We look forward to seeing you virtually on Thursday, November 10, at 9:30am for the CS Complier Seminar.
Zoom link: https://illinois.zoom.us/j/89400978467?pwd=NjV3ZFQrQ1JidTNyS0ZNUEVOcEtpUT09
Abstract:
Leveraging machine-learning (ML) techniques for compiler optimizations has been widely studied and explored in academia. However, the adoption of ML in general-purpose, industry strength compilers has yet to happen. MLGO is a framework for integrating ML techniques systematically in an industrial compiler -- LLVM. There are currently two successful applications under the MLGO framework: 1) inlining-for-size; 2) register allocation (for performance). They improve over current industry state of the art (profile guided + link-time optimizations), and have been landed in Google and Open Source production systems. This work is the first full integration of ML in complex compiler passes that is deployed in the real-world setting. They are both available in the main LLVM repository and github.com/google/ml_compiler_opt. In this talk, we will cover the general MLGO framework, the two applications, and the experience of deploying them in the production system.
Bio:
Yundi Qian is a software engineer at Google Brain. In this role, she focuses on applying Machine Learning to System optimization problems, especially ML for compiler optimization problems.
Mircea Trofin is a compiler engineer with the Compiler Optimization team at Google.
