HintPilot: LLM-Generated Compiler Hints Bridge Neural Reasoning and Traditional Optimization Without Source Modification

arXiv·medium signal

HintPilot uses LLMs to synthesize compiler optimization hints rather than directly modifying source code, avoiding semantic errors while capturing fine-grained optimization opportunities that direct LLM-based code rewriting misses. The approach bridges LLM reasoning with traditional compiler infrastructure, letting the model suggest optimization strategies while the compiler ensures correctness — a practical division of labor for production code optimization.