PRNGine: Massively Parallel Pseudo-Random Number Generation and Probability Distribution Approximations on AMD AI Engines [CGRA4HPC, IPDPS'25].

This project enables porting PRNGs to AMD Versal AIEs by analyzing the algorithmic requirements of three PRNGs: XORWOW, XOROSHIRO, and SFMT. Two execution models are proposed: a co-processor model (M1), where AIEs generate random numbers and send them to the host processor, and a standalone dataflow accelerator model (M2), where random number generation is pipelined with computation within a dataflow overlay on AIEs. M2 uses the Acklam's ICDF approximation as a subsequent computation step in the dataflow to random number generation.

Installation

To run PRNGine, clone the repository:

git clone https://github.com/medbzkst/PRNGine
cd PRNGine

Then depending on which execution model: co-processor model (M1) or standalone dataflow accelerator (M2), locate within the corresponding directory:

cd M1 # Co-processor model

Each model employs 3 PRNGs: SFMT, XORWOW, XOROSHIRO128. Locate in the corresponding directory and launch the build (this may take a few hours).

cd SFMT # SFMT PRNG
make all

M2 under the directory M2 runs the Acklam's ICDF approximation as a subsequent computation step in the dataflow to random number generation.M2_template can be used as a template to M2 to incorporate a subsequent computation step to the random number generation. This can be done by coding the computation kernel under M2_template/PRNG_name/aie/src/subsequent.cpp.

Contributing

Contributions are welcome! Please fork the repository and submit a pull request with your changes. Make sure to follow the coding standards and include tests for any new features.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For any questions or suggestions, please open an issue or contact at mohamed.bouaziz@kaust.edu.sa.

Reference

M. Bouaziz and S. A. Fahmy, "PRNGine: Massively Parallel Pseudo-Random Number Generation and Probability Distribution Approximations on AMD AI Engines," 2025 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW), 2025.

@INPROCEEDINGS{PRNGine,
  author={Bouaziz, Mohamed and Fahmy, Suhaib A.},
  booktitle={2025 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)}, 
  title={PRNGine: Massively Parallel Pseudo-Random Number Generation and Probability Distribution Approximations on AMD AI Engines}, 
  year={2025}
}