CUDA-LLM: LLMs Can Write Efficient CUDA Kernels

TL;DR

This paper introduces CUDA-LLM, a framework that uses large language models combined with a novel optimization method to generate high-performance, architecture-aware CUDA kernels that outperform human-written code in speed.

Contribution

The paper presents FSR, a new framework that jointly optimizes correctness and performance of CUDA code generated by LLMs, enabling automated, efficient GPU kernel creation.

Findings

LLMs with FSR achieve high correctness rates.

Generated kernels outperform human code by up to 179× in speed.

Framework effectively tailors CUDA code to specific GPU architectures.

Abstract

Large Language Models (LLMs) have demonstrated strong capabilities in general-purpose code generation. However, generating the code which is deeply hardware-specific, architecture-aware, and performance-critical, especially for massively parallel GPUs, remains a complex challenge. In this work, we explore the use of LLMs for the automated generation and optimization of CUDA programs, with the goal of producing high-performance GPU kernels that fully exploit the underlying hardware. To address this challenge, we propose a novel framework called \textbf{Feature Search and Reinforcement (FSR)}. FSR jointly optimizes compilation and functional correctness, as well as the runtime performance, which are validated through extensive and diverse test cases, and measured by actual kernel execution latency on the target GPU, respectively. This approach enables LLMs not only to generate syntactically and semantically correct CUDA code but also to iteratively refine it for efficiency, tailored to the characteristics of the GPU architecture. We evaluate FSR on representative CUDA kernels, covering AI workloads and computational intensive algorithms. Our results show that LLMs augmented with FSR consistently guarantee correctness rates. Meanwhile, the automatically generated kernels can outperform general human-written code by a factor of up to 179$\times$ in execution speeds. These findings highlight the potential of combining LLMs with performance reinforcement to automate GPU programming for hardware-specific, architecture-sensitive, and performance-critical applications.