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

[Joe Nash 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
+      // just bail out.
+      if (NewOpc == AMDGPU::V_FMAAK_F16_t16)
----------------
Sisyph wrote:

Bail out for FMAMK_F16_t16 as well, above?

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