Mosaic: Exploiting Instruction-Level Parallelism on Deep Learning Accelerators with <i>iTex</i> Tessellation
Deep learning has achieved great success in numerous application areas at the cost of high computational complexity. To meet the ever-increasing computational demand, commodity hardware platforms (e.g., CPUs and GPUs) offer abundant computing resources including scalar, vector, and tensor units for deep learning that could execute in parallel. However, existing top-down tiling-based deep learning compilers often generate a homogeneous mapping from the given tensor computation task to hardware arithmetic instructions, failing to utilize different computing units simultaneously to achieve higher performance.
Authors
- Yuanbo Wen (ORCID: https://orcid.org/0000-0002-7775-2724)
- Yifan Hao (ORCID: https://orcid.org/0009-0002-9823-2573)
- Jie Jiang (ORCID: https://orcid.org/0000-0001-9658-5127)
- Zidong Du (ORCID: https://orcid.org/0000-0002-7603-4210)
- Tingting Ruan (ORCID: https://orcid.org/0000-0002-3348-0421)
- Xinkai Song (ORCID: https://orcid.org/0000-0003-3458-1555)
- Rui Zhang (ORCID: https://orcid.org/0000-0001-8691-8549)
- Xing Hu (ORCID: https://orcid.org/0000-0002-9979-0561)
- Di Huang (ORCID: https://orcid.org/0000-0002-2370-0072)
- Qi Guo (ORCID: https://orcid.org/0000-0003-2530-5874)
- Ruiyang Xu (ORCID: https://orcid.org/0000-0002-8337-1565)
- Chongqing Zhao
- Jianxing Xu (ORCID: https://orcid.org/0000-0002-0373-411X)
- Zikang Liu
- Jun Bi (ORCID: https://orcid.org/0000-0001-9956-7039)
Institutions
- University of Science and Technology of China (CN)
- Chinese Academy of Sciences (CN)
- Tencent (China) (CN)
- Institute of Computing Technology (CN)
- University of Chinese Academy of Sciences (CN)
Publication Details
- Published
- 2025-03-27
- DOI
- https://doi.org/10.1145/3676641.3716262
- Citations
- 1
- Primary Topic
- CCD and CMOS Imaging Sensors
- Type
- article
- Field-Weighted Citation Impact
- 0.73