diff options
| author | Gabriel Schneider <[email protected]> | 2026-11-17 05:38:26 -0300 |
|---|---|---|
| committer | Gabriel Schneider <[email protected]> | 2026-11-17 12:15:01 -0300 |
| commit | d3742ddf2b3a85eb3935cb35e958fa27f1293a6b (patch) | |
| tree | b47dce7b82e72673432f6c8746517386c67ab742 /shaders/ui.frag.glsl | |
| parent | b65048ac19f9f47b6faa21d1a9f3d06ba3ff1a95 (diff) | |
| parent | 81c4157b6731ee64a7bf9194c5df0818e2977f5d (diff) | |
| download | pardes-d3742ddf2b3a85eb3935cb35e958fa27f1293a6b.tar.gz pardes-d3742ddf2b3a85eb3935cb35e958fa27f1293a6b.zip | |
integration: the GPU move whole, with the terminal a module of its own, the plumber and the diff work
A merge of pardes-next (uoqyunnw) and gpu-s4b (zxswllzu). One conflict, build.zig: core-test's root is the core without a terminal AND the Cells and Atlas tests join the unit suite; gui.zig merged by itself and is formatted (two role casts spelled as zig fmt spells them, two blank lines). Gated against a baseline recorded from pardes-next with the gate's test hooks and the renderer still in gui.zig (.scratch/sketch/gpu/mbase): frame corpus 77/77 identical in Debug and in ReleaseFast, the gated window -0.032% instructions (threshold 0.5%), unit-test on gui 1583/1591 with no failure the baseline has not, every GPU counter 0; on tui core-test 810/949 (139 skipped), shell-test 989/992, behaviour-test 269/273 (the four addon ABI tests the base fails).
Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNBECMWcJD5pdS4SR6T
Diffstat (limited to 'shaders/ui.frag.glsl')
| -rw-r--r-- | shaders/ui.frag.glsl | 39 |
1 files changed, 0 insertions, 39 deletions
diff --git a/shaders/ui.frag.glsl b/shaders/ui.frag.glsl deleted file mode 100644 index 7f70629a..00000000 --- a/shaders/ui.frag.glsl +++ /dev/null @@ -1,39 +0,0 @@ -#version 450 - -// Sample the glyph atlas (R = coverage 0..1) and mix the cell's bg up to its fg -// by that coverage. Output is PREMULTIPLIED: an opaque cell is unchanged. -// WindowOpacity changes only the ground's coverage, never the glyph's ink. - -layout(set = 2, binding = 0) uniform sampler2D u_atlas; -layout(std140, set = 3, binding = 0) uniform UiUniforms { - float background_opacity; -} u_ui; - -layout(location = 0) in vec2 v_uv; -layout(location = 1) in vec3 v_fg; -layout(location = 2) in vec3 v_bg; -layout(location = 4) flat in uint v_effect; -layout(location = 5) flat in vec4 v_clip; - -layout(location = 0) out vec4 o_col; - -void main() { - if (v_clip.z >= v_clip.x && (gl_FragCoord.x < v_clip.x || gl_FragCoord.x >= v_clip.z || - gl_FragCoord.y < v_clip.y || gl_FragCoord.y >= v_clip.w)) discard; - float a = texture(u_atlas, v_uv).r; - // A block cursor is foreground chrome even though its fill occupies the - // cell background: keep it visible at WindowOpacity 0. Otherwise the - // ground's coverage is the setting's. The flag is shared with - // ui.vert.glsl and src/gui/gui.zig. - float bgA = (v_effect & 0x20000000u) != 0u ? 1.0 : clamp(u_ui.background_opacity, 0.0, 1.0); - - // Ink over ground, both premultiplied: v_fg*a + (v_bg*bgA)*(1-a), which is - // mix() with the ground already premultiplied. At bgA = 1 that is exactly - // the mix(v_bg, v_fg, a) at alpha 1 this shader always wrote. - vec4 under = vec4(v_bg * bgA, bgA); - vec4 cell = vec4(mix(under.rgb, v_fg, a), mix(under.a, 1.0, a)); - - // FreeType's hinted grayscale bitmap is already the intended coverage. - // Preserve it exactly; weight-changing gamma hacks blur small stems. - o_col = cell; -} |
