[llvm] [LoopIdiom] Use costing to determine CRC strategy (PR #211040)

[Sean Clarke via llvm-commits]

xarkenz wrote:

> > AMD EPYC 9654, x86_64 (using current version of #211113):
> 
> It looks like it fixes TC=16, but breaks TC=32.

It does, so either this patch or #211113 could probably use some adjustment. I think part of the problem is that the speed of `pclmulqdq` varies quite a lot across different X86 CPUs, and EPYC is on the faster side. At the very least, this isn't a performance *regression*.

https://github.com/llvm/llvm-project/pull/211040