What type of study is this?

This is a Experimental Study study.

October 19, 2025

Rethinking Tiling and Dataflow for SpMM Acceleration: A Graph Transformation Framework

View Full Paper

Ask AI

Bookmark

Authors

A novel graph transformation framework improves spmm performance in diverse datasets, indicating better data locality and efficiency.

Aquila improves execution time for SpMM by 4.3× and energy efficiency by 4.8× compared to existing accelerators.
The framework reinterprets spmm computations, addressing limitations in data locality and sparsity handling.
Vertex decomposition and adaptive depth traversal allow for non-contiguous tiling, enhancing data reuse without rigid alignment.
The new pull-after-push dataflow overcomes synchronization challenges, enabling optimized output matrix accumulation.

Ahsaei et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark