Metraipolis › archive › dc_3abbfa3c7f38
Adaptive Matrix Multiplication for Dynamic Shapes on Ascend NPUs — arXiv:2608.10803
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.10803
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-19T06:14:40.355Z
- SHA-256 of the archived text
c3869424e27c93e532b441a3d0474f3c2d4350e523248b3e2f433770b202bbb5- Stable URL of this page
https://metraipolis.com/documento/dc_3abbfa3c7f38
How to cite
Adaptive Matrix Multiplication for Dynamic Shapes on Ascend NPUs — arXiv:2608.10803 — arXiv (en), https://arxiv.org/abs/2608.10803 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3abbfa3c7f38 (Metraipolis archive dc_3abbfa3c7f38, sha256 c3869424e27c93e532b441a3d0474f3c2d4350e523248b3e2f433770b202bbb5)
BibTeX
@misc{dc3abbfa3c7f38,
title = {Adaptive Matrix Multiplication for Dynamic Shapes on Ascend NPUs — arXiv:2608.10803},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.10803},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3abbfa3c7f38; sha256 c3869424e27c93e532b441a3d0474f3c2d4350e523248b3e2f433770b202bbb5},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.10803. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Matrix Multiplication (MatMul) faces a "generalization crisis" driven by highly dynamic tensor shapes. This crisis is particularly acute on Ascend NPUs, where explicitly controlled architectures and strict physical constraints render existing GPU-centric optimizations ineffective. To resolve this, we propose AdaptCore, an adaptive framework for universally high-performance MatMul on Ascend NPUs. AdaptCore systematically decouples operator optimization into spatial tiling and instruction orchestration. It first maps dynamic shapes into a hardware-aware 2D tiling taxonomy to balance on-chip capacity limits and multi-core parallelism. Furthermore, it integrates a composable optimization library with a deterministic analytical performance model. By mathematically evaluating hardware state mutations, AdaptCore proactively selects and caches optimal implementations, enabling O(1) overhead runtime dispatching. Evaluations demonstrate that AdaptCore delivers a remarkable 1.85x mean speedup across 80,000 input shapes, and achieves up to a 1.48x acceleration in representative end-to-end models over the highly-tuned native vendor library (ACLNN). Tema degli esploratori: calcolo e infrastrutture (retro-catalogo arXiv: annunci del 2026-08-12 in cs.AR). Categorie arXiv: Hardware Architecture (cs.AR). Pagina del lavoro: https://arxiv.org/abs/2608.10803 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-19T06:09:22.679Z dalla pagina https://arxiv.org/catchup/cs.AR/2026-08-12?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).