SPARK00

SPARK00数据集由莱顿大学创建，专注于评估优化编译器处理不规则访问模式的能力。数据集包含5个矩阵，分别来自Harwell-Boeing矩阵集合和SPARSE库，用于模拟真实世界问题中的矩阵结构。这些矩阵具有不同的应用领域、非零密度和结构，如add32、utm5940、sherman3、codecs4812.dc和bcsstk13。数据集的创建旨在通过提供不同复杂度的访问模式，评估编译器在处理不规则代码时的效果，解决现代架构在遇到不规则内存访问时性能下降的问题。

SPARK00 dataset was developed by Leiden University, focusing on evaluating the capability of optimizing compilers to handle irregular memory access patterns. The dataset comprises 5 matrices sourced from both the Harwell-Boeing matrix collection and the SPARSE library, which are used to simulate matrix structures in real-world problems. These matrices vary across application domains, non-zero densities and structural characteristics, including add32, utm5940, sherman3, codecs4812.dc and bcsstk13. The dataset is constructed to evaluate compiler performance when handling irregular code by providing access patterns with varying complexities, addressing the performance degradation issue in modern architectures caused by irregular memory accesses.