Reproduce “Automatic Hardware Pragma Insertion in High-Level Synthesis: A Non-Linear Programming Approach” 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
The NLP-DSE framework automates hardware pragma insertion for High-Level Synthesis in regular loop-based programs by formulating the problem as a Non-Linear Program. It achieved an average throughput improvement of 17.24x and was 5.69x faster in design space exploration compared to AutoDSE, providing a provable lower bound on latency for efficient pruning.
Key insights
- Formulating HLS pragma insertion for affine programs as an NLP allows for rigorous mathematical optimization and the derivation of a provable lower bound on latency.
- The lower-bound property of the analytical model enables aggressive yet safe pruning of the design space, significantly reducing DSE time.
- A hybrid DSE approach, which blends theoretical NLP-driven optimization with selective empirical HLS validation, effectively addresses compiler quirks while maintaining high QoR.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.