p = q + n; // advance past header + // make sure we've read the segment table + if (p > f->stream_end) return error(f, VORBIS_need_more_data); + for (s=0; s < n; ++s) { + p += q[s]; + if (q[s] < 255) + break; + } + if (s == n) + s = -1; // set 'crosses page' flag + if (p > f->stream_end) return error(f, VORBIS_need_more_data); + first = FALSE; + } + return TRUE; +} +#endif // !STB_VORBIS_NO_PUSHDATA_API + +static int start_decoder(vorb *f) +{ + uint8 header[6], x,y; + int len,i,j,k, max_submaps = 0; + int longest_floorlist=0; + + // first page, first packet + f->first_decode = TRUE; + + if (!start_page(f)) return FALSE; + // validate page flag + if (!(f->page_flag & PAGEFLAG_first_page)) return error(f, VORBIS_invalid_first_page); + if (f->page_flag & PAGEFLAG_last_page) return error(f, VORBIS_invalid_first_page); + if (f->page_flag & PAGEFLAG_continued_packet) return error(f, VORBIS_invalid_first_page); + // check for expected packet length + if (f->segment_count != 1) return error(f, VORBIS_invalid_first_page); + if (f->segments[0] != 30) { + // check for the Ogg skeleton fishead identifying header to refine our error + if (f->segments[0] == 64 && + getn(f, header, 6) && + header[0] == 'f' && + header[1] == 'i' && + header[2] == 's' && + header[3] == 'h' && + header[4] == 'e' && + header[5] == 'a' && + get8(f) == 'd' && + get8(f) == '\0') return error(f, VORBIS_ogg_skeleton_not_supported); + else + return error(f, VORBIS_invalid_first_page); + } + + // read packet + // check packet header + if (get8(f) != VORBIS_packet_id) return error(f, VORBIS_invalid_first_page); + if (!getn(f, header, 6)) return error(f, VORBIS_unexpected_eof); + if (!vorbis_validate(header)) return error(f, VORBIS_invalid_first_page); + // vorbis_version + if (get32(f) != 0) return error(f, VORBIS_invalid_first_page); + f->channels = get8(f); if (!f->channels) return error(f, VORBIS_invalid_first_page); + if (f->channels > STB_VORBIS_MAX_CHANNELS) return error(f, VORBIS_too_many_channels); + f->sample_rate = get32(f); if (!f->sample_rate) return error(f, VORBIS_invalid_first_page); + get32(f); // bitrate_maximum + get32(f); // bitrate_nominal + get32(f); // bitrate_minimum + x = get8(f); + { + int log0,log1; + log0 = x & 15; + log1 = x >> 4; + f->blocksize_0 = 1 << log0; + f->blocksize_1 = 1 << log1; + if (log0 < 6 || log0 > 13) return error(f, VORBIS_invalid_setup); + if (log1 < 6 || log1 > 13) return error(f, VORBIS_invalid_setup); + if (log0 > log1) return error(f, VORBIS_invalid_setup); + } + + // framing_flag + x = get8(f); + if (!(x & 1)) return error(f, VORBIS_invalid_first_page); + + // second packet! + if (!start_page(f)) return FALSE; + + if (!start_packet(f)) return FALSE; + + if (!next_segment(f)) return FALSE; + + if (get8_packet(f) != VORBIS_packet_comment) return error(f, VORBIS_invalid_setup); + for (i=0; i < 6; ++i) header[i] = get8_packet(f); + if (!vorbis_validate(header)) return error(f, VORBIS_invalid_setup); + //file vendor + len = get32_packet(f); + f->vendor = (char*)setup_malloc(f, sizeof(char) * (len+1)); + if (f->vendor == NULL) return error(f, VORBIS_outofmem); + for(i=0; i < len; ++i) { + f->vendor[i] = get8_packet(f); + } + f->vendor[len] = (char)'\0'; + //user comments + f->comment_list_length = get32_packet(f); + f->comment_list = NULL; + if (f->comment_list_length > 0) + { + f->comment_list = (char**) setup_malloc(f, sizeof(char*) * (f->comment_list_length)); + if (f->comment_list == NULL) return error(f, VORBIS_outofmem); + } + + for(i=0; i < f->comment_list_length; ++i) { + len = get32_packet(f); + f->comment_list[i] = (char*)setup_malloc(f, sizeof(char) * (len+1)); + if (f->comment_list[i] == NULL) return error(f, VORBIS_outofmem); + + for(j=0; j < len; ++j) { + f->comment_list[i][j] = get8_packet(f); + } + f->comment_list[i][len] = (char)'\0'; + } + + // framing_flag + x = get8_packet(f); + if (!(x & 1)) return error(f, VORBIS_invalid_setup); + + + skip(f, f->bytes_in_seg); + f->bytes_in_seg = 0; + + do { + len = next_segment(f); + skip(f, len); + f->bytes_in_seg = 0; + } while (len); + + // third packet! + if (!start_packet(f)) return FALSE; + + #ifndef STB_VORBIS_NO_PUSHDATA_API + if (IS_PUSH_MODE(f)) { + if (!is_whole_packet_present(f)) { + // convert error in ogg header to write type + if (f->error == VORBIS_invalid_stream) + f->error = VORBIS_invalid_setup; + return FALSE; + } + } + #endif + + crc32_init(); // always init it, to avoid multithread race conditions + + if (get8_packet(f) != VORBIS_packet_setup) return error(f, VORBIS_invalid_setup); + for (i=0; i < 6; ++i) header[i] = get8_packet(f); + if (!vorbis_validate(header)) return error(f, VORBIS_invalid_setup); + + // codebooks + + f->codebook_count = get_bits(f,8) + 1; + f->codebooks = (Codebook *) setup_malloc(f, sizeof(*f->codebooks) * f->codebook_count); + if (f->codebooks == NULL) return error(f, VORBIS_outofmem); + memset(f->codebooks, 0, sizeof(*f->codebooks) * f->codebook_count); + for (i=0; i < f->codebook_count; ++i) { + uint32 *values; + int ordered, sorted_count; + int total=0; + uint8 *lengths; + Codebook *c = f->codebooks+i; + CHECK(f); + x = get_bits(f, 8); if (x != 0x42) return error(f, VORBIS_invalid_setup); + x = get_bits(f, 8); if (x != 0x43) return error(f, VORBIS_invalid_setup); + x = get_bits(f, 8); if (x != 0x56) return error(f, VORBIS_invalid_setup); + x = get_bits(f, 8); + c->dimensions = (get_bits(f, 8)<<8) + x; + x = get_bits(f, 8); + y = get_bits(f, 8); + c->entries = (get_bits(f, 8)<<16) + (y<<8) + x; + ordered = get_bits(f,1); + c->sparse = ordered ? 0 : get_bits(f,1); + + if (c->dimensions == 0 && c->entries != 0) return error(f, VORBIS_invalid_setup); + + if (c->sparse) + lengths = (uint8 *) setup_temp_malloc(f, c->entries); + else + lengths = c->codeword_lengths = (uint8 *) setup_malloc(f, c->entries); + + if (!lengths) return error(f, VORBIS_outofmem); + + if (ordered) { + int current_entry = 0; + int current_length = get_bits(f,5) + 1; + while (current_entry < c->entries) { + int limit = c->entries - current_entry; + int n = get_bits(f, ilog(limit)); + if (current_length >= 32) return error(f, VORBIS_invalid_setup); + if (current_entry + n > (int) c->entries) { return error(f, VORBIS_invalid_setup); } + memset(lengths + current_entry, current_length, n); + current_entry += n; + ++current_length; + } + } else { + for (j=0; j < c->entries; ++j) { + int present = c->sparse ? get_bits(f,1) : 1; + if (present) { + lengths[j] = get_bits(f, 5) + 1; + ++total; + if (lengths[j] == 32) + return error(f, VORBIS_invalid_setup); + } else { + lengths[j] = NO_CODE; + } + } + } + + if (c->sparse && total >= c->entries >> 2) { + // convert sparse items to non-sparse! + if (c->entries > (int) f->setup_temp_memory_required) + f->setup_temp_memory_required = c->entries; + + c->codeword_lengths = (uint8 *) setup_malloc(f, c->entries); + if (c->codeword_lengths == NULL) return error(f, VORBIS_outofmem); + memcpy(c->codeword_lengths, lengths, c->entries); + setup_temp_free(f, lengths, c->entries); // note this is only safe if there have been no intervening temp mallocs! + lengths = c->codeword_lengths; + c->sparse = 0; + } + + // compute the size of the sorted tables + if (c->sparse) { + sorted_count = total; + } else { + sorted_count = 0; + #ifndef STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH + for (j=0; j < c->entries; ++j) + if (lengths[j] > STB_VORBIS_FAST_HUFFMAN_LENGTH && lengths[j] != NO_CODE) + ++sorted_count; + #endif + } + + c->sorted_entries = sorted_count; + values = NULL; + + CHECK(f); + if (!c->sparse) { + c->codewords = (uint32 *) setup_malloc(f, sizeof(c->codewords[0]) * c->entries); + if (!c->codewords) return error(f, VORBIS_outofmem); + } else { + unsigned int size; + if (c->sorted_entries) { + c->codeword_lengths = (uint8 *) setup_malloc(f, c->sorted_entries); + if (!c->codeword_lengths) return error(f, VORBIS_outofmem); + c->codewords = (uint32 *) setup_temp_malloc(f, sizeof(*c->codewords) * c->sorted_entries); + if (!c->codewords) return error(f, VORBIS_outofmem); + values = (uint32 *) setup_temp_malloc(f, sizeof(*values) * c->sorted_entries); + if (!values) return error(f, VORBIS_outofmem); + } + size = c->entries + (sizeof(*c->codewords) + sizeof(*values)) * c->sorted_entries; + if (size > f->setup_temp_memory_required) + f->setup_temp_memory_required = size; + } + + if (!compute_codewords(c, lengths, c->entries, values)) { + if (c->sparse) setup_temp_free(f, values, 0); + return error(f, VORBIS_invalid_setup); + } + + if (c->sorted_entries) { + // allocate an extra slot for sentinels + c->sorted_codewords = (uint32 *) setup_malloc(f, sizeof(*c->sorted_codewords) * (c->sorted_entries+1)); + if (c->sorted_codewords == NULL) return error(f, VORBIS_outofmem); + // allocate an extra slot at the front so that c->sorted_values[-1] is defined + // so that we can catch that case without an extra if + c->sorted_values = ( int *) setup_malloc(f, sizeof(*c->sorted_values ) * (c->sorted_entries+1)); + if (c->sorted_values == NULL) return error(f, VORBIS_outofmem); + ++c->sorted_values; + c->sorted_values[-1] = -1; + compute_sorted_huffman(c, lengths, values); + } + + if (c->sparse) { + setup_temp_free(f, values, sizeof(*values)*c->sorted_entries); + setup_temp_free(f, c->codewords, sizeof(*c->codewords)*c->sorted_entries); + setup_temp_free(f, lengths, c->entries); + c->codewords = NULL; + } + + compute_accelerated_huffman(c); + + CHECK(f); + c->lookup_type = get_bits(f, 4); + if (c->lookup_type > 2) return error(f, VORBIS_invalid_setup); + if (c->lookup_type > 0) { + uint16 *mults; + c->minimum_value = float32_unpack(get_bits(f, 32)); + c->delta_value = float32_unpack(get_bits(f, 32)); + c->value_bits = get_bits(f, 4)+1; + c->sequence_p = get_bits(f,1); + if (c->lookup_type == 1) { + int values = lookup1_values(c->entries, c->dimensions); + if (values < 0) return error(f, VORBIS_invalid_setup); + c->lookup_values = (uint32) values; + } else { + c->lookup_values = c->entries * c->dimensions; + } + if (c->lookup_values == 0) return error(f, VORBIS_invalid_setup); + mults = (uint16 *) setup_temp_malloc(f, sizeof(mults[0]) * c->lookup_values); + if (mults == NULL) return error(f, VORBIS_outofmem); + for (j=0; j < (int) c->lookup_values; ++j) { + int q = get_bits(f, c->value_bits); + if (q == EOP) { setup_temp_free(f,mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_invalid_setup); } + mults[j] = q; + } + +#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK + if (c->lookup_type == 1) { + int len, sparse = c->sparse; + float last=0; + // pre-expand the lookup1-style multiplicands, to avoid a divide in the inner loop + if (sparse) { + if (c->sorted_entries == 0) goto skip; + c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->sorted_entries * c->dimensions); + } else + c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->entries * c->dimensions); + if (c->multiplicands == NULL) { setup_temp_free(f,mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_outofmem); } + len = sparse ? c->sorted_entries : c->entries; + for (j=0; j < len; ++j) { + unsigned int z = sparse ? c->sorted_values[j] : j; + unsigned int div=1; + for (k=0; k < c->dimensions; ++k) { + int off = (z / div) % c->lookup_values; + float val = mults[off]*c->delta_value + c->minimum_value + last; + c->multiplicands[j*c->dimensions + k] = val; + if (c->sequence_p) + last = val; + if (k+1 < c->dimensions) { + if (div > UINT_MAX / (unsigned int) c->lookup_values) { + setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values); + return error(f, VORBIS_invalid_setup); + } + div *= c->lookup_values; + } + } + } + c->lookup_type = 2; + } + else +#endif + { + float last=0; + CHECK(f); + c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->lookup_values); + if (c->multiplicands == NULL) { setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_outofmem); } + for (j=0; j < (int) c->lookup_values; ++j) { + float val = mults[j] * c->delta_value + c->minimum_value + last; + c->multiplicands[j] = val; + if (c->sequence_p) + last = val; + } + } +#ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK + skip:; +#endif + setup_temp_free(f, mults, sizeof(mults[0])*c->lookup_values); + + CHECK(f); + } + CHECK(f); + } + + // time domain transfers (notused) + + x = get_bits(f, 6) + 1; + for (i=0; i < x; ++i) { + uint32 z = get_bits(f, 16); + if (z != 0) return error(f, VORBIS_invalid_setup); + } + + // Floors + f->floor_count = get_bits(f, 6)+1; + f->floor_config = (Floor *) setup_malloc(f, f->floor_count * sizeof(*f->floor_config)); + if (f->floor_config == NULL) return error(f, VORBIS_outofmem); + for (i=0; i < f->floor_count; ++i) { + f->floor_types[i] = get_bits(f, 16); + if (f->floor_types[i] > 1) return error(f, VORBIS_invalid_setup); + if (f->floor_types[i] == 0) { + Floor0 *g = &f->floor_config[i].floor0; + g->order = get_bits(f,8); + g->rate = get_bits(f,16); + g->bark_map_size = get_bits(f,16); + g->amplitude_bits = get_bits(f,6); + g->amplitude_offset = get_bits(f,8); + g->number_of_books = get_bits(f,4) + 1; + for (j=0; j < g->number_of_books; ++j) + g->book_list[j] = get_bits(f,8); + return