CUDA-Harness: Harnessing Agentic CUDA Kernel Generation and Optimization from Natural Language

Researchers introduced CUDA-Harness, a framework for generating and optimizing CUDA kernels directly from natural language (Text2CUDA), addressing the expertise barrier in high-performance kernel development. The framework introduces Intermediate-Structured Generation to bridge semantic understanding with low-level kernel generation, Synthesis-Based Verification to prevent reward hacking, and Feedback-Adaptive Evolution to prioritize correctness while optimizing performance. Experiments demonstrate effectiveness across LLMs, hardware platforms, and C-to-CUDA transpilation.

arXiv:2609.00058v1 Announce Type: new Abstract: 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.