Add x86-specific DAG combine to simplify:

 x == -y --> x+y == 0
 x != -y --> x+y != 0

On x86, the generated code goes from
   negl    %esi
   cmpl    %esi, %edi
   je    .LBB0_2
to
   addl    %esi, %edi
   je    .L4

This case is correctly handled for ARM with "