Beyond Pass-by-Pass Optimization: Intent-Driven IR Optimization with Large Language Models

TL;DR

IntOpt introduces an intent-driven IR optimization framework that explicitly captures high-level optimization goals, enabling globally coordinated transformations, resulting in significant performance improvements over existing LLM-based and traditional compiler optimizations.

Contribution

This work presents the first explicit intent-driven IR optimizer, separating high-level strategy from low-level transformations to improve correctness and performance.

Findings

Achieves 90.5% verified correctness

Attains 2.660x average speedup on 200 programs

Surpasses modern compiler -O3 on 37 benchmarks with up to 272.60x speedup

Abstract

Modern compilers optimize programs through a sequence of modular passes over intermediate representations (IR). While this pass-by-pass paradigm offers engineering benefits, it suffers from a pass coordination problem: locally beneficial transformations may block more profitable optimizations in later stages. This limitation stems from the lack of an explicit notion of optimization intent, defined as a holistic strategy for coordinating multiple transformations toward a global performance objective. Recent LLM-based approaches formulate IR optimization as an end-to-end generation task, thereby avoiding the traditional pass-by-pass structure. However, optimization intent remains implicit in these methods, forcing models to jointly infer optimization strategy and generate low-level transformations, which limits both correctness and performance. We propose IntOpt, the first intent-driven IR optimizer that explicitly separates high-level optimization intent from low-level analysis and transformation. IntOpt organizes IR optimization into three stages: intent formulation, intent refinement, and intent realization, enabling globally coordinated transformations. Experiments show that IntOpt achieves 90.5% verified correctness and 2.660x average speedup on 200-program test set, outperforming state-of-the-art LLM-based optimizers in both correctness and performance, and surpassing modern compiler with the -O3 option on 37 benchmarks with speedups of up to 272.60x.