mining-bench

Artifact Description for "vectorizing sparse matrix codes using partially strided codelets" at SC22. The source code is available at the codelet mining repository. The artifact is available as:

Quick Instruction

The mining-bench repository should be cloned:

    git clone https://github.com/cheshmi/mining-bench.git
    cd mining-bench

Running the script to perform all experiments on server

bash run_all.sh 1

You should pass 0, if you want to run experiments locally. The script assumes, singularity is loaded and can be used. Otherwise, please modify common.sh.

Plotting

bash plot_all.sh

Plots should be generated as PDFs in the current directory. You will need python3 for running the script successfully.

How to run the artifact using the image step by step

The mining-bench repository should be cloned:

    git clone https://github.com/cheshmi/mining-bench.git
    cd mining-bench

The singularity image should be pulled to the same directory that the code is cloned using:

   singularity pull <link is provided in the AD>

You can test the image by running the following command from the current directory:

singularity exec sparse.sif /source/codelet_mining/build/demo/spmv_demo --matrix ./LFAT5.mtx --numerical_operation SPMV --storage_format CSR

The output is set of comma separated values such matrix specification and execution time of different tools. Also, plots will be created as PDF files upon completion of all experiments.

The datasets should be downloaded by calling:

    python ssgetpy/dl_matrices.py
    python ssgetpy/dl_SPD_matrices.py

Matrices are downloaded into the mm and SPD directories in the current directory (This might take several hours and requires internet connection).

The SpMV experiment can be executed by emitting:

    bash run_spmv.sh

For running on compute node:

sbatch bash run_spmv.sh

You might need to update scripts with new absolute paths to the dataset and the image file.

SpTRSV experiment can be done by running:

bash run_sptrsv.sh

SpMM experiment can be reproduced by calling:

bash run_spmm.sh 0
bash run_spmm.sh 1
bash run_spmm.sh 2

all results should be stored as CSV files under the ./logs/ directory. Plots should be created in the current directory as PDF files by calling:

bash plot_all.sh