Commit 8782dd42 authored by Tomáš Oberhuber's avatar Tomáš Oberhuber
Browse files

Modifying the matrix assembler for stationary problems. 

Adding --fast-build optin for the build script to speed-up compilation.
parent 9b588df4

CMakeLists.txt

+42 −7
Original line number Diff line number Diff line
@@ -18,7 +18,8 @@ if( CMAKE_BUILD_TYPE STREQUAL "Debug") 
    set( LIBRARY_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/Debug/lib )

    set( EXECUTABLE_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/Debug/bin )

    set( debugExt -dbg )

    AddCompilerFlag( "-g" )

    set( CXX_FLAGS "${CXXFLAGS} -g ")

    #AddCompilerFlag( "-g" )

else()

    set( PROJECT_BUILD_PATH ${PROJECT_SOURCE_DIR}/Release/src )

    set( PROJECT_TESTS_PATH ${PROJECT_SOURCE_DIR}/Release/tests )
@@ -28,10 +29,6 @@ else() 
    AddCompilerFlag( "-O3 -march=native -DNDEBUG" )

endif()



if( WITH_TEMPLATE_EXPLICIT_INSTANTIATION STREQUAL "yes" )

   AddCompilerFlag( "-DTEMPLATE_EXPLICIT_INSTANTIATION " )

endif()   



#####

# Check for CUDA

#
@@ -44,10 +41,20 @@ if( WITH_CUDA STREQUAL "yes" ) 
        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 " )          

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

        set( CUDA_ADD_EXECUTABLE_OPTIONS -gencode arch=compute_20,code=sm_20 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 )

        set( CUDA_ADD_LIBRARY_OPTIONS -gencode arch=compute_20,code=sm_20 -gencode arch=compute_30,code=sm_30 -gencode arch=compute_35,code=sm_35 -shared )

        set( CUDA_LINKER_OPTIONS "-arch sm_20 -shared " )



        ####

        # Check for cuBLAS

        #

        if( WITH_CUBLAS STREQUAL "yes" ) 

            message( "Enabling CUBLAS." )

            set( HAVE_CUBLAS TRUE)

            set( HAVE_CUBLAS "#define HAVE_CUBLAS" )

        endif( WITH_CUBLAS STREQUAL "yes" )       



        ####

        # Check for CUSP

        #
@@ -94,7 +101,8 @@ endif( WITH_CUDA STREQUAL "yes" ) 
#

find_package( OpenMP ) 

if( OPENMP_FOUND )

   AddCompilerFlag( "-DHAVE_OPENMP -fopenmp" )

   #AddCompilerFlag( "-DHAVE_OPENMP -fopenmp" )

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

# TODO: finish this

endif()
@@ -215,6 +223,33 @@ endif( WITH_TESTS STREQUAL "yes" ) 
#   endif()

#endif()



####

# Explicit template instantiation

#

if( WITH_TEMPLATE_INSTANTIATION STREQUAL "yes" )

   AddCompilerFlag( "-DTEMPLATE_EXPLICIT_INSTANTIATION " )

endif()   



if( INSTANTIATE_INT STREQUAL "yes" )

   AddCompilerFlag( "-DINSTANTIATE_INT " )

endif()   



if( INSTANTIATE_LONG_INT STREQUAL "yes" )

   AddCompilerFlag( "-DINSTANTIATE_LONG_INT " )

endif()   



if( INSTANTIATE_FLOAT STREQUAL "yes" )

   AddCompilerFlag( "-DINSTANTIATE_FLOAT " )

endif()   



if( INSTANTIATE_DOUBLE STREQUAL "yes" )

   AddCompilerFlag( "-DINSTANTIATE_DOUBLE " )

endif()   



if( INSTANTIATE_LONG_DOUBLE STREQUAL "yes" )

   AddCompilerFlag( "-DINSTANTIATE_LONG_DOUBLE " )

endif()   



set( CXX_TEST_FLAGS "-fprofile-arcs -ftest-coverage" )

set( LD_TEST_FLAGS "-lgcov -coverage" )

build

+25 −1
Original line number Diff line number Diff line
@@ -4,7 +4,14 @@ TARGET=TNL 
PREFIX=${HOME}/local

WITH_CUDA="yes"

WITH_TESTS="yes"



WITH_CUBLAS="no"

WITH_TEMPLATE_INSTANTIATION="yes"

INSTANTIATE_LONG_INT="yes"

INSTANTIATE_INT="yes"

INSTANTIATE_LONG_DOUBLE="yes"

INSTANTIATE_DOUBLE="yes"

INSTANTIATE_FLOAT="yes"

CMAKE="cmake"

CMAKE_ONLY="no"

