[llvm] [SLP] Prefer copyable vectorization over alternate opcodes (PR #153684)

[Alexey Bataev via llvm-commits]

alexey-bataev wrote:

> > MultiSource/Benchmarks/Prolangs-C/simulator - small regressions, caused by early optimizations, should improved with future patches
> 
> What do these regressions look like and can they be yak shaved first?

They depend on other copyables opcodes support, not sure it is possible to fix it right now. The regression is pretty small, like we're missing something like single 2 x vectorization

https://github.com/llvm/llvm-project/pull/153684