PaperScope
LIVE · 2026-09-29 05:40 UTC

Hardware-Aware Features for CUTLASS Kernel Selection

Shriram Chandran, Dominic Rinderer, Yakup Budanaz, Alexandru Calotoiu, Marcin Copik, Torsten Hoefler

Latestcs.CLcs.LGcs.AIcs.CV
arXiv ID
2609.35587 v1
Category
Submitted
2026-09-28

Abstract

GPU libraries such as CUTLASS expose tens of thousands of semantically equivalent kernels for a single operation, making exhaustive autotuning expensive and execution-free selection difficult. Existing analytical selectors require hand-designed performance rules, while learned selectors operate on raw configuration parameters and must infer hardware consequences from data. We introduce a hardware-aware representation for CUTLASS kernel selection that augments candidate configurations with statically computable estimates of induced hardware behavior. We construct a dataset of 4.9 million CUTLASS kernels and train gradient-boosted and neural learning-to-rank models to rank candidates within each problem. On held-out exhaustive evaluation problems, hardware-aware representations reduce selection regret by up to 40\% relative to structural baselines and 64.2\% relative to NVIDIA's matrix-multiply heuristics. We further evaluate data-efficient cross-precision and epilogue-fusion transfer within CUTLASS GEMM, showing that explicitly representing candidate-induced hardware behavior provides a useful inductive bias for learned kernel selection.

Comment: 20 pages, 19 figures

arXiv abs page · PDF · same-day batch