Vectorizer: Vectorizing NumPy Programs with Shape-Guided Rewrite

Vectorizer is a source‑to‑source tool that transforms NumPy programs containing explicit loops into vectorized array operations by rewriting loop bodies from the inside out. It uses array shapes and dataflow analysis to guide a set of rewrite rules that are correct by construction, achieving fast transformations—averaging 0.53 seconds per benchmark. In tests on 150 benchmarks, Vectorizer successfully vectorized 142 directly and 2 with minor edits, producing code that runs on average 74.83× faster than the original loop‑based implementations.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at Hugging Face Trending Papers.

• benchmarks