Building a domain-specific compiler for emerging processors with a reusable approach
Mingzhen Li,Yi Liu,Bangduo Chen,Hailong Yang,Zhongzhi Luan,Depei Qian
DOI: https://doi.org/10.1007/s11432-022-3727-6
2024-01-07
Science China Information Sciences
Abstract:High-performance computing and deep learning domains have been motivating the design of domain-specific processors. Although these processors can provide promising computation capability, they are notorious for exotic programming paradigms. To improve programming productivity and fully exploit the performance potential of these processors, domain-specific compilers (DSCs) have been proposed. However, building DSCs for emerging processors requires tremendous engineering efforts because the commonly used compilation stack is difficult to be reused. Owing to the advent of multilevel intermediate representation (MLIR), DSC developers can leverage reusable infrastructure to extend their customized functionalities without rebuilding the entire compilation stack. In this paper, we further demonstrate the effectiveness of MLIR by extending its reusable infrastructure to embrace a heterogeneous many-core processor (Sunway processor). In particular, we design a new Sunway dialect and corresponding backend for the Sunway processor, fully exploiting its architectural advantage and hiding its programming complexity. To show the ease of building a DSC, we leverage the Sunway dialect and existing MLIR dialects to build a stencil compiler for the Sunway processor. The experimental results show that our stencil compiler, built with a reusable approach, can even perform better than state-of-the-art stencil compilers.
computer science, information systems,engineering, electrical & electronic