Give _fake16 record a vOPDName so that it now has a suffix in VOPDName.

The record now looks like:

  def V_CNDMASK_B16_e32 {
    ...
    Instruction BaseVOP = V_CNDMASK_B16_e32;
    string VOPDName = "v_dual_";
    bits<5> VOPDOp = { 1, 1, 1, 1, 1 };
    bit CanBeVOPDX = 0;
    -->
    Instruction BaseVOP = V_CNDMASK_B16_e32;
    string VOPDName = "v_dual_cndmask_b16";
    bits<5> VOPDOp = { 1, 1, 1, 1, 1 };
    bit CanBeVOPDX = 0;
  }

Whereas before, these fields were omitted entirely.

Author: jofrn

llvm/lib/Target/AMDGPU/VOP2Instructions.td

@@ -319,7 +319,7 @@ multiclass
 multiclass
     VOP2eInst<string opName, VOPProfile P, SDPatternOperator node = null_frag,
               string revOp = opName, bit useSGPRInput = !eq(P.NumSrcArgs, 3)>
-    : VOP2eInst_Base<opName, P, 0, "", node, revOp, useSGPRInput>;
+    : VOP2eInst_Base<opName, P, -1, "v_cndmask_b16", node, revOp, useSGPRInput>;
 
 multiclass
     VOP2eInst_VOPD<string opName, VOPProfile P, bits<5> VOPDOp, string VOPDName,

llvm/lib/Target/AMDGPU/VOPInstructions.td

@@ -37,7 +37,7 @@ defvar VOPDX_Max_Index = 12;
 
 class VOPD_Component<bits<5> OpIn, string vOPDName> {
   Instruction BaseVOP = !cast<Instruction>(NAME);
-  string VOPDName = "v_dual_" # !if(!le(!size(vOPDName), 2), "", !substr(vOPDName, 2));
+  string VOPDName = "v_dual_" # !substr(vOPDName, 2);
   bits<5> VOPDOp = OpIn;
   bit CanBeVOPDX = !le(VOPDOp, VOPDX_Max_Index);
 }