diff options
| author | jsmall-nvidia <jsmall@nvidia.com> | 2020-02-18 14:14:16 -0500 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2020-02-18 14:14:16 -0500 |
| commit | 8ee39e08c48a315163fe1850dbb12ca292020d4d (patch) | |
| tree | 5041064a194849399aa587ac13b46db2088bdb05 /tools/render-test/cuda/cuda-compute-util.cpp | |
| parent | e109985375712b449d365450b3d3e39416a171ce (diff) | |
First pass Texture Array support on CUDA/CPU (#1225)
* Add cubemap support.
* Add CUDA fence instrinsics.
* Added Gather for CUDA.
* Use the CUDA driver API as much as possible.
* * Support 1D texture on CPU
* WIP on 1D texture on CUDA
* Added simplified texture test
* Fix test.
* Improve texture-simple tests.
* * Add CPU support for 3d textures
* Add support for mip maps to CUDA
* Disable warnings in nvrtc
* Update CUDA docs
* WIP on 3d texture support.
* Add support for 3d textures for CPU and CUDA.
* CPU and CUDA support for cube maps.
* Add CPU support for Texture1DArray.
* Support CUDA Layered/Array type in meta library.
Diffstat (limited to 'tools/render-test/cuda/cuda-compute-util.cpp')
| -rw-r--r-- | tools/render-test/cuda/cuda-compute-util.cpp | 52 |
1 files changed, 46 insertions, 6 deletions
diff --git a/tools/render-test/cuda/cuda-compute-util.cpp b/tools/render-test/cuda/cuda-compute-util.cpp index 59b5e65f6..779cfc96a 100644 --- a/tools/render-test/cuda/cuda-compute-util.cpp +++ b/tools/render-test/cuda/cuda-compute-util.cpp @@ -528,6 +528,7 @@ static SlangResult _compute(CUcontext context, CUmodule module, const ShaderComp CUDA_ARRAY3D_DESCRIPTOR arrayDesc; memset(&arrayDesc, 0, sizeof(arrayDesc)); + // If we have a cubemap the depth is 6 arrayDesc.Depth = depth; arrayDesc.Height = height; arrayDesc.Width = width; @@ -536,6 +537,12 @@ static SlangResult _compute(CUcontext context, CUmodule module, const ShaderComp arrayDesc.Flags = 0; + if (baseShape == SLANG_TEXTURE_CUBE) + { + arrayDesc.Depth = 6; + arrayDesc.Flags |= CUDA_ARRAY3D_CUBEMAP; + } + SLANG_CUDA_RETURN_ON_FAIL(cuArray3DCreate(&tex->m_cudaArray, &arrayDesc)); } else @@ -554,6 +561,9 @@ static SlangResult _compute(CUcontext context, CUmodule module, const ShaderComp } } + // Work space for holding data for uploading if it needs to be rearranged + List<uint8_t> workspace; + for (int mipLevel = 0; mipLevel < mipLevels; ++mipLevel) { int mipWidth = width >> mipLevel; @@ -564,6 +574,12 @@ static SlangResult _compute(CUcontext context, CUmodule module, const ShaderComp mipHeight = (mipHeight == 0) ? 1 : mipHeight; mipDepth = (mipDepth == 0) ? 1 : mipDepth; + // If it's a cubemap then the depth is always 6 + if (baseShape == SLANG_TEXTURE_CUBE) + { + mipDepth = 6; + } + auto dstArray = tex->m_cudaArray; if (tex->m_cudaMipMappedArray) { @@ -572,9 +588,6 @@ static SlangResult _compute(CUcontext context, CUmodule module, const ShaderComp } SLANG_ASSERT(dstArray); - const auto& srcData = texData.dataBuffer[mipLevel]; - - SLANG_ASSERT(mipWidth * mipHeight * mipDepth == srcData.getCount()); // Check using the desc to see if it's plausible { @@ -582,7 +595,34 @@ static SlangResult _compute(CUcontext context, CUmodule module, const ShaderComp SLANG_CUDA_RETURN_ON_FAIL(cuArrayGetDescriptor(&arrayDesc, dstArray)); SLANG_ASSERT(mipWidth == arrayDesc.Width); - SLANG_ASSERT(mipHeight == arrayDesc.Height); + SLANG_ASSERT(mipHeight == arrayDesc.Height || (mipHeight == 1 && arrayDesc.Height == 0)); + } + + const void* srcDataPtr = nullptr; + + if (baseShape == SLANG_TEXTURE_CUBE) + { + size_t faceSizeInBytes = elementSize * mipWidth * mipHeight; + + workspace.setCount(faceSizeInBytes * 6); + + // Copy the data over to make contiguous + for (Index j = 0; j < 6; j++) + { + const auto& srcData = texData.dataBuffer[mipLevels * j + mipLevel]; + SLANG_ASSERT(mipWidth * mipHeight == srcData.getCount()); + + ::memcpy(workspace.getBuffer() + faceSizeInBytes * j, srcData.getBuffer(), faceSizeInBytes); + } + + srcDataPtr = workspace.getBuffer(); + } + else + { + const auto& srcData = texData.dataBuffer[mipLevel]; + SLANG_ASSERT(mipWidth * mipHeight * mipDepth == srcData.getCount()); + + srcDataPtr = srcData.getBuffer(); } switch (baseShape) @@ -595,7 +635,7 @@ static SlangResult _compute(CUcontext context, CUmodule module, const ShaderComp copyParam.dstMemoryType = CU_MEMORYTYPE_ARRAY; copyParam.dstArray = dstArray; copyParam.srcMemoryType = CU_MEMORYTYPE_HOST; - copyParam.srcHost = srcData.getBuffer(); + copyParam.srcHost = srcDataPtr; copyParam.srcPitch = mipWidth * elementSize; copyParam.WidthInBytes = copyParam.srcPitch; copyParam.Height = mipHeight; @@ -612,7 +652,7 @@ static SlangResult _compute(CUcontext context, CUmodule module, const ShaderComp copyParam.dstArray = dstArray; copyParam.srcMemoryType = CU_MEMORYTYPE_HOST; - copyParam.srcHost = srcData.getBuffer(); + copyParam.srcHost = srcDataPtr; copyParam.srcPitch = mipWidth * elementSize; copyParam.WidthInBytes = copyParam.srcPitch; copyParam.Height = mipHeight; |