Commit 00409974 authored by Tomáš Oberhuber's avatar Tomáš Oberhuber Committed by Jakub Klinkovský
Browse files

Added both strategies of threads mapping in CSR Light to SpMV benchmark.

parent 47606aff

src/Benchmarks/SpMV/spmv.h

+13 −4
Original line number Diff line number Diff line
@@ -437,10 +437,19 @@ benchmarkSpMVCSRLight( BenchmarkType& benchmark, 
      cudaMatrix.vectorProduct( cudaInVector, cudaOutVector );

   };



   {

      cudaMatrix.getSegments().getKernel().setThreadsMapping( Algorithms::Segments::CSRLightAutomaticThreads );

   String format = MatrixInfo< HostMatrix >::getFormat();

      String format = MatrixInfo< HostMatrix >::getFormat() + " Automatic";

      SpmvBenchmarkResult< Real, Devices::Cuda, int > cudaBenchmarkResults( format, csrResultVector, cudaOutVector, cudaMatrix.getNonzeroElementsCount() );

      benchmark.time< Devices::Cuda >( resetCudaVectors, "GPU", spmvCuda, cudaBenchmarkResults );

   };



   {

      cudaMatrix.getSegments().getKernel().setThreadsMapping( Algorithms::Segments::CSRLightAutomaticThreadsLightSpMV );

      String format = MatrixInfo< HostMatrix >::getFormat() + " Automatic Light";

      SpmvBenchmarkResult< Real, Devices::Cuda, int > cudaBenchmarkResults( format, csrResultVector, cudaOutVector, cudaMatrix.getNonzeroElementsCount() );

      benchmark.time< Devices::Cuda >( resetCudaVectors, "GPU", spmvCuda, cudaBenchmarkResults );

   };



   /*for( auto threadsPerRow : std::vector< int >{ 1, 2, 4, 8, 16, 32 } )

   {