ML/AI LLVM Methods to Map Code to Core Architectures and Optimize

Modern compilers have increasingly large number of complex optimizations to meet the prevalent demand of using Machine Learning (ML) and Artificial Intelligence (AI) in gaming and other applications. Optimization passes are program and architecture depend. Therefore, selecting the best optimizations in the most optimal ordering is a difficult task. While leveraging ML methods in compiler optimization has become a prominent field of study, integration in production-level compiler for manycore architectures has yet to become standard. This project aims to find optimal methods for running sequential code onto the core architectures to find the most optimal performance and explore the trade-offs of performance versus ease of adoption by game developers of different solutions. The project would increase the hardware performance for ML/AI operations.

Faculty Supervisor:

Nandita Vijaykumar

Student:

Partner:

AMD Canada

Discipline:

Computer science

Sector:

Manufacturing; Professional, scientific and technical services

University:

University of Toronto

Program: