Reproduce “SIMTurn: Lightweight Divergent Subgroup Scheduling for Intra-Warp Synchronization” 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

Nikhil Rout, Joe Samuel Rosh, Blaise Tine

SIMTurn lets divergent GPU subgroups yield and resume around potentially blocking loops, using compiler-inserted yield points and bounded scheduling state while retaining the reconvergence stack. On evaluated Vortex workloads, it is 2.7× faster than IPDOM_SW on BH-ST but takes 2.2× the cycles on single-contention TL-CG, where there is no independent subgroup to schedule.

Summary by alphaXiv.

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