The following benchmarks have been tested with and without affinity extenstion for the OpenMP task construct.
- ./stream
- STREAM version that uses tasks instead of work sharing construct
- ./bots/omp-tasks/sort
- Parallel merge sort algorithm from bots benchmark suite
- ./bots/omp-tasks/health
- health benchmark from bots benchmark suite
- ./sparse_CG/tasking
- CG solver that uses SPMXV
git clone [email protected]:jklinkenberg/openmp.git
cd openmp
git checkout task-affinity
cd ..
git clone [email protected]:jklinkenberg/task-affinity-codes.git
cd task-affinity-codes
export RUNTIME_AFF_REL_DIR=$(pwd)/runtime_affinity_rel
export RUNTIME_AFF_DEB_DIR=$(pwd)/runtime_affinity_deb
make
2.4 Set your C compiler either to intel or clang (Should be a compiler that is compatible with LLVM OpenMP runtime)
export CC=icc
2.5 Move to desired benchmark folder, e.g., stream and execute fast run, that will compare llvm basline with affinity versions
cd stream
sh fast_run.sh