[llvm] [AArch64][GlobalISel] Add pattern to prevent scalar uqxtn fallback (PR #201546)

[Joshua Rodriguez via llvm-commits]

JoshdRod wrote:

> Seems OK to me. There is presumably a sqxtn and sqxtun too that could work the same way. Did you want to add them here or do it in a followup?

I was planning to get feedback here and do the rest in a followup - as I'm pretty sure the rest will be exactly the same fix.

> I wouldn't expect to write a new pattern for an intrinsic which works with selectiondag already

The vector version of this intrinsic lowers using a generic opcode. I feel like it doesn't make sense to create a new GI node for this one case - what do you think?


https://github.com/llvm/llvm-project/pull/201546