Refactor `thrust::reduce_by_key` to use `cub::DeviceReduce::ReduceByKey`
Open
- Dominant language
- C++
- Stars
- 2.5k
- Forks
- 486
- Avg merge
- 2d 6h
- Merged PRs (30d)
- 295
Description
- [x] https://github.com/NVIDIA/cccl/issues/5438
- [x] Refactor `thrust::reduce_by_key` to use `cub::DeviceReduce::ReduceByKey`
- [x] Add tests that checks whether equality_op is invoked for oob items
- [x] Add tests that verifies accumulator type handling is kept unchanged
- [x] Add sanity tests for large number of items to thrust
- [x] Benchmark CUB performance versus original thrust performance (see https://github.com/NVIDIA/cccl/issues/3942)
Contributor guide
Assessment
This issue has not been assessed yet.