Basic CUDA benchmark was added.

tests/benchmarks/CMakeLists.txt

 ADD_SUBDIRECTORY( share )
 
 IF( BUILD_CUDA )
+    CUDA_ADD_EXECUTABLE( tnl-cuda-benchmarks${debugExt} tnl-cuda-benchmarks.cu
+                         OPTIONS ${CUDA_ADD_EXECUTABLE_OPTIONS} )
+    SET_TARGET_PROPERTIES( tnl-cuda-benchmarks${debugExt} PROPERTIES CUDA_COMPILE_FLAGS "${CXX_OPTIMIZE_FLAGS}" )
+    TARGET_LINK_LIBRARIES( tnl-cuda-benchmarks${debugExt} tnl${debugExt}-${tnlVersion} ${CUSPARSE_LIBRARY} )                        
     CUDA_ADD_EXECUTABLE( tnl-benchmark-spmv${debugExt} tnl-benchmark-spmv.cu
                          OPTIONS ${CUDA_ADD_EXECUTABLE_OPTIONS} )
     SET_TARGET_PROPERTIES( tnl-benchmark-spmv${debugExt} PROPERTIES CUDA_COMPILE_FLAGS "${CXX_OPTIMIZE_FLAGS}" )
@@ -18,7 +22,12 @@ ELSE()
 ENDIF()
 
 TARGET_LINK_LIBRARIES( tnl-benchmark-linear-solvers${debugExt} tnl${debugExt}-${tnlVersion} )
-                                                              
+
+if( BUILD_CUDA )                                                              
+   INSTALL( TARGETS tnl-cuda-benchmarks${debugExt}
+            RUNTIME DESTINATION bin )
+endif()
+
 INSTALL( TARGETS tnl-benchmark-spmv${debugExt}
                  tnl-benchmark-linear-solvers${debugExt}
          RUNTIME DESTINATION bin )

tests/benchmarks/tnl-benchmarks.cpp → tests/benchmarks/tnl-cuda-benchmarks.cu

 /***************************************************************************
-                          tnl-benchmarks.cpp  -  description
+                          tnl-cuda-benchmarks.cu  -  description
                              -------------------
-    begin                : Nov 25, 2010
-    copyright            : (C) 2010 by Tomas Oberhuber
+    begin                : May 28, 2015
+    copyright            : (C) 2015 by Tomas Oberhuber
     email                : tomas.oberhuber@fjfi.cvut.cz
  ***************************************************************************/
 
@@ -15,43 +15,38 @@
  *                                                                         *
  ***************************************************************************/
 
-
-#include <core/vectors/tnlVectorHost.h>
-#include <core/vectors/tnlVectorCUDA.h>
-#include <tnl-benchmarks.h>
-
+#include <core/vectors/tnlVector.h>
+#include <core/tnlTimerRT.h>
 
 int main( int argc, char* argv[] )
 {
-
 #ifdef HAVE_CUDA
-   cout << "Benchmarking memory bandwidth when transfering int ..." << endl;
-
-   const int size = 1 << 22;
-   double host_to_host_band_width;
-   double host_to_device_band_width;
-   double device_to_host_band_width;
-   double device_to_device_band_width;
 
-   transferBenchmark< int >( size,
-                             host_to_host_band_width,
-                             host_to_device_band_width,
-                             device_to_host_band_width,
-                             device_to_device_band_width );
+    tnlTimerRT timer;
+    const double oneGB = 1024.0 * 1024.0 * 1024.0;
 
+    cout << "Benchmarking memory bandwidth when transfering int ..." << endl;
 
-   cout << "Benchmarking reduction of int ..." << endl;
-   for( int i = 0; i <= 6; i ++ )
-      reductionBenchmark< int >( size, i );
+    const int size = 1 << 22;
+    
+    tnlVector< int, tnlHost > hostVector;
+    tnlVector< int, tnlCuda > deviceVector;
+    hostVector.setSize( size );
+    deviceVector.setSize( size );
 
-   cout << "Benchmarking reduction of float ..." << endl;
-   for( int i = 0; i <= 6; i ++ )
-      reductionBenchmark< float >( size, i );
+    hostVector.setValue( 1.0 );
+    deviceVector.setValue( 0.0 );
+    
+    timer.reset();
+    timer.start();
+    deviceVector = hostVector;
+    timer.stop();
+    
+    double bandwidth = ( double ) ( size ) * sizeof( int ) / timer.getTime() / oneGB;
 
-   cout << "Benchmarking reduction of double ..." << endl;
-   for( int i = 0; i <= 6; i ++ )
-      reductionBenchmark< double >( size / 2, i );
+    cout << bandwidth << " GB/sec." << endl;
 
+   
 #endif
    return EXIT_SUCCESS;
 }