Work

JOVS: Joint Optimization of Vectorization and Scheduling for DNN on AI DSPs

Year: 2025

Type: conference-paper

Abstract: Recent embedded devices have integrated digital signal processors (DSPs) to balance performance and power when executing complex Deep Neural Network (DNN) workloads. With modern AI DSPs providing specialized tensor computation vector instructions and limited on-chip memory, fully releasing the potential of these DSPs remains a significant challenge. The performance of AI DSPs relies heavily on vendor-provided libraries and compilers. In practice, vendor-provided libraries are inflexible and prevent further optimization. State-of-the-art compilers usually focus on a single optimization (vectorization or scheduling), which is insufficient to address this challenge.

Source: ACM Symposium on Parallelism in Algorithms and Architectures (SPAA)

DOI: 10.1145/3694906.3743309

Authors: Yaochen Han, Hongxu Jiang, Runhua Zhang, Rui She

Institution: Beihang University

Language: English

Topic: Parallel Computing and Optimization Techniques

Subfield: Hardware and Architecture

Field: Computer Science

Domain: Physical Sciences

Open Access status: bronze

Funders: Universitas Brawijaya, National Key Research and Development Program of China

Awards: G2050173416, G8574546177