# Hardware-Aware Features for CUTLASS Kernel Selection

arXiv: https://arxiv.org/abs/2609.35587 · HTML: https://mlsanity.com/paper/2609.35587/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 29 references that are in the map.
- First posted on arXiv 28 Sep 2026.

## Builds on (most cited first, 12 of 12)

- [FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision](https://mlsanity.com/paper/2407.08608/) (2024)
- [Ansor: Generating High-Performance Tensor Programs for Deep Learning](https://mlsanity.com/paper/2006.06762/) (2020)
- [Learning to Optimize Tensor Programs](https://mlsanity.com/paper/1805.08166/) (2018)
- [Stream-K: Work-centric Parallel Decomposition for Dense Matrix-Matrix Multiplication on the GPU](https://mlsanity.com/paper/2301.03598/) (2023)
- [TLP: A Deep Learning-based Cost Model for Tensor Program Tuning](https://mlsanity.com/paper/2211.03578/) (2022)
- [MetaTune: Meta-Learning Based Cost Model for Fast and Efficient Auto-tuning Frameworks](https://mlsanity.com/paper/2102.04199/) (2021)
- [Input-Aware Auto-Tuning of Compute-Bound HPC Kernels](https://mlsanity.com/paper/1802.05371/) (2018)
- [Optimal Software Pipelining and Warp Specialization for Tensor Core GPUs](https://mlsanity.com/paper/2512.18134/) (2025)
- [Hexcute: A Compiler Framework for Automating Layout Synthesis in GPU Programs](https://mlsanity.com/paper/2504.16214/) (2025)
- [Learning from distinctive candidates to optimize reduced-precision convolution program on tensor cores](https://mlsanity.com/paper/2202.06819/) (2022)
- [WaveTune: Wave-aware Bilinear Modeling for Efficient GPU Kernel Auto-tuning](https://mlsanity.com/paper/2604.10187/) (2026)
- [Microbenchmark-Driven Analytical Performance Modeling Across Modern GPU Architectures](https://mlsanity.com/paper/2605.04178/) (2026)

## Abstract (arXiv)

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.