Commit 345e290c authored by Tomáš Oberhuber's avatar Tomáš Oberhuber Committed by Jakub Klinkovský
Browse files

Fixing SpMV benchmark with CUDA.

parent 2b69cfe1

src/Benchmarks/BLAS/tnl-benchmark-blas.h

+2 −2
Original line number Diff line number Diff line
@@ -52,7 +52,7 @@ runBlasBenchmarks( Benchmark<> & benchmark, 
   benchmark.newBenchmark( String("Array operations (") + precision + ", host allocator = CudaHost)",

                           metadata );

   for( std::size_t size = minSize; size <= maxSize; size *= 2 ) {

      benchmark.setMetadataColumns( Benchmark::MetadataColumns({

      benchmark.setMetadataColumns( Benchmark<>::MetadataColumns({

         { "size", convertToString( size ) },

      } ));

      benchmarkArrayOperations< Real, int, Allocators::CudaHost >( benchmark, size );
@@ -60,7 +60,7 @@ runBlasBenchmarks( Benchmark<> & benchmark, 
   benchmark.newBenchmark( String("Array operations (") + precision + ", host allocator = CudaManaged)",

                           metadata );

   for( std::size_t size = minSize; size <= maxSize; size *= 2 ) {

      benchmark.setMetadataColumns( Benchmark::MetadataColumns({

      benchmark.setMetadataColumns( Benchmark<>::MetadataColumns({

         { "size", convertToString( size ) },

      } ));

      benchmarkArrayOperations< Real, int, Allocators::CudaManaged >( benchmark, size );

src/Benchmarks/SpMV/spmv.h

+1 −1
Original line number Diff line number Diff line
@@ -479,7 +479,7 @@ benchmarkSpmv( BenchmarkType& benchmark, 
       cusparseMatrix.vectorProduct( cudaInVector, cudaOutVector );

   };



   SpmvBenchmarkResult< Real, Devices::Cuda, int > cudaBenchmarkResults( String( "cusprase" ), hostOutVector, hostOutVector, csrHostMatrix.getNonzeroElementsCount() );

   SpmvBenchmarkResult< Real, Devices::Cuda, int > cudaBenchmarkResults( String( "cusparse" ), hostOutVector, cudaOutVector, csrHostMatrix.getNonzeroElementsCount() );

   benchmark.time< Devices::Cuda >( resetCusparseVectors, "GPU", spmvCusparse, cudaBenchmarkResults );



#ifdef HAVE_CSR5

src/Benchmarks/SpMV/tnl-benchmark-spmv.h

+5 −5
Original line number Diff line number Diff line
@@ -32,8 +32,8 @@ using namespace TNL::Benchmarks; 


template< typename Real >

void

runSpMVBenchmarks( SpMV::BenchmarkType & benchmark,

                   SpMV::BenchmarkType::MetadataMap metadata,

runSpMVBenchmarks( TNL::Benchmarks::SpMV::BenchmarkType & benchmark,

                   TNL::Benchmarks::SpMV::BenchmarkType::MetadataMap metadata,

                   const String & inputFileName,

                   const Config::ParameterContainer& parameters,

                   bool verboseMR = false )
@@ -46,7 +46,7 @@ runSpMVBenchmarks( SpMV::BenchmarkType & benchmark, 
                           metadata );

   // Start the actual benchmark in spmv.h

   try {

      SpMV::benchmarkSpmv< Real >( benchmark, inputFileName, parameters, verboseMR );

      TNL::Benchmarks::SpMV::benchmarkSpmv< Real >( benchmark, inputFileName, parameters, verboseMR );

   }

   catch( const std::exception& ex ) {

      std::cerr << ex.what() << std::endl;
@@ -133,10 +133,10 @@ main( int argc, char* argv[] ) 
   std::ofstream logFile( logFileName.getString(), mode );



   // init benchmark and common metadata

   SpMV::BenchmarkType benchmark( loops, verbose, outputMode );

   TNL::Benchmarks::SpMV::BenchmarkType benchmark( loops, verbose, outputMode );



   // prepare global metadata

   SpMV::BenchmarkType::MetadataMap metadata = getHardwareMetadata< Logging >();

   TNL::Benchmarks::SpMV::BenchmarkType::MetadataMap metadata = getHardwareMetadata< Logging >();



   // Initiate setup of benchmarks

   if( precision == "all" || precision == "float" )