NVIDIA
[FEA]: Use `CUB_IF_CONSTEXPR` in checks on compile-time constants
Is this a duplicate?
- [X] I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct
Area
CUB
Is your feature request related to a problem? Please describe.
There are many checks in the codebase that only depend on compile-time parameters.
A few examples: https://github.com/NVIDIA/cccl/blob/7c53bbd8991effb05707583474c3b59756812678/cub/cub/device/dispatch/dispatch_radix_sort.cuh#L659 https://github.com/NVIDIA/cccl/blob/7c53bbd8991effb05707583474c3b59756812678/cub/cub/agent/agent_merge_sort.cuh#L178 https://github.com/NVIDIA/cccl/blob/7c53bbd8991effb05707583474c3b59756812678/cub/cub/agent/agent_scan.cuh#L357
Describe the solution you'd like
When compiling with C++17 or newer, if constexpr can be used to specify that the branch should be evaluated at compile time.
The cub macro CUB_IF_CONSTEXPR uses if constexpr when supported.
Describe alternatives you've considered
No response
Additional context
No response
Note, we should also move this to a central place inside the __cccl folder inside libcu++ as this is a generic facility that should be reused throughout the whole project.
We could even consider expanding support for clang / msvc or other compiler that allow using this feature in earlier standard modes
