Reproduce “Protean Compiler: An Agile Framework to Drive Fine-grain Phase Ordering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Amir H. Ashouri, Shayan Shirahmad Gale Bagi, Kavin Satheeskumar, Tejas Srikanth, Jonathan Zhao, Ibrahim Saidoun, Ziwen Wang, Bryan Chan and 1 more
The Protean Compiler integrates a machine learning-driven framework directly into LLVM, enabling fine-grained, iterative optimization of compiler pass ordering at various code granularities. It achieves average speedups of up to 4.1% on CBench and 1.1% on SPEC CPU 2017 benchmarks over -O3, demonstrating improved performance with controlled compilation overhead.
Key insights
- Integrating an iterative, ML-driven phase ordering mechanism directly into the compiler infrastructure at fine-grained code granularities can yield measurable performance improvements over fixed optimization levels.
- A hybrid recommender system combining Simulated Annealing with a Genetic Algorithm effectively balances exploration and exploitation, navigating the vast optimization sequence space efficiently.
- Handcrafted static features extracted from Intermediate Representation, when paired with a Transformer model, can accurately predict the performance impact of optimization sequences, enabling efficient, runtime-agnostic optimization decisions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.