error(f, VORBIS_feature_not_supported); + } else { + stbv__floor_ordering p[31*8+2]; + Floor1 *g = &f->floor_config[i].floor1; + int max_class = -1; + g->partitions = get_bits(f, 5); + for (j=0; j < g->partitions; ++j) { + g->partition_class_list[j] = get_bits(f, 4); + if (g->partition_class_list[j] > max_class) + max_class = g->partition_class_list[j]; + } + for (j=0; j <= max_class; ++j) { + g->class_dimensions[j] = get_bits(f, 3)+1; + g->class_subclasses[j] = get_bits(f, 2); + if (g->class_subclasses[j]) { + g->class_masterbooks[j] = get_bits(f, 8); + if (g->class_masterbooks[j] >= f->codebook_count) return error(f, VORBIS_invalid_setup); + } + for (k=0; k < 1 << g->class_subclasses[j]; ++k) { + g->subclass_books[j][k] = (int16)get_bits(f,8)-1; + if (g->subclass_books[j][k] >= f->codebook_count) return error(f, VORBIS_invalid_setup); + } + } + g->floor1_multiplier = get_bits(f,2)+1; + g->rangebits = get_bits(f,4); + g->Xlist[0] = 0; + g->Xlist[1] = 1 << g->rangebits; + g->values = 2; + for (j=0; j < g->partitions; ++j) { + int c = g->partition_class_list[j]; + for (k=0; k < g->class_dimensions[c]; ++k) { + g->Xlist[g->values] = get_bits(f, g->rangebits); + ++g->values; + } + } + // precompute the sorting + for (j=0; j < g->values; ++j) { + p[j].x = g->Xlist[j]; + p[j].id = j; + } + qsort(p, g->values, sizeof(p[0]), point_compare); + for (j=0; j < g->values-1; ++j) + if (p[j].x == p[j+1].x) + return error(f, VORBIS_invalid_setup); + for (j=0; j < g->values; ++j) + g->sorted_order[j] = (uint8) p[j].id; + // precompute the neighbors + for (j=2; j < g->values; ++j) { + int low = 0,hi = 0; + neighbors(g->Xlist, j, &low,&hi); + g->neighbors[j][0] = low; + g->neighbors[j][1] = hi; + } + + if (g->values > longest_floorlist) + longest_floorlist = g->values; + } + } + + // Residue + f->residue_count = get_bits(f, 6)+1; + f->residue_config = (Residue *) setup_malloc(f, f->residue_count * sizeof(f->residue_config[0])); + if (f->residue_config == NULL) return error(f, VORBIS_outofmem); + memset(f->residue_config, 0, f->residue_count * sizeof(f->residue_config[0])); + for (i=0; i < f->residue_count; ++i) { + uint8 residue_cascade[64]; + Residue *r = f->residue_config+i; + f->residue_types[i] = get_bits(f, 16); + if (f->residue_types[i] > 2) return error(f, VORBIS_invalid_setup); + r->begin = get_bits(f, 24); + r->end = get_bits(f, 24); + if (r->end < r->begin) return error(f, VORBIS_invalid_setup); + r->part_size = get_bits(f,24)+1; + r->classifications = get_bits(f,6)+1; + r->classbook = get_bits(f,8); + if (r->classbook >= f->codebook_count) return error(f, VORBIS_invalid_setup); + for (j=0; j < r->classifications; ++j) { + uint8 high_bits=0; + uint8 low_bits=get_bits(f,3); + if (get_bits(f,1)) + high_bits = get_bits(f,5); + residue_cascade[j] = high_bits*8 + low_bits; + } + r->residue_books = (short (*)[8]) setup_malloc(f, sizeof(r->residue_books[0]) * r->classifications); + if (r->residue_books == NULL) return error(f, VORBIS_outofmem); + for (j=0; j < r->classifications; ++j) { + for (k=0; k < 8; ++k) { + if (residue_cascade[j] & (1 << k)) { + r->residue_books[j][k] = get_bits(f, 8); + if (r->residue_books[j][k] >= f->codebook_count) return error(f, VORBIS_invalid_setup); + } else { + r->residue_books[j][k] = -1; + } + } + } + // precompute the classifications[] array to avoid inner-loop mod/divide + // call it 'classdata' since we already have r->classifications + r->classdata = (uint8 **) setup_malloc(f, sizeof(*r->classdata) * f->codebooks[r->classbook].entries); + if (!r->classdata) return error(f, VORBIS_outofmem); + memset(r->classdata, 0, sizeof(*r->classdata) * f->codebooks[r->classbook].entries); + for (j=0; j < f->codebooks[r->classbook].entries; ++j) { + int classwords = f->codebooks[r->classbook].dimensions; + int temp = j; + r->classdata[j] = (uint8 *) setup_malloc(f, sizeof(r->classdata[j][0]) * classwords); + if (r->classdata[j] == NULL) return error(f, VORBIS_outofmem); + for (k=classwords-1; k >= 0; --k) { + r->classdata[j][k] = temp % r->classifications; + temp /= r->classifications; + } + } + } + + f->mapping_count = get_bits(f,6)+1; + f->mapping = (Mapping *) setup_malloc(f, f->mapping_count * sizeof(*f->mapping)); + if (f->mapping == NULL) return error(f, VORBIS_outofmem); + memset(f->mapping, 0, f->mapping_count * sizeof(*f->mapping)); + for (i=0; i < f->mapping_count; ++i) { + Mapping *m = f->mapping + i; + int mapping_type = get_bits(f,16); + if (mapping_type != 0) return error(f, VORBIS_invalid_setup); + m->chan = (MappingChannel *) setup_malloc(f, f->channels * sizeof(*m->chan)); + if (m->chan == NULL) return error(f, VORBIS_outofmem); + if (get_bits(f,1)) + m->submaps = get_bits(f,4)+1; + else + m->submaps = 1; + if (m->submaps > max_submaps) + max_submaps = m->submaps; + if (get_bits(f,1)) { + m->coupling_steps = get_bits(f,8)+1; + if (m->coupling_steps > f->channels) return error(f, VORBIS_invalid_setup); + for (k=0; k < m->coupling_steps; ++k) { + m->chan[k].magnitude = get_bits(f, ilog(f->channels-1)); + m->chan[k].angle = get_bits(f, ilog(f->channels-1)); + if (m->chan[k].magnitude >= f->channels) return error(f, VORBIS_invalid_setup); + if (m->chan[k].angle >= f->channels) return error(f, VORBIS_invalid_setup); + if (m->chan[k].magnitude == m->chan[k].angle) return error(f, VORBIS_invalid_setup); + } + } else + m->coupling_steps = 0; + + // reserved field + if (get_bits(f,2)) return error(f, VORBIS_invalid_setup); + if (m->submaps > 1) { + for (j=0; j < f->channels; ++j) { + m->chan[j].mux = get_bits(f, 4); + if (m->chan[j].mux >= m->submaps) return error(f, VORBIS_invalid_setup); + } + } else + // @SPECIFICATION: this case is missing from the spec + for (j=0; j < f->channels; ++j) + m->chan[j].mux = 0; + + for (j=0; j < m->submaps; ++j) { + get_bits(f,8); // discard + m->submap_floor[j] = get_bits(f,8); + m->submap_residue[j] = get_bits(f,8); + if (m->submap_floor[j] >= f->floor_count) return error(f, VORBIS_invalid_setup); + if (m->submap_residue[j] >= f->residue_count) return error(f, VORBIS_invalid_setup); + } + } + + // Modes + f->mode_count = get_bits(f, 6)+1; + for (i=0; i < f->mode_count; ++i) { + Mode *m = f->mode_config+i; + m->blockflag = get_bits(f,1); + m->windowtype = get_bits(f,16); + m->transformtype = get_bits(f,16); + m->mapping = get_bits(f,8); + if (m->windowtype != 0) return error(f, VORBIS_invalid_setup); + if (m->transformtype != 0) return error(f, VORBIS_invalid_setup); + if (m->mapping >= f->mapping_count) return error(f, VORBIS_invalid_setup); + } + + flush_packet(f); + + f->previous_length = 0; + + for (i=0; i < f->channels; ++i) { + f->channel_buffers[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1); + f->previous_window[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2); + f->finalY[i] = (int16 *) setup_malloc(f, sizeof(int16) * longest_floorlist); + if (f->channel_buffers[i] == NULL || f->previous_window[i] == NULL || f->finalY[i] == NULL) return error(f, VORBIS_outofmem); + memset(f->channel_buffers[i], 0, sizeof(float) * f->blocksize_1); + #ifdef STB_VORBIS_NO_DEFER_FLOOR + f->floor_buffers[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2); + if (f->floor_buffers[i] == NULL) return error(f, VORBIS_outofmem); + #endif + } + + if (!init_blocksize(f, 0, f->blocksize_0)) return FALSE; + if (!init_blocksize(f, 1, f->blocksize_1)) return FALSE; + f->blocksize[0] = f->blocksize_0; + f->blocksize[1] = f->blocksize_1; + +#ifdef STB_VORBIS_DIVIDE_TABLE + if (integer_divide_table[1][1]==0) + for (i=0; i < DIVTAB_NUMER; ++i) + for (j=1; j < DIVTAB_DENOM; ++j) + integer_divide_table[i][j] = i / j; +#endif + + // compute how much temporary memory is needed + + // 1. + { + uint32 imdct_mem = (f->blocksize_1 * sizeof(float) >> 1); + uint32 classify_mem; + int i,max_part_read=0; + for (i=0; i < f->residue_count; ++i) { + Residue *r = f->residue_config + i; + unsigned int actual_size = f->blocksize_1 / 2; + unsigned int limit_r_begin = r->begin < actual_size ? r->begin : actual_size; + unsigned int limit_r_end = r->end < actual_size ? r->end : actual_size; + int n_read = limit_r_end - limit_r_begin; + int part_read = n_read / r->part_size; + if (part_read > max_part_read) + max_part_read = part_read; + } + #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE + classify_mem = f->channels * (sizeof(void*) + max_part_read * sizeof(uint8 *)); + #else + classify_mem = f->channels * (sizeof(void*) + max_part_read * sizeof(int *)); + #endif + + // maximum reasonable partition size is f->blocksize_1 + + f->temp_memory_required = classify_mem; + if (imdct_mem > f->temp_memory_required) + f->temp_memory_required = imdct_mem; + } + + + if (f->alloc.alloc_buffer) { + assert(f->temp_offset == f->alloc.alloc_buffer_length_in_bytes); + // check if there's enough temp memory so we don't error later + if (f->setup_offset + sizeof(*f) + f->temp_memory_required > (unsigned) f->temp_offset) + return error(f, VORBIS_outofmem); + } + + // @TODO: stb_vorbis_seek_start expects first_audio_page_offset to point to a page + // without PAGEFLAG_continued_packet, so this either points to the first page, or + // the page after the end of the headers. It might be cleaner to point to a page + // in the middle of the headers, when that's the page where the first audio packet + // starts, but we'd have to also correctly skip the end of any continued packet in + // stb_vorbis_seek_start. + if (f->next_seg == -1) { + f->first_audio_page_offset = stb_vorbis_get_file_offset(f); + } else { + f->first_audio_page_offset = 0; + } + + return TRUE; +} + +static void vorbis_deinit(stb_vorbis *p) +{ + int i,j; + + setup_free(p, p->vendor); + for (i=0; i < p->comment_list_length; ++i) { + setup_free(p, p->comment_list[i]); + } + setup_free(p, p->comment_list); + + if (p->residue_config) { + for (i=0; i < p->residue_count; ++i) { + Residue *r = p->residue_config+i; + if (r->classdata) { + for (j=0; j < p->codebooks[r->classbook].entries; ++j) + setup_free(p, r->classdata[j]); + setup_free(p, r->classdata); + } + setup_free(p, r->residue_books); + } + } + + if (p->codebooks) { + CHECK(p); + for (i=0; i < p->codebook_count; ++i) { + Codebook *c = p->codebooks + i; + setup_free(p, c->codeword_lengths); + setup_free(p, c->multiplicands); + setup_free(p, c->codewords); + setup_free(p, c->sorted_codewords); + // c->sorted_values[-1] is the first entry in the array + setup_free(p, c->sorted_values ? c->sorted_values-1 : NULL); + } + setup_free(p, p->codebooks); + } + setup_free(p, p->floor_config); + setup_free(p, p->residue_config); + if (p->mapping) { + for (i=0; i < p->mapping_count; ++i) + setup_free(p, p->mapping[i].chan); + setup_free(p, p->mapping); + } + CHECK(p); + for (i=0; i < p->channels && i < STB_VORBIS_MAX_CHANNELS; ++i) { + setup_free(p, p->channel_buffers[i]); + setup_free(p, p->previous_window[i]); + #ifdef STB_VORBIS_NO_DEFER_FLOOR + setup_free(p, p->floor_buffers[i]); + #endif + setup_free(p, p->finalY[i]); + } + for (i=0; i < 2; ++i) { + setup_free(p, p->A[i]); + setup_free(p, p->B[i]); + setup_free(p, p->C[i]); + setup_free(p, p->window[i]); + setup_free(p, p->bit_reverse[i]); + } + #ifndef STB_VORBIS_NO_STDIO + if (p->close_on_free) fclose(p->f); + #endif +} + +void stb_vorbis_close(stb_vorbis *p) +{ + if (p == NULL) return; + vorbis_deinit(p); + setup_free(p,p); +} + +static void vorbis_init(stb_vorbis *p, const stb_vorbis_alloc *z) +{ + memset(p, 0, sizeof(*p)); // NULL out all malloc'd pointers to start + if (z) { + p->alloc = *z; + p->alloc.alloc_buffer_length_in_bytes &= ~7; + p->temp_offset = p->alloc.alloc_buffer_length_in_bytes; + } + p->eof = 0; + p->error = VORBIS__no_error; + p->stream = NULL; + p->codebooks = NULL; + p->page_crc_tests = -1; + #ifndef STB_VORBIS_NO_STDIO + p->close_on_free = FALSE; + p->f = NULL; + #endif +} + +int stb_vorbis_get_sample_offset(stb_vorbis *f) +{ + if (f->current_loc_valid) + return f->current_loc; + else + return -1; +} + +stb_vorbis_info stb_vorbis_get_info(stb_vorbis *f) +{ + stb_vorbis_info d; + d.channels = f->channels; + d.sample_rate = f->sample_rate; + d.setup_memory_required = f->setup_memory_required; + d.setup_temp_memory_required = f->setup_temp_memory_required; + d.temp_memory_required = f->temp_memory_required; + d.max_frame_size = f->blocksize_1 >> 1; + return d; +} + +stb_vorbis_comment stb_vorbis_get_comment(stb_vorbis *f) +{ + stb_vorbis_comment d; + d.vendor = f->vendor; + d.comment_list_length = f->comment_list_length; + d.comment_list = f->comment_list; + return d; +} + +int stb_vorbis_get_error(stb_vorbis *f) +{ + int e = f->error; + f->error = VORBIS__no_error; + return e; +} + +static stb_vorbis * vorbis_alloc(stb_vorbis *f) +{ + stb_vorbis *p = (stb_vorbis *) setup_malloc(f, sizeof(*p)); + return p; +} + +#ifndef STB_VORBIS_NO_PUSHDATA_API + +void stb_vorbis_flush_pushdata(stb_vorbis *f) +{ + f->previous_length = 0; + f->page_crc_tests = 0; + f->discard_samples_deferred = 0; + f->current_loc_valid = FALSE; + f->first_decode = FALSE; + f->samples_output = 0; + f->channel_buffer_start = 0; + f->channel_buffer_end = 0; +} + +static int vorbis_search_for_page_pushdata(vorb *f, uint8 *data, int data_len) +{ + int i,n; + for (i=0; i < f->page_crc_tests; ++i) + f->scan[i].bytes_done = 0; + + // if we have room for more scans, search for them first, because + // they may cause us to stop early if their header is incomplete + if (f->page_crc_tests < STB_VORBIS_PUSHDATA_CRC_COUNT) { + if (data_len < 4) return 0; + data_len -= 3; // need to look for 4-byte sequence, so don't miss + // one that straddles a boundary + for (i=0; i < data_len; ++i) { + if (data[i] == 0x4f) { + if (0==memcmp(data+i, ogg_page_header, 4)) { + int j,len; + uint32 crc; + // make sure we have the whole page header + if (i+26 >= data_len || i+27+data[i+26] >= data_len) { + // only read up to this page start, so hopefully we'll + // have the whole page header start next time + data_len = i; + break; + } + // ok, we have it all; compute the length of the page + len = 27 + data[i+26]; + for (j=0; j < data[i+26]; ++j) + len += data[i+27+j]; + // scan everything up to the embedded crc (which we must 0) + crc = 0; + for (j=0; j < 22; ++j) + crc = crc32_update(crc, data[i+j]); + // now process 4 0-bytes + for ( ; j < 26; ++j) + crc = crc32_update(crc, 0); + // len is the total number of bytes we need to scan + n = f->page_crc_tests++; + f->scan[n].bytes_left = len-j; + f->scan[n].crc_so_far = crc; + f->scan[n].goal_crc = data[i+22] + (data[i+23] << 8) + (data[i+24]<<16) + (data[i+25]<<24); + // if the last frame on a page is continued to the next, then + // we can't recover the sample_loc immediately + if (data[i+27+data[i+26]-1] == 255) + f->scan[n].sample_loc = ~0; + else + f->scan[n].sample_loc = data[i+6] + (data[i+7] << 8) + (data[i+ 8]<<16) + (data[i+ 9]<<24); + f->scan[n].bytes_done = i+j; + if (f->page_crc_tests == STB_VORBIS_PUSHDATA_CRC_COUNT) + break; + // keep going if we still have room for more + } + } + } + } + + for (i=0; i < f->page_crc_tests;) { + uint32 crc; + int j; + int n = f->scan[i].bytes_done; + int m = f->scan[i].bytes_left; + if (m > data_len - n) m = data_len - n; + // m is the bytes to scan in the current chunk + crc = f->scan[i].crc_so_far; + for (j=0; j < m; ++j) + crc = crc32_update(crc, data[n+j]); + f->scan[i].bytes_left -= m; + f->scan[i].crc_so_far = crc; + if (f->scan[i].bytes_left == 0) { + // does it match? + if (f->scan[i].crc_so_far == f->scan[i].goal_crc) { + // Houston, we have page + data_len = n+m; // consumption amount is wherever that scan ended + f->page_crc_tests = -1; // drop out of page scan mode + f->previous_length = 0; // decode-but-don't-output one frame + f->next_seg = -1; // start a new page + f->current_loc = f->scan[i].sample_loc; // set the current sample location + // to the amount we'd have decoded had we decoded this page + f->current_loc_valid = f->current_loc != ~0U; + return data_len; + } + // delete entry + f->scan[i] = f->scan[--f->page_crc_tests]; + } else { + ++i; + } + } + + return data_len; +} + +// return value: number of bytes we used +int stb_vorbis_decode_frame_pushdata( + stb_vorbis *f, // the file we're decoding + const uint8 *data, int data_len, // the memory available for decoding + int *channels, // place to write number of float * buffers + float ***output, // place to write float ** array of float * buffers + int *samples // place to write number of output samples + ) +{ + int i; + int len,right,left; + + if (!IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing); + + if (f->page_crc_tests >= 0) { + *samples = 0; + return vorbis_search_for_page_pushdata(f, (uint8 *) data, data_len); + } + + f->stream = (uint8 *) data; + f->stream_end = (uint8 *) data + data_len; + f->error = VORBIS__no_error; + + // check that we have the entire packet in memory + if (!is_whole_packet_present(f)) { + *samples = 0; + return 0; + } + + if (!vorbis_decode_packet(f, &len, &left, &right)) { + // save the actual error we encountered + enum STBVorbisError error = f->error; + if (error == VORBIS_bad_packet_type) { + // flush and resynch + f->error = VORBIS__no_error; + while (get8_packet(f) != EOP) + if (f->eof) break; + *samples = 0; + return (int) (f->stream - data); + } + if (error == VORBIS_continued_packet_flag_invalid) { + if (f->previous_length == 0) { + // we may be resynching, in which case it's ok to hit one + // of these; just discard the packet + f->error = VORBIS__no_error; + while (get8_packet(f) != EOP) + if (f->eof) break; + *samples = 0; + return (int) (f->stream - data); + } + } + // if we get an error while parsing, what to do? + // well, it DEFINITELY won't work to continue from where we are! + stb_vorbis_flush_pushdata(f); + // restore the error that actually made us bail + f->error = error; + *samples = 0; + return 1; + } + + // success! + len = vorbis_finish_frame(f, len, left, right); + for (i=0; i < f->channels; ++i) + f->outputs[i] = f->channel_buffers[i] + left; + + if (channels) *channels = f->channels; + *samples = len; + *output = f->outputs; + return (int) (f->stream - data); +} + +stb_vorbis *stb_vorbis_open_pushdata( + const unsigned char *data, int data_len, // the memory available for decoding + int *data_used, // only defined if result is not NULL + int *error, const stb_vorbis_alloc *alloc) +{ + stb_vorbis *f, p; + vorbis_init(&p, alloc); + p.stream = (uint8 *) data; + p.stream_end = (uint8 *) data + data_len; + p.push_mode = TRUE; + if (!start_decoder(&p)) { + if (p.eof) + *error = VORBIS_need_more_data; + else + *error = p.error; + vorbis_deinit(&p); + return NULL; + } + f = vorbis_alloc(&p); + if (f) { + *f = p; + *data_used = (int) (f->stream - data); + *error = 0; + return f; + } else { + vorbis_deinit(&p); + return NULL; + } +} +#endif // STB_VORBIS_NO_PUSHDATA_API + +unsigned int stb_vorbis_get_file_offset(stb_vorbis *f) +{ + #ifndef STB_VORBIS_NO_PUSHDATA_API + if (f->push_mode) return 0; + #endif + if (USE_MEMORY(f)) return (unsigned int) (f->stream - f->stream_start); + #ifndef STB_VORBIS_NO_STDIO + return (unsigned int) (ftell(f->f) - f->f_start); + #endif +} + +#ifndef STB_VORBIS_NO_PULLDATA_API +// +// DATA-PULLING API +// + +static uint32 vorbis_find_page(stb_vorbis *f, uint32 *end, uint32 *last) +{ + for(;;) { + int n; + if (f->eof) return 0; + n = get8(f); + if (n == 0x4f) { // page header candidate + unsigned int retry_loc = stb_vorbis_get_file_offset(f); + int i; + // check if we're off the end of a file_section stream + if (retry_loc - 25 > f->stream_len) + return 0; + // check the rest of the header + for (i=1; i < 4; ++i) + if (get8(f) != ogg_page_header[i]) + break; + if (f->eof) return 0; + if (i == 4) { + uint8 header[27]; + uint32 i, crc, goal, len; + for (i=0; i < 4; ++i) + header[i] = ogg_page_header[i]; + for (; i < 27; ++i) + header[i] = get8(f); + if (f->eof) return 0; + if (header[4] != 0) goto invalid; + goal = header[22] + (header[23] << 8) + (header[24]<<16) + ((uint32)header[25]<<24); + for (i=22; i < 26; ++i) + header[i] = 0; + crc = 0; + for (i=0; i < 27; ++i) + crc = crc32_update(crc, header[i]); + len = 0; + for (i=0; i < header[26]; ++i) { + int s = get8(f); + crc = crc32_update(crc, s); + len += s; + } + if (len && f->eof) return 0; + for (i=0; i < len; ++i) + crc = crc32_update(crc, get8(f)); + // finished parsing probable page + if (crc == goal) { + // we could now check that it's either got the last + // page flag set, OR it's followed by the capture + // pattern, but I guess TECHNICALLY you could have + // a file with garbage between each ogg page and recover + // from it automatically? So even though that paranoia + // might decrease the chance of an invalid decode by + // another 2^32, not worth it since it would hose those + // invalid-but-useful files? + if (end) + *end = stb_vorbis_get_file_offset(f); + if (last) { + if (header[5] & 0x04) + *last = 1; + else + *last = 0; + } + set_file_offset(f, retry_loc-1); + return 1; + } + } + invalid: + // not a valid page, so rewind and look for next one + set_file_offset(f, retry_loc); + } + } +} + + +#define SAMPLE_unknown 0xffffffff + +// seeking is implemented with a binary search, which narrows down the range to +// 64K, before using a linear search (because finding the synchronization +// pattern can be expensive, and the chance we'd find the end page again is +// relatively high for small ranges) +// +// two initial interpolation-style probes are used at the start of the search +// to try to bound either side of the binary search sensibly, while still +// working in O(log n) time if they fail. + +static int get_seek_page_info(stb_vorbis *f, ProbedPage *z) +{ + uint8 header[27], lacing[255]; + int i,len; + + // record where the page starts + z->page_start = stb_vorbis_get_file_offset(f); + + // parse the header + getn(f, header, 27); + if (header[0] != 'O' || header[1] != 'g' || header[2] != 'g' || header[3] != 'S') + return 0; + getn(f, lacing, header[26]); + + // determine the length of the payload + len = 0; + for (i=0; i < header[26]; ++i) + len += lacing[i]; + + // this implies where the page ends + z->page_end = z->page_start + 27 + header[26] + len; + + // read the last-decoded sample out of the data + z->last_decoded_sample = header[6] + (header[7] << 8) + (header[8] << 16) + (header[9] << 24); + + // restore file state to where we were + set_file_offset(f, z->page_start); + return 1; +} + +// rarely used function to seek back to the preceding page while finding the +// start of a packet +static int go_to_page_before(stb_vorbis *f, unsigned int limit_offset) +{ + unsigned int previous_safe, end; + + // now we want to seek back 64K from the limit + if (limit_offset >= 65536 && limit_offset-65536 >= f->first_audio_page_offset) + previous_safe = limit_offset - 65536; + else + previous_safe = f->first_audio_page_offset; + + set_file_offset(f, previous_safe); + + while (vorbis_find_page(f, &end, NULL)) { + if (end >= limit_offset && stb_vorbis_get_file_offset(f) < limit_offset) + return 1; + set_file_offset(f, end); + } + + return 0; +} + +// implements the search logic for finding a page and starting decoding. if +// the function succeeds, current_loc_valid will be true and current_loc will +// be less than or equal to the provided sample number (the closer the +// better). +static int seek_to_sample_coarse(stb_vorbis *f, uint32 sample_number) +{ + ProbedPage left, right, mid; + int i, start_seg_with_known_loc, end_pos, page_start; + uint32 delta, stream_length, padding, last_sample_limit; + double offset = 0.0, bytes_per_sample = 0.0; + int probe = 0; + + // find the last page and validate the target sample + stream_length = stb_vorbis_stream_length_in_samples(f); + if (stream_length == 0) return error(f, VORBIS_seek_without_length); + if (sample_number > stream_length) return error(f, VORBIS_seek_invalid); + + // this is the maximum difference between the window-center (which is the + // actual granule position value), and the right-start (which the spec + // indicates should be the granule position (give or take one)). + padding = ((f->blocksize_1 - f->blocksize_0) >> 2); + if (sample_number < padding) + last_sample_limit = 0; + else + last_sample_limit = sample_number - padding; + + left = f->p_first; + while (left.last_decoded_sample == ~0U) { + // (untested) the first page does not have a 'last_decoded_sample' + set_file_offset(f, left.page_end); + if (!get_seek_page_info(f, &left)) goto error; + } + + right = f->p_last; + assert(right.last_decoded_sample != ~0U); + + // starting from the start is handled differently + if (last_sample_limit <= left.last_decoded_sample) { + if (stb_vorbis_seek_start(f)) { + if (f->current_loc > sample_number) + return error(f, VORBIS_seek_failed); + return 1; + } + return 0; + } + + while (left.page_end != right.page_start) { + assert(left.page_end < right.page_start); + // search range in bytes + delta = right.page_start - left.page_end; + if (delta <= 65536) { + // there's only 64K left to search - handle it linearly + set_file_offset(f, left.page_end); + } else { + if (probe < 2) { + if (probe == 0) { + // first probe (interpolate) + double data_bytes = right.page_end - left.page_start; + bytes_per_sample = data_bytes / right.last_decoded_sample; + offset = left.page_start + bytes_per_sample * (last_sample_limit - left.last_decoded_sample); + } else { + // second probe (try to bound the other side) + double error = ((double) last_sample_limit - mid.last_decoded_sample) * bytes_per_sample; + if (error >= 0 && error < 8000) error = 8000; + if (error < 0 && error > -8000) error = -8000; + offset += error * 2; + } + + // ensure the offset is valid + if (offset < left.page_end) + offset = left.page_end; + if (offset > right.page_start - 65536) + offset = right.page_start - 65536; + + set_file_offset(f, (unsigned int) offset); + } else { + // binary search for large ranges (offset by 32K to ensure + // we don't hit the right page) + set_file_offset(f, left.page_end + (delta / 2) - 32768); + } + + if (!vorbis_find_page(f, NULL, NULL)) goto error; + } + + for (;;) { + if (!get_seek_page_info(f, &mid)) goto error; + if (mid.last_decoded_sample != ~0U) break; + // (untested) no frames end on this page + set_file_offset(f, mid.page_end); + assert(mid.page_start < right.page_start); + } + + // if we've just found the last page again then we're in a tricky file, + // and we're close enough (if it wasn't an interpolation probe). + if (mid.page_start == right.page_start) { + if (probe >= 2 || delta <= 65536) + break; + } else { + if (last_sample_limit < mid.last_decoded_sample) + right = mid; + else + left = mid; + } + + ++probe; + } + + // seek back to start of the last packet + page_start = left.page_start; + set_file_offset(f, page_start); + if (!start_page(f)) return error(f, VORBIS_seek_failed); + end_pos = f->end_seg_with_known_loc; + assert(end_pos >= 0); + + for (;;) { + for (i = end_pos; i > 0; --i) + if (f->segments[i-1] != 255) + break; + + start_seg_with_known_loc = i; + + if (start_seg_with_known_loc > 0 || !