diff options
Diffstat (limited to 'tools/bench.zig')
| -rw-r--r-- | tools/bench.zig | 96 |
1 files changed, 96 insertions, 0 deletions
diff --git a/tools/bench.zig b/tools/bench.zig new file mode 100644 index 0000000..faa769e --- /dev/null +++ b/tools/bench.zig @@ -0,0 +1,96 @@ +//! How long a frame of vector graphics takes, per stage, in release mode: +//! parsing the SVG, turning it into polygons and coverage (CPU backend), or +//! into a GPU pass description (GPU backend; the GPU's own time is not +//! measured here). +const std = @import("std"); +const vector = @import("vector"); +const scene = @import("scene").scene; +const typst = @import("typst"); + +const width = 1280; +const height = 720; +const runs = 300; +/// Drawings up to this many shapes count as simple: they must take < 1ms. +const simple_shapes = 30; + +/// Stand-in for `sketch.Host`: the renderers only need images (and to free +/// the GPU backend's paint buffer, which this bench never makes). +const Host = struct { + pub fn newImage(_: Host, _: u32, _: u32) @import("abi").Image { + return @fromBackingInt(@intCast(1)); + } + pub fn freeImage(_: Host, _: @import("abi").Image) void {} + pub fn freeBuffer(_: Host, _: @import("abi").Buffer) void {} +}; + +pub fn main(init: std.process.Init) !void { + const gpa = std.heap.smp_allocator; + const io = init.io; + var failed = false; + std.debug.print("{s:<24} {s:>7} {s:>10} {s:>10} {s:>10} {s:>10}\n", .{ "object", "shapes", "parse", "cpu draw", "gpu record", "cpu total" }); + // Every Typst object of the scene being played (a live scene adds them + // all whatever the time). + var objects_arena: std.heap.ArenaAllocator = .init(gpa); + defer objects_arena.deinit(); + const shot = try scene.shot(objects_arena.allocator(), .{ .t = 0 }); + for (shot.objects, 0..) |object, index| if (object.source == .typst) { + const svg = typst.find(object.source.typst).?; + var arena: std.heap.ArenaAllocator = .init(gpa); + defer arena.deinit(); + var cpu: vector.Renderer(.cpu) = try .init(gpa, Host{}, width, height); + defer cpu.deinit(Host{}); + var gpu: vector.Renderer(.gpu) = try .init(gpa, Host{}, width, height); + defer gpu.deinit(Host{}); + const Frame = struct { scratch: std.mem.Allocator }; + + var parse_ns: [runs]u64 = undefined; + var cpu_ns: [runs]u64 = undefined; + var gpu_ns: [runs]u64 = undefined; + var shapes: usize = 0; + for (0..runs) |i| { + _ = arena.reset(.retain_capacity); + const t0 = now(io); + const d = try vector.svg.parse(arena.allocator(), svg); + const t1 = now(io); + // Fit to the canvas, as large as it goes: the worst case. + const m = vector.fit(d, .{ .w = width, .h = height }); + cpu.begin(Frame{ .scratch = arena.allocator() }, .{ 0, 0, 0, 1 }); + try cpu.draw(d, m, 1); + const t2 = now(io); + gpu.begin(Frame{ .scratch = arena.allocator() }, .{ 0, 0, 0, 1 }); + try gpu.draw(d, m, 1); + std.mem.doNotOptimizeAway(gpu.target.pass()); + const t3 = now(io); + parse_ns[i] = t1 - t0; + cpu_ns[i] = t2 - t1; + gpu_ns[i] = t3 - t2; + shapes = d.shapes.len; + } + const p = median(&parse_ns); + const c = median(&cpu_ns); + const g = median(&gpu_ns); + std.debug.print("{d:>2} {s:<21} {d:>7} {d:>8.3}ms {d:>8.3}ms {d:>8.3}ms {d:>8.3}ms\n", .{ index, label(object.source.typst), shapes, ms(p), ms(c), ms(g), ms(p + c) }); + if (shapes <= simple_shapes and (p + c > std.time.ns_per_ms or p + g > std.time.ns_per_ms)) failed = true; + }; + std.debug.print("\nup to {d} shapes must parse and draw in under 1ms at {d}x{d}, filling the canvas.\n", .{ simple_shapes, width, height }); + if (failed) return error.TooSlow; +} + +/// The start of a Typst source's first line. +fn label(source: []const u8) []const u8 { + const line = source[0 .. std.mem.indexOfScalar(u8, source, '\n') orelse source.len]; + return line[0..@min(line.len, 21)]; +} + +fn now(io: std.Io) u64 { + return @intCast(std.Io.Timestamp.now(io, .awake).nanoseconds); +} + +fn median(samples: []u64) u64 { + std.mem.sort(u64, samples, {}, std.sort.asc(u64)); + return samples[samples.len / 2]; +} + +fn ms(ns: u64) f64 { + return @as(f64, @floatFromInt(ns)) / std.time.ns_per_ms; +} |
