[llvm] [AMDGPU][True16][CodeGen] 16bit inline constant check for COPY (PR #150317)

[Brox Chen via llvm-commits]

broxigarchen wrote:

> > > > It seems we don't have any lit test that generates a COPY with an inline constant.
> > > 
> > > 
> > > If this is also the case downstream, including at the revision where this logic was brought in, it should be removed there.
> > 
> > 
> > Oh I mean I am not sure if we need this isCopyInstr check. Do we generate COPY with inline constant?
> 
> COPY only has register inputs. This is figuring out the mov instruction that will be produced if the mov is folded into the copy. i.e. this needs to be something like `%reg = v_mov_b16 imm; %copy = COPY %reg`

It seems we should merge this then. Thanks Matt!

https://github.com/llvm/llvm-project/pull/150317