diff options
| author | Yong He <yonghe@outlook.com> | 2021-06-08 07:44:05 -0700 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2021-06-08 10:44:05 -0400 |
| commit | 6cee1eeda28c1ce1e5d326a0c43427b4776a1d09 (patch) | |
| tree | 6e4559e48fecf5f71aece8b128184925b2d0f790 | |
| parent | fb50fab76a723f46026474ea5bb0226c297d1fd5 (diff) | |
Various fixes to CUDA backend. (#1877)
- Fix emitting `StructuredBuffer<ISomething>::Load`, which triggers emitting for `IROp_WrapExistential` that is previously unhandled.
- Fix cuda layout around vectors, they should be aligned to 1,2,4,8,16 bytes instead of just using element type's alignment. That means `float4` has alignment of 16 instead of 4.
- Fix `SLANG_CUDA_HANDLE_ERROR` macro definition.
- Fix navis sometimes fail to find `Slang::kIROp_*` enum values when debugging external projects.
Co-authored-by: Yong He <yhe@nvidia.com>
Co-authored-by: jsmall-nvidia <jsmall@nvidia.com>
| -rw-r--r-- | source/slang/slang-emit-c-like.cpp | 11 | ||||
| -rw-r--r-- | source/slang/slang-ir.cpp | 12 | ||||
| -rw-r--r-- | source/slang/slang-type-layout.cpp | 16 | ||||
| -rw-r--r-- | source/slang/slang.natvis | 22 | ||||
| -rw-r--r-- | tools/gfx/cuda/render-cuda.cpp | 21 |
5 files changed, 61 insertions, 21 deletions
diff --git a/source/slang/slang-emit-c-like.cpp b/source/slang/slang-emit-c-like.cpp index 1b0fe7c44..47c584251 100644 --- a/source/slang/slang-emit-c-like.cpp +++ b/source/slang/slang-emit-c-like.cpp @@ -1794,6 +1794,17 @@ void CLikeSourceEmitter::defaultEmitInstExpr(IRInst* inst, const EmitOpInfo& inO } break; + case kIROp_WrapExistential: + { + // Normally `WrapExistential` shouldn't exist in user code at this point. + // The only exception is when the user is calling a stdlib generic + // function that has an existential type argument, for example + // `StructuredBuffer<ISomething>.Load()`. + // We can safely ignore the `wrapExistential` operation in this case. + emitOperand(inst->getOperand(0), outerPrec); + } + break; + case kIROp_Select: { diff --git a/source/slang/slang-ir.cpp b/source/slang/slang-ir.cpp index 1f68623bb..90d0181f7 100644 --- a/source/slang/slang-ir.cpp +++ b/source/slang/slang-ir.cpp @@ -6019,3 +6019,15 @@ namespace Slang } } // namespace Slang +#if SLANG_VC +#ifdef _DEBUG +// Natvis sometimes cannot find enum values. +// Export symbols for them to make sure natvis works correctly when debugging external projects. +SLANG_API const int __SlangIROpNameHint = Slang::kIROp_NameHintDecoration; +SLANG_API const int __SlangIROpExport = Slang::kIROp_ExportDecoration; +SLANG_API const int __SlangIROpImport = Slang::kIROp_ImportDecoration; +SLANG_API const int __SlangIROpStringLit = Slang::kIROp_StringLit; +SLANG_API const int __SlangIROpIntLit = Slang::kIROp_IntLit; +#endif +#endif + diff --git a/source/slang/slang-type-layout.cpp b/source/slang/slang-type-layout.cpp index 2568547a4..02b5b5bdc 100644 --- a/source/slang/slang-type-layout.cpp +++ b/source/slang/slang-type-layout.cpp @@ -465,6 +465,18 @@ struct CUDALayoutRulesImpl : DefaultLayoutRulesImpl return arrayInfo; } + // Given `size` between [0, 16] return the smallest power-of-2 that is greater than or equal to `size`. + uint32_t getVectorAlignment(uint32_t size) + { + SLANG_ASSERT(size <= 16); + --size; + // Set every bit after the highest bit. + size |= (size >> 1); + size |= (size >> 2); + ++size; + return size; + } + SimpleLayoutInfo GetVectorLayout(BaseType elementType, SimpleLayoutInfo elementInfo, size_t elementCount) override { // Special case bool @@ -479,8 +491,8 @@ struct CUDALayoutRulesImpl : DefaultLayoutRulesImpl SimpleLayoutInfo vectorInfo; vectorInfo.kind = elementInfo.kind; vectorInfo.size = elementInfo.size * elementCount; - vectorInfo.alignment = elementInfo.alignment; - + vectorInfo.alignment = getVectorAlignment( + (uint32_t)(elementInfo.size.getFiniteValue() * elementCount)); return vectorInfo; } diff --git a/source/slang/slang.natvis b/source/slang/slang.natvis index ea010c187..5af533715 100644 --- a/source/slang/slang.natvis +++ b/source/slang/slang.natvis @@ -83,20 +83,20 @@ <If Condition="child == 0"> <Break/> </If> - <If Condition="child->m_op == Slang::kIROp_NameHintDecoration"> + <If Condition="child->m_op == __SlangIROpNameHint"> <Item Name="[name]">((Slang::IRStringLit*)(((Slang::IRUse*)(child + 1))->usedValue))->value.stringVal.chars,[((Slang::IRStringLit*)(((Slang::IRUse*)(child + 1))->usedValue))->value.stringVal.numChars]s8</Item> </If> - <If Condition="child->m_op == Slang::kIROp_ExportDecoration"> + <If Condition="child->m_op == __SlangIROpExport"> <Item Name="[exportName]">((Slang::IRStringLit*)(((Slang::IRUse*)(child + 1))->usedValue))->value.stringVal.chars,[((Slang::IRStringLit*)(((Slang::IRUse*)(child + 1))->usedValue))->value.stringVal.numChars]s8</Item> </If> - <If Condition="child->m_op == Slang::kIROp_ImportDecoration"> + <If Condition="child->m_op == __SlangIROpImport"> <Item Name="[importName]">((Slang::IRStringLit*)(((Slang::IRUse*)(child + 1))->usedValue))->value.stringVal.chars,[((Slang::IRStringLit*)(((Slang::IRUse*)(child + 1))->usedValue))->value.stringVal.numChars]s8</Item> </If> <Exec>child = child->next</Exec> </Loop> </CustomListItems> - <Item Name="[value]" Condition="m_op == Slang::kIROp_StringLit">((IRStringLit*)this)->value.stringVal.chars,[((IRStringLit*)this)->value.stringVal.numChars]s8</Item> - <Item Name="[value]" Condition="m_op == Slang::kIROp_IntLit">((IRIntLit*)this)->value.intVal</Item> + <Item Name="[value]" Condition="m_op == __SlangIROpStringLit">((IRStringLit*)this)->value.stringVal.chars,[((IRStringLit*)this)->value.stringVal.numChars]s8</Item> + <Item Name="[value]" Condition="m_op == __SlangIROpIntLit">((IRIntLit*)this)->value.intVal</Item> <!-- <Synthetic Name="[operands]"> <DisplayString>{{count = {operandCount}}}</DisplayString> @@ -121,14 +121,14 @@ <Exec>child = pOperandInst->m_decorationsAndChildren.first</Exec> <Exec>nameDecoration = 0</Exec> <Loop Condition="child != 0"> - <If Condition="child->m_op == Slang::kIROp_NameHintDecoration"> + <If Condition="child->m_op == __SlangIROpNameHint"> <Exec>nameDecoration = child</Exec> <Break/> </If> - <If Condition="child->m_op == Slang::kIROp_ExportDecoration && (nameDecoration == 0 || nameDecoration->m_op != Slang::kIROp_NameHintDecoration)"> + <If Condition="child->m_op == __SlangIROpExport && (nameDecoration == 0 || nameDecoration->m_op != __SlangIROpNameHint)"> <Exec>nameDecoration = child</Exec> </If> - <If Condition="child->m_op == Slang::kIROp_ImportDecoration && (nameDecoration == 0 || nameDecoration->m_op != Slang::kIROp_NameHintDecoration)"> + <If Condition="child->m_op == __SlangIROpImport && (nameDecoration == 0 || nameDecoration->m_op != __SlangIROpNameHint)"> <Exec>nameDecoration = child</Exec> </If> <Exec>child = child->next</Exec> @@ -150,14 +150,14 @@ <Exec>child = pItem->m_decorationsAndChildren.first </Exec> <Exec>nameDecoration = 0</Exec> <Loop Condition="child != 0"> - <If Condition="child->m_op == Slang::kIROp_NameHintDecoration"> + <If Condition="child->m_op == __SlangIROpNameHint"> <Exec>nameDecoration = child</Exec> <Break/> </If> - <If Condition="child->m_op == Slang::kIROp_ExportDecoration && (nameDecoration == 0 || nameDecoration->m_op != Slang::kIROp_NameHintDecoration)"> + <If Condition="child->m_op == __SlangIROpExport && (nameDecoration == 0 || nameDecoration->m_op != __SlangIROpNameHint)"> <Exec>nameDecoration = child</Exec> </If> - <If Condition="child->m_op == Slang::kIROp_ImportDecoration && (nameDecoration == 0 || nameDecoration->m_op != Slang::kIROp_NameHintDecoration)"> + <If Condition="child->m_op == __SlangIROpImport && (nameDecoration == 0 || nameDecoration->m_op != __SlangIROpNameHint)"> <Exec>nameDecoration = child</Exec> </If> <Exec>child = child->next</Exec> diff --git a/tools/gfx/cuda/render-cuda.cpp b/tools/gfx/cuda/render-cuda.cpp index 64da1721d..ed7f44ed2 100644 --- a/tools/gfx/cuda/render-cuda.cpp +++ b/tools/gfx/cuda/render-cuda.cpp @@ -98,7 +98,7 @@ static SlangResult _handleCUDAError(cudaError_t error, const char* file, int lin return CUDAErrorInfo(file, line, cudaGetErrorName(error), cudaGetErrorString(error)).handle(); } -# define SLANG_CUDA_HANDLE_ERROR(x) _handleCUDAError(_res, __FILE__, __LINE__) +# define SLANG_CUDA_HANDLE_ERROR(x) _handleCUDAError(x, __FILE__, __LINE__) # define SLANG_CUDA_RETURN_ON_FAIL(x) \ { \ @@ -431,7 +431,7 @@ public: Slang::RefPtr<MemoryCUDAResource> m_bufferResource; Slang::RefPtr<CUDAResourceView> m_bufferView; Slang::List<uint8_t> m_cpuBuffer; - void setCount(Index count) + Result setCount(Index count) { if (isHostOnly) { @@ -444,7 +444,7 @@ public: m_bufferView->proxyBuffer = m_cpuBuffer.getBuffer(); m_bufferView->desc = viewDesc; } - return; + return SLANG_OK; } if (!m_bufferResource) @@ -454,7 +454,9 @@ public: desc.sizeInBytes = count; m_bufferResource = new MemoryCUDAResource(desc); if (count) - cudaMalloc(&m_bufferResource->m_cudaMemory, (size_t)count); + { + SLANG_CUDA_RETURN_ON_FAIL(cudaMalloc(&m_bufferResource->m_cudaMemory, (size_t)count)); + } IResourceView::Desc viewDesc = {}; viewDesc.type = IResourceView::Type::UnorderedAccess; m_bufferView = new CUDAResourceView(); @@ -467,20 +469,21 @@ public: void* newMemory = nullptr; if (count) { - cudaMalloc(&newMemory, (size_t)count); + SLANG_CUDA_RETURN_ON_FAIL(cudaMalloc(&newMemory, (size_t)count)); } if (oldSize) { - cudaMemcpy( + SLANG_CUDA_RETURN_ON_FAIL(cudaMemcpy( newMemory, m_bufferResource->m_cudaMemory, Math::Min((size_t)count, oldSize), - cudaMemcpyDefault); + cudaMemcpyDefault)); } cudaFree(m_bufferResource->m_cudaMemory); m_bufferResource->m_cudaMemory = newMemory; m_bufferResource->getDesc()->sizeInBytes = count; } + return SLANG_OK; } Slang::Index getCount() @@ -1021,7 +1024,9 @@ public: virtual SLANG_NO_THROW void SLANG_MCALL wait() override { - cuStreamSynchronize(stream); + auto resultCode = cuStreamSynchronize(stream); + if (resultCode != cudaSuccess) + SLANG_CUDA_HANDLE_ERROR(resultCode); } public: |
