Learning GPU Code structure using Transformers for Translation and Performance Optimization

This research project aims to improve the performance of GPUs, which are important for running machine learning algorithms. GPUs are a fundamental architecture in machine learning, and this project will use transformer-based models to learn the program structure of GPU kernels for various downstream tasks like performance projection and metrics such as GPU utilization. The research will also potentially allow bidirectional translation from assembly to source code, significantly improving code optimization and generation. The proposed research has the capacity to significantly improve code optimization and generation, leading to more sustainable and efficient practices and will also be beneficial to society for reducing the carbon footprint by reducing the number of optimizations runs.

Faculty Supervisor:

Maryam Mehri Dehnavi

Student:

Partner:

AMD Canada

Discipline:

Computer science

Sector:

Manufacturing; Professional, scientific and technical services

University:

University of Toronto

Program: