Predictable GPUs for Safety-critical Systems

GPU with predictable wavefront splitting, from the DATE 2024 paper.

GPUs offer large speedups for massively parallel workloads such as the perception tasks in autonomous driving. Using them in safety-critical systems requires worst-case execution time (WCET) estimates, and branch divergence, where threads in a wavefront take different paths and execute serially, makes those estimates large.

This project develops predictable wavefront splitting (PWS), which splits diverging wavefronts into separately schedulable entities while keeping the WCET analyzable.

Architecture. PWS adds micro-architectural enhancements to an AMD-based GPU and a compiler pass that removes branch serialization, improving performance while lowering the WCET compared with prior wavefront-splitting work.
Compiler. A compiler phase selects, with an optimal algorithm, the branches at which to split wavefronts so that the kernel’s WCET is as low as possible. It is evaluated on a gem5 model of the AMD GCN3 GPU, where it reduces the WCET by 34% to 52% compared with five alternative approaches.

The 2025 ACM TECS article studies optimal split-point placement for predictable wavefront splitting.

Publications

TECS

Optimal Split Point Placement for Predictable GPU Wavefront Splitting

, Mahesh Tripunitara and Hiren Patel

ACM Transactions on Embedded Computing Systems, pp. 1–25, October, 2025

DOI

BibTeX

@article{klashtorny_25_tecs,
  author = {Artem Klashtorny and Mahesh Tripunitara and Hiren Patel},
  journal = {ACM Transactions on Embedded Computing Systems},
  volume = {24},
  number = {6},
  pages = {1--25},
  publisher = {ACM},
  month = {oct},
  doi = {10.1145/3769118},
  title = {{Optimal Split Point Placement for Predictable GPU Wavefront Splitting}},
  year = {2025}
}

DATE

A Compiler Phase to Optimally Split GPU Wavefronts for Safety-critical Systems

, Mahesh Tripunitara and Hiren Patel

In proceedings of IEEE Design Automation and Test in Europe (DATE), pp. 1–6, March, 2024

PDF DOI

BibTeX

@inproceedings{klashtorny_24_optpws,
  author = {Artem Klashtorny and Mahesh Tripunitara and Hiren Patel},
  booktitle = {{proceedings of IEEE Design Automation and Test in Europe (DATE)}},
  pages = {1–-6},
  numpages = {6},
  publisher = {IEEE},
  month = {mar},
  doi = {10.23919/DATE58400.2024.10546600},
  title = {{A Compiler Phase to Optimally Split GPU Wavefronts for
           Safety-critical Systems}},
  year = {2024}
}

CASES/TECS

Predictable GPU Wavefront Splitting for Safety-Critical Systems

, , Anirudh Mohan Kaushik and Hiren Patel

ACM Trans. Embed. Comput. Syst., pp. 1–25, September, 2023

PDF DOI

BibTeX

@article{klashtorny_23_pws,
  author = {Klashtorny, Artem and Wu, Zhuanhao and Kaushik, Anirudh Mohan and
            Patel, Hiren},
  title = {Predictable GPU Wavefront Splitting for Safety-Critical Systems},
  year = {2023},
  issue_date = {October 2023},
  publisher = {Association for Computing Machinery},
  address = {New York, NY, USA},
  pages = {1--25},
  volume = {22},
  number = {5s},
  issn = {1539-9087},
  url = {https://doi.org/10.1145/3609102},
  doi = {10.1145/3609102},
  journal = {ACM Trans. Embed. Comput. Syst.},
  month = {sep},
  articleno = {107},
  numpages = {25},
  keywords = {safety-critical systems, GPU}
}