[llvm] [AArch64][ISel] Prefer SVE over NEON for fixed-width ASL LSR (PR #199003)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

The code looks OK to me but It looks like the neon ushl has more register freedom than the sve lsr. Unfortunately the SVE instructions can require an extra movprfx which either isnt always free or is at least more instructions in the loop.

Which means that whilst this is profitable outside of loops it might not be profitable inside one, depending on if there are other later uses of the value being shifted. We might need to move this later in the pipeline so that we can only use the new instruction when profitable.

https://github.com/llvm/llvm-project/pull/199003