paperbot · PL 论文追踪

全部论文 待读 收藏

P2 RASSM: Residue-based Acceleration of Single Sparse Matrix Computation via Adaptive Tiling

RASSM：基于残差的自适应分块单稀疏矩阵计算加速
ASPLOS Proceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 12025匹配度 2/10 · 论文聚焦稀疏矩阵核的高性能优化与缓存分块，属于体系结构/性能优化方向，与用户核心关注的静态分析、LLM/Agent 交叉、浮点程序分析、贝叶斯程序分析无直接关联；仅在应用背景中提及 GNN/LLM，属于间接相关。引用 4 inbox
Anirudh Jain, Pulkit Gupta, Thomas M. Conte
语言与体系结构
crossref: 2026-09-12dblp: 2026-09-12

AI 速览

一句话：提出基于残差矩阵的自适应二维分块技术 RASSM，为 SpMM 等单稀疏矩阵核提升缓存占用与性能。

单稀疏矩阵核（SSMK）如 SpMM、SDDMM、SpMV、SpTS 是数据分析、图处理、有限元分析和机器学习等应用的核心，但稀疏矩阵不规则的非零分布会造成缓存占用低与数据移动开销大。本文提出 RASSM，一种输入相关的自适应二维分块技术；其核心是残差矩阵，可紧凑捕获稀疏矩阵非零模式，并据此智能选择 tile 尺寸，兼顾空间与时间局部性，改善数据移动与整体性能。在商用服务器 CPU 上，RASSM 对 SpMM 的空间分析比 MKL、J-Stream、ASpT、CSF-4 等有 1.10×–1.32× 加速，加入时间分析后可达 1.16×–1.38×。

问题
SSMK 在数据分析、图处理、机器学习等应用中至关重要，但稀疏矩阵不规则的非零模式导致缓存占用低、数据移动开销大。本文研究如何针对输入自适应地优化 SSMK 的分块与数据局部性。
方法
提出 RASSM，用残差矩阵紧凑表示非零模式，并据此自适应选择二维 tile 尺寸，同时利用空间与时间局部性优化数据移动。
评估
在商用服务器 CPU 上针对 SpMM 等 SSMK 与 MKL、J-Stream、ASpT、CSF-4 等实现对比，报告 1.10×–1.38× 的加速。
原文摘要（Abstract）

Single-Sparse-Matrix Kernels (SSMKs) such as SpMM, SDDMM, SpMV, and SpTS form the backbone of applications such as data analytics, graph processing, finite-element analysis, machine learning (including GNNs and LLMs), etc. This paper introduces Residue-based Acceleration of Single Sparse Matrix Computation via Adaptive Tiling (RASSM), an input-dependent, adaptive 2-dimensional tiling technique for SSMKs. The adaptation leverages the concept of a residue matrix: a data structure that compactly captures the pattern of non-zeros in the sparse matrix. With residues, we show it is possible to make intelligent decisions on adaptive tile sizes, resulting in increased cache occupancy. Residues allow for optimizations across both spatial and temporal locality. RASSM improves data movement and overall performance as compared to prior techniques. For example, using spatial analysis for SpMM on commodity server CPUs, RASSM has 1.30X speedup over MKL, 1.32X over J-Stream, 1.20X over ASpT, 1.11X over CSF-4 uniform-shape, and 1.10X over CSF-4 uniform-occupancy. RASSM with temporal analysis improves this to 1.36X (vs. MKL), 1.38X (vs. J-Stream), 1.26X (vs. ASpT), 1.17X (vs. CSF-4 uniform-shape), and 1.16X (vs. CSF-4 uniform-occupancy).

链接与引用

DOI 原文 · PDF（开放获取） · DBLP

BibTeX 
@inproceedings{JainGC25,
  title = {RASSM: Residue-based Acceleration of Single Sparse Matrix Computation via Adaptive Tiling},
  author = {Anirudh Jain and Pulkit Gupta and Thomas M. Conte},
  booktitle = {ASPLOS},
  year = {2025},
  doi = {10.1145/3669940.3707219}
}