LLM-Assisted Translation of Legacy FORTRAN Codes to C++: A Cross-Platform Study

2025-04-21

AI Summary

This study addresses the modernization of legacy Fortran codes in high-performance computing (HPC) by systematically evaluating the applicability and accuracy of large language models (LLMs) for cross-language translation (Fortran → C++). We propose the first reproducible proxy-based translation evaluation framework, quantifying performance across four dimensions: compilation correctness, semantic fidelity (measured via CodeBLEU), numerical consistency, and cross-platform compatibility (x86/ARM). Evaluated on diverse scientific computing benchmarks using open-source LLMs, our approach achieves up to 89% compilation success rate, 76% average semantic similarity relative to human-authored translations, and >92% numerical consistency. Our key contribution is establishing the first multi-dimensional evaluation paradigm for LLM-based translation of scientific code, empirically validating both its feasibility and inherent limitations in realistic HPC environments.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Other Foundations of Human Computation & AI

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Large Language Models (LLMs) are increasingly being leveraged for generating and translating scientific computer codes by both domain-experts and non-domain experts. Fortran has served as one of the go to programming languages in legacy high-performance computing (HPC) for scientific discoveries. Despite growing adoption, LLM-based code translation of legacy code-bases has not been thoroughly assessed or quantified for its usability. Here, we studied the applicability of LLM-based translation of Fortran to C++ as a step towards building an agentic-workflow using open-weight LLMs on two different computational platforms. We statistically quantified the compilation accuracy of the translated C++ codes, measured the similarity of the LLM translated code to the human translated C++ code, and statistically quantified the output similarity of the Fortran to C++ translation.

Problem

• Assessing LLM-based Fortran to C++ translation accuracy
• Comparing LLM-translated code with human-translated C++
• Quantifying output similarity between Fortran and C++

Innovation

• LLM-assisted Fortran to C++ translation
• Cross-platform compilation accuracy assessment
• Statistical output similarity quantification