[llvm] [AMDGPU] Fix VGPR lowering for V_DUAL_FMAMK_F32 (PR #170567)

[Stanislav Mekhanoshin via llvm-commits]

rampitec wrote:

> In another part of the docs I see:
> 
> > Instructions with non-standard operands
> > 
> > * FMAAK: uses SRC0 and SRC1 MSBs
> > * FMAMK: uses SRC0 and SRC2 MSBs
> 
> so do we need a special case for FMAAK too?

I think FMAAK is fine. The operand order is src0, src1, kimm, so it is natural.

https://github.com/llvm/llvm-project/pull/170567