[llvm] [AMDGPU][NFC] Simplify t16/fake16 TableGen definitions. (PR #122693)

[Matt Arsenault via llvm-commits]

================
@@ -973,12 +973,10 @@ multiclass VOP1_Real_FULL_with_name_gfx11_gfx12<bits<9> op, string opName,
   VOP1_Real_FULL_with_name<GFX11Gen, op, opName, asmName>,
   VOP1_Real_FULL_with_name<GFX12Gen, op, opName, asmName>;
 
-multiclass VOP1_Real_FULL_t16_and_fake16_gfx11_gfx12<bits<9> op, string asmName,
-                                          string opName = NAME> {
-  defm opName#"_t16" :
-	VOP1_Real_FULL_with_name_gfx11_gfx12<op, opName#"_t16", asmName>;
-  defm opName#"_fake16":
-	VOP1_Real_FULL_with_name_gfx11_gfx12<op, opName#"_fake16", asmName>;
+multiclass VOP1_Real_FULL_t16_and_fake16_gfx11_gfx12<bits<9> op> {
----------------
arsenm wrote:

I think I'd rather preserve the default parameter, and just fix the default not using to lower. Unfortunately they keep hitting us with new instruction renames 

https://github.com/llvm/llvm-project/pull/122693