| struct Params { |
| filterDim : u32, |
| blockDim : u32, |
| } |
| |
| @group(0) @binding(0) var samp : sampler; |
| |
| @group(0) @binding(1) var<uniform> params : Params; |
| |
| @group(1) @binding(1) var inputTex : texture_2d<f32>; |
| |
| @group(1) @binding(2) var outputTex : texture_storage_2d<rgba8unorm, write>; |
| |
| struct Flip { |
| value : u32, |
| } |
| |
| @group(1) @binding(3) var<uniform> flip : Flip; |
| |
| var<workgroup> tile : array<array<vec3<f32>, 256>, 4>; |
| |
| @compute @workgroup_size(64, 1, 1) |
| fn main(@builtin(workgroup_id) WorkGroupID : vec3<u32>, @builtin(local_invocation_id) LocalInvocationID : vec3<u32>) { |
| let filterOffset : u32 = ((params.filterDim - 1u) / 2u); |
| let dims = textureDimensions(inputTex, 0); |
| let baseIndex = (((WorkGroupID.xy * vec2(params.blockDim, 4)) + (LocalInvocationID.xy * vec2(4u, 1u))) - vec2(filterOffset, 0)); |
| for(var r : u32 = 0u; (r < 4u); r = (r + 1u)) { |
| for(var c : u32 = 0u; (c < 4u); c = (c + 1u)) { |
| var loadIndex = (baseIndex + vec2(c, r)); |
| if ((flip.value != 0u)) { |
| loadIndex = loadIndex.yx; |
| } |
| tile[r][((4u * LocalInvocationID.x) + c)] = textureSampleLevel(inputTex, samp, ((vec2<f32>(loadIndex) + vec2<f32>(0.25, 0.25)) / vec2<f32>(dims)), 0.0).rgb; |
| } |
| } |
| workgroupBarrier(); |
| for(var r : u32 = 0u; (r < 4u); r = (r + 1u)) { |
| for(var c : u32 = 0u; (c < 4u); c = (c + 1u)) { |
| var writeIndex = (baseIndex + vec2(c, r)); |
| if ((flip.value != 0u)) { |
| writeIndex = writeIndex.yx; |
| } |
| let center : u32 = ((4u * LocalInvocationID.x) + c); |
| if ((((center >= filterOffset) && (center < (256u - filterOffset))) && all((writeIndex < dims)))) { |
| var acc : vec3<f32> = vec3<f32>(0.0, 0.0, 0.0); |
| for(var f : u32 = 0u; (f < params.filterDim); f = (f + 1u)) { |
| var i : u32 = ((center + f) - filterOffset); |
| acc = (acc + ((1.0 / f32(params.filterDim)) * tile[r][i])); |
| } |
| textureStore(outputTex, writeIndex, vec4<f32>(acc, 1.0)); |
| } |
| } |
| } |
| } |