AgRefactor: Self-Evolving Agentic Workflow for HLS Compatibility and Performance

2026-06-29 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Hardware Architecture & Design · Depth: Expert, quick

Summary

AgRefactor introduces an LLM-based multi-agent workflow designed to refactor real-world software into High-Level Synthesis (HLS)-compatible programs. This system addresses the significant challenges of HLS conversion, which stem from restrictive language support and the disparity between software and hardware programming practices, issues that existing automated and LLM-based approaches often fail to resolve efficiently or robustly. AgRefactor incorporates a self-evolving memory system that accumulates and retrieves knowledge, enhancing its robustness and efficiency on unseen programs. It also integrates automated refactoring tools, allowing agents to strategically combine LLM-driven rewrites with efficient tool-based transformations to reduce cost and improve scalability. Benchmarking on 9 out of 11 challenging real-world cases, which are 5-10x longer than previous studies, AgRefactor outperforms or matches state-of-the-art tools. It further achieves a 6.51x geometric mean speedup over the SoTA pragma tuning tool and a 1.20x speedup over optimized open-source designs with under 20% extra resources. AgRefactor is fully-automated and open-sourced.

Key takeaway

For AI Hardware Engineers or HLS developers struggling with converting complex software into synthesizable HLS code, AgRefactor provides a significant advancement. If you are facing challenges with restrictive language support or optimizing performance, this open-sourced, fully-automated multi-agent LLM workflow can dramatically improve compatibility and achieve substantial speedups. You should explore integrating AgRefactor to streamline your HLS development process, potentially gaining a 6.51x speedup and 1.20x over optimized designs with minimal extra resources.

Key insights

AgRefactor employs a self-evolving multi-agent LLM workflow and integrated tools to efficiently refactor software for HLS compatibility and performance.

Principles

• Combine LLM agents with automated tools for efficiency.
• Self-evolving memory improves robustness and scalability.
• Address HLS compatibility and performance simultaneously.

Method

AgRefactor employs an LLM-based multi-agent workflow with a self-evolving memory system. It integrates automated refactoring tools to balance LLM-driven rewrites with efficient tool-based transformations for HLS compatibility and performance.

In practice

• Refactor complex software for HLS compatibility.
• Optimize HLS designs for speedup and resource efficiency.
• Utilize multi-agent LLM systems for code transformation.

Topics

• High-Level Synthesis
• LLM Refactoring
• Multi-Agent Systems
• Code Optimization
• Hardware Acceleration
• Self-Evolving Memory

Best for: Research Scientist, AI Scientist, AI Engineer, AI Hardware Engineer

Related on AIssential

• Autogenesis: A Self-Evolving Agent Protocol — Autogenesis Protocol enables self-evolving LLM agents through structured resource management and auditable closed-loop evolution.
• HyEvo for Topological Reasoning Graph Optimization — A self-evolving system autonomously designs optimal hybrid LLM-code agent workflows, eliminating manual architecture design.
• 🤖 AI Agents Weekly: Hyperagents, Multi-Agent Harness Design, Chroma Context-1, Composer 2, ARC-AGI-3, and More — Hyperagents enable recursive self-improvement by allowing a meta agent to edit its own modification strategy.
• lsdefine / GenericAgent — GenericAgent is a minimal, self-evolving autonomous agent framework that builds skills through task execution.
• From Data to Theory: Autonomous Large Language Model Agents for Materials Science — Autonomous LLM agents can perform end-to-end scientific theory development, from equation selection to code execution and validation.
• When an AI can Rewrite Itself (Darwin-Gödel HyperAgent) — Hyper Agents merge task and meta-agents into a single Python script for meta-cognitive self-modification.

Open in AIssential →

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