From 813892cd023e216f6f6560eb47566522d3a82609 Mon Sep 17 00:00:00 2001 From: Yong He Date: Thu, 13 Apr 2023 09:49:22 -0700 Subject: Set sharedMem argument to 0 when launching cuda kernel. (#2799) Co-authored-by: Yong He --- tools/gfx/cuda/cuda-command-queue.cpp | 8 +------- 1 file changed, 1 insertion(+), 7 deletions(-) (limited to 'tools/gfx/cuda/cuda-command-queue.cpp') diff --git a/tools/gfx/cuda/cuda-command-queue.cpp b/tools/gfx/cuda/cuda-command-queue.cpp index 0c17a418e..4b0ab7d94 100644 --- a/tools/gfx/cuda/cuda-command-queue.cpp +++ b/tools/gfx/cuda/cuda-command-queue.cpp @@ -93,12 +93,6 @@ void CommandQueueImpl::dispatchCompute(int x, int y, int z) UInt threadGroupSize[3]; programLayout->getKernelThreadGroupSize(kernelId, threadGroupSize); - int sharedSizeInBytes; - cuFuncGetAttribute( - &sharedSizeInBytes, - CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES, - currentPipeline->shaderProgram->cudaKernel); - // Copy global parameter data to the `SLANG_globalParams` symbol. { CUdeviceptr globalParamsSymbol = 0; @@ -144,7 +138,7 @@ void CommandQueueImpl::dispatchCompute(int x, int y, int z) int(threadGroupSize[0]), int(threadGroupSize[1]), int(threadGroupSize[2]), - sharedSizeInBytes, + 0, stream, nullptr, extraOptions); -- cgit v1.2.3