Release CUB 2.1.0 · NVIDIA/cub

Breaking Changes

#553: Deprecate the CUB_USE_COOPERATIVE_GROUPS macro, as all supported CTK distributions provide CG. This macro will be removed in a future version of CUB.

#359: Add new DeviceBatchMemcpy algorithm.
#565: Add DeviceMergeSort::StableSortKeysCopy API. Thanks to David Wendt (@davidwendt) for this contribution.
#585: Add SM90 tuning policy for DeviceRadixSort. Thanks to Andy Adinets (@canonizer) for this contribution.
#586: Introduce a new mechanism to opt-out of compiling CDP support in CUB algorithms by defining CUB_DISABLE_CDP.
#589: Support 64-bit indexing in DeviceReduce.
#607: Support 128-bit integers in radix sort.

#547: Resolve several long-running issues resulting from using multiple versions of CUB within the same process. Adds an inline namespace that encodes CUB version and targeted PTX architectures.
#562: Fix bug in BlockShuffle resulting from an invalid thread offset. Thanks to @sjfeng1999 for this contribution.
#564: Fix bug in BlockRadixRank when used with blocks that are not a multiple of 32 threads.
#579: Ensure that all threads in the logical warp participate in the index-shuffle for BlockRadixRank. Thanks to Andy Adinets (@canonizer) for this contribution.
#582: Fix reordering in CUB member initializer lists.
#589: Fix DeviceSegmentedSort when used with bool keys.
#590: Fix CUB’s CMake install rules. Thanks to Robert Maynard (@robertmaynard) for this contribution.
#592: Fix overflow in DeviceReduce.
#598: Fix DeviceRunLengthEncode when the first item is a NaN.
#611: Fix WarpScanExclusive for vector types.

#537: Add detailed and expanded version of a [CUB developer overview](https://github.com/NVIDIA/cub/blob/main/docs/developer_overview.md).
#549: Fix BlockReduceRaking docs for non-commutative operations. Thanks to Tobias Ribizel (@upsj) for this contribution.
#606: Optimize CUB’s decoupled-lookback implementation.