Commit 3ccd032e authored by Tomáš Oberhuber's avatar Tomáš Oberhuber
Browse files

C++ compiler log identification was improved. 

RowLengths was renamed to CompressedRowsLengths.
OpenMP support wads added to matrix-vector multiplication and to scalar
products.
CSR format can use Cusparse for SpMV.
parent 945ff662

AUTHORS

+6 −0
Original line number Diff line number Diff line 
Oberhuber Tomas <tomas.oberhuber@fjfi.cvut.cz>

Zabka Vitezslav <zabkavit@fjfi.cvut.cz>

Vladimir Klement

Tomáš Sobotík

Ondřej Székely

Jiří Kafka

Libor Bakajsa

Jakub Klinkovský

Vacata Jan

Heller Martin

Novotny Matej

CMakeLists.txt

+5 −10
Original line number Diff line number Diff line
@@ -41,13 +41,12 @@ if( WITH_CUDA STREQUAL "yes" ) 
        set(CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE OFF)

        set(BUILD_SHARED_LIBS ON)

        set(CUDA_SEPARABLE_COMPILATION ON)

        set(CUSPARSE_LIBRARY /usr/local/cuda/lib64/libcusparse.so) # TODO: fix this              

        set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} ; -DHAVE_CUDA )

        AddCompilerFlag( "-DHAVE_NOT_CXX11" ) # -U_GLIBCXX_ATOMIC_BUILTINS -U_GLIBCXX_USE_INT128 " )

        set( ALL_CUDA_ARCHS -gencode arch=compute_20,code=sm_20

                            -gencode arch=compute_30,code=sm_30

                            -gencode arch=compute_32,code=sm_32 

                            -gencode arch=compute_37,code=sm_37 

                            -gencode arch=compute_35,code=sm_35 

                            -gencode arch=compute_37,code=sm_37 

                            -gencode arch=compute_50,code=sm_50 

                            -gencode arch=compute_52,code=sm_52 )
@@ -87,7 +86,6 @@ if( WITH_CUDA STREQUAL "yes" ) 
        set( CUDA_ADD_LIBRARY_OPTIONS ${CUDA_ARCH} -shared )

        set( CUDA_LINKER_OPTIONS "-arch sm_20 -shared" )





        ####

        # Check for cuBLAS

        #
@@ -127,6 +125,7 @@ if( WITH_CUDA STREQUAL "yes" ) 
               message( "CUSPARSE found. -- ${CUSPARSE_INCLUDE_DIR}" )

               set( HAVE_CUSPARSE "#define HAVE_CUSPARSE" )

               cuda_include_directories( ${CUSPARSE_INCLUDE_DIR} )

               set( CUSPARSE_LIBRARY "${CUDA_cusparse_LIBRARY}" )

           endif()            

        endif( NOT WITH_CUSPARSE STREQUAL "no" )
@@ -143,9 +142,8 @@ endif( WITH_CUDA STREQUAL "yes" ) 
#

find_package( OpenMP ) 

if( OPENMP_FOUND )

   #AddCompilerFlag( "-DHAVE_OPENMP -fopenmp" )

   message( "Compiler supports OpenMP." )

   set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DHAVE_OPENMP -fopenmp")

# TODO: finish this

endif()



####
@@ -184,9 +182,6 @@ else() 
    set( HAVE_SYS_RESOURCE_H "#define HAVE_SYS_RESOURCE_H" )

endif()



  





####

# Check for cppunit

#

examples/heat-equation/CMakeLists.txt

+4 −2
Original line number Diff line number Diff line
@@ -5,13 +5,15 @@ set( tnl_heat_equation_SOURCES 
IF( BUILD_CUDA )

   CUDA_ADD_EXECUTABLE(tnl-heat-equation${debugExt} tnl-heat-equation.cu)

   CUDA_ADD_EXECUTABLE(tnl-heat-equation-eoc-test${debugExt} tnl-heat-equation-eoc.cu)

   target_link_libraries (tnl-heat-equation${debugExt} tnl${debugExt}-${tnlVersion}  ${CUSPARSE_LIBRARY} )

   target_link_libraries (tnl-heat-equation-eoc-test${debugExt} tnl${debugExt}-${tnlVersion}  ${CUSPARSE_LIBRARY} )

ELSE(  BUILD_CUDA )               

   ADD_EXECUTABLE(tnl-heat-equation${debugExt} tnl-heat-equation.cpp)     

   ADD_EXECUTABLE(tnl-heat-equation-eoc-test${debugExt} tnl-heat-equation-eoc.cpp)   

ENDIF( BUILD_CUDA )



   target_link_libraries (tnl-heat-equation${debugExt} tnl${debugExt}-${tnlVersion} )

   target_link_libraries (tnl-heat-equation-eoc-test${debugExt} tnl${debugExt}-${tnlVersion} )

ENDIF( BUILD_CUDA )





INSTALL( TARGETS tnl-heat-equation${debugExt}

                 tnl-heat-equation-eoc-test${debugExt}

install

+1 −2
Original line number Diff line number Diff line
@@ -6,7 +6,6 @@ BUILD_RELEASE="yes" 
OPTIONS=""



CMAKE_TEST=`which cmake`    

echo ${CMAKE_TEST}

if test x${CMAKE_TEST} = "x";

then

    echo "Cmake is not installed on your system. Please install it by:"

src/core/cuda/cuda-reduction_impl.h

+4 −1
Original line number Diff line number Diff line
@@ -44,7 +44,7 @@ using namespace std; 
 */

const int minGPUReductionDataSize = 128;//65536; //16384;//1024;//256;



static tnlCudaReductionBuffer cudaReductionBuffer( 8 * minGPUReductionDataSize );

//static tnlCudaReductionBuffer cudaReductionBuffer( 8 * minGPUReductionDataSize );



#ifdef HAVE_CUDA
@@ -74,6 +74,9 @@ typename Operation::IndexType reduceOnCudaDevice( const Operation& operation, 
   dim3 blockSize( 256 ), gridSize( 0 );   

   gridSize.x = Min( tnlCuda::getNumberOfBlocks( size, blockSize.x ), desGridSize );

  

   // create reference to the reduction buffer singleton and set default size

   tnlCudaReductionBuffer & cudaReductionBuffer = tnlCudaReductionBuffer::getInstance( 8 * minGPUReductionDataSize );

   

   //tnlCudaReductionBuffer cudaReductionBuffer( 8 * minGPUReductionDataSize );

   if( ! cudaReductionBuffer.setSize( gridSize.x * sizeof( ResultType ) ) )

      return false;