Reproduce “HLSClaw: Bringing Vibe Coding to High-Level Synthesis through Skill-Based Agentic Harness” 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
Jinming Ge, Shang Liu, Jiawei Liang, Shangkun Li, Afzal Ahmad, Jiang Xu, Linfeng Du, Wei Zhang
HLSClaw coordinates an LLM agent’s profiling, code restructuring, functional checks, and directive search, using synthesis feedback to revisit earlier structural choices. On a 3×3 convolution, its line-buffer rewrite reduces latency from 4,502 to 1,147 cycles at a 5 ns target clock. Across 12 benchmarks, it reports 39.12× geometric-mean cycle-latency speedup and 1.98× over ScaleHLS, with a 15.2× normalized resource score.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.