Expose intra-warp races through binary instrumentation

Develop a perturbation technique or analysis capable of exposing synchronization faults and races occurring between threads within a single GPU warp using binary-level instrumentation.

Background

RESOLVE’s binary-level determinism tester cannot currently expose intra-warp races. The instrumentation sites are bracketed by NVIDIA SASS BSSY and BSYNC instructions, preventing the independent thread scheduler from producing a sufficiently different interleaving. In the mutation study, 20 synchronization mutants remained undetected for this reason; the authors identify compiler support or static analysis as necessary future directions.

References

We attempted to perturb intra-warp races and could not: every instrumentation site is bracketed by BSSY and BSYNC, the SASS instructions that synchronize and reconverge warp control flow, so the independent thread scheduler never gets the chance to interleave differently.

— RESOLVE: Language-Agnostic Validation of GPU Kernels Through Testing, Reduction, and Proof  (2610.05683 - Gargary et al., 5 Oct 2026) in Section 2.3, "Limitations"; Section 3.2, "Catching synchronization bugs"