AI-PROPELLER: Warehouse-Scale Interprocedural Code Layout Optimization with AlphaEvolve
Researchers have developed AI-PROPELLER, a novel system for optimizing code layout at an interprocedural level, a task previously considered intractable due to its complex search space. The system utilizes an agentic workflow named Magellan, which evolves compiler heuristics and fine-tunes policy hyperparameters. By generating multiple layout variants and measuring their performance on actual hardware, AI-PROPELLER achieves significant performance improvements ranging from 0.23% to 1.6% on large-scale warehouse applications. AI
IMPACT Introduces a new method for interprocedural code optimization, potentially improving performance for large-scale applications.