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

libcu++

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

Default mdspan accessor is not sufficient to provide the best performance in some scenarios where specialized CUDA features can be beneficial

Describe the solution you'd like

Provide optimized mdspan accessors:

• cache modifiers/evict policies
• restrict pointers
• assume aligned

Describe alternatives you've considered

No response

Additional context

No response