BAAS

BAAS, a high-performance framework designed for parallelizing sparse matrix computations. This work is featured in our IPDPS 2026 paper: “BAAS: A Bidirectional Aggregation and Affinity-Aware Scheduling Framework for Parallelizing Sparse Matrix Computations.”

Prerequisites

• Intel MKL: For optimized BLAS/Sparse kernels.

• Metis: For graph partitioning and matrix reordering.

• PAPI: For hardware performance counter monitoring and profiling.

• CMake: Version 3.10 or higher

• GCC: With C++17 support and OpenMP.

Getting Started

1. Configuration First, you need to point BAAS to your local library installations. Navigate to the script directory and edit env.sh:

cd script
# Open env.sh and update the paths for MKL, Metis, and PAPI
vi env.sh

2. Build

We provide convenient scripts to handle the build process:

# Generate Build files
bash cmake.sh

# Compile the project
bash make.sh

3. Data Preparation BAAS expects matrix files to be located in the /data directory.

• Place your .mtx files in /data.

• Update matrixList.txt to include the filenames of the matrices you wish to test.

4. Running We have provided specialized scripts for different sparse solvers. You can run them directly from the script directory.

./run_trsv.sh # SpTRSV 
./run_ic.sh   # SpIC0
./run_ilu.sh  # SpILU0
./run_iccg.sh # ICCG

Note: You can customize the thread count and iteration rounds within these scripts to match your hardware.

5. Output & Results: All generated files are stored in the output/ directory:

output/csv/: Contains the measured performance metrics and benchmark results.

output/log/: Contains detailed program execution logs.

output/out/: Contains standard output redirected from the running scripts.