SA‐Cost: Structure‐Aware Cost Model With Attention for Tensor Program Tuning
Concurrency and Computation: Practice and Experience · 2026-04 · 38 (8)
DOI: 10.1002/cpe.70713 ↗Kurz gefasst
Fachartikel in der Zeitschrift Concurrency and Computation: Practice and Experience, online erschienen am 19.04.2026. Verfasst von Junguo Liao und 6 weiteren Autor:innen. Die Arbeit nennt 38 Literaturangaben. Gefördert von: National Science and Technology Major Project, National Natural Science Foundation of China, Natural Science Foundation of Hunan Province.
Abstract (Originalsprache)
Automatic scheduling optimization is a critical technology for improving the performance of deep learning compilers (DLCs). As neural network models grow in scale and hardware platforms diversify, the automatic generation of high‐performance tensor programs has become a focal point of research. A robust cost model is essential for automatic scheduling, as it enables the selection of the most optimal scheduling scheme from a multitude of possibilities. This paper introduces SA‐Cost, a novel cost model for DLCs. Unlike conventional approaches that treat tensor programs as a whole during feature extraction, our method reformulates feature extraction as an analysis of the inter‐relationships and semantic correlations among scheduling primitives. By capturing the semantic correlations among scheduling decisions via an attention mechanism, we construct a more precise evaluation model. The model performs a comprehensive assessment across four dimensions: decision space distribution, primitive significance, hardware awareness, and resource utilization efficiency. Furthermore, we incorporate an attention mechanism to identify key scheduling decisions and their contributions, improving the cost model's prediction accuracy. Experimental results demonstrate that on the GPU platform, SA‐Cost achieves an average speedup of over Ansor and over AMOS.
Förderung
- National Science and Technology Major Project
- National Natural Science Foundation of China
- Natural Science Foundation of Hunan Province
Zitieren
APA
Liao, J., Cheng, Z., Hu, Y., Xie, A., Xia, Y., Zhang, Y., & Zhao, Z. (2026). SA‐Cost: Structure‐Aware Cost Model With Attention for Tensor Program Tuning. Concurrency and Computation: Practice and Experience, 38(8). https://doi.org/10.1002/cpe.70713
Vancouver
Liao J, Cheng Z, Hu Y, Xie A, Xia Y, Zhang Y, et al. SA‐Cost: Structure‐Aware Cost Model With Attention for Tensor Program Tuning. Concurrency and Computation: Practice and Experience. 2026;38(8). doi:10.1002/cpe.70713
BibTeX
@article{Liao2026,
author = {Liao, Junguo and Cheng, Zenghua and Hu, Yonghua and Xie, Anxin and Xia, Yimin and Zhang, Yuanjie and Zhao, Zhezhuo},
title = {SA‐Cost: Structure‐Aware Cost Model With Attention for Tensor Program Tuning},
journal = {Concurrency and Computation: Practice and Experience},
year = {2026},
volume = {38},
number = {8},
doi = {10.1002/cpe.70713}
}
EUVIMED