发表机构
Shanghai Jiao Tong University（上海交通大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对自然语言生成CUDA内核的专业壁垒与奖励黑客问题，本文提出CUDA-Harness框架，通过中间结构化生成、基于合成的验证及反馈自适应进化实现内核生成优化，且泛化性良好。
AI 中文摘要
开发高性能CUDA内核需要算法实现、正确性验证及硬件感知并行优化的专业知识，形成了显著的专业壁垒，使得从自然语言生成CUDA内核（Text2CUDA）成为刚需。同时，大型语言模型（LLM）的通用代码生成能力推动了一系列基于LLM的CUDA内核生成研究，这些研究主要聚焦于从PyTorch等高级框架到CUDA的转译（Torch2CUDA），而非Text2CUDA，后者要求模型理解高级输入语义并处理低级内核实现与验证。此外，这些方法因依赖预定义测试输入而易受奖励黑客攻击。本文提出CUDA-Harness框架，用于利用智能体从自然语言生成与优化CUDA内核。具体而言，引入中间结构化生成（Intermediate-Structured Generation）以连接高级语义理解与低级内核生成；为缓解Text2CUDA中的奖励黑客攻击，构建基于合成的验证（Synthesis-Based Verification）以提供隔离测试数据与渐进式验证；进一步提出反馈自适应进化（Feedback-Adaptive Evolution），一种优先保证正确性同时优化性能的内核进化策略。最后，通过大量实验证明了CUDA-Harness的有效性，进一步评估显示其在不同LLM、硬件平台及C转译CUDA（C-to-CUDA）任务上具备泛化能力。
英文摘要
Developing high-performance CUDA kernels demands specialized knowledge in algorithm implementation, correctness validation, and hardware-aware parallel optimization, creating a substantial expertise barrier and making generating CUDA kernels directly from natural language (Text2CUDA) essential. Meanwhile, the general-purpose code generation capability of Large Language Models (LLMs) prompts a series of works exploring LLM-based CUDA kernel generation. They mainly focus on transpilation from high-level frameworks such as PyTorch to CUDA (Torch2CUDA) rather than Text2CUDA, where models must understand the high-level input semantics and handle low-level kernel implementation and validation. Additionally, these methods are vulnerable to reward hacking due to reliance on predefined test inputs. In this paper, we propose CUDA-Harness, a framework for harnessing agentic CUDA kernel generation and optimization from natural language. Specifically, we introduce Intermediate-Structured Generation to connect high-level semantic understanding with low-level kernel generation. To dilute reward hacking in Text2CUDA, we construct Synthesis-Based Verification to provide isolated test data and progressive validation. Furthermore, we propose Feedback-Adaptive Evolution, a kernel evolution strategy that prioritizes correctness while optimizing performance. Finally, through extensive experiments, we demonstrate the effectiveness of CUDA-Harness, with further evaluations illustrating generalization across LLMs, hardware platforms, and to C-to-CUDA transpilation.
