Reproduce “A Unified Framework for Automated Code Transformation and Pragma Insertion” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Stéphane Pouget, Louis-Noël Pouchet, Jason Cong
Sisyphus is a unified framework that automates code transformation and pragma insertion for High-Level Synthesis (HLS) on FPGAs. It achieves geometric mean speedups of 9.17x over AutoDSE, 3.47x over NLP-DSE, and 16.15x over ScaleHLS, significantly improving performance while reducing design exploration time from hours to minutes.
Key insights
- Unifying code transformations and pragma insertion within a single NLP-based optimization framework leads to superior QoR for HLS designs compared to disjoint approaches.
- A hierarchical, three-level strip-mining strategy combined with polyhedral analysis effectively exposes and manages different granularities of parallelism and data locality for HLS.
- Efficiently solving the NLP problem allows for rapid design space exploration, identifying high-quality solutions within minutes, a significant improvement over iterative HLS synthesis methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.