Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

C2HLSC: Leveraging Large Language Models to Bridge the Software-to-Hardware Design Gap

Luca Collini, Siddharth Garg, Ramesh Karri

TL;DR

This paper tackles the software-to-hardware gap by using large language models to refactor generic C code into HLS-synthesizable C, enabling hardware acceleration via high-level synthesis. It introduces the C2HLSC framework, a hierarchical, automated pipeline that splits designs into function-level tasks, iteratively rewrites code, and then auto-inserts pragmas to optimize area or throughput. Case studies and a fully automated evaluation on real-world benchmarks (AES, DES, Present, NIST randomness tests, and QuickSort) demonstrate high success rates on substantially more complex designs than prior LLM-to-Verilog efforts, while highlighting the need for synthesis-feedback loops to refine optimization. The approach promises substantial time and cost savings for hardware design and provides an open-source platform for broader adoption and extension to more complex C/C++ codebases. The work advances hardware software co-design by leveraging LLMs at a higher abstraction level and integrating automated testing and synthesis feedback to produce viable HLS-ready C code.

Abstract

High-Level Synthesis (HLS) tools offer rapid hardware design from C code, but their compatibility is limited by code constructs. This paper investigates Large Language Models (LLMs) for automatically refactoring C code into HLS-compatible formats. We present a case study using an LLM to rewrite C code for NIST 800-22 randomness tests, a QuickSort algorithm, and AES-128 into HLS-synthesizable C. The LLM iteratively transforms the C code guided by the system prompt and tool's feedback, implementing functions like streaming data and hardware-specific signals. With the hindsight obtained from the case study, we implement a fully automated framework to refactor C code into HLS-compatible formats using LLMs. To tackle complex designs, we implement a preprocessing step that breaks down the hierarchy in order to approach the problem in a divide-and-conquer bottom-up way. We validated our framework on three ciphers, one hash function, five NIST 800-22 randomness tests, and a QuickSort algorithm. Our results show a high success rate on benchmarks that are orders of magnitude more complex than what has been achieved generating Verilog with LLMs.

C2HLSC: Leveraging Large Language Models to Bridge the Software-to-Hardware Design Gap

TL;DR

This paper tackles the software-to-hardware gap by using large language models to refactor generic C code into HLS-synthesizable C, enabling hardware acceleration via high-level synthesis. It introduces the C2HLSC framework, a hierarchical, automated pipeline that splits designs into function-level tasks, iteratively rewrites code, and then auto-inserts pragmas to optimize area or throughput. Case studies and a fully automated evaluation on real-world benchmarks (AES, DES, Present, NIST randomness tests, and QuickSort) demonstrate high success rates on substantially more complex designs than prior LLM-to-Verilog efforts, while highlighting the need for synthesis-feedback loops to refine optimization. The approach promises substantial time and cost savings for hardware design and provides an open-source platform for broader adoption and extension to more complex C/C++ codebases. The work advances hardware software co-design by leveraging LLMs at a higher abstraction level and integrating automated testing and synthesis feedback to produce viable HLS-ready C code.

Abstract

High-Level Synthesis (HLS) tools offer rapid hardware design from C code, but their compatibility is limited by code constructs. This paper investigates Large Language Models (LLMs) for automatically refactoring C code into HLS-compatible formats. We present a case study using an LLM to rewrite C code for NIST 800-22 randomness tests, a QuickSort algorithm, and AES-128 into HLS-synthesizable C. The LLM iteratively transforms the C code guided by the system prompt and tool's feedback, implementing functions like streaming data and hardware-specific signals. With the hindsight obtained from the case study, we implement a fully automated framework to refactor C code into HLS-compatible formats using LLMs. To tackle complex designs, we implement a preprocessing step that breaks down the hierarchy in order to approach the problem in a divide-and-conquer bottom-up way. We validated our framework on three ciphers, one hash function, five NIST 800-22 randomness tests, and a QuickSort algorithm. Our results show a high success rate on benchmarks that are orders of magnitude more complex than what has been achieved generating Verilog with LLMs.

Paper Structure

This paper contains 25 sections, 12 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Background
  3. High-Level Synthesis: capabilities and limitations
  4. Large Language Models (LLMs)
  5. Related works
  6. Interactive Approach: a Case Study
  7. Overview
  8. Methodology
  9. Case Study Results
  10. C2HLSC of NIST 800-22 Randomness Tests
  11. C2HLSC of QuickSort
  12. C2HLSC of AES 128 Encrypt
  13. Case Study Discussion
  14. Fully Automated C2HLSC Framework
  15. Hierarchical Preprocessing
  16. ...and 10 more sections

Figures (12)

  • Figure 1: Flow used in the C2HLSC case-study.
  • Figure 2: Transforming into a streaming pipeline. On the left, in green, human prompt. On the right, in purple, LLM response
  • Figure 3: Removing pointers and inlining. On the left, in green, human prompt. On the right, in purple, LLM response.
  • Figure 4: Removing pointers. On the left, in green, human prompt. On the right, in purple, LLM response
  • Figure 5: Automated C2HLSC LLM-based prototype flow.
  • ...and 7 more figures