[llvm-branch-commits] [llvm] release/18.x: [llvm][LoongArch] Improve loongarch_lasx_xvpermi_q instrinsic (#82984) (PR #83540)

[Lu Weining via llvm-branch-commits]

SixWeining wrote:

> I have some doubt about this change.
> 
> To me if the user requests `xvpermi.q` via the `loongarch_lasx_xvpermi_q` intrinsic, we should give her/him the `xvpermi.q` instruction. If (s)he is passing an invalid operand then (s)he is invoking the undefined behavior herself/himself and we don't need to guarantee a thing.
> 
> So to me we should not merge this and we should revert this change for main. Or am I missing something? @xen0n @heiher @SixWeining @MaskRay

Yes, it can be argued. But I know gcc has similar change.

https://github.com/llvm/llvm-project/pull/83540