一句话：Heron 通过自动刻画 DLA 固有约束来约束搜索空间，从而生成高性能深度学习加速器库。
深度学习加速器（DLA）能提升计算密集型深度学习算法的性能与能效。利用具有成熟 API 的高性能软件库是发挥 DLA 能力的一种灵活且可移植的方式；这类库通常由人工实现，或通过基于探索的编译方法自动生成。然而，基于探索的方法虽然显著降低编程开销，却难以从庞大但质量低下的搜索空间中找到最优或近最优程序，因为 DLA 的大量固有约束无法被准确刻画。Heron 的核心思路是自动地对 DLA 库生成过程施加约束，以更准确地表达这些固有约束并缩小/优化搜索空间，从而自动生成高性能库。
Deep Learning Accelerators (DLAs) are effective to improve both performance and energy efficiency of compute-intensive deep learning algorithms. A flexible and portable mean to exploit DLAs is using high-performance software libraries with well-established APIs, which are typically either manually implemented or automatically generated by exploration-based compilation approaches. Though exploration-based approaches significantly reduce programming efforts, they fail to find optimal or near-optimal programs from a large but low-quality search space because the massive inherent constraints of DLAs cannot be accurately characterized.
@inproceedings{paperbot10084,
title = {Heron: Automatically Constrained High-Performance Library Generation for Deep Learning Accelerators},
author = {Jun Bi and Qi Guo and Xiaqing Li and Yongwei Zhao and Yuanbo Wen and Yuxuan Guo and Enshuai Zhou and Xing Hu and Zidong Du and Ling Li and Huaping Chen and Tianshi Chen},
booktitle = {ASPLOS},
year = {2023},
doi = {10.1145/3582016.3582061}
}