HELP="no"
@@ -19,7 +26,18 @@ do 
        --build=*                      ) BUILD="${option#*=}" ;;

        --with-tests=*                 ) WITH_TESTS="${option#*=}" ;;

        --with-cuda=*                  ) WITH_CUDA="${option#*=}" ;;

        --with-cublas=*                ) WITH_CUBLAS="${option#*=}" ;;

        --with-templates-instantiation ) WITH_TEMPLATE_INSTANTIATION="${option#*=}" ;;

        --instantiate-long-int=*       ) INSTANTIATE_LONG_INT="${option#*=}" ;;

        --instantiate-int=*            ) INSTANTIATE_INT="${option#*=}" ;;

        --instantiate-long-double=*    ) INSTANTIATE_LONG_DOUBLE="${option#*=}" ;;

        --instantiate-double=*         ) INSTANTIATE_DOUBLE="${option#*=}" ;;

        --instantiate-float=*          ) INSTANTIATE_FLOAT="${option#*=}" ;;

        --fast-build                   ) INSTANTIATE_LONG_INT="no"

                                         INSTANTIATE_INT="yes"

                                         INSTANTIATE_LONG_DOUBLE="no"

                                         INSTANTIATE_DOUBLE="yes"

                                         INSTANTIATE_FLOAT="no" ;;

        --with-cmake=*                 ) CMAKE="${option#*=}" ;;

        --build-jobs=*                 ) BUILD_JOBS="${option#*=}" ;;

        --cmake-only=*                 ) CMAKE_ONLY="${option#*=}" ;;
@@ -55,9 +73,15 @@ ${CMAKE} ${ROOT_DIR} \ 
         -DCMAKE_BUILD_TYPE=${BUILD} \

         -DCMAKE_INSTALL_PREFIX=${PREFIX} \

         -DWITH_CUDA=${WITH_CUDA} \

         -DWITH_CUBLAS=${WITH_CUBLAS} \

         -DWITH_TESTS=${WITH_TESTS} \

         -DPETSC_DIR=${PETSC_DIR} \

         -DWITH_TEMPLATE_INSTANTIATION=${WITH_TEMPLATE_INSTANTIATION}

         -DWITH_TEMPLATE_INSTANTIATION=${WITH_TEMPLATE_INSTANTIATION} \

         -DINSTANTIATE_FLOAT=${INSTANTIATE_FLOAT} \

         -DINSTANTIATE_DOUBLE=${INSTANTIATE_DOUBLE} \

         -DINSTANTIATE_LONG_DOUBLE=${INSTANTIATE_LONG_DOUBLE} \

         -DINSTANTIATE_INT=${INSTANTIATE_INT} \

         -DINSTANTIATE_LONG_INT=${INSTANTIATE_LONG_INT}



if test ${CMAKE_ONLY} = "yes";

then

src/CMakeLists.txt

+6 −0
Original line number Diff line number Diff line
@@ -35,6 +35,9 @@ set( tnl_CUDA__SOURCES ${tnl_functions_CUDA__SOURCES} 
if( BUILD_CUDA )

   CUDA_ADD_LIBRARY( tnl${debugExt}-${tnlVersion} SHARED ${tnl_CUDA__SOURCES}

                                                  OPTIONS ${CUDA_ADD_LIBRARY_OPTIONS} )

    if( HAVE_CUBLAS )

       CUDA_ADD_CUBLAS_TO_TARGET( tnl${debugExt}-${tnlVersion} )

    endif( HAVE_CUBLAS )

else( BUILD_CUDA )

   ADD_LIBRARY( tnl${debugExt}-${tnlVersion} SHARED 

                ${tnl_SOURCES} )
@@ -51,6 +54,9 @@ IF( BUILD_MPI ) 
   if( BUILD_CUDA )

      CUDA_ADD_LIBRARY( tnl-mpi${debugExt}-${tnlVersion} SHARED ${tnl_CUDA__SOURCES} 

                                                         OPTIONS ${CUDA_ADD_LIBRARY_OPTIONS} )

      if( HAVE_CUBLAS )

         CUDA_ADD_CUBLAS_TO_TARGET( tnl-mpi${debugExt}-${tnlVersion} )

      endif( HAVE_CUBLAS )

   else( BUILD_CUDA )

         ADD_LIBRARY( tnl-mpi${debugExt}-${tnlVersion} SHARED

                      ${tnl_SOURCES} )

src/core/arrays/tnlArrayOperationsCuda_impl.cpp

+101 −24

File changed.

Preview size limit exceeded, changes collapsed.

src/core/arrays/tnlArrayOperationsCuda_impl.cu

+101 −24

File changed.

Preview size limit exceeded, changes collapsed.