compute-analysis-pipeline.md

Compute Shader Analysis Pipeline

Custom compute shaders can act as analysis effects, producing typed output fields that are read back to the CPU and can drive downstream effect properties via data bindings.

Analysis Output Types

Type	Pixels	Packing
float	1	.x used
float2	1	.xy used
float3	1	.xyz used
float4	1	all 4 components
floatarray	ceil(N/4)	4 floats packed per pixel
float2array	N	.xy per pixel
float3array	N	.xyz per pixel
float4array	N	all 4 per pixel

D2D Compute Shader Conventions

D2D evaluates compute effects in tiles. Key conventions:

• _TileOffset (int2): Auto-injected at cbuffer offset 0 in CalculateThreadgroups. Gives the tile's origin in the full image.
• Source.GetDimensions(): Returns the full source image size (not tile size).
• SampleLevel(): Must use normalized UVs via SampleLevel(). Load() is not available in D2D compute shaders.
• Output.GetDimensions(): Returns the tile size, not the full image.
• Constant buffer upload: Done in CalculateThreadgroups (not PrepareForRender) for correct per-tile values.

Shader Pattern

cbuffer Constants : register(b0) {
    int2 _TileOffset;  // Auto-injected per tile
    // User parameters here...
};
Texture2D Source : register(t0);
RWTexture2D<float4> Output : register(u0);
SamplerState Sampler0 : register(s0);

[numthreads(8, 8, 1)]
void main(uint3 DTid : SV_DispatchThreadID) {
    uint srcW, srcH;
    Source.GetDimensions(srcW, srcH);
    uint2 globalPos = DTid.xy + uint2(_TileOffset);
    if (globalPos.x >= srcW || globalPos.y >= srcH) return;
    
    float2 uv = (float2(globalPos) + 0.5) / float2(srcW, srcH);
    float4 color = Source.SampleLevel(Sampler0, uv, 0);
    Output[DTid.xy] = color;
}

---

Back to docs/ • Repo root