Loading src/TNL/Containers/Algorithms/ArrayOperationsCuda.hpp +8 −6 Original line number Diff line number Diff line Loading @@ -112,6 +112,7 @@ setMemory( Element* data, Index blocksNumber = TNL::ceil( ( double ) size / ( double ) blockSize.x ); gridSize.x = TNL::min( blocksNumber, Devices::Cuda::getMaxGridSize() ); setArrayValueCudaKernel<<< gridSize, blockSize >>>( data, size, value ); cudaStreamSynchronize(0); TNL_CHECK_CUDA_DEVICE; #else throw Exceptions::CudaSupportMissing(); Loading Loading @@ -164,6 +165,7 @@ copyMemory( DestinationElement* destination, Index blocksNumber = TNL::ceil( ( double ) size / ( double ) blockSize.x ); gridSize.x = min( blocksNumber, Devices::Cuda::getMaxGridSize() ); copyMemoryCudaToCudaKernel<<< gridSize, blockSize >>>( destination, source, size ); cudaStreamSynchronize(0); TNL_CHECK_CUDA_DEVICE; } #else Loading Loading
src/TNL/Containers/Algorithms/ArrayOperationsCuda.hpp +8 −6 Original line number Diff line number Diff line Loading @@ -112,6 +112,7 @@ setMemory( Element* data, Index blocksNumber = TNL::ceil( ( double ) size / ( double ) blockSize.x ); gridSize.x = TNL::min( blocksNumber, Devices::Cuda::getMaxGridSize() ); setArrayValueCudaKernel<<< gridSize, blockSize >>>( data, size, value ); cudaStreamSynchronize(0); TNL_CHECK_CUDA_DEVICE; #else throw Exceptions::CudaSupportMissing(); Loading Loading @@ -164,6 +165,7 @@ copyMemory( DestinationElement* destination, Index blocksNumber = TNL::ceil( ( double ) size / ( double ) blockSize.x ); gridSize.x = min( blocksNumber, Devices::Cuda::getMaxGridSize() ); copyMemoryCudaToCudaKernel<<< gridSize, blockSize >>>( destination, source, size ); cudaStreamSynchronize(0); TNL_CHECK_CUDA_DEVICE; } #else Loading