From 7a05f4d33b611ba1e9b6c68e0d0bd67d6ea011ee Mon Sep 17 00:00:00 2001
From: skal <pascal.massimino@gmail.com>
Date: Tue, 10 Feb 2026 21:11:05 +0100
Subject: refactor: Optimize CNN grayscale computation

Compute gray once per fragment using dot() instead of per-layer.
Pass gray as f32 parameter to conv functions instead of vec4 original.

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---
 workspaces/main/shaders/cnn/cnn_conv5x5.wgsl | 14 +++++---------
 1 file changed, 5 insertions(+), 9 deletions(-)

(limited to 'workspaces/main/shaders/cnn/cnn_conv5x5.wgsl')

diff --git a/workspaces/main/shaders/cnn/cnn_conv5x5.wgsl b/workspaces/main/shaders/cnn/cnn_conv5x5.wgsl
index 0f261dd..5570589 100644
--- a/workspaces/main/shaders/cnn/cnn_conv5x5.wgsl
+++ b/workspaces/main/shaders/cnn/cnn_conv5x5.wgsl
@@ -1,5 +1,5 @@
 // 5×5 variant for 7→4 channels (RGBD output)
-// Assumes 'tex' and 'original' are already normalized to [-1,1]
+// Assumes 'tex' is already normalized to [-1,1]
 // UV coordinates remain in [0,1] and are normalized internally
 // weights: array<array<f32, 8>, 100> (25 positions × 4 channels, each with 7 weights + bias)
 fn cnn_conv5x5_7to4(
@@ -7,12 +7,10 @@ fn cnn_conv5x5_7to4(
   samp: sampler,
   uv: vec2<f32>,
   resolution: vec2<f32>,
-  original: vec4<f32>,
+  gray: f32,
   weights: array<array<f32, 8>, 100>
 ) -> vec4<f32> {
   let step = 1.0 / resolution;
-
-  let gray = 0.2126*original.r + 0.7152*original.g + 0.0722*original.b;
   let uv_norm = (uv - 0.5) * 2.0;
 
   var sum = vec4<f32>(0.0);
@@ -44,7 +42,7 @@ fn cnn_conv5x5_7to4(
 }
 
 // 5×5 variant for 7→1 channel (scalar output)
-// Assumes 'tex' and 'original' are already normalized to [-1,1]
+// Assumes 'tex' is already normalized to [-1,1]
 // UV coordinates remain in [0,1] and are normalized internally
 // weights: array<array<f32, 8>, 25> (25 positions, each with 7 weights + bias)
 fn cnn_conv5x5_7to1(
@@ -52,12 +50,10 @@ fn cnn_conv5x5_7to1(
   samp: sampler,
   uv: vec2<f32>,
   resolution: vec2<f32>,
-  original: vec4<f32>,
+  gray: f32,
   weights: array<array<f32, 8>, 25>
 ) -> f32 {
   let step = 1.0 / resolution;
-
-  let gray = 0.2126*original.r + 0.7152*original.g + 0.0722*original.b;
   let uv_norm = (uv - 0.5) * 2.0;
 
   var sum = 0.0;
@@ -96,7 +92,7 @@ fn cnn_conv5x5_7to4_src(
   let step = 1.0 / resolution;
 
   let original = (textureSample(tex, samp, uv) - 0.5) * 2.0;
-  let gray = 0.2126*original.r + 0.7152*original.g + 0.0722*original.b;
+  let gray = dot(original.rgb, vec3<f32>(0.2126, 0.7152, 0.0722));
   let uv_norm = (uv - 0.5) * 2.0;
 
   var sum = vec4<f32>(0.0);
-- 
cgit v1.2.3