Back to search

Georgia Institute of Technology

Compiler-Runtime Co-Design for Performance-Portable GPU Programming on CPUs

Abstract

dc:description.abstract

With the rapid development of Artificial Intelligence and High-Performance Computing, an increasing number of programs are being implemented for GPU execution. However, due to the limited supply and high cost, GPUs are not always readily accessible to all developers. As a result, it is common for developers to wait hours for their jobs to be scheduled on available GPU resources in data centers. On the other hand, CPUs are the most ubiquitous and accessible architecture in data centers. As a result, researchers have proposed executing GPU programs on CPUs. Using CPUs as an alternative backend for GPU programs opens new opportunities for improving hardware utilization and reducing data center costs. Additionally, CPU-based execution enables developers to leverage mature CPU debugging toolkits to debug GPU programs. CPUs can also serve as an accessible educational platform for students who do not have access to GPUs. Although GPU-to-CPU migration has been studied for a long time, it remains an unsolved problem. Specifically, two aspects are critical: coverage and performance. In other words, the goal is to execute as many GPU programs on CPUs as possible, and ensure that the migrated programs efficiently utilize CPU computational resources to achieve high performance. This dissertation presents a compiler-runtime co-design solution for GPU-to-CPU migration that improves both coverage and performance. The proposed solution is end-to-end, accepting CUDA source code and automatically generating CPU executable files without requiring any manual pre- or post-processing. The codebase for the corresponding project is available as open source on GitHub(https://github.com/cupbop/CuPBoP).

Degree

thesis:*
Level thesis:degree_level
Doctoral
Department dc:contributor.department
Computer Science
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Han, Ruobing
Advisor dc:contributor.advisor
  • Kim, Hyesoon
Committee members dc:contributor.committeemember
  • Pande, Santosh
  • Lin, Yingyan
  • Vuduc, Richard
  • Sim, Jaewoong
  • Zhou, Zongwei

Subjects

dc:subject × 4

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/77874

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Han, Ruobing. Compiler-Runtime Co-Design for Performance-Portable GPU Programming on CPUs. Doctoral thesis, Georgia Institute of Technology, 2025. https://hdl.handle.net/1853/77874