High-performance attention kernels are essential for Large Language Models. This paper presents analysis of CuTile-based Flash Attention memory behavior and a technique to improve its cache performance. In particular, our analysis on the NVIDIA GB10 (Grace Blackwell) identifies the main cause of L2 cache miss. Leveraging this insight, we introduce a new programming technique called Sawtooth Wavefront Reordering that reduces L2 misses. We validate it in both CUDA and CuTile, observing 50\% or greater reduction in L2 misses and up to 60\% increase in throughput on GB10.
Cite
@article{arxiv.2601.16032,
title = {Sawtooth Wavefront Reordering: Enhanced CuTile FlashAttention on NVIDIA GB10},
author = {Yifan Zhu and Yekai Pan and Chen Ding},
journal= {arXiv preprint arXiv:2601.16032},
year = {2026}
}