Domain-specific framework for performance analysis of parallel programs
PerFlow is dependent on
Dyninst and PAPI need user to build themselves. igraph
has been integrated into PerFlow as submodule.
git submodule update --init
There are two ways to build PerFlow. One is to build dependency from source and specify their location when building PerFlow, the other is to use spack to build these.
cmake .. -DBOOST_ROOT=/path_to_your_boost_install_dir -DDyninst_DIR=/path_to_your_dyninst_install_dir/lib/cmake/Dyninst -DPAPI_PREFIX=/path_to_your_papi_install_dir
# You should make sure that there is `DyninstConfig.cmake` in /path_to_your_dyninst_install_dir/lib/cmake/Dyninst
# And there is `include` `lib` in /path_to_your_papi_install_dir
# And there is `include` `lib` in /path_to_your_boost_install_dir, `boost` in /path_to_your_boost_install_dir/include
Note that if dyninst is built from source, the boost will be downloaded and installed automatically with it, in the install directory of dyninst.
In this case, the cmake commands will be like
cmake .. -DBOOST_ROOT=/path_to_your_dyninst_install_dir -DDyninst_DIR=/path_to_your_dyninst_install_dir/lib/cmake/Dyninst -DPAPI_PREFIX=/path_to_your_papi_install_dir
The recommended way to build Dyninst (with Boost) and PAPI is to use Spack
spack install dyninst # boost will be installed at the same time
spack install papi
# before building PerFlow
spack load dyninst # boost will be loaded at the same time
spack load papi
# build
mkdir build
cd build
cmake ..