Vectorizer: Vectorizing NumPy Programs with Shape-Guided Rewrite

DOI10.48550/arXiv.2609.08088MaRDI QIDQ7288247arXivFDO

Authors Yue-Peng Wang, Jingqian Liu, Xiaoyu Liu, Yuepeng Wang

Publication date 8 September 2026

Published in arXiv

Tldr Vectorizer is a source-to-source transformation tool that vectorizes NumPy loops using shape-guided rewriting and achieves a speedup of up to 75x on average for 150 benchmarks.

Full work available at URL https://doi.org/10.48550/arXiv.2609.08088

arXiv classification

cs.CLcs.SE

Cites work

• Analysis of Programs for Parallel Processing
• Array programming with NumPy
• Automatic translation of FORTRAN programs to vector form
• Building program optimizers with rewriting strategies
• Generating performance portable code using rewrite rules: from high-level functional expressions to high-performance OpenCL code
• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Strategies for source-to-source constant propagation
• Synthesizing imperative programs from examples guided by static analysis
• Vectorizer: Vectorizing NumPy Programs with Shape-Guided Rewrite

Cited in

(1)

• Vectorizer: Vectorizing NumPy Programs with Shape-Guided Rewrite

This page was built for publication: Vectorizer: Vectorizing NumPy Programs with Shape-Guided Rewrite

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7288247)