.png){kind=icon}
.png){kind=icon}
Return
Improving MPI Error Detection and Repair with Large Language Models and Bug References
.png){kind=icon}
PRE
AI
S
Y
S
L
DOI：10.1016/j.jpdc.2026.105255.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Proposes a novel MPI error detection and repair framework leveraging Large Language Models (LLMs) with Few-Shot Learning, Chain-of-Thought reasoning, and bug references. • Improves error detection accuracy in MPI programs from 44.39% to 77.84%, with a 96% reduction in false negatives compared to baseline ChatGPT usage. • Demonstrates superior performance over state-of-the-art static and dynamic MPI analysis tools in both precision and recall. • Automatically repairs 84.5% of defective MPI programs, including challenging bugs such as deadlocks, resource leaks, and parameter mismatches. • Introduces a reusable, domain-informed prompt engineering methodology that enhances LLM performance in parallel programming contexts.
Keywords:
MPI error detection
Large Language Models
bug references
parallel programming
prompt engineering
Journal
IF:
4
Papers:
3.8K
Citations:
4.8K
Organization
Cited Papers