[llvm] [X86] combineConcatVectorOps - concat per-lane v2f64/v4f64 shuffles into vXf64 vshufpd (PR #143017)

[Slava Zakharin via llvm-commits]

vzakhari wrote:

> Can you provide a asm diff?

Sure, I will.  I just need to verify that the regression happens exactly after this commit.  On the above range of commits, I do see changes in the shuffling, but I also see some reordering of FP math operations (which might be caused by some other pass kicking in after your change).  I suppose `-fassociative-math` allows that reordering, but I will double check.

https://github.com/llvm/llvm-project/pull/143017