Domain-Specific Multi-Level IR Rewriting for GPU

TL;DR

This paper introduces a multi-level IR rewriting approach tailored for GPU programming in the weather and climate domain, demonstrating significant performance improvements over existing solutions.

Contribution

It presents a novel multi-level IR rewriting framework with domain-specific dialects and optimizations, outperforming state-of-the-art compilers in the weather and climate domain.

Findings

Two domain-specific optimizations outperform existing solutions

Development of stencil- and GPU-specific dialects

Multi-level rewriting enhances compiler specialization

Abstract

Traditional compilers operate on a single generic intermediate representation (IR). These IRs are usually low-level and close to machine instructions. As a result, optimizations relying on domain-specific information are either not possible or require complex analysis to recover the missing information. In contrast, multi-level rewriting instantiates a hierarchy of dialects (IRs), lowers programs level-by-level, and performs code transformations at the most suitable level. We demonstrate the effectiveness of this approach for the weather and climate domain. In particular, we develop a prototype compiler and design stencil- and GPU-specific dialects based on a set of newly introduced design principles. We find that two domain-specific optimizations (500 lines of code) realized on top of LLVM's extensible MLIR compiler infrastructure suffice to outperform state-of-the-art solutions. In essence, multi-level rewriting promises to herald the age of specialized compilers composed from domain- and target-specific dialects implemented on top of a shared infrastructure.