Vibe Coding HPC: AI-Assisted Development of DTC for Emerging Architectures¶
Johannes Gebert (HLRS)
Abstract¶
"Vibe coding" steers software development through natural-language prompts to AI coding assistants. It is rapidly entering high-performance computing, promising to compress the effort needed to port and maintain code across fragmented hardware landscapes. This talk examines what vibe coding means in a strict HPC sense, using the struct-process/DTC codebase as a case study. We trace DTC's development on Fugaku (R-CCS) and its evolution into pyDTC, a Python-based reimplementation targeting future computing hardware. We present a real-world demonstration of AI-assisted porting, discuss the cost relative to traditional manual porting, and weigh the possibilities it opens for adapting scientific software to new architectures. The talk closes with a forward-looking estimate of how AI-assisted development is likely to reshape HPC software engineering over the coming years, including implications for portability, developer productivity, and code longevity across current and next-generation systems.