[FEA]: Guarantee monotonic-increating temp_storage_bytes (for d_temp_storage) wrt num_items

[akeley98NV]

Is this a duplicate?

• 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.

For many of my algorithms I will be executing a series of cub functions with varying num_items values and it is cost-prohibitive to allocate d_temp_storage each time. I cannot predict ahead-of-time what num_items will be but I can guarantee a tight bound on the max value. I currently query temp_storage_bytes with this bound value and allocate a block of memory of the queried size to use for all subsequent cub functions. This works in practice but so far as I can tell, is not formally guaranteed to work.

Describe the solution you'd like

Guarantee that (for the same function and same template parameters), the output temp_storage_bytes value is a non-strict monotonic increasing function of num_items.

Describe alternatives you've considered

Nvidians: don't suggest workarounds and alternative solutions before reading internal discussion on Slack cdd-cub.

Additional context

No response

[github-actions]

Hi @akeley98NV!

Thanks for submitting this issue - the CCCL team has been notified and we'll get back to you as soon as we can!
In the mean time, feel free to add any relevant information to this issue.