LS-CAT

LS-CAT数据集由挪威科技大学计算机科学系创建，包含19,683个CUDA内核，主要关注线性代数。数据集大小为5,028,536个关联运行时，来源于GitHub，通过GH Archive项目收集。创建过程中，通过自动化脚本从GitHub下载并处理源代码，确保内核的可执行性。数据集应用于机器学习辅助的自动调优，旨在通过分析源代码特征与线程块大小的最优选择之间的相关性，提高软件性能。

The LS-CAT dataset was developed by the Department of Computer Science at the Norwegian University of Science and Technology. It encompasses 19,683 CUDA kernels and focuses primarily on linear algebra. The dataset comprises 5,028,536 associated runtime records, which are sourced from GitHub and collected via the GH Archive project. During the dataset construction process, automated scripts were utilized to download and process source code from GitHub, ensuring the executability of the included kernels. This dataset is applied to machine learning-assisted automatic tuning, with the goal of enhancing software performance by analyzing the correlation between source code features and the optimal selection of thread block sizes.