Install dependencies

System dependencies:

• Python3
• GNU parallel
• MPI

Install Intel MKL. Make sure setvars.sh has been executed in current working shell.

Install Python dependencies:

python3 -m pip install -r requirements.txt

In src/consts.py set the correct paths.

To get the block choosing model:

• Run studies/threshold.py.
• Run studies/find_threshold.py.

In bench_parallel_launcher.py:

• set mtx_dir to the Suitesparse directory.
• the list eval is the list of matrices that you want to generate code for and compile.

In eval.py:

• the list eval is the list of matrices that you want to evaluate.
• Set the number of threads in the call to eval_single_proc().
• Account for NUMA nodes and such while setting the cores to execute our programs on, in line 30.
• Evaluated values will be stored in results/res_{}.csv.

To get SABLE numbers:

• python3 bench_parallel_launcher.py
• taskset -a -c <comma separated cores> python3 eval.py (the cores should match the cores on line 30 in eval.py)

For MKL baseline, in bench_mkl.py:

• Set mtx_dir to the Suitesparse directory.
• the list eval is the list of matrices that you want to evaluate.
• Correctly set the value of threads = [].
• Evaluated values will be stored in results/mkl-spmv-merged-results.csv.
• Execute it using taskset -a -c <comma separated cores> python3 eval.py