Commit e8600bdc authored by Tomáš Oberhuber's avatar Tomáš Oberhuber
Browse files

Improving buildAll script. 

Small fix in RgCSR kernel.
parent 07502840

buildAll

+7 −4
Original line number Diff line number Diff line 
#!/bin/bash



INSTALL_PREFIX=${HOME}/local

WITH_CUDA=yes

CUDA_ARCHITECTURE=2.0

VERBOSE=0



CPUS=`grep -c processor /proc/cpuinfo`
@@ -16,13 +19,13 @@ then 
fi



cd Debug

cmake .. -DCMAKE_BUILD_TYPE=Debug -DCMAKE_INSTALL_PREFIX=${HOME}/local -DCUDA_ARCHITECTURE=1.2 -DWITH_CUDA=no

make -j${CPUS} #VERBOSE=1

cmake .. -DCMAKE_BUILD_TYPE=Debug -DCMAKE_INSTALL_PREFIX=${HOME}/local -DCUDA_ARCHITECTURE=${CUDA_ARCHITECTURE} -DWITH_CUDA=${WITH_CUDA}

make -j${CPUS} #VERBOSE=${VERBOSE}

make -j${CPUS} test

make -j${CPUS} install



cd ../Release

cmake .. -DCMAKE_INSTALL_PREFIX=${HOME}/local -DCUDA_ARCHITECTURE=1.2 -DWITH_CUDA=no

make -j${CPUS} #VERBOSE=1

cmake .. -DCMAKE_INSTALL_PREFIX=${HOME}/local -DCUDA_ARCHITECTURE=${CUDA_ARCHITECTURE} -DWITH_CUDA=${WITH_CUDA}

make -j${CPUS} #VERBOSE=${VERBOSE}

make -j${CPUS} test

make -j${CPUS} install

src/matrix/tnlRgCSRMatrix.h

+3 −3
Original line number Diff line number Diff line
@@ -1107,9 +1107,9 @@ __global__ void tnlRgCSRMatrixAdpativeGroupSizeVectorProductKernel( const Index 
   for( Index i = threadIndexInRow; i < nonzeros; i += threadsPerRow )

   {

      const Index column = columns[ pos ];

      if( column == -1 )

         printf( "* rowInMatrix = %d blockIdx. x = %d threadIdx. x = %d threadIndexInRow = %d i = %d \n",

                rowInMatrix, blockIndex, threadIdx. x, threadIndexInRow, i );

      //if( column == -1 )

      //   printf( "* rowInMatrix = %d blockIdx. x = %d threadIdx. x = %d threadIndexInRow = %d i = %d \n",

      //          rowInMatrix, blockIndex, threadIdx. x, threadIndexInRow, i );

      if( column != -1 )

         partialSums[ threadIdx. x ] += nonzeroElements[ pos ] * vec_x[ column ];

      //if( rowInMatrix == 0 )