An MLIR Lowering Pipeline for Stencils at Wafer-Scale

TL;DR

This paper introduces a compiler pipeline that automatically transforms stencil-based HPC kernels into optimized code for the Cerebras WSE, achieving significant performance gains without application code modifications.

Contribution

The paper presents a novel MLIR-based lowering pipeline that enables automatic targeting of the WSE for stencil computations, bridging the gap between mathematical models and hardware execution.

Findings

Performance on WSE3 is 14x faster than 128 Nvidia A100 GPUs.

Performance on WSE3 is 20x faster than 128 nodes of a CPU supercomputer.

The approach matches or exceeds manually optimized code performance.

Abstract

The Cerebras Wafer-Scale Engine (WSE) delivers performance at an unprecedented scale of over 900,000 compute units, all connected via a single-wafer on-chip interconnect. Initially designed for AI, the WSE architecture is also well-suited for High Performance Computing (HPC). However, its distributed asynchronous programming model diverges significantly from the simple sequential or bulk-synchronous programs that one would typically derive for a given mathematical program description. Targeting the WSE requires a bespoke re-implementation when porting existing code. The absence of WSE support in compilers such as MLIR, meant that there was little hope for automating this process. Stencils are ubiquitous in HPC, and in this paper we explore the hypothesis that domain specific information about stencils can be leveraged by the compiler to automatically target the WSE without requiring application-level code changes. We present a compiler pipeline that transforms stencil-based kernels into highly optimized CSL code for the WSE, bridging the semantic gap between the mathematical representation of the problem and the WSE's asynchronous execution model. Based upon five benchmarks across three HPC programming technologies, running on both the Cerebras WSE2 and WSE3, our approach delivers comparable, if not slightly better, performance than manually optimized code. Furthermore, without requiring any application level code changes, performance on the WSE3 is around 14 times faster than 128 Nvidia A100 GPUs and 20 times faster than 128 nodes of a CPU-based Cray-EX supercomputer when using our approach.