[llvm] [AMDGPU][GFX11] Do not rewrite V_FMA/FMAC_* to V_FMAAK_F16_t16 on operand legalization. (PR #66202)

[Ivan Kosarev via llvm-commits]

================
@@ -3318,6 +3318,12 @@ bool SIInstrInfo::FoldImmediate(MachineInstr &UseMI, MachineInstr &DefMI,
       if (pseudoToMCOpcode(NewOpc) == -1)
         return false;
 
+      // V_FMAAK_F16_t16 takes VGPR_32_Lo128 operands, so the rewrite
+      // would also require restricting their register classes. For now
----------------
kosarev wrote:

I'm not sure it is of any practical value. For fake True16 instructions, it seems we already have enough support and generally don't want to invest more in that. For real True16 instructions that new code wouldn't be used anyway; VGPR_16_Lo128 is not even allocatable downstream.

https://github.com/llvm/llvm-project/pull/66202