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.

Original languageEnglish (US)
StatePublished - Jan 1 2018
Event6th International Conference on Learning Representations, ICLR 2018 - Vancouver, Canada
Duration: Apr 30 2018May 3 2018

Conference

Conference6th International Conference on Learning Representations, ICLR 2018
CountryCanada
CityVancouver
Period4/30/185/3/18

Fingerprint

DSL
Learning systems
infrastructure
learning software
learning
interpreter
Linear algebra
Tensors
present
performance
Deep learning
Learning Systems
Software

ASJC Scopus subject areas

  • Education
  • Computer Science Applications
  • Linguistics and Language
  • Language and Linguistics

Cite this

Wei, R., Schwartz, L. O., & Adve, V. S. (2018). DLVM: A modern compiler infrastructure for deep learning systems. Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada.

DLVM : A modern compiler infrastructure for deep learning systems. / Wei, Richard; Schwartz, Lane Oscar; Adve, Vikram Sadanand.

2018. Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada.

Research output: Contribution to conferencePaper

Wei, R, Schwartz, LO & Adve, VS 2018, 'DLVM: A modern compiler infrastructure for deep learning systems' Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada, 4/30/18 - 5/3/18, .
Wei R, Schwartz LO, Adve VS. DLVM: A modern compiler infrastructure for deep learning systems. 2018. Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada.
Wei, Richard ; Schwartz, Lane Oscar ; Adve, Vikram Sadanand. / DLVM : A modern compiler infrastructure for deep learning systems. Paper presented at 6th International Conference on Learning Representations, ICLR 2018, Vancouver, Canada.
@conference{9ccced73cf224212ac799dc511f65a73,
title = "DLVM: A modern compiler infrastructure for deep learning systems",
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.",
author = "Richard Wei and Schwartz, {Lane Oscar} and Adve, {Vikram Sadanand}",
year = "2018",
month = "1",
day = "1",
language = "English (US)",
note = "6th International Conference on Learning Representations, ICLR 2018 ; Conference date: 30-04-2018 Through 03-05-2018",

}

TY - CONF

T1 - DLVM

T2 - A modern compiler infrastructure for deep learning systems

AU - Wei, Richard

AU - Schwartz, Lane Oscar

AU - Adve, Vikram Sadanand

PY - 2018/1/1

Y1 - 2018/1/1

N2 - 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.

AB - 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.

UR - http://www.scopus.com/inward/record.url?scp=85071012995&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85071012995&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85071012995

ER -