This HiRSE Seminar Series webinar will introduce CompilerGPT, a framework that connects compilers, large language models, and evaluation workflows to help programmers act on compiler optimization reports.
| Event Information | Details |
|---|---|
| Webinar Title | CompilerGPT: Leveraging Large Language Models to Analyze and Act on Compiler Optimization Reports |
| Date and Time | 2026-10-22 4:00 pm CEST |
| Presenter | Dr. Peter Pirkelbauer (Lawrence Livermore National Laboratory Center for Applied Scientific Computing) |
| Request Connection Information | hirse@fz-juelich.de |
| Series Information | https://www.helmholtz-hirse.de/series.html |
| Archives | Recordings (YouTube), Slides (Zenodo) |
This event is part of the HiRSE Seminar Series, a series of online talks on topics relevant to research software engineering and research software engineers.
Abstract
Compiler optimization reports contain valuable insights, but their technical complexity often makes them difficult for programmers to interpret and act upon. This talk presents CompilerGPT, a framework that connects compilers, large language models, and a user-definable evaluation harness in an iterative compile, optimize, and evaluate workflow. CompilerGPT uses optimization reports, source code, compiler diagnostics, and performance measurements to guide LLM-generated code transformations while checking correctness and tracking results.
The seminar will describe the framework, its configurable compiler and model interfaces, and experimental results using Clang and GCC, multiple LLMs, and representative sequential and OpenMP benchmarks. It will review demonstrated speedups of up to 6.5x, along with the variability, cost, and user involvement observed across workloads. The talk will also present updated results and examine CompilerGPT in the broader context of coding agents, discussing how compiler feedback, testing, profiling, and performance evaluation can serve as tools within agentic software-development systems.
Presenter
- Dr. Peter Pirkelbauer (Lawrence Livermore National Laboratory Center for Applied Scientific Computing)
Please check the event website for full details and updates.
