Structure-aware AI improves high-level code optimization significantly
Structure-augmented LLMs for High-Level Synthesis Pragma Optimization
Hardware Architecture
Summary
Writing special instructions called pragmas helps make hardware designs work better, but it's tricky and needs expert knowledge. The authors designed an AI model named PRISM that understands the structure of code better by looking at how the code’s parts connect and flow. This helps PRISM suggest better pragmas than other models, making hardware designs faster and more efficient. It works well even without extra training on new examples.
What this means in practice
- •For hardware design engineers: Use the PRISM model to automatically add pragmas that improve hardware component performance during design.
- •For software tooling developers: Integrate PRISM’s structure-aware approach into code generation tools to enhance optimization suggestions for complex software.
Authors
Haocheng Xu, Ye Qiao, Phyo Pyae Moe Aung, Alok Mishra, Pavana Prakash, Rolando Pablo Hong Enriquez, Adam Han Wu, Zhiheng Chen, Dejan Milojicic, Sitao Huang
Abstract
Pragma insertion drives the quality of high-level synthesis (HLS) designs. Choosing the right directives demands expert knowledge and reasoning about loop nesting, data dependences, and memory layout. While existing large language models (LLMs) show promise in code generation, they lack explicit program-structure awareness, limiting their ability to suggest effective pragmas. We present PRISM, a novel structure-augmented LLM that closes this gap by adding compiler-grade structural reasoning to a pretrained, frozen code LLM. It combines three hierarchical program representations, Abstract Syntax Tree (AST), Control-Flow Graph (CFG), and Data-Flow Graph (DFG), injecting them into a specific transformer layer while keeping original code tokens in a separate stream. The cross-attention gate at the injection point allows falling back to the pretrained representation when its structural signal is unhelpful. On zero-shot evaluation in HLS-Eval, PRISM synthesizes 3.5\times as many kernels as Llama3-8B (26.9\% vs. 7.7\%), and on the kernels where it does succeed, it produces designs that are 2.31\times faster (geomean) than GPT-5-mini's. In the agentic flow, the PRISM codegen outperforms other baselines when optimizing complex code and drives the average normalized improvement across the HLS-Eval suite to 26.4\%.