[llvm] [SLPVectorizer][NVPTX] Customize getBuildVectorCost for NVPTX (PR #128077)

[via llvm-commits]

peterbell10 wrote:

I think your example is a bit misleading because it includes the argument passing convention, if we read the values from gmem instead the mov becomes a single SASS instruction:
https://godbolt.org/z/rKeGTrj96

I am a bit surprised though that ptxas keeps those extra `PRMT` calls in for the argument passing case. AFAICT it's zeroing out the top 16 bits, then immediately discarding them.

https://github.com/llvm/llvm-project/pull/128077