DLVM: A modern compiler infrastructure for deep learning systems

TL;DR

DLVM is a modern, modular compiler infrastructure designed for deep learning systems, offering high expressivity, safety, and performance through a linear algebra IR, algorithmic differentiation, and GPU code generation.

Contribution

It introduces DLVM, a flexible and modular compiler infrastructure inspired by LLVM, supporting tensor DSLs and enabling safer, more performant deep learning frameworks.

Findings

DLVM achieves high performance on GPU targets.

Supports expressive tensor domain-specific languages.

Enables safer and more modular deep learning software development.

Abstract

Deep learning software demands reliability and performance. However, many of the existing deep learning frameworks are software libraries that act as an unsafe DSL in Python and a computation graph interpreter. We present DLVM, a design and implementation of a compiler infrastructure with a linear algebra intermediate representation, algorithmic differentiation by adjoint code generation, domain-specific optimizations and a code generator targeting GPU via LLVM. Designed as a modern compiler infrastructure inspired by LLVM, DLVM is more modular and more generic than existing deep learning compiler frameworks, and supports tensor DSLs with high expressivity. With our prototypical staged DSL embedded in Swift, we argue that the DLVM system enables a form of modular, safe and performant frameworks for deep learning.