Research Area
Building the compiler infrastructure that empowers scientists and engineers to transform, analyze, and optimize large-scale scientific codes — without needing to become compiler experts themselves. The ROSE compiler framework is Chunhua's flagship contribution to this space, enabling a new generation of HPC tooling.
Compiler technology is the invisible backbone of every high-performance application — but historically, building compiler tools required deep, specialized expertise that put them out of reach for most developers. Chunhua's work on the ROSE compiler infrastructure fundamentally changed this by providing a mature, open-source framework for constructing program analysis and transformation tools for C, C++, Fortran, Ada, and Java. ROSE empowers domain scientists, HPC engineers, and tool builders to write powerful source-to-source transformations without needing to understand the full complexity of compiler internals.
A key insight behind ROSE's design is the source-to-source transformation approach: rather than operating on low-level intermediate representations (IRs), ROSE works at the source-code level. This preserves developer intent, keeps generated code readable and maintainable, and allows transformations to be validated by human experts. This philosophy makes ROSE uniquely suited for scientific computing, where code longevity and human understandability are paramount.
More recently, Chunhua has expanded this vision by integrating Large Language Models (LLMs) into compiler workflows — using AI to interpret compiler optimization reports, suggest code transformations, and assist with language migrations. This intersection of classical compiler theory and modern AI represents the frontier of his current research, culminating in tools like CompilerGPT that make compiler intelligence accessible to every HPC developer.
Introduces CompilerGPT, a system that uses large language models to interpret, explain, and act upon compiler optimization reports — bridging the gap between cryptic compiler feedback and actionable developer guidance. Directly extends ROSE's mission of making compiler technology accessible.
Extends ROSE's language coverage to Ada, enabling static analysis and transformation of safety-critical and legacy Ada codebases used in defense and aerospace applications. Demonstrates ROSE's extensibility as a multi-language platform.
A novel compiler-AI integration technique that uses reductive analysis — progressively simplifying program context — to guide LLMs in identifying and applying code optimizations in complex scientific programs. Published at PLDI, the premier programming languages venue.
The foundational work describing the ROSE compiler infrastructure for source-to-source analysis and transformation. Recognized with the prestigious R&D 100 Award in 2009, ROSE has been deployed in numerous DOE laboratory projects and academic institutions worldwide.
An open-source compiler framework for building source-level program analysis, transformation, and optimization tools for C, C++, Fortran, Ada, Java, and Python. Used at LLNL, DOE labs, and universities worldwide. Enables non-compiler-experts to build powerful static analysis tools. Winner of the R&D 100 Award in 2009.
LLM-powered tool that translates compiler optimization reports into plain-English explanations — helping HPC developers understand what the compiler did (or couldn't do) and how to improve their code. Built on top of ROSE analysis.
Compiler technology should not require a PhD in compilers. ROSE was built on the principle that scientists and engineers — not just compiler researchers — should be able to analyze and transform their own code.