Reproduce “SOLAR: AI-Powered Speed-of-Light Performance Analysis” 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

Qijing Huang, Sana Damani, Zhifan Ye, Athinagoras Skiadopoulos, Siva Kumar Sastry Hari, Jason Clemons, Sahil Modi, Jingquan Wang and 4 more

SOLAR turns a program into an estimate of its speed-of-light bound: the shortest idealized computation and data movement time on specified hardware. It has an LLM translate source into executable Affine Loop IR, validates outputs, then applies deterministic graph and roofline analysis. On KernelBench L3, measured PyTorch eager runtime was 54.6× the fused bound, indicating modeled headroom, not achieved speedup.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.