Acc-SpMM

• This repository is the official implementation of Acc-SpMM: Accelerating General-purpose Sparse Matrix-Matrix Multiplication with GPU Tensor Cores, PPoPP2025.
  • arXiv
  • ACM Digital Library

Requirements

• Supported GPU:
  • A800
  • H100
  • RTX4090
• CUDA >= 11.8

Usage

1. Clone the repository

   git clone https://github.com/AI4SClab/AccSpMM.git
   cd AccSpMM
   

2. Run the code

   mkdir build && cd build
   cmake .. && make
   cd ..
   ./mma $file_path $feature_dim   # you can use the matrix in folder dataset/test
   

Citation

To cite this project, you can use the following BibTex citation.

@inproceedings{zhao2025acc,
  title={Acc-SpMM: Accelerating General-purpose Sparse Matrix-Matrix Multiplication with GPU Tensor Cores},
  author={Zhao, Haisha and Li, San and Wang, Jiaheng and Zhou, Chunbao and Wang, Jue and Xin, Zhikuang and Li, Shunde and Liang, Zhiqiang and Pan, Zhijie and Liu, Fang and others},
  booktitle={Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming},
  pages={326--338},
  year={2025}
}