vello_hybrid: Use vec3 instead of arrays for BlurParams

[LaurenzV]

This was tested on a Vivo y21. For some reason, this phone seems to have problems accepting structs that contain more than one array. Something like this:

#version 300 es
precision highp float;
precision highp int;
out vec4 outColor;
struct Params {
  float ar1[2];
  float ar2[2];
};
Params make_params() {
  float[2] a = float[2](0.0, 0.0);
  float[2] b = float[2](0.0, 0.0);
  return Params(a, b);
}
void main() {
  Params params = make_params();
  outColor = vec4(0.0, params.ar1[0], params.ar1[0], 1.0);
}

If I replace this with a single float array of larger size or try other stuff, it seems to work fine, it really seems to just stop working if there are at least two array fields. I don't think this is something actionable to report to naga.

However, luckily we can work around this from our side easily by instead using vec3, which the phone does support fine. While Vello Hybrid still doesn't work on that device, at least it fails with a different error now. 😆

[LaurenzV]

I guess Windows doesn't like that change. 🙃

[taj-p]

I guess Windows doesn't like that change. 🙃

Looks related to: gfx-rs/wgpu#4460

Maybe you can do something like:

diff --git a/sparse_strips/vello_sparse_shaders/shaders/filters.wgsl b/sparse_strips/vello_sparse_shaders/shaders/filters.wgsl
index 5413ce3d..2f08f9cc 100644
--- a/sparse_strips/vello_sparse_shaders/shaders/filters.wgsl
+++ b/sparse_strips/vello_sparse_shaders/shaders/filters.wgsl
@@ -87,13 +87,16 @@ fn unpack_flood_filter(data: GpuFilterData) -> FloodFilter {
 fn unpack_blur_params(data: GpuFilterData) -> BlurParams {
     let n_linear_taps = unpack_header_n_linear_taps(data.data[0]);
     let center_weight = bitcast<f32>(data.data[1]);
-    var weights = vec3<f32>(0.0);
-    var offsets = vec3<f32>(0.0);
-
-    for (var i = 0u; i < MAX_TAPS_PER_SIDE; i++) {
-        weights[i] = bitcast<f32>(data.data[2u + i]);
-        offsets[i] = bitcast<f32>(data.data[2u + MAX_TAPS_PER_SIDE + i]);
-    }
+    let weights = vec3<f32>(
+        bitcast<f32>(data.data[2u]),
+        bitcast<f32>(data.data[3u]),
+        bitcast<f32>(data.data[4u]),
+    );
+    let offsets = vec3<f32>(
+        bitcast<f32>(data.data[2u + MAX_TAPS_PER_SIDE]),
+        bitcast<f32>(data.data[3u + MAX_TAPS_PER_SIDE]),
+        bitcast<f32>(data.data[4u + MAX_TAPS_PER_SIDE]),
+    );
 
     return BlurParams(n_linear_taps, center_weight, weights, offsets);
 }
@@ -283,10 +286,13 @@ fn convolve(
     // First compute the color contribution of the center pixel.
     var color = textureSampleLevel(in_tex, linear_sampler, (src_texel + 0.5) / tex_size, 0.0) * center_weight;
 
+    let weights_arr = array<f32, 3>(weights.x, weights.y, weights.z);
+    let offsets_arr = array<f32, 3>(offsets.x, offsets.y, offsets.z);
+
     // Then, compute and sum the contributions of the adjacent pixels.
     for (var i = 0u; i < n_linear_taps; i++) {
-        let w = weights[i];
-        let d = dir * offsets[i];
+        let w = weights_arr[i];
+        let d = dir * offsets_arr[i];
         color += textureSampleLevel(in_tex, linear_sampler, (src_texel + d + 0.5) / tex_size, 0.0) * w;
         color += textureSampleLevel(in_tex, linear_sampler, (src_texel - d + 0.5) / tex_size, 0.0) * w;
     }

[taj-p]

LGTM after Windows CI passes

[LaurenzV]

Unfortuantely I'm getting this now on a Realme 8 phone. 😔