arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2602.22240cs.PLcs.AIcs.DC
收藏

从提示到性能:评估LLM在基于任务的并行代码生成中的表现

From Prompts to Performance: Evaluating LLMs for Task-based Parallel Code Generation

Institute for Parallel and Distributed Systems(并行与分布式系统研究所)

Linus Bantel, Moritz Strack, Alexander Strack, Dirk Pflüger

AI总结:

本文评估了LLM在基于任务的并行代码生成中的表现,通过三种输入提示和三个编程框架,分析了LLM在正确性和可扩展性方面的优缺点。

摘要:

大型语言模型(LLM)在代码生成方面表现出强大的能力,但其在创建高效并行程序方面的技能研究较少。本文探讨了LLM如何从三种类型的输入提示生成基于任务的并行代码:自然语言问题描述、顺序参考实现和并行伪代码。我们专注于三个编程框架:OpenMP Tasking、C++标准并行性和异步多任务运行时HPX。每个框架都为任务执行提供了不同的抽象层次和控制程度。我们评估LLM生成的解决方案在正确性和可扩展性方面的表现。我们的结果揭示了LLM在问题复杂性和框架方面的优缺点。最后,我们讨论这些发现对未来LLM辅助的高性能和科学计算开发的意义。

英文摘要:

Large Language Models (LLM) show strong abilities in code generation, but their skill in creating efficient parallel programs is less studied. This paper explores how LLMs generate task-based parallel code from three kinds of input prompts: natural language problem descriptions, sequential reference implementations, and parallel pseudo code. We focus on three programming frameworks: OpenMP Tasking, C++ standard parallelism, and the asynchronous many-task runtime HPX. Each framework offers different levels of abstraction and control for task execution. We evaluate LLM-generated solutions for correctness and scalability. Our results reveal both strengths and weaknesses of LLMs with regard to problem complexity and framework. Finally, we discuss what these findings mean for future LLM-assisted development in high-performance and scientific computing.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