Abstract
There is a growing interest in enhancing compiler optimizations with ML models. Yet compilers remain challenging environments for interacting with ML frameworks. Some optimizations require tightly coupled models and compiler internals, raising issues with modularity, performance and ML framework independence. Practical deployment and transparency for the end-user are also important concerns. We propose a library allowing ML model development within a traditional Python framework while enabling efficient, end-to-end integration with an optimizing compiler. We evaluate it on both research and production use cases, for training and inference, over four optimization problems.
Authors
S. VenkataKeerthy*, Siddharth Jain*, Umesh Kalvakuntla*, Gorantla Pranav Sai*, Albert Cohen, Eugene Brevdo, Mircea Trofin*, Ramakrishna Upadrasta*
Venue
ACM SIGPLAN 2024 International Conference on Compiler Construction