(f->page_flag & PAGEFLAG_continued_packet)) + break; + + // (untested) the final packet begins on an earlier page + if (!go_to_page_before(f, page_start)) + goto error; + + page_start = stb_vorbis_get_file_offset(f); + if (!start_page(f)) goto error; + end_pos = f->segment_count - 1; + } + + // prepare to start decoding + f->current_loc_valid = FALSE; + f->last_seg = FALSE; + f->valid_bits = 0; + f->packet_bytes = 0; + f->bytes_in_seg = 0; + f->previous_length = 0; + f->next_seg = start_seg_with_known_loc; + + for (i = 0; i < start_seg_with_known_loc; i++) + skip(f, f->segments[i]); + + // start decoding (optimizable - this frame is generally discarded) + if (!vorbis_pump_first_frame(f)) + return 0; + if (f->current_loc > sample_number) + return error(f, VORBIS_seek_failed); + return 1; + +error: + // try to restore the file to a valid state + stb_vorbis_seek_start(f); + return error(f, VORBIS_seek_failed); +} + +// the same as vorbis_decode_initial, but without advancing +static int peek_decode_initial(vorb *f, int *p_left_start, int *p_left_end, int *p_right_start, int *p_right_end, int *mode) +{ + int bits_read, bytes_read; + + if (!vorbis_decode_initial(f, p_left_start, p_left_end, p_right_start, p_right_end, mode)) + return 0; + + // either 1 or 2 bytes were read, figure out which so we can rewind + bits_read = 1 + ilog(f->mode_count-1); + if (f->mode_config[*mode].blockflag) + bits_read += 2; + bytes_read = (bits_read + 7) / 8; + + f->bytes_in_seg += bytes_read; + f->packet_bytes -= bytes_read; + skip(f, -bytes_read); + if (f->next_seg == -1) + f->next_seg = f->segment_count - 1; + else + f->next_seg--; + f->valid_bits = 0; + + return 1; +} + +int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number) +{ + uint32 max_frame_samples; + + if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing); + + // fast page-level search + if (!seek_to_sample_coarse(f, sample_number)) + return 0; + + assert(f->current_loc_valid); + assert(f->current_loc <= sample_number); + + // linear search for the relevant packet + max_frame_samples = (f->blocksize_1*3 - f->blocksize_0) >> 2; + while (f->current_loc < sample_number) { + int left_start, left_end, right_start, right_end, mode, frame_samples; + if (!peek_decode_initial(f, &left_start, &left_end, &right_start, &right_end, &mode)) + return error(f, VORBIS_seek_failed); + // calculate the number of samples returned by the next frame + frame_samples = right_start - left_start; + if (f->current_loc + frame_samples > sample_number) { + return 1; // the next frame will contain the sample + } else if (f->current_loc + frame_samples + max_frame_samples > sample_number) { + // there's a chance the frame after this could contain the sample + vorbis_pump_first_frame(f); + } else { + // this frame is too early to be relevant + f->current_loc += frame_samples; + f->previous_length = 0; + maybe_start_packet(f); + flush_packet(f); + } + } + // the next frame should start with the sample + if (f->current_loc != sample_number) return error(f, VORBIS_seek_failed); + return 1; +} + +int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number) +{ + if (!stb_vorbis_seek_frame(f, sample_number)) + return 0; + + if (sample_number != f->current_loc) { + int n; + uint32 frame_start = f->current_loc; + stb_vorbis_get_frame_float(f, &n, NULL); + assert(sample_number > frame_start); + assert(f->channel_buffer_start + (int) (sample_number-frame_start) <= f->channel_buffer_end); + f->channel_buffer_start += (sample_number - frame_start); + } + + return 1; +} + +int stb_vorbis_seek_start(stb_vorbis *f) +{ + if (IS_PUSH_MODE(f)) { return error(f, VORBIS_invalid_api_mixing); } + set_file_offset(f, f->first_audio_page_offset); + f->previous_length = 0; + f->first_decode = TRUE; + f->next_seg = -1; + return vorbis_pump_first_frame(f); +} + +unsigned int stb_vorbis_stream_length_in_samples(stb_vorbis *f) +{ + unsigned int restore_offset, previous_safe; + unsigned int end, last_page_loc; + + if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing); + if (!f->total_samples) { + unsigned int last; + uint32 lo,hi; + char header[6]; + + // first, store the current decode position so we can restore it + restore_offset = stb_vorbis_get_file_offset(f); + + // now we want to seek back 64K from the end (the last page must + // be at most a little less than 64K, but let's allow a little slop) + if (f->stream_len >= 65536 && f->stream_len-65536 >= f->first_audio_page_offset) + previous_safe = f->stream_len - 65536; + else + previous_safe = f->first_audio_page_offset; + + set_file_offset(f, previous_safe); + // previous_safe is now our candidate 'earliest known place that seeking + // to will lead to the final page' + + if (!vorbis_find_page(f, &end, &last)) { + // if we can't find a page, we're hosed! + f->error = VORBIS_cant_find_last_page; + f->total_samples = 0xffffffff; + goto done; + } + + // check if there are more pages + last_page_loc = stb_vorbis_get_file_offset(f); + + // stop when the last_page flag is set, not when we reach eof; + // this allows us to stop short of a 'file_section' end without + // explicitly checking the length of the section + while (!last) { + set_file_offset(f, end); + if (!vorbis_find_page(f, &end, &last)) { + // the last page we found didn't have the 'last page' flag + // set. whoops! + break; + } + //previous_safe = last_page_loc+1; // NOTE: not used after this point, but note for debugging + last_page_loc = stb_vorbis_get_file_offset(f); + } + + set_file_offset(f, last_page_loc); + + // parse the header + getn(f, (unsigned char *)header, 6); + // extract the absolute granule position + lo = get32(f); + hi = get32(f); + if (lo == 0xffffffff && hi == 0xffffffff) { + f->error = VORBIS_cant_find_last_page; + f->total_samples = SAMPLE_unknown; + goto done; + } + if (hi) + lo = 0xfffffffe; // saturate + f->total_samples = lo; + + f->p_last.page_start = last_page_loc; + f->p_last.page_end = end; + f->p_last.last_decoded_sample = lo; + + done: + set_file_offset(f, restore_offset); + } + return f->total_samples == SAMPLE_unknown ? 0 : f->total_samples; +} + +float stb_vorbis_stream_length_in_seconds(stb_vorbis *f) +{ + return stb_vorbis_stream_length_in_samples(f) / (float) f->sample_rate; +} + + + +int stb_vorbis_get_frame_float(stb_vorbis *f, int *channels, float ***output) +{ + int len, right,left,i; + if (IS_PUSH_MODE(f)) return error(f, VORBIS_invalid_api_mixing); + + if (!vorbis_decode_packet(f, &len, &left, &right)) { + f->channel_buffer_start = f->channel_buffer_end = 0; + return 0; + } + + len = vorbis_finish_frame(f, len, left, right); + for (i=0; i < f->channels; ++i) + f->outputs[i] = f->channel_buffers[i] + left; + + f->channel_buffer_start = left; + f->channel_buffer_end = left+len; + + if (channels) *channels = f->channels; + if (output) *output = f->outputs; + return len; +} + +#ifndef STB_VORBIS_NO_STDIO + +stb_vorbis * stb_vorbis_open_file_section(FILE *file, int close_on_free, int *error, const stb_vorbis_alloc *alloc, unsigned int length) +{ + stb_vorbis *f, p; + vorbis_init(&p, alloc); + p.f = file; + p.f_start = (uint32) ftell(file); + p.stream_len = length; + p.close_on_free = close_on_free; + if (start_decoder(&p)) { + f = vorbis_alloc(&p); + if (f) { + *f = p; + vorbis_pump_first_frame(f); + return f; + } + } + if (error) *error = p.error; + vorbis_deinit(&p); + return NULL; +} + +stb_vorbis * stb_vorbis_open_file(FILE *file, int close_on_free, int *error, const stb_vorbis_alloc *alloc) +{ + unsigned int len, start; + start = (unsigned int) ftell(file); + fseek(file, 0, SEEK_END); + len = (unsigned int) (ftell(file) - start); + fseek(file, start, SEEK_SET); + return stb_vorbis_open_file_section(file, close_on_free, error, alloc, len); +} + +stb_vorbis * stb_vorbis_open_filename(const char *filename, int *error, const stb_vorbis_alloc *alloc) +{ + FILE *f; +#if defined(_WIN32) && defined(__STDC_WANT_SECURE_LIB__) + if (0 != fopen_s(&f, filename, "rb")) + f = NULL; +#else + f = fopen(filename, "rb"); +#endif + if (f) + return stb_vorbis_open_file(f, TRUE, error, alloc); + if (error) *error = VORBIS_file_open_failure; + return NULL; +} +#endif // STB_VORBIS_NO_STDIO + +stb_vorbis * stb_vorbis_open_memory(const unsigned char *data, int len, int *error, const stb_vorbis_alloc *alloc) +{ + stb_vorbis *f, p; + if (!data) { + if (error) *error = VORBIS_unexpected_eof; + return NULL; + } + vorbis_init(&p, alloc); + p.stream = (uint8 *) data; + p.stream_end = (uint8 *) data + len; + p.stream_start = (uint8 *) p.stream; + p.stream_len = len; + p.push_mode = FALSE; + if (start_decoder(&p)) { + f = vorbis_alloc(&p); + if (f) { + *f = p; + vorbis_pump_first_frame(f); + if (error) *error = VORBIS__no_error; + return f; + } + } + if (error) *error = p.error; + vorbis_deinit(&p); + return NULL; +} + +#ifndef STB_VORBIS_NO_INTEGER_CONVERSION +#define PLAYBACK_MONO 1 +#define PLAYBACK_LEFT 2 +#define PLAYBACK_RIGHT 4 + +#define L (PLAYBACK_LEFT | PLAYBACK_MONO) +#define C (PLAYBACK_LEFT | PLAYBACK_RIGHT | PLAYBACK_MONO) +#define R (PLAYBACK_RIGHT | PLAYBACK_MONO) + +static int8 channel_position[7][6] = +{ + { 0 }, + { C }, + { L, R }, + { L, C, R }, + { L, R, L, R }, + { L, C, R, L, R }, + { L, C, R, L, R, C }, +}; + + +#ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT + typedef union { + float f; + int i; + } float_conv; + typedef char stb_vorbis_float_size_test[sizeof(float)==4 && sizeof(int) == 4]; + #define FASTDEF(x) float_conv x + // add (1<<23) to convert to int, then divide by 2^SHIFT, then add 0.5/2^SHIFT to round + #define MAGIC(SHIFT) (1.5f * (1 << (23-SHIFT)) + 0.5f/(1 << SHIFT)) + #define ADDEND(SHIFT) (((150-SHIFT) << 23) + (1 << 22)) + #define FAST_SCALED_FLOAT_TO_INT(temp,x,s) (temp.f = (x) + MAGIC(s), temp.i - ADDEND(s)) + #define check_endianness() +#else + #define FAST_SCALED_FLOAT_TO_INT(temp,x,s) ((int) ((x) * (1 << (s)))) + #define check_endianness() + #define FASTDEF(x) +#endif + +static void copy_samples(short *dest, float *src, int len) +{ + int i; + check_endianness(); + for (i=0; i < len; ++i) { + FASTDEF(temp); + int v = FAST_SCALED_FLOAT_TO_INT(temp, src[i],15); + if ((unsigned int) (v + 32768) > 65535) + v = v < 0 ? -32768 : 32767; + dest[i] = v; + } +} + +static void compute_samples(int mask, short *output, int num_c, float **data, int d_offset, int len) +{ + #define STB_BUFFER_SIZE 32 + float buffer[STB_BUFFER_SIZE]; + int i,j,o,n = STB_BUFFER_SIZE; + check_endianness(); + for (o = 0; o < len; o += STB_BUFFER_SIZE) { + memset(buffer, 0, sizeof(buffer)); + if (o + n > len) n = len - o; + for (j=0; j < num_c; ++j) { + if (channel_position[num_c][j] & mask) { + for (i=0; i < n; ++i) + buffer[i] += data[j][d_offset+o+i]; + } + } + for (i=0; i < n; ++i) { + FASTDEF(temp); + int v = FAST_SCALED_FLOAT_TO_INT(temp,buffer[i],15); + if ((unsigned int) (v + 32768) > 65535) + v = v < 0 ? -32768 : 32767; + output[o+i] = v; + } + } + #undef STB_BUFFER_SIZE +} + +static void compute_stereo_samples(short *output, int num_c, float **data, int d_offset, int len) +{ + #define STB_BUFFER_SIZE 32 + float buffer[STB_BUFFER_SIZE]; + int i,j,o,n = STB_BUFFER_SIZE >> 1; + // o is the offset in the source data + check_endianness(); + for (o = 0; o < len; o += STB_BUFFER_SIZE >> 1) { + // o2 is the offset in the output data + int o2 = o << 1; + memset(buffer, 0, sizeof(buffer)); + if (o + n > len) n = len - o; + for (j=0; j < num_c; ++j) { + int m = channel_position[num_c][j] & (PLAYBACK_LEFT | PLAYBACK_RIGHT); + if (m == (PLAYBACK_LEFT | PLAYBACK_RIGHT)) { + for (i=0; i < n; ++i) { + buffer[i*2+0] += data[j][d_offset+o+i]; + buffer[i*2+1] += data[j][d_offset+o+i]; + } + } else if (m == PLAYBACK_LEFT) { + for (i=0; i < n; ++i) { + buffer[i*2+0] += data[j][d_offset+o+i]; + } + } else if (m == PLAYBACK_RIGHT) { + for (i=0; i < n; ++i) { + buffer[i*2+1] += data[j][d_offset+o+i]; + } + } + } + for (i=0; i < (n<<1); ++i) { + FASTDEF(temp); + int v = FAST_SCALED_FLOAT_TO_INT(temp,buffer[i],15); + if ((unsigned int) (v + 32768) > 65535) + v = v < 0 ? -32768 : 32767; + output[o2+i] = v; + } + } + #undef STB_BUFFER_SIZE +} + +static void convert_samples_short(int buf_c, short **buffer, int b_offset, int data_c, float **data, int d_offset, int samples) +{ + int i; + if (buf_c != data_c && buf_c <= 2 && data_c <= 6) { + static int channel_selector[3][2] = { {0}, {PLAYBACK_MONO}, {PLAYBACK_LEFT, PLAYBACK_RIGHT} }; + for (i=0; i < buf_c; ++i) + compute_samples(channel_selector[buf_c][i], buffer[i]+b_offset, data_c, data, d_offset, samples); + } else { + int limit = buf_c < data_c ? buf_c : data_c; + for (i=0; i < limit; ++i) + copy_samples(buffer[i]+b_offset, data[i]+d_offset, samples); + for ( ; i < buf_c; ++i) + memset(buffer[i]+b_offset, 0, sizeof(short) * samples); + } +} + +int stb_vorbis_get_frame_short(stb_vorbis *f, int num_c, short **buffer, int num_samples) +{ + float **output = NULL; + int len = stb_vorbis_get_frame_float(f, NULL, &output); + if (len > num_samples) len = num_samples; + if (len) + convert_samples_short(num_c, buffer, 0, f->channels, output, 0, len); + return len; +} + +static void convert_channels_short_interleaved(int buf_c, short *buffer, int data_c, float **data, int d_offset, int len) +{ + int i; + check_endianness(); + if (buf_c != data_c && buf_c <= 2 && data_c <= 6) { + assert(buf_c == 2); + for (i=0; i < buf_c; ++i) + compute_stereo_samples(buffer, data_c, data, d_offset, len); + } else { + int limit = buf_c < data_c ? buf_c : data_c; + int j; + for (j=0; j < len; ++j) { + for (i=0; i < limit; ++i) { + FASTDEF(temp); + float f = data[i][d_offset+j]; + int v = FAST_SCALED_FLOAT_TO_INT(temp, f,15);//data[i][d_offset+j],15); + if ((unsigned int) (v + 32768) > 65535) + v = v < 0 ? -32768 : 32767; + *buffer++ = v; + } + for ( ; i < buf_c; ++i) + *buffer++ = 0; + } + } +} + +int stb_vorbis_get_frame_short_interleaved(stb_vorbis *f, int num_c, short *buffer, int num_shorts) +{ + float **output; + int len; + if (num_c == 1) return stb_vorbis_get_frame_short(f,num_c,&buffer, num_shorts); + len = stb_vorbis_get_frame_float(f, NULL, &output); + if (len) { + if (len*num_c > num_shorts) len = num_shorts / num_c; + convert_channels_short_interleaved(num_c, buffer, f->channels, output, 0, len); + } + return len; +} + +int stb_vorbis_get_samples_short_interleaved(stb_vorbis *f, int channels, short *buffer, int num_shorts) +{ + float **outputs; + int len = num_shorts / channels; + int n=0; + while (n < len) { + int k = f->channel_buffer_end - f->channel_buffer_start; + if (n+k >= len) k = len - n; + if (k) + convert_channels_short_interleaved(channels, buffer, f->channels, f->channel_buffers, f->channel_buffer_start, k); + buffer += k*channels; + n += k; + f->channel_buffer_start += k; + if (n == len) break; + if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) break; + } + return n; +} + +int stb_vorbis_get_samples_short(stb_vorbis *f, int channels, short **buffer, int len) +{ + float **outputs; + int n=0; + while (n < len) { + int k = f->channel_buffer_end - f->channel_buffer_start; + if (n+k >= len) k = len - n; + if (k) + convert_samples_short(channels, buffer, n, f->channels, f->channel_buffers, f->channel_buffer_start, k); + n += k; + f->channel_buffer_start += k; + if (n == len) break; + if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) break; + } + return n; +} + +#ifndef STB_VORBIS_NO_STDIO +int stb_vorbis_decode_filename(const char *filename, int *channels, int *sample_rate, short **output) +{ + int data_len, offset, total, limit, error; + short *data; + stb_vorbis *v = stb_vorbis_open_filename(filename, &error, NULL); + if (v == NULL) return -1; + limit = v->channels * 4096; + *channels = v->channels; + if (sample_rate) + *sample_rate = v->sample_rate; + offset = data_len = 0; + total = limit; + data = (short *) malloc(total * sizeof(*data)); + if (data == NULL) { + stb_vorbis_close(v); + return -2; + } + for (;;) { + int n = stb_vorbis_get_frame_short_interleaved(v, v->channels, data+offset, total-offset); + if (n == 0) break; + data_len += n; + offset += n * v->channels; + if (offset + limit > total) { + short *data2; + total *= 2; + data2 = (short *) realloc(data, total * sizeof(*data)); + if (data2 == NULL) { + free(data); + stb_vorbis_close(v); + return -2; + } + data = data2; + } + } + *output = data; + stb_vorbis_close(v); + return data_len; +} +#endif // NO_STDIO + +int stb_vorbis_decode_memory(const uint8 *mem, int len, int *channels, int *sample_rate, short **output) +{ + int data_len, offset, total, limit, error; + short *data; + stb_vorbis *v = stb_vorbis_open_memory(mem, len, &error, NULL); + if (v == NULL) return -1; + limit = v->channels * 4096; + *channels = v->channels; + if (sample_rate) + *sample_rate = v->sample_rate; + offset = data_len = 0; + total = limit; + data = (short *) malloc(total * sizeof(*data)); + if (data == NULL) { + stb_vorbis_close(v); + return -2; + } + for (;;) { + int n = stb_vorbis_get_frame_short_interleaved(v, v->channels, data+offset, total-offset); + if (n == 0) break; + data_len += n; + offset += n * v->channels; + if (offset + limit > total) { + short *data2; + total *= 2; + data2 = (short *) realloc(data, total * sizeof(*data)); + if (data2 == NULL) { + free(data); + stb_vorbis_close(v); + return -2; + } + data = data2; + } + } + *output = data; + stb_vorbis_close(v); + return data_len; +} +#endif // STB_VORBIS_NO_INTEGER_CONVERSION + +int stb_vorbis_get_samples_float_interleaved(stb_vorbis *f, int channels, float *buffer, int num_floats) +{ + float **outputs; + int len = num_floats / channels; + int n=0; + int z = f->channels; + if (z > channels) z = channels; + while (n < len) { + int i,j; + int k = f->channel_buffer_end - f->channel_buffer_start; + if (n+k >= len) k = len - n; + for (j=0; j < k; ++j) { + for (i=0; i < z; ++i) + *buffer++ = f->channel_buffers[i][f->channel_buffer_start+j]; + for ( ; i < channels; ++i) + *buffer++ = 0; + } + n += k; + f->channel_buffer_start += k; + if (n == len) + break; + if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) + break; + } + return n; +} + +int stb_vorbis_get_samples_float(stb_vorbis *f, int channels, float **buffer, int num_samples) +{ + float **outputs; + int n=0; + int z = f->channels; + if (z > channels) z = channels; + while (n < num_samples) { + int i; + int k = f->channel_buffer_end - f->channel_buffer_start; + if (n+k >= num_samples) k = num_samples - n; + if (k) { + for (i=0; i < z; ++i) + memcpy(buffer[i]+n, f->channel_buffers[i]+f->channel_buffer_start, sizeof(float)*k); + for ( ; i < channels; ++i) + memset(buffer[i]+n, 0, sizeof(float) * k); + } + n += k; + f->channel_buffer_start += k; + if (n == num_samples) + break; + if (!stb_vorbis_get_frame_float(f, NULL, &outputs)) + break; + } + return n; +} +#endif // STB_VORBIS_NO_PULLDATA_API + +/* Version history + 1.17 - 2019-07-08 - fix CVE-2019-13217, -13218, -13219, -13220, -13221, -13222, -13223 + found with Mayhem by ForAllSecure + 1.16 - 2019-03-04 - fix warnings + 1.15 - 2019-02-07 - explicit failure if Ogg Skeleton data is found + 1.14 - 2018-02-11 - delete bogus dealloca usage + 1.13 - 2018-01-29 - fix truncation of last frame (hopefully) + 1.12 - 2017-11-21 - limit residue begin/end to blocksize/2 to avoid large temp allocs in bad/corrupt files + 1.11 - 2017-07-23 - fix MinGW compilation + 1.10 - 2017-03-03 - more robust seeking; fix negative ilog(); clear error in open_memory + 1.09 - 2016-04-04 - back out 'avoid discarding last frame' fix from previous version + 1.08 - 2016-04-02 - fixed multiple warnings; fix setup memory leaks; + avoid discarding last frame of audio data + 1.07 - 2015-01-16 - fixed some warnings, fix mingw, const-correct API + some more crash fixes when out of memory or with corrupt files + 1.06 - 2015-08-31 - full, correct support for seeking API (Dougall Johnson) + some crash fixes when out of memory or with corrupt files + 1.05 - 2015-04-19 - don't define __forceinline if it's redundant + 1.04 - 2014-08-27 - fix missing const-correct case in API + 1.03 - 2014-08-07 - Warning fixes + 1.02 - 2014-07-09 - Declare qsort compare function _cdecl on windows + 1.01 - 2014-06-18 - fix stb_vorbis_get_samples_float + 1.0 - 2014-05-26 - fix memory leaks; fix warnings; fix bugs in multichannel + (API change) report sample rate for decode-full-file funcs + 0.99996 - bracket #include <malloc.h> for macintosh compilation by Laurent Gomila + 0.99995 - use union instead of pointer-cast for fast-float-to-int to avoid alias-optimization problem + 0.99994 - change fast-float-to-int to work in single-precision FPU mode, remove endian-dependence + 0.99993 - remove assert that fired on legal files with empty tables + 0.99992 - rewind-to-start + 0.99991 - bugfix to stb_vorbis_get_samples_short by Bernhard Wodo + 0.9999 - (should have been 0.99990) fix no-CRT support, compiling as C++ + 0.9998 - add a full-decode function with a memory source + 0.9997 - fix a bug in the read-from-FILE case in 0.9996 addition + 0.9996 - query length of vorbis stream in samples/seconds + 0.9995 - bugfix to another optimization that only happened in certain files + 0.9994 - bugfix to one of the optimizations that caused significant (but inaudible?) errors + 0.9993 - performance improvements; runs in 99% to 104% of time of reference implementation + 0.9992 - performance improvement of IMDCT; now performs close to reference implementation + 0.9991 - performance improvement of IMDCT + 0.999 - (should have been 0.9990) performance improvement of IMDCT + 0.998 - no-CRT support from Casey Muratori + 0.997 - bugfixes for bugs found by Terje Mathisen + 0.996 - bugfix: fast-huffman decode initialized incorrectly for sparse codebooks; fixing gives 10% speedup - found by Terje Mathisen + 0.995 - bugfix: fix to 'effective' overrun detection - found by Terje Mathisen + 0.994 - bugfix: garbage decode on final VQ symbol of a non-multiple - found by Terje Mathisen + 0.993 - bugfix: pushdata API required 1 extra byte for empty page (failed to consume final page if empty) - found by Terje Mathisen + 0.992 - fixes for MinGW warning + 0.991 - turn fast-float-conversion on by default + 0.990 - fix push-mode seek recovery if you seek into the headers + 0.98b - fix to bad release of 0.98 + 0.98 - fix push-mode seek recovery; robustify float-to-int and support non-fast mode + 0.97 - builds under c++ (typecasting, don't use 'class' keyword) + 0.96 - somehow MY 0.95 was right, but the web one was wrong, so here's my 0.95 rereleased as 0.96, fixes a typo in the clamping code + 0.95 - clamping code for 16-bit functions + 0.94 - not publically released + 0.93 - fixed all-zero-floor case (was decoding garbage) + 0.92 - fixed a memory leak + 0.91 - conditional compiles to omit parts of the API and the infrastructure to support them: STB_VORBIS_NO_PULLDATA_API, STB_VORBIS_NO_PUSHDATA_API, STB_VORBIS_NO_STDIO, STB_VORBIS_NO_INTEGER_CONVERSION + 0.90 - first public release +*/ + +#endif // STB_VORBIS_HEADER_ONLY + + +/* +------------------------------------------------------------------------------ +This software is available under 2 licenses -- choose whichever you prefer. +------------------------------------------------------------------------------ +ALTERNATIVE A - MIT License +Copyright (c) 2017 Sean Barrett +Permission is hereby granted, free of charge, to any person obtaining a copy of +this software and associated documentation files (the "Software"), to deal in +the Software without restriction, including without limitation the rights to +use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies +of the Software, and to permit persons to whom the Software is furnished to do +so, subject to the following conditions: +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. +------------------------------------------------------------------------------ +ALTERNATIVE B - Public Domain (www.unlicense.org) +This is free and unencumbered software released into the public domain. +Anyone is free to copy, modify, publish, use, compile, sell, or distribute this +software, either in source code form or as a compiled binary, for any purpose, +commercial or non-commercial, and by any means. +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. +------------------------------------------------------------------------------ +*/ diff --git a/vendor/text.c b/vendor/text.c new file mode 100644 index 0000000..9a81558 --- /dev/null +++ b/vendor/text.c @@ -0,0 +1,228 @@ +/* Native text: Pardes's FreeType/HarfBuzz quality target, via Pango's complete + script itemization, shaping, bidi and Fontconfig fallback instead of its + terminal-cell-specific shaping layer. Grayscale coverage avoids color fringes. */ +#define _GNU_SOURCE +#include "text.h" +#include <fontconfig/fontconfig.h> +#include <pango/pangofc-fontmap.h> +#include <sys/mman.h> +#include <unistd.h> +#include <errno.h> +#include <pango/pangocairo.h> +#include <math.h> +#include <stdint.h> +#include <stdlib.h> +#include <string.h> + +#define CACHE_ENTRIES 512 +#define CACHE_BYTES (16 * 1024 * 1024) +#define CACHE_PROPERTY "sketchbook.native-text" +static const unsigned char go_mono[] = { +#embed "fonts/Go-Mono.ttf" +}; + +typedef struct { + SDL_Texture *texture; + char *text; + size_t len, bytes; + uint64_t hash, used; + int size, width, height, left, top; + Uint32 color; +} Text; +typedef struct { + PangoFontMap *map; + PangoContext *context; + PangoLayout *layout; + PangoFontDescription *font; + FcConfig *config; + int font_fd; + Text entries[CACHE_ENTRIES]; + size_t