v6.0

Release 6.0 introduces several key enhancements to our software. The update brings new APIs for partial sorting of key-value pairs, along with comprehensive AVX2 support across all routines. Additionally, PyTorch now leverages AVX-512 and AVX2 key-value sort routines to boost the performance of torch.sort and torch.argsort functions by up-to 10x (see pytorch/pytorch#127936). We thank the community for all their contributions. Here is a quick summary of all the changes:

New features:

• Support for qselect and partial sort for key-value data types #151.
• Accelerate key-value sort using openMP pragmas #153.
• AVX2 support for key-value sort, partial sort and objsort methods #145, .
• Simply use of ISA specific sort methods with a unified static API #143.
• Support for descending order sort for all the sort routines #140, #144, #151.
• Intel LLVM compiler support #131.

Improvements:

• Improved pivot selection for better performance on data with few unique values #127, #168.

New Contributors

• @cmsxbc made their first contribution in #155
• @ron-at-swgy made their first contribution in #157
• @yixuan made their first contribution in #159
• @WilliamTambellini made their first contribution in #164
• @yinqiwen made their first contribution in #165

Full Changelog: v5.0...v6.0