Authors Jianguo Yao, Haibing Guan, Jinghao Wang, Qiqi Gu, Chenpeng Wu, Xijun Li
Publication date 21 August 2026
Tldr This paper introduces HIERA, a hierarchical search-space planning framework for GPU kernel optimization that selects an appropriate implementation space across PyTorch operators, CUDA libraries, and custom CUDA kernels based on workload characteristics and expert knowledge. By doing so, it delivers stronger overall performance than existing training-free methods while remaining competitive with the training-based CUDA-L1 without additional model training.
This page was built for publication: HIERA: Workload-Aware Planning Across Implementation Spaces for GPU Kernel Optimization
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7285062)
