[FEA]: Analyze segment_sort/keys benchmark performance

[danielfrg]

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

cuda.compute (Python)

Is your feature request related to a problem? Please describe.

Follow up from the new nvbench Python benchmarks and comparison with C++ ones: #7341

We are getting about 90% performance of C++ on the tranform/heavy benchmark.

Describe the solution you'd like

We have a target of 5% performance gap with C++ for large amount of items

Describe alternatives you've considered

No response

Additional context

No response