bytes; + uint64_t frame; +} TextCache; + +static void release_text(Text *t) { + if (t->texture) SDL_DestroyTexture(t->texture); + free(t->text); + memset(t, 0, sizeof(*t)); +} +static void SDLCALL release_cache(void *userdata, void *value) { + (void)userdata; + TextCache *c = value; + /* SDL destroys its owned textures before renderer property cleanup. */ + for (size_t i = 0; i < CACHE_ENTRIES; ++i) free(c->entries[i].text); + if (c->font) pango_font_description_free(c->font); + if (c->layout) g_object_unref(c->layout); + if (c->context) g_object_unref(c->context); + if (c->map) g_object_unref(c->map); + if (c->config) FcConfigDestroy(c->config); + if (c->font_fd >= 0) close(c->font_fd); + free(c); +} +static TextCache *cache_for(SDL_Renderer *renderer) { + SDL_PropertiesID props = SDL_GetRendererProperties(renderer); + TextCache *c = SDL_GetPointerProperty(props, CACHE_PROPERTY, NULL); + if (c) return c; + c = calloc(1, sizeof(*c)); + if (!c) return NULL; + c->font_fd = memfd_create("sketchbook-go-mono", MFD_CLOEXEC); + if (c->font_fd < 0) { free(c); return NULL; } + size_t offset = 0; + while (offset < sizeof(go_mono)) { + ssize_t n = write(c->font_fd, go_mono + offset, sizeof(go_mono) - offset); + if (n < 0 && errno == EINTR) continue; + if (n <= 0) { release_cache(NULL, c); return NULL; } + offset += (size_t)n; + } + char path[64]; + snprintf(path, sizeof(path), "/proc/self/fd/%d", c->font_fd); + c->config = FcInitLoadConfigAndFonts(); + if (!c->config || !FcConfigAppFontAddFile(c->config, (const FcChar8 *)path)) { + release_cache(NULL, c); + return NULL; + } + c->map = pango_cairo_font_map_new_for_font_type(CAIRO_FONT_TYPE_FT); + pango_fc_font_map_set_config(PANGO_FC_FONT_MAP(c->map), c->config); + c->context = pango_font_map_create_context(c->map); + c->layout = pango_layout_new(c->context); + c->font = pango_font_description_from_string("Go Mono, Noto Sans Mono, DejaVu Sans Mono, Noto Sans Arabic, Noto Sans Hebrew, IBM Plex Sans JP, Noto Color Emoji, Noto Sans Symbols 2, DejaVu Sans, monospace"); + cairo_font_options_t *options = cairo_font_options_create(); + cairo_font_options_set_antialias(options, CAIRO_ANTIALIAS_GRAY); + cairo_font_options_set_hint_style(options, CAIRO_HINT_STYLE_SLIGHT); + cairo_font_options_set_hint_metrics(options, CAIRO_HINT_METRICS_ON); + pango_cairo_context_set_font_options(c->context, options); + cairo_font_options_destroy(options); + pango_context_set_round_glyph_positions(c->context, TRUE); + pango_layout_set_single_paragraph_mode(c->layout, TRUE); + c->frame = 1; + if (!SDL_SetPointerPropertyWithCleanup(props, CACHE_PROPERTY, c, release_cache, NULL)) { + release_cache(NULL, c); + return NULL; + } + return c; +} + +void sk_text_begin_frame(SDL_Renderer *renderer) { + TextCache *c = SDL_GetPointerProperty(SDL_GetRendererProperties(renderer), CACHE_PROPERTY, NULL); + /* A sketch may clear twice in a frame. Flush first so any eviction of a + previous epoch can never destroy a texture referenced by queued draws. */ + if (c) { + SDL_FlushRenderer(renderer); + ++c->frame; + } +} + +static float density(SDL_Renderer *renderer) { + float sx = 1, sy = 1; + SDL_GetRenderScale(renderer, &sx, &sy); + int w = 0, h = 0; + SDL_RendererLogicalPresentation mode; + SDL_FRect rect; + if (SDL_GetRenderLogicalPresentation(renderer, &w, &h, &mode) && + mode != SDL_LOGICAL_PRESENTATION_DISABLED && w > 0 && h > 0 && + SDL_GetRenderLogicalPresentationRect(renderer, &rect)) { + sx *= rect.w / w; + sy *= rect.h / h; + } + return fmaxf(0.25f, fmaxf(sx, sy)); +} +static uint64_t hash_text(const char *bytes, size_t len) { + uint64_t hash = UINT64_C(14695981039346656037); + for (size_t i = 0; i < len; ++i) hash = (hash ^ (unsigned char)bytes[i]) * UINT64_C(1099511628211); + return hash; +} +static bool rasterize(TextCache *c, SDL_Renderer *renderer, Text *out, + const char *bytes, size_t len, int size, Uint32 color) { + if (len > INT_MAX || !g_utf8_validate(bytes, (gssize)len, NULL)) return false; + pango_font_description_set_absolute_size(c->font, (double)size * PANGO_SCALE / 64.0); + pango_layout_set_font_description(c->layout, c->font); + pango_layout_set_text(c->layout, bytes, (int)len); + PangoRectangle ink, logical; + pango_layout_get_pixel_extents(c->layout, &ink, &logical); + const int left = MIN(0, ink.x) - 1; + const int top = MIN(0, ink.y) - 1; + const int width = MAX(logical.x + logical.width, ink.x + ink.width) - left + 1; + const int height = MAX(logical.y + logical.height, ink.y + ink.height) - top + 1; + if (width <= 0 || height <= 0 || width > 16384 || height > 4096 || + (size_t)width * height * 4 > CACHE_BYTES) return false; + cairo_surface_t *surface = cairo_image_surface_create(CAIRO_FORMAT_ARGB32, width, height); + if (cairo_surface_status(surface) != CAIRO_STATUS_SUCCESS) { + cairo_surface_destroy(surface); + return false; + } + cairo_t *cr = cairo_create(surface); + cairo_set_source_rgb(cr, ((color >> 16) & 255) / 255.0, ((color >> 8) & 255) / 255.0, (color & 255) / 255.0); + cairo_move_to(cr, -left, -top); + pango_cairo_show_layout(cr, c->layout); + cairo_destroy(cr); + cairo_surface_flush(surface); + SDL_Texture *texture = SDL_CreateTexture(renderer, SDL_PIXELFORMAT_ARGB8888, + SDL_TEXTUREACCESS_STATIC, width, height); + const bool okay = texture && SDL_UpdateTexture(texture, NULL, cairo_image_surface_get_data(surface), + cairo_image_surface_get_stride(surface)); + cairo_surface_destroy(surface); + if (!okay) { if (texture) SDL_DestroyTexture(texture); return false; } + SDL_SetTextureBlendMode(texture, SDL_BLENDMODE_BLEND_PREMULTIPLIED); + SDL_SetTextureScaleMode(texture, SDL_SCALEMODE_LINEAR); + char *key = malloc(len + 1); + if (!key) { SDL_DestroyTexture(texture); return false; } + memcpy(key, bytes, len); key[len] = 0; + *out = (Text){.texture=texture,.text=key,.len=len,.bytes=(size_t)width*height*4, + .hash=hash_text(bytes,len),.used=c->frame,.size=size,.width=width,.height=height, + .left=left,.top=top,.color=color}; + return true; +} + +bool sk_text_draw(SDL_Renderer *renderer, float x, float y, float scale, + const char *utf8, size_t len, Uint8 r, Uint8 g, Uint8 b, Uint8 a) { + if (!len || !a) return true; + if (!isfinite(scale) || scale <= 0) return false; + TextCache *c = cache_for(renderer); + if (!c) return false; + const float dpi = density(renderer); + const int size = (int)lroundf(fminf(512, 13.0f * scale * dpi) * 64); + const Uint32 color = ((Uint32)r << 16) | ((Uint32)g << 8) | b; + const uint64_t hash = hash_text(utf8, len); + Text *entry = NULL; + for (size_t i = 0; i < CACHE_ENTRIES; ++i) { + Text *t = &c->entries[i]; + if (t->texture && t->hash == hash && t->len == len && t->size == size && + t->color == color && !memcmp(t->text, utf8, len)) { entry = t; break; } + } + Text made = {0}; + if (!entry) { + if (!rasterize(c, renderer, &made, utf8, len, size, color)) return false; + /* Only evict textures unused this epoch. SDL can still have queued draws + referring to current-frame entries; those stay alive through present. */ + while (!entry) { + Text *oldest = NULL; + for (size_t i = 0; i < CACHE_ENTRIES; ++i) { + Text *t = &c->entries[i]; + if (!t->texture) { if (!entry) entry = t; } + else if (t->used < c->frame && (!oldest || t->used < oldest->used)) oldest = t; + } + if (entry && c->bytes + made.bytes <= CACHE_BYTES) break; + entry = NULL; + if (!oldest) break; + c->bytes -= oldest->bytes; + release_text(oldest); + } + if (entry) { *entry = made; c->bytes += made.bytes; } + else entry = &made; + } + entry->used = c->frame; + /* Cairo pixels are premultiplied; modulate RGB and alpha together. Color + glyphs retain their intrinsic paint, while ordinary text uses its ink. */ + SDL_SetTextureColorMod(entry->texture, a, a, a); + SDL_SetTextureAlphaMod(entry->texture, a); + SDL_FRect dest = {.x=(roundf(x*dpi)+entry->left)/dpi,.y=(roundf(y*dpi)+entry->top)/dpi, + .w=entry->width/dpi,.h=entry->height/dpi}; + const bool okay = SDL_RenderTexture(renderer, entry->texture, NULL, &dest); + if (entry == &made) { + /* Exceptional very dense frames remain bounded instead of retaining + every unique string. Flush before releasing this temporary texture. */ + SDL_FlushRenderer(renderer); + release_text(&made); + } + return okay; +} diff --git a/vendor/text.h b/vendor/text.h new file mode 100644 index 0000000..bf4f363 --- /dev/null +++ b/vendor/text.h @@ -0,0 +1,10 @@ +#ifndef SKETCHBOOK_TEXT_H +#define SKETCHBOOK_TEXT_H +#include <SDL3/SDL.h> +#include <stddef.h> +/* x/y are the logical line-box top. Scale 1 is 13 px Go Mono (roughly + eight-pixel ASCII advance and a 16-pixel line box); fallback uses Fontconfig. */ +bool sk_text_draw(SDL_Renderer *, float x, float y, float scale, + const char *utf8, size_t len, Uint8 r, Uint8 g, Uint8 b, Uint8 a); +void sk_text_begin_frame(SDL_Renderer *); +#endif diff --git a/web/host.js b/web/host.js index 9f92a75..aa45840 100644 --- a/web/host.js +++ b/web/host.js @@ -1,15 +1,19 @@ -// sketchbook web host: forwards input to the sketch and draws what it asks. -// It knows nothing about any particular sketch. +// sketchbook web host: forwards input to the sketch and does what it asks: +// drawing on a 2D canvas, GPU work on WebGPU (shaders arrive as WGSL), sound +// and the microphone on Web Audio, files through a picker. It knows nothing +// about any particular sketch. const canvas = document.querySelector('canvas'); const status = document.querySelector('#status'); const ctx = canvas.getContext('2d', {alpha: false}); const decoder = new TextDecoder(); +const encoder = new TextEncoder(); const images = [null]; const byPath = new Map(); const loads = []; -let wasm, clipped = false, size = [960, 600]; +let wasm, clipped = false, size = [960, 600], frameNumber = 0; -const bytes = (ptr, len) => new Uint8Array(wasm.memory.buffer, ptr, len); +// An empty Zig slice may point anywhere, even past the memory. +const bytes = (ptr, len) => len ? new Uint8Array(wasm.memory.buffer, ptr, len) : new Uint8Array(0); const string = (ptr, len) => decoder.decode(bytes(ptr, len)); const rgba = (r, g, b, a) => `rgba(${r},${g},${b},${a / 255})`; const Kind = {move: 1, down: 2, up: 3, wheel: 4, keyDown: 5, keyUp: 6, text: 7, cancel: 8}; @@ -27,6 +31,380 @@ function fit() { ctx.imageSmoothingEnabled = false; } +// Hands bytes to the sketch: into memory it allocates, which the export +// it is passed to takes over. Views are made after allocating, which may +// grow (and so replace) the memory. +function give(source) { + const view = source instanceof Float32Array ? new Uint8Array(source.buffer, source.byteOffset, source.byteLength) : source; + const ptr = wasm.sk_alloc(view.length); + bytes(ptr, view.length).set(view); + return ptr; +} + +// --------------------------------------------------------------------------- +// GPU: abi.gpu on WebGPU. Buffers and images by handle; shader modules and +// pipelines cached by their description; one queue submission per dispatch +// or pass, in order, as on native. + +const gpu = { + device: null, + buffers: [null], + modules: new Map(), + pipelines: new Map(), + // Uniforms and vertices go through rings written before each submission; + // the queue orders those writes with the work, so wrapping is safe. + rings: {}, + spare: [], + + // Why there is no device, for the page to say. + missing: '', + + async start() { + if (!window.isSecureContext) return void (gpu.missing = `WebGPU needs a secure context: open the page on localhost or over https, or tell the browser to trust ${location.origin}.`); + if (!navigator.gpu) return void (gpu.missing = 'This browser has WebGPU turned off. In Firefox: about:config, set dom.webgpu.enabled to true.'); + try { + const adapter = await navigator.gpu.requestAdapter(); + if (!adapter) return void (gpu.missing = 'WebGPU found no usable GPU: the browser may block this one (in Firefox: about:config, gfx.webgpu.ignore-blocklist).'); + const limits = adapter.limits; + gpu.device = await adapter.requestDevice({requiredLimits: { + maxComputeWorkgroupStorageSize: limits.maxComputeWorkgroupStorageSize, + maxStorageBufferBindingSize: limits.maxStorageBufferBindingSize, + maxBufferSize: limits.maxBufferSize, + }}); + gpu.device.addEventListener('uncapturederror', e => console.error('WebGPU:', e.error.message)); + gpu.device.lost.then(info => { console.error('WebGPU device lost:', info.message); gpu.device = null; }); + } catch (error) { + gpu.missing = `WebGPU failed to start: ${error.message ?? error}`; + } + }, + // Says why, once, the first time a sketch asks for the GPU and finds none. + absent() { + if (gpu.missing && !gpu.told) { + gpu.told = true; + console.warn(gpu.missing); + status.textContent = gpu.missing; + status.hidden = false; + } + return 0; + }, + + ring(name, usage, need) { + let r = gpu.rings[name]; + if (!r || r.size < need) { + const size = Math.max(need, 4 << 20); + r = gpu.rings[name] = {buffer: gpu.device.createBuffer({size, usage: usage | GPUBufferUsage.COPY_DST}), size, at: 0}; + } + return r; + }, + // Writes `data` into a ring, returning where. + put(name, usage, data, align) { + const r = gpu.ring(name, usage, data.length); + let at = Math.ceil(r.at / align) * align; + if (at + data.length > r.size) at = 0; + gpu.device.queue.writeBuffer(r.buffer, at, data); + r.at = at + data.length; + return {buffer: r.buffer, offset: at, size: data.length}; + }, + uniforms([ptr, len]) { + return len ? gpu.put('uniforms', GPUBufferUsage.UNIFORM, bytes(ptr, len).slice(), 256) : null; + }, + + module(shader) { + let m = gpu.modules.get(shader.hash); + if (!m) { + m = gpu.device.createShaderModule({code: string(...shader.code)}); + m.getCompilationInfo().then(info => { for (const msg of info.messages) if (msg.type === 'error') console.error(`WGSL ${msg.lineNum}:${msg.linePos} ${msg.message}`); }); + gpu.modules.set(shader.hash, m); + } + return m; + }, + + // Bind groups 0..3 in SDL_GPU's numbering; an empty one where a later + // one is used. + bind(encoder, pipeline, groups) { + for (let i = 0; i < groups.length; i++) { + const entries = groups[i] ?? []; + if (!entries.length && !groups.slice(i + 1).some(g => g?.length)) break; + encoder.setBindGroup(i, gpu.device.createBindGroup({layout: pipeline.getBindGroupLayout(i), entries})); + } + }, + storage(ids) { return ids.map((id, binding) => ({binding, resource: {buffer: gpu.buffers[id]}})); }, + uniform(u) { return u ? [{binding: 0, resource: u}] : []; }, + + newBuffer(size) { + const usage = GPUBufferUsage.STORAGE | GPUBufferUsage.COPY_DST | GPUBufferUsage.COPY_SRC; + gpu.buffers.push(gpu.device.createBuffer({size: Math.ceil(size / 4) * 4, usage})); + return gpu.buffers.length - 1; + }, + + compute(d) { + const k = d.kernel; + const key = `compute ${k.hash}`; + let pipeline = gpu.pipelines.get(key); + if (!pipeline) { + pipeline = gpu.device.createComputePipeline({layout: 'auto', compute: {module: gpu.module(k)}}); + gpu.pipelines.set(key, pipeline); + } + const uniforms = gpu.uniforms(d.uniforms); + const commands = gpu.device.createCommandEncoder(); + const pass = commands.beginComputePass(); + pass.setPipeline(pipeline); + gpu.bind(pass, pipeline, [gpu.storage(d.readonly), gpu.storage(d.readwrite), gpu.uniform(uniforms)]); + pass.dispatchWorkgroups(...d.groups); + pass.end(); + gpu.device.queue.submit([commands.finish()]); + }, + + // The texture a pass renders into, and the canvas that shows it. + target(id, w, h) { + const entry = images[id]; + if (!entry) return null; + if (!entry.texture || entry.texture.width !== w || entry.texture.height !== h) { + entry.texture?.destroy(); + entry.texture = gpu.device.createTexture({size: [w, h], format: 'rgba8unorm', usage: GPUTextureUsage.RENDER_ATTACHMENT | GPUTextureUsage.COPY_SRC | GPUTextureUsage.TEXTURE_BINDING}); + entry.shown = document.createElement('canvas'); + entry.shown.width = w; entry.shown.height = h; + entry.view = entry.shown.getContext('webgpu'); + entry.view.configure({device: gpu.device, format: 'rgba8unorm', usage: GPUTextureUsage.COPY_DST | GPUTextureUsage.RENDER_ATTACHMENT, alphaMode: 'premultiplied'}); + entry.w = w; entry.h = h; + entry.attachments = {}; + } + return entry; + }, + // Multisample and stencil attachments, kept per image. + attachment(entry, kind, format, samples) { + const key = `${kind} ${samples}`; + if (!entry.attachments[key]) entry.attachments[key] = gpu.device.createTexture({size: [entry.w, entry.h], format, sampleCount: samples, usage: GPUTextureUsage.RENDER_ATTACHMENT}); + return entry.attachments[key]; + }, + // Copies the latest render into the image's canvas, once a frame, for + // the 2D canvas to draw. + present(entry) { + if (entry.presented === frameNumber) return; + entry.presented = frameNumber; + const commands = gpu.device.createCommandEncoder(); + commands.copyTextureToTexture({texture: entry.texture}, {texture: entry.view.getCurrentTexture()}, [entry.w, entry.h]); + gpu.device.queue.submit([commands.finish()]); + }, + + pipeline(p, samples, stencil) { + const key = JSON.stringify([p.vertex.hash, p.fragment.hash, p.stride, p.attributes, p.primitive, p.blend, p.write_color, p.stencil, samples, stencil]); + let pipeline = gpu.pipelines.get(key); + if (pipeline) return pipeline; + const formats = {float: 'float32', float2: 'float32x2', float3: 'float32x3', float4: 'float32x4', ubyte4_norm: 'unorm8x4'}; + const blends = { + none: undefined, + premultiplied: {color: {srcFactor: 'one', dstFactor: 'one-minus-src-alpha'}, alpha: {srcFactor: 'one', dstFactor: 'one-minus-src-alpha'}}, + additive: {color: {srcFactor: 'one', dstFactor: 'one'}, alpha: {srcFactor: 'one', dstFactor: 'one'}}, + }; + const dashed = s => s.replaceAll('_', '-').replace('-or-equal', '-equal').replace('-and-clamp', '-clamp').replace('-and-wrap', '-wrap'); + const face = f => ({compare: dashed(f.compare), passOp: dashed(f.pass), failOp: dashed(f.fail), depthFailOp: 'keep'}); + const keep = {compare: 'always', passOp: 'keep', failOp: 'keep', depthFailOp: 'keep'}; + pipeline = gpu.device.createRenderPipeline({ + layout: 'auto', + vertex: { + module: gpu.module(p.vertex), + buffers: p.attributes.length ? [{arrayStride: p.stride, attributes: p.attributes.map(a => ({shaderLocation: a.location, offset: a.offset, format: formats[a.format]}))}] : [], + }, + fragment: {module: gpu.module(p.fragment), targets: [{format: 'rgba8unorm', blend: blends[p.blend], writeMask: p.write_color ? GPUColorWrite.ALL : 0}]}, + primitive: {topology: dashed(p.primitive)}, + multisample: {count: samples}, + depthStencil: stencil ? { + format: 'depth24plus-stencil8', depthWriteEnabled: false, depthCompare: 'always', + stencilFront: p.stencil ? face(p.stencil.front) : keep, stencilBack: p.stencil ? face(p.stencil.back) : keep, + } : undefined, + }); + gpu.pipelines.set(key, pipeline); + return pipeline; + }, + + pass(p) { + const entry = gpu.target(p.image, p.width, p.height); + if (!entry) return; + const samples = p.samples > 1 ? 4 : 1; // WebGPU offers 1 and 4. + const view = entry.texture.createView(); + const color = { + view: samples > 1 ? gpu.attachment(entry, 'msaa', 'rgba8unorm', samples).createView() : view, + resolveTarget: samples > 1 ? view : undefined, + loadOp: p.clear ? 'clear' : 'load', + clearValue: p.clear ? {r: p.clear[0], g: p.clear[1], b: p.clear[2], a: p.clear[3]} : undefined, + storeOp: 'store', + }; + const depthStencil = p.stencil ? { + view: gpu.attachment(entry, 'stencil', 'depth24plus-stencil8', samples).createView(), + depthLoadOp: 'clear', depthClearValue: 1, depthStoreOp: 'discard', + stencilLoadOp: 'clear', stencilClearValue: 0, stencilStoreOp: 'discard', + } : undefined; + // Every draw's bytes are written before the pass is submitted. + const draws = p.draws.map(d => ({ + d, + vertices: d.vertices[1] ? gpu.put('vertices', GPUBufferUsage.VERTEX, bytes(...d.vertices).slice(), 4) : null, + vertexUniforms: gpu.uniforms(d.vertex_uniforms), + fragmentUniforms: gpu.uniforms(d.fragment_uniforms), + })); + const commands = gpu.device.createCommandEncoder(); + const pass = commands.beginRenderPass({colorAttachments: [color], depthStencilAttachment: depthStencil}); + for (const {d, vertices, vertexUniforms, fragmentUniforms} of draws) { + const pipeline = gpu.pipeline(p.pipelines[d.pipeline], samples, p.stencil); + pass.setPipeline(pipeline); + gpu.bind(pass, pipeline, [gpu.storage(d.vertex_storage), gpu.uniform(vertexUniforms), gpu.storage(d.fragment_storage), gpu.uniform(fragmentUniforms)]); + if (vertices) pass.setVertexBuffer(0, vertices.buffer, vertices.offset, vertices.size); + if (p.stencil) pass.setStencilReference(d.stencil_reference); + const s = d.scissor; + if (s) { + const x = Math.max(0, Math.floor(s.x)), y = Math.max(0, Math.floor(s.y)); + pass.setScissorRect(x, y, Math.max(0, Math.min(p.width, Math.ceil(s.x + s.w)) - x), Math.max(0, Math.min(p.height, Math.ceil(s.y + s.h)) - y)); + } else pass.setScissorRect(0, 0, p.width, p.height); + pass.draw(d.vertex_count); + } + pass.end(); + gpu.device.queue.submit([commands.finish()]); + }, + + // Reads a buffer back; it reaches the sketch before a later frame. + arrived: [], + download(id, offset, size, key) { + const staging = gpu.spare.find(b => b.size >= size) ?? gpu.device.createBuffer({size: Math.max(size, 256), usage: GPUBufferUsage.MAP_READ | GPUBufferUsage.COPY_DST}); + gpu.spare = gpu.spare.filter(b => b !== staging); + const commands = gpu.device.createCommandEncoder(); + commands.copyBufferToBuffer(gpu.buffers[id], offset, staging, 0, size); + gpu.device.queue.submit([commands.finish()]); + staging.mapAsync(GPUMapMode.READ, 0, size).then(() => { + gpu.arrived.push({key, bytes: new Uint8Array(staging.getMappedRange(0, size)).slice()}); + staging.unmap(); + gpu.spare.push(staging); + }, () => {}); + }, +}; + +// --------------------------------------------------------------------------- +// Sound: streams queued and mixed on the main thread, the microphone +// collected the same way, both at the context's rate (48 kHz if the browser +// agrees). Browsers start sound only after a click or a key. + +const sound = { + context: null, + queues: new Map(), + heard: [], + microphone: null, + + start() { + // Safari: play as a media app does, through the ringer's silent switch. + if (navigator.audioSession) navigator.audioSession.type = 'playback'; + try { sound.context = new AudioContext({sampleRate: 48000, latencyHint: 'interactive'}); } catch { sound.context = new AudioContext(); } + const player = sound.context.createScriptProcessor(1024, 1, 1); + player.onaudioprocess = e => { + sound.pulled = performance.now(); + const out = e.outputBuffer.getChannelData(0); + out.fill(0); + for (const q of sound.queues.values()) { + let i = 0; + while (i < out.length && q.chunks.length) { + const chunk = q.chunks[0]; + const n = Math.min(out.length - i, chunk.length - q.at); + for (let j = 0; j < n; j++) out[i + j] += chunk[q.at + j]; + i += n; q.at += n; q.queued -= n; + if (q.at === chunk.length) { q.chunks.shift(); q.at = 0; } + } + } + }; + player.connect(sound.context.destination); + const resume = () => sound.context.resume().then(() => { + if (sound.hinted && !gpu.told) status.hidden = true; + }); + // On touch screens browsers count only a tap's end (or the click after + // it) as the gesture that may start sound. + for (const type of ['pointerdown', 'pointerup', 'touchend', 'click', 'keydown']) window.addEventListener(type, resume); + }, + // Browsers start sound only after a gesture: say so, once, while a + // sketch that plays or listens waits for one. + hinted: false, + hint() { + if (sound.context.state === 'running' || sound.hinted || gpu.told) return; + sound.hinted = true; + status.textContent = 'Click, tap or press a key to start sound and the microphone.'; + status.hidden = false; + }, + // While sound is not playing (no tap yet, the page hidden, or a browser + // whose mixer has stopped asking for samples), the queues drain by the + // clock as if heard, so sketches that pace themselves by what is queued + // keep running, silently. + pulled: 0, + drain(seconds) { + if (sound.context.state === 'running' && performance.now() - sound.pulled < 300) return; + for (const q of sound.queues.values()) { + let n = Math.round(seconds * sound.context.sampleRate); + while (n > 0 && q.chunks.length) { + const take = Math.min(n, q.chunks[0].length - q.at); + q.at += take; q.queued -= take; n -= take; + if (q.at === q.chunks[0].length) { q.chunks.shift(); q.at = 0; } + } + } + }, + queue(key) { + if (!sound.queues.has(key)) sound.queues.set(key, {chunks: [], at: 0, queued: 0}); + return sound.queues.get(key); + }, + listen() { + sound.hint(); + if (sound.microphone) return; + sound.microphone = 'asked'; + if (!navigator.mediaDevices) return console.warn('The microphone needs a secure context: localhost or https.'); + navigator.mediaDevices.getUserMedia({audio: {echoCancellation: false, noiseSuppression: false, autoGainControl: false}}).then(stream => { + const source = sound.context.createMediaStreamSource(stream); + const recorder = sound.context.createScriptProcessor(1024, 1, 1); + recorder.onaudioprocess = e => { + sound.heard.push(e.inputBuffer.getChannelData(0).slice()); + e.outputBuffer.getChannelData(0).fill(0); + }; + source.connect(recorder); + recorder.connect(sound.context.destination); + sound.microphone = stream; + }, error => console.warn('No microphone:', error)); + }, +}; + +// A file the user picked: decoded by the browser, handed to the sketch by +// name, then announced as the console line `<command> <name>`. +async function openFile(command, pattern) { + const input = document.createElement('input'); + input.type = 'file'; + if (pattern) input.accept = pattern.split(';').map(e => `.${e}`).join(','); + input.onchange = async () => { + const file = input.files[0]; + if (!file) return; + try { + const decoded = await sound.context.decodeAudioData(await file.arrayBuffer()); + const mono = new Float32Array(decoded.length); + for (let c = 0; c < decoded.numberOfChannels; c++) { + const channel = decoded.getChannelData(c); + for (let i = 0; i < mono.length; i++) mono[i] += channel[i] / decoded.numberOfChannels; + } + const name = encoder.encode(file.name); + wasm.sk_sound(give(name), name.length, give(mono), mono.length); + const line = encoder.encode(`${command} ${file.name}`); + wasm.sk_command(give(line), line.length); + } catch (error) { + console.error(`Cannot open ${file.name}:`, error); + } + }; + input.click(); +} + +// Microphone samples and readbacks, into the sketch before each frame. +function deliver() { + if (sound.heard.length) { + const total = sound.heard.reduce((n, c) => n + c.length, 0); + const all = new Float32Array(total); + let at = 0; + for (const chunk of sound.heard) { all.set(chunk, at); at += chunk.length; } + sound.heard = []; + wasm.sk_microphone(give(all), total); + } + for (const {key, bytes: b} of gpu.arrived) wasm.sk_download(key, give(b), b.length); + gpu.arrived = []; +} + const imports = { host: { log(ptr, len) { console.log(string(ptr, len)); }, @@ -52,12 +430,29 @@ const imports = { images.push({source, w, h, ctx: source.getContext('2d')}); return images.length - 1; }, + freeImage(id) { + if (!images[id]) return; + images[id].texture?.destroy(); + for (const t of Object.values(images[id].attachments ?? {})) t.destroy(); + images[id] = null; + }, + openFile(command, commandLen, pattern, patternLen) { openFile(string(command, commandLen), string(pattern, patternLen)); }, }, canvas: { clear(r, g, b, a) { ctx.fillStyle = rgba(r, g, b, a); ctx.fillRect(0, 0, size[0], size[1]); }, fill(x, y, w, h, r, g, b, a) { ctx.fillStyle = rgba(r, g, b, a); ctx.fillRect(x, y, w, h); }, image(id, x, y, w, h, sx, sy, sw, sh, alpha) { const entry = images[id]; + if (entry?.texture) { + // GPU renders scale smoothly, as on native. + gpu.present(entry); + ctx.globalAlpha = alpha / 255; + ctx.imageSmoothingEnabled = true; + if (sw > 0) ctx.drawImage(entry.shown, sx, sy, sw, sh, x, y, w, h); else ctx.drawImage(entry.shown, x, y, w, h); + ctx.imageSmoothingEnabled = false; + ctx.globalAlpha = 1; + return; + } if (!entry?.source || (entry.source instanceof Image && !entry.w)) { ctx.fillStyle = '#6e1e46'; ctx.fillRect(x, y, w, h); return; } ctx.globalAlpha = alpha / 255; if (sw > 0) ctx.drawImage(entry.source, sx, sy, sw, sh, x, y, w, h); else ctx.drawImage(entry.source, x, y, w, h); @@ -65,9 +460,10 @@ const imports = { }, text(x, y, scale, ptr, len, r, g, b, a) { ctx.fillStyle = rgba(r, g, b, a); - ctx.font = `${12 * scale}px monospace`; + // The native font is a fixed 8px cell: the same advance, and a size + // whose capitals fill the cell as native ones do. + ctx.font = `${10 * scale}px monospace`; ctx.textBaseline = 'top'; - // The native font is a fixed 8px cell; keep the same geometry. for (const [i, ch] of Array.from(string(ptr, len)).entries()) ctx.fillText(ch, x + i * 8 * scale, y); }, clip(enabled, x, y, w, h) { @@ -87,6 +483,26 @@ const imports = { entry.ctx.putImageData(new ImageData(data, w, h), 0, 0); }, }, + gpu: { + available: () => gpu.device ? 1 : gpu.absent(), + newBuffer: size => gpu.device ? gpu.newBuffer(size) : 0, + freeBuffer(id) { gpu.buffers[id]?.destroy(); gpu.buffers[id] = null; }, + write(id, offset, ptr, len) { if (gpu.buffers[id]) gpu.device.queue.writeBuffer(gpu.buffers[id], offset, bytes(ptr, len).slice()); }, + compute(ptr, len) { if (gpu.device) gpu.compute(JSON.parse(string(ptr, len))); }, + pass(ptr, len) { if (gpu.device) gpu.pass(JSON.parse(string(ptr, len))); }, + download(id, offset, size, key) { if (gpu.buffers[id]) gpu.download(id, offset, size, key); }, + }, + audio: { + rate: () => sound.context.sampleRate, + push(key, ptr, len, clear) { + if (len) sound.hint(); + const q = sound.queue(key); + if (clear) { q.chunks = []; q.at = 0; q.queued = 0; } + if (len) { q.chunks.push(new Float32Array(wasm.memory.buffer.slice(ptr, ptr + len * 4))); q.queued += len; } + }, + queued: key => sound.queues.get(key)?.queued ?? 0, + listen: () => sound.listen(), + }, }; // DOM `code` to USB HID usage, the numbering recordings and sketches use. @@ -101,39 +517,99 @@ const hid = code => { const floatBits = f => new Uint32Array(new Float32Array([f]).buffer)[0]; try { + await gpu.start(); + sound.start(); const response = await fetch('sketch.wasm'); if (!response.ok) throw new Error('The sketch could not be loaded.'); wasm = (await WebAssembly.instantiate(await response.arrayBuffer(), imports)).instance.exports; + // The room for the canvas, which responsive sketches fill. + const room = () => wasm.sk_window(window.visualViewport?.width ?? window.innerWidth, window.visualViewport?.height ?? window.innerHeight); + room(); + window.addEventListener('resize', room); + window.visualViewport?.addEventListener('resize', room); wasm.sk_init(); wasm.sk_frame(0); await Promise.all(loads); - status.hidden = true; + if (!gpu.told && !(sound.hinted && sound.context.state !== 'running')) status.hidden = true; - const send = (kind, x = 0, y = 0, code = 0, value = 0) => wasm.sk_event(kind, code, x, y, value); + // `flags`: for pointer events, the modifiers held (abi.Event.withModifiers). + const send = (kind, x = 0, y = 0, code = 0, value = 0, flags = 0) => wasm.sk_event(kind, code, x, y, value, flags); + const held = e => ((e.shiftKey ? 1 : 0) | (e.ctrlKey ? 2 : 0) | (e.altKey ? 4 : 0) | (e.metaKey ? 8 : 0)) << 4; const point = e => { const r = canvas.getBoundingClientRect(); return [(e.clientX - r.left) * size[0] / r.width, (e.clientY - r.top) * size[1] / r.height]; }; + // Mouse and pen: the pointer, one at a time. let pointer = null; + // Touch: one finger is the mouse's left button. Two fingers pan, sent as + // the wheel (60 px a notch), and pinch, sent as the wheel with ctrl in its + // flags (one notch per 20 % of spread), the way browsers report a + // trackpad's pinch; both around the point between the fingers. Once two fingers are down, the + // rest of the touch is a gesture until every finger lifts. + const touches = new Map(); + let gesture = null; + const measure = () => { + const [a, b] = [...touches.values()]; + return {mid: [(a[0] + b[0]) / 2, (a[1] + b[1]) / 2], spread: Math.hypot(a[0] - b[0], a[1] - b[1])}; + }; + const ctrl = 2 << 4; + // Synthetic pointers (tests) cannot be captured; real ones always can. + const capture = id => { try { canvas.setPointerCapture(id); } catch {} }; canvas.addEventListener('pointerdown', e => { - if (pointer !== null) return; e.preventDefault(); - pointer = e.pointerId; canvas.setPointerCapture(pointer); - send(Kind.down, ...point(e), e.button + 1); + if (e.pointerType === 'touch') { + capture(e.pointerId); + touches.set(e.pointerId, point(e)); + if (touches.size === 1 && !gesture) { + send(Kind.move, ...point(e)); + send(Kind.down, ...point(e), 1); + } else if (touches.size === 2) { + if (!gesture) send(Kind.cancel); + gesture = measure(); + } + return; + } + if (pointer !== null) return; + pointer = e.pointerId; capture(pointer); + send(Kind.down, ...point(e), e.button + 1, 0, held(e)); + }); + canvas.addEventListener('pointermove', e => { + if (e.pointerType !== 'touch') return send(Kind.move, ...point(e), 0, 0, held(e)); + if (!touches.has(e.pointerId)) return; + touches.set(e.pointerId, point(e)); + if (!gesture) return send(Kind.move, ...point(e)); + if (touches.size < 2) return; + const now = measure(); + const pan = now.mid[1] - gesture.mid[1]; + if (pan) send(Kind.wheel, ...now.mid, 0, floatBits(pan / 60)); + if (gesture.spread > 0 && now.spread > 0 && now.spread !== gesture.spread) { + send(Kind.wheel, ...now.mid, 0, floatBits(Math.log(now.spread / gesture.spread) / Math.log(1.2)), ctrl); + } + gesture = now; }); - canvas.addEventListener('pointermove', e => send(Kind.move, ...point(e))); - canvas.addEventListener('pointerup', e => { + const lift = e => { + if (e.pointerType === 'touch') { + if (!touches.delete(e.pointerId)) return; + if (!gesture) send(e.type === 'pointercancel' ? Kind.cancel : Kind.up, ...point(e), 1); + else if (touches.size === 0) gesture = null; + return; + } if (e.pointerId !== pointer) return; - send(Kind.up, ...point(e), e.button + 1); + if (e.type === 'pointercancel') send(Kind.cancel); else send(Kind.up, ...point(e), e.button + 1, 0, held(e)); const id = pointer; pointer = null; if (canvas.hasPointerCapture(id)) canvas.releasePointerCapture(id); - }); - const cancel = () => { pointer = null; send(Kind.cancel); }; - canvas.addEventListener('pointercancel', cancel); - canvas.addEventListener('lostpointercapture', () => { if (pointer !== null) cancel(); }); - canvas.addEventListener('pointerleave', () => { if (pointer === null) send(Kind.move, -1, -1); }); - canvas.addEventListener('wheel', e => { e.preventDefault(); send(Kind.wheel, ...point(e), 0, floatBits(-Math.sign(e.deltaY))); }, {passive: false}); + }; + canvas.addEventListener('pointerup', lift); + canvas.addEventListener('pointercancel', lift); + canvas.addEventListener('contextmenu', e => e.preventDefault()); + const cancel = () => { pointer = null; touches.clear(); gesture = null; send(Kind.cancel); }; + canvas.addEventListener('lostpointercapture', e => { if (e.pointerId === pointer) cancel(); }); + canvas.addEventListener('pointerleave', e => { if (pointer === null && e.pointerType !== 'touch') send(Kind.move, -1, -1); }); + // A trackpad's pinch arrives as the wheel with ctrlKey set. + canvas.addEventListener('wheel', e => { e.preventDefault(); send(Kind.wheel, ...point(e), 0, floatBits(-Math.sign(e.deltaY)), held(e)); }, {passive: false}); window.addEventListener('keydown', e => { + // Keys the sketch may use, not the page. + if (['Space', 'ArrowUp', 'ArrowDown', 'ArrowLeft', 'ArrowRight', 'PageUp', 'PageDown', 'Home', 'End', 'Backspace'].includes(e.code)) e.preventDefault(); send(Kind.keyDown, 0, 0, hid(e.code), 0); if (e.key.length === 1 && !e.ctrlKey && !e.metaKey) send(Kind.text, 0, 0, 0, e.key.codePointAt(0)); }); @@ -145,23 +621,38 @@ try { // Tests read state by field path, like the native probe. window.sketch = { field(path) { - const encoded = new TextEncoder().encode(path); + const encoded = encoder.encode(path); new Uint8Array(wasm.memory.buffer, wasm.sk_buffer(), encoded.length).set(encoded); return wasm.sk_field(encoded.length); }, + // Runs a console line; with `sound`, opens sounds without the picker. + command(line) { const encoded = encoder.encode(line); wasm.sk_command(give(encoded), encoded.length); }, + sound(name, samples) { const encoded = encoder.encode(name); wasm.sk_sound(give(encoded), encoded.length, give(samples), samples.length); }, size: () => size, + // 'running', or 'suspended' until a tap or key starts sound. + audio: () => sound.context.state, }; + let previous = performance.now(); const frame = ms => { try { + frameNumber++; + sound.drain(Math.min((ms - previous) / 1000, 0.25)); + previous = ms; + deliver(); wasm.sk_frame(ms); requestAnimationFrame(frame); } catch (error) { - status.textContent = 'The sketch stopped. Reload to start again.'; status.hidden = false; console.error(error); + status.textContent = `The sketch stopped (${error?.message ?? error}). Reload to start again.`; status.hidden = false; console.error(error); } }; requestAnimationFrame(frame); + + // Built as an installable app (a manifest in the page): keep an offline copy. + if (document.querySelector('link[rel="manifest"]') && navigator.serviceWorker) { + navigator.serviceWorker.register('sw.js').catch(error => console.warn('No offline copy:', error)); + } } catch (error) { - status.textContent = 'The sketch could not start. Reload to try again.'; + status.textContent = `The sketch could not start (${error?.message ?? error}). Reload to try again.`; console.error(error); } diff --git a/web/index.html b/web/index.html index 4b55799..bb2e463 100644 --- a/web/index.html +++ b/web/index.html @@ -4,6 +4,7 @@ <meta name="viewport" content="width=device-width, initial-scale=1, viewport-fit=cover"> <title>{title}</title> <link rel="icon" href="data:,"> +{head} <style> * { box-sizing: border-box; } html, body { margin: 0; width: 100%; height: 100%; background: #0b1112; color: #e8e8d6; } diff --git a/web/sw.js b/web/sw.js new file mode 100644 index 0000000..8298b85 --- /dev/null +++ b/web/sw.js @@ -0,0 +1,24 @@ +// sketchbook's service worker, for sketches built as installable apps: it +// keeps a copy of everything the page fetches, so the app opens offline. +// Network first, so a new build shows as soon as there is a network; the +// copy only when there is none. +const cache = `sketchbook ${self.registration.scope}`; +const core = ['./', 'index.html', 'host.js', 'sketch.wasm', 'manifest.webmanifest']; + +self.addEventListener('install', event => { + event.waitUntil(caches.open(cache).then(c => Promise.all(core.map(url => c.add(url).catch(() => {}))))); + self.skipWaiting(); +}); +self.addEventListener('activate', event => event.waitUntil(self.clients.claim())); + +self.addEventListener('fetch', event => { + const request = event.request; + if (request.method !== 'GET' || !request.url.startsWith(self.registration.scope)) return; + event.respondWith(fetch(request).then(response => { + if (response.ok) { + const copy = response.clone(); + event.waitUntil(caches.open(cache).then(c => c.put(request, copy))); + } + return response; + }, () => caches.open(cache).then(c => c.match(request, {ignoreSearch: true})).then(found => found ?? Response.error()))); +}); |
