[llvm] [RISCV] Add FMA support to combineOp_VLToVWOp_VL. (PR #100454)

[Craig Topper via llvm-commits]

================
@@ -14883,19 +14904,26 @@ static SDValue combineBinOp_VLToVWBinOp_VL(SDNode *N,
 
   while (!Worklist.empty()) {
     SDNode *Root = Worklist.pop_back_val();
-    if (!NodeExtensionHelper::isSupportedRoot(Root, Subtarget))
-      return SDValue();
 
     NodeExtensionHelper LHS(Root, 0, DAG, Subtarget);
     NodeExtensionHelper RHS(Root, 1, DAG, Subtarget);
-    auto AppendUsersIfNeeded = [&Worklist,
+    auto AppendUsersIfNeeded = [&Worklist, &Subtarget,
                                 &Inserted](const NodeExtensionHelper &Op) {
       if (Op.needToPromoteOtherUsers()) {
-        for (SDNode *TheUse : Op.OrigOperand->uses()) {
+        for (SDNode::use_iterator UI = Op.OrigOperand->use_begin(),
+                                  UE = Op.OrigOperand->use_end();
+             UI != UE; ++UI) {
+          SDNode *TheUse = *UI;
+          if (!NodeExtensionHelper::isSupportedRoot(TheUse, Subtarget))
+            return false;
+          // We only support the first 2 operands of FMA.
+          if (UI.getOperandNo() >= 2)
----------------
topperc wrote:

The only undef check I see so far is in fillUpExtensionSupportForSplat, but that's only for RISCVISD::VMV_V_X_VL

https://github.com/llvm/llvm-project/pull/100454