A Performance Comparison of Sort and Scan Libraries for GPUs
A Performance Comparison of Sort and Scan Libraries for GPUs
复制标题
GPU 排序和扫描库的性能比较
DOI：
10.1142/s0129626415500073
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
Bruce Merry
中科院分区：
文献类型：
--
作者：
Bruce Merry
排序和扫描是构造高度并行算法的两个基本原语。许多库现在提供了这些原语的GPU实现，但关于这些实现的性能的信息相对较少。我们对七个库进行了32位整数扫描和排序的基准测试，并通过32位整数键对32位值进行排序。我们发现，有一个很大的变化，性能之间的图书馆，没有一个图书馆既有最佳的性能和可移植性。
Sorting and scanning are two fundamental primitives for constructing highly parallel algorithms. A number of libraries now provide implementations of these primitives for GPUs, but there is relatively little information about the performance of these implementations. We benchmark seven libraries for 32-bit integer scan and sort, and sorting 32-bit values by 32-bit integer keys. We show that there is a large variation in performance between the libraries, and that no one library has both optimal performance and portability.