Loading src/Benchmarks/SpMV/spmv.h +1 −1 Original line number Diff line number Diff line Loading @@ -437,7 +437,7 @@ benchmarkSpMVCSRLight( BenchmarkType& benchmark, cudaMatrix.vectorProduct( cudaInVector, cudaOutVector ); }; cudaMatrix.getSegments().getKernel().setThreadsMapping( Algorithms::Segments::CSRLightAutomaticThreadsLightSpMV ); cudaMatrix.getSegments().getKernel().setThreadsMapping( Algorithms::Segments::CSRLightAutomaticThreads ); String format = MatrixInfo< HostMatrix >::getFormat(); SpmvBenchmarkResult< Real, Devices::Cuda, int > cudaBenchmarkResults( format, csrResultVector, cudaOutVector, cudaMatrix.getNonzeroElementsCount() ); benchmark.time< Devices::Cuda >( resetCudaVectors, "GPU", spmvCuda, cudaBenchmarkResults ); Loading src/TNL/Algorithms/Segments/Kernels/CSRLightKernel.h +1 −1 Original line number Diff line number Diff line Loading @@ -68,7 +68,7 @@ struct CSRLightKernel protected: LightCSRSThreadsMapping mapping = LightCSRConstantThreads; LightCSRSThreadsMapping mapping = CSRLightAutomaticThreads; int threadsPerSegment = 32; }; Loading Loading
src/Benchmarks/SpMV/spmv.h +1 −1 Original line number Diff line number Diff line Loading @@ -437,7 +437,7 @@ benchmarkSpMVCSRLight( BenchmarkType& benchmark, cudaMatrix.vectorProduct( cudaInVector, cudaOutVector ); }; cudaMatrix.getSegments().getKernel().setThreadsMapping( Algorithms::Segments::CSRLightAutomaticThreadsLightSpMV ); cudaMatrix.getSegments().getKernel().setThreadsMapping( Algorithms::Segments::CSRLightAutomaticThreads ); String format = MatrixInfo< HostMatrix >::getFormat(); SpmvBenchmarkResult< Real, Devices::Cuda, int > cudaBenchmarkResults( format, csrResultVector, cudaOutVector, cudaMatrix.getNonzeroElementsCount() ); benchmark.time< Devices::Cuda >( resetCudaVectors, "GPU", spmvCuda, cudaBenchmarkResults ); Loading
src/TNL/Algorithms/Segments/Kernels/CSRLightKernel.h +1 −1 Original line number Diff line number Diff line Loading @@ -68,7 +68,7 @@ struct CSRLightKernel protected: LightCSRSThreadsMapping mapping = LightCSRConstantThreads; LightCSRSThreadsMapping mapping = CSRLightAutomaticThreads; int threadsPerSegment = 32; }; Loading