Reproduce “Equivalence Checking of ML GPU Kernels” 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
Benjamin Driscoll, Kshitij Dubey, Anjiang Wei, Neeraj Kayal, Rahul Sharma, Alex Aiken
Volta checks whether optimized GPU kernels preserve a reference computation. It symbolically executes PTX, tracking memory accesses and synchronization, checks for races and deadlocks, then compares output expressions. The authors verify naive and online softmax as mathematically equivalent. Its guarantees require statically resolvable kernels and model tensor values as real numbers, not bitwise IEEE-754 equality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.