[X86] Simplify part of the isel for X86ISD::FCMP/STRICT_FCMP/STRICT_FCMPS.

We don't need to have the compare output a value and then copy it
to FPSW for use by FNSTSW. Instead we can just have the compare
output Glue and glue the FNSTSW to it. InstrEmitter effectively
performed this optimization when emitting the Machine IR. Doing
it directly simplifies the codes and reduces the work in
InstrEmitter. There's no change in the machine IR at the end of
isel before and after this change.

diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
index 0a5bae3..c263c62 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -5370,24 +5370,20 @@ void X86DAGToDAGISel::Select(SDNode *Node) {
       break;
     }
 
-    SDValue Cmp;
     SDValue Chain =
         IsStrictCmp ? Node->getOperand(0) : CurDAG->getEntryNode();
+    SDValue Glue;
     if (IsStrictCmp) {
-      SDVTList VTs = CurDAG->getVTList(MVT::i16, MVT::Other);
-      Cmp = SDValue(CurDAG->getMachineNode(Opc, dl, VTs, {N0, N1, Chain}), 0);
-      Chain = Cmp.getValue(1);
+      SDVTList VTs = CurDAG->getVTList(MVT::Other, MVT::Glue);
+      Chain = SDValue(CurDAG->getMachineNode(Opc, dl, VTs, {N0, N1, Chain}), 0);
+      Glue = Chain.getValue(1);
     } else {
-      Cmp = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::i16, N0, N1), 0);
+      Glue = SDValue(CurDAG->getMachineNode(Opc, dl, MVT::Glue, N0, N1), 0);
     }
 
     // Move FPSW to AX.
-    SDValue FPSW = CurDAG->getCopyToReg(Chain, dl, X86::FPSW, Cmp, SDValue());
-    Chain = FPSW;
     SDValue FNSTSW =
-        SDValue(CurDAG->getMachineNode(X86::FNSTSW16r, dl, MVT::i16, FPSW,
-                                       FPSW.getValue(1)),
-                0);
+        SDValue(CurDAG->getMachineNode(X86::FNSTSW16r, dl, MVT::i16, Glue), 0);
 
     // Extract upper 8-bits of AX.
     SDValue Extract =