Compack3D

This work was supported by the U.S. National Science Foundation (NSF-OAC-2103509). The project is still active and will receive long-term support.

• Temporary GitHub Repository: https://github.com/songhanglucky/Compack3D

• Temporary Documentation: https://fpal-stanford-university.github.io/Compack3D-Site/ 

Compack3D is a open-source software package that provides scalable parallel linear solvers for numerical simulations using compact finite difference methods. It solves the tridiagonal and penta-diagonal systems arising from compact numerical schemes on 3D computational meshes across distributed memory partitions. Compact3D is implemented in C++ following the "MPI+X" programming paradigm that supports multi-GPU computing. The solver is highly optimized using state-of-the art algorithms developed by our research group, and code implementation is also highly optimized for absolute throughput with high memory bandwidth and low latency. Compack3D provides programming interfaces for both C++ and Fortran programming. The software package is designed to be lightweight and can be easily integrated into an existing code. It also provides abstractions for application code development without going through the unnecessary efforts of low-level data structure design and performance optimization.

Currently, more features are being added to Compack3D. We welcome early-stage users to provide valuable feedback.

Related publications:

• Song, H., Matsuno, K. V., West, J. R., Subramaniam, A., Ghate, A. S., & Lele, S. K. (2022). Scalable parallel linear solver for compact banded systems on heterogeneous architectures. Journal of Computational Physics, 468, 111443.
• Lele, S., Song, H., Matsuno, K., Ghate, A., Subramaniam, A., & Aiken, A. (2023). Towards adaptive high-order simulations of multiphase compressible turbulent flows at exa-scale. In APS March Meeting Abstracts (Vol. 2023, pp. F60-001).
• Song, H., Matsuno, K. V., West, J. R., Subramaniam, A, Ghate, A.S., & Lele, S.K. Scalable parallel linear solver for compact banded systems. In NASA Advanced Modeling and Simulation (AMS) Seminars.

Primary development team: Hang Song, Akshay Subramaniam, Andy Wu, and Anjini Chandra

Core collaborators:

• Dr. Britton J. Olson (Lawrence Livermore National Laboratory)
• Prof. Spencer H. Bryngelson (Georgia Institute of Technology)