[llvm] [AArch64] Fold vector select with power-of-2 bit-test to CMTST+BSP (PR #209100)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

> I tried updating the test using the update test script. Currently, the GI instruction selection framework is still generating the AND+CMPEQ sequence, since our optimization only applies to the SD framework.
> 
> I’m not very familiar with GI yet. Would it be possible to add the corresponding optimization for GI separately at a separate PR?

Yeah sounds good. GISel is currently still under constructions for anything that isn't -O0, so we can easily add it to the list of things that we need to fix up.

> Also, one more regression in combinemul test, I will add guard to fix that

It might be OK if it is just a degenerate case.

https://github.com/llvm/llvm-project/pull/209100