README.md

spmv-benchmarking

Build Dependencies

Build MatrixMarket IO

# Start from spmv-benchmarking root folder

git clone -b dev https://github.com/ozusrl/MMMatrixIO.git

mkdir -p lib/mmmatrixio
cd lib/mmmatrixio

cmake ../../MMMatrixIO/src
make

Build dockopt.cpp

# Start from spmv-benchmarking root folder

git clone https://github.com/docopt/docopt.cpp.git

mkdir -p lib/docopt.cpp
cd lib/docopt.cpp

cmake ../../docopt.cpp
make

Build spmv-benchmarking

Simple build with reference methods only

# Start from spmv-benchmarking root folder

mkdir build
cd build

cmake ../src
make

Enabling code specializer methods

Build asmjit

# Start from spmv-benchmarking root folder
git clone https://github.com/asmjit/asmjit

mkdir -p asmjit/build
cd asmjit/build

cmake ../
make

Build spmv-benchmarking with code specializers enabled

# Start from spmv-benchmarking root folder
cd build

cmake -DCODEGEN=true ../src
make

Enable ViennaCL

Build ViennaCL

Download ViennaCL from http://viennacl.sourceforge.net/viennacl-download.html to ${VIENNACL_PATH}

Please refer to ViennaCL documentation at http://viennacl.sourceforge.net/doc/manual-installation.html for more details on build options and available backends.

cd ${VIENNACL_PATH}/build
cmake ../
make

Build spmv-benchmarking with ViennaCL

# Start from spmv-benchmarking root folder
cd build

cmake -DVIENNACL=${VIENNACL_PATH} ../src
make

Enable Intel MKL

Intel MKL is automatically enabled if it is installed.

Enable cuSPARSE

cuSPARSE is automatically enabled if CUDA library is installed.

Enable CUSP

# Go to your prefferd folder for CUSP library
git clone https://github.com/cusplibrary/cusplibrary

# Go to spmv_benchmarking root folder
cmake -DCUSP=${CUSP_PATH} ../src
make