Composable and Modular Code Generation in MLIR: A Structured and Retargetable Approach to Tensor Compiler Construction

TL;DR

This paper introduces a structured, modular approach to building tensor compilers using MLIR, enhancing flexibility, reusability, and productivity for compiler developers and users.

Contribution

It presents a novel, structured design for tensor compiler code generators that improves modularity, composability, and genericity leveraging tensor algebra.

Findings

Improved modularity and composability in tensor compiler construction.

Preliminary experimental results demonstrate potential benefits.

Design facilitates progressive lowering paths in MLIR.

Abstract

Despite significant investment in software infrastructure, machine learning systems, runtimes and compilers do not compose properly. We propose a new design aiming at providing unprecedented degrees of modularity, composability and genericity. This paper discusses a structured approach to the construction of domain-specific code generators for tensor compilers, with the stated goal of improving the productivity of both compiler engineers and end-users. The approach leverages the natural structure of tensor algebra. It has been the main driver for the design of progressive lowering paths in \MLIR. The proposed abstractions and transformations span data structures and control flow with both functional (SSA form) and imperative (side-effecting) semantics. We discuss the implications of this infrastructure on compiler construction and present preliminary experimental results.