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.

The following functions

• Countl_zero
• Countl_one
• Countr_zero
• Countr_one
• popcount
• rotl
• rotr
• bit_floor
• bit_ceil
• has_single_bit
• byteswap

require to be constexpr which prevents potential optimizations such as intrinsics and asm.

Describe the solution you'd like

Such functions should be mapped to GPU ISA via PTX and/or intrinsics

• Countl_zero → FLO + IADD3 (__clz) / FLO.SH (PTX: bfind_shift)
• Countr_zero → Countl_zero(__brev(x))
• Countl_one → Countl_zero (~x)
• Countr_one → Countl_one(__brev(x))
• popcount → __popc
• rotl/rotr → PTX: funnel shift
• bit_floor, bit_ceil → FLO + shift
• has_single_bit → __popc
• byteswap → PRMT

Describe alternatives you've considered

No response

Additional context

No response