一句话：提出动态反射式分块（DRT），在运行时按输入张量当前稀疏度持续重分块，提升稀疏张量核的数据复用与加速器缓冲区利用率。
多稀疏操作数张量代数严重受内存带宽约束，不规则稀疏性使传统分块难以有效缓解瓶颈。现有稀疏分块方案对稀疏度不敏感，将张量切成统一的坐标空间形状，导致分块占用率低、可复用性差。本文提出动态反射式分块（DRT）：一种动态且稀疏度感知的分块方法。DRT 在运行时根据所有输入张量活跃区域的当前稀疏度持续对稀疏张量重新分块，以最大化加速器缓冲区利用率，同时保留对不同张量的分块进行协同迭代的能力。该方法相较现有稀疏分块方案提升了数据复用，并释放出显著的性能优化空间。
Tensor algebra involving multiple sparse operands is severely memory bound, making it a challenging target for acceleration. Furthermore, irregular sparsity complicates traditional techniques—such as tiling—for ameliorating memory bottlenecks. Prior sparse tiling schemes are sparsity unaware: they carve tensors into uniform coordinate-space shapes, which leads to low-occupancy tiles and thus lower exploitable reuse. To address these challenges, this paper proposes dynamic reflexive tiling (DRT), a novel tiling method that improves data reuse over prior art for sparse tensor kernels, unlocking significant performance improvement opportunities. DRT’s key idea is dynamic sparsity-aware tiling. DRT continuously re-tiles sparse tensors at runtime based on the current sparsity of the active regions of all input tensors, to maximize accelerator buffer utilization while retaining the ability to co-iterate through tiles of distinct tensors.
@inproceedings{paperbot10064,
title = {Accelerating Sparse Data Orchestration via Dynamic Reflexive Tiling},
author = {Toluwanimi O. Odemuyiwa and Hadi Asghari-Moghaddam and Michael Pellauer and Kartik Hegde and Po-An Tsai and Neal C. Crago and Aamer Jaleel and John D. Owens and Edgar Solomonik and Joel S. Emer and Christopher W. Fletcher},
booktitle = {ASPLOS},
year = {2023},
doi = {10.1145/3582016.3582064}
}