Loading tests/benchmarks/benchmarks.h +1 −6 Original line number Diff line number Diff line Loading @@ -62,13 +62,8 @@ benchmarkCuda( const int & loops, CheckFunction check = trueFunc, ResetFunction reset = voidFunc ) { // timers are constructed zero-initialized and stopped tnlTimerRT timerHost, timerCuda, timerCudaSync; timerHost.reset(); timerHost.stop(); timerCuda.reset(); timerCuda.stop(); timerCudaSync.reset(); timerCudaSync.stop(); for(int i = 0; i < loops; ++i) { timerHost.start(); Loading tests/benchmarks/tnl-cuda-benchmarks.h +0 −15 Original line number Diff line number Diff line Loading @@ -28,26 +28,11 @@ using namespace tnl::benchmarks; #ifdef HAVE_CUBLAS //#include <cublas.h> #endif // silly alias to match the number of template parameters with other formats template< typename Real, typename Device, typename Index > using SlicedEllpackMatrix = tnlSlicedEllpackMatrix< Real, Device, Index >; // TODO: // check operations with the timer: // - reset() clears the timer and starts it again // - getTime() stops the timer and starts it again !!! // - data members are not zero-initialized - reset has to be called manually, but it immediately starts the timer // FIXME: // - scalarProduct is not const method // - cudaThreadSynchronize() should be called from all CUDA methods template< typename Matrix > int setHostTestMatrix( Matrix& matrix, const int elementsPerRow ) Loading tests/benchmarks/vector-operations.h +4 −0 Original line number Diff line number Diff line Loading @@ -4,6 +4,10 @@ #include <core/vectors/tnlVector.h> #ifdef HAVE_CUBLAS //#include <cublas.h> #endif namespace tnl { namespace benchmarks Loading Loading
tests/benchmarks/benchmarks.h +1 −6 Original line number Diff line number Diff line Loading @@ -62,13 +62,8 @@ benchmarkCuda( const int & loops, CheckFunction check = trueFunc, ResetFunction reset = voidFunc ) { // timers are constructed zero-initialized and stopped tnlTimerRT timerHost, timerCuda, timerCudaSync; timerHost.reset(); timerHost.stop(); timerCuda.reset(); timerCuda.stop(); timerCudaSync.reset(); timerCudaSync.stop(); for(int i = 0; i < loops; ++i) { timerHost.start(); Loading
tests/benchmarks/tnl-cuda-benchmarks.h +0 −15 Original line number Diff line number Diff line Loading @@ -28,26 +28,11 @@ using namespace tnl::benchmarks; #ifdef HAVE_CUBLAS //#include <cublas.h> #endif // silly alias to match the number of template parameters with other formats template< typename Real, typename Device, typename Index > using SlicedEllpackMatrix = tnlSlicedEllpackMatrix< Real, Device, Index >; // TODO: // check operations with the timer: // - reset() clears the timer and starts it again // - getTime() stops the timer and starts it again !!! // - data members are not zero-initialized - reset has to be called manually, but it immediately starts the timer // FIXME: // - scalarProduct is not const method // - cudaThreadSynchronize() should be called from all CUDA methods template< typename Matrix > int setHostTestMatrix( Matrix& matrix, const int elementsPerRow ) Loading
tests/benchmarks/vector-operations.h +4 −0 Original line number Diff line number Diff line Loading @@ -4,6 +4,10 @@ #include <core/vectors/tnlVector.h> #ifdef HAVE_CUBLAS //#include <cublas.h> #endif namespace tnl { namespace benchmarks Loading