AutoPass: Evidence-Guided LLM Agents for Compiler Performance Tuning

2026-06-18 · Source: Artificial Intelligence · Field: Technology & Digital — Software Development & Engineering, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

AutoPass is a multi-agent framework designed for compiler performance tuning, leveraging Large Language Models (LLMs) to navigate complex microarchitectural effects and noisy runtime measurements. Unlike prior black-box auto-tuning schemes, AutoPass integrates LLMs directly with the compiler, allowing them to query internal optimization states and analyze intermediate representations to orchestrate compiler options. This inference-only, training-free system iteratively refines optimization configurations using measured runtime feedback to diagnose regressions and guide latency-improving edits. Implemented on the LLVM compiler, AutoPass was evaluated on server-grade x86-64 and embedded ARM64 systems. It achieved geometric-mean speedups of 1.043x and 1.117x over LLVM -O3 on x86-64 and ARM64, respectively, outperforming expert-tuned heuristics and classical autotuning methods without requiring offline training or task-specific fine-tuning.

Key takeaway

For Machine Learning Engineers or Compiler Developers optimizing code performance, AutoPass offers a significant advancement. You should consider integrating this evidence-guided LLM agent framework. It achieves notable speedups on x86-64 and ARM64 systems without extensive offline training. This approach moves beyond traditional black-box auto-tuning. It leverages compiler-internal insights for more effective, adaptable optimization strategies across diverse platforms and benchmarks.

Key insights

AutoPass guides LLM agents with compiler and runtime evidence for superior, training-free compiler performance tuning.

Principles

• Compiler-internal visibility enhances LLM optimization.
• Iterative refinement improves optimization configurations.
• Runtime feedback diagnoses regressions effectively.

Method

AutoPass uses LLM agents to query compiler states and analyze IR, iteratively refining optimization configurations with runtime feedback to guide latency-improving edits.

In practice

• Apply to LLVM for x86-64 and ARM64.
• Achieve speedups over LLVM -O3.
• Use for new benchmarks/platforms.

Topics

• LLM Agents
• Compiler Optimization
• Performance Tuning
• LLVM
• x86-64
• ARM64

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Software Engineer

Related on AIssential

• AutoPass: Evidence-Guided LLM Agents for Compiler Performance Tuning — LLMs effectively guide compiler optimization by integrating compiler-internal signals and runtime feedback without offline training.
• Code Loops and A Self Learning AI Harness — An AI self-optimizes its operational harness by iteratively identifying and correcting failure patterns without human intervention.
• When AI Learns to Tune Itself: How ML Is Rewriting the Rules of Compiler Optimization — Published on June 11th, 2026, by Ankit Sinha, a Tech Lead at Google specializing in ML/AI infrastructure, this article examines the profound impact of machine learning on compiler optimization.
• Winning a Kaggle Competition with Generative AI–Assisted Coding — LLM agents, paired with GPU acceleration, dramatically speed up ML experimentation by automating code generation and execution.
• Prior Knowledge or Search? A Study of LLM Agents in Hardware-Aware Code Optimization — LLM agents in code optimization primarily leverage pretrained priors over iterative feedback or agentic exploration.
• Exploring LLM-based Verilog Code Generation with Data-Efficient Fine-Tuning and Testbench Automation — Automated testbench generation enables data-efficient LLM fine-tuning for high-quality Verilog code generation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.