Reproduce “Ascend-SpMM: Accelerating Sparse Matrix-Dense Matrix Multiplication on Ascend NPUs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiahe Wang, Yingxia Shao, Shuhan Wang, Qingbo Ge, Hongzheng Li, Yucheng Zhang
Ascend-SpMM accelerates sparse-dense multiplication for graph neural network message passing on Ascend NPUs. It assigns denser matrix windows to the Cube unit and sparser work to the Vector unit, coordinating output ownership and pipelining data transfers. Across 820 SuiteSparse matrices, it outperforms the strongest evaluated NPU baseline in every tested setting; at width 256 in FP32, its geometric-mean speedup is 12.70× on the tested system.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.