Hello guys,

I would like to use MinMax filtering (D3D12_FILTER_MAXIMUM_MIN_MAG_MIP_POINT) in compute shader.

I am trying to compile compute shader (cs_5_0) and encounter an error: "error X4532: cannot map expression to cs_5_0 instruction set".

I tried to compile the shader in cs_6_0 mode and got "unrecognized compiler target cs_6_0". I do not really understand the error as cs_6_0 is supposed to be supported.

According to MSDN, D3D12_FILTER_MAXIMUM_MIN_MAG_MIP_POINT should "Fetch the same set of texels as D3D12_FILTER_MIN_MAG_MIP_POINT and instead of filtering them return the maximum of the texels. Texels that are weighted 0 during filtering aren't counted towards the maximum. You can query support for this filter type from the MinMaxFiltering member in the D3D11_FEATURE_DATA_D3D11_OPTIONS1 structure".

Not sure if this is valid documentation as it is talking about Direct3D 11. D3D12_FEATURE_DATA_D3D12_OPTIONS does not seem to provide this kind of check.

Direct3D device is created with feature level D3D_FEATURE_LEVEL_12_0 and I am using VS 2015.

Thanks!

Shader model 6 is a whole new compiler generating DXIL bytecode instead of DXBC, it is on github, you also need to turn on the experimental feature on a creator update or newer windows.

I don't think the sampler is the problem, are you passing it in a descriptor table or static sampler by the way ? 

Your real issue is that you call Sample instead of SampleLevel or SampleGrad and it is invalid in a Compute Shader because it can't produce the derivatives.

@galop1n Yep, you are right. I was using Sample instead of SampleLevel. The issue has been solved.

Do you know if MinMax filtering is supported by default in D3D12? How do you check if it is supported otherwise?

Thanks!

That's correct, Min/Max filtering is tied to TILED_RESOURCES_TIER_2 in D3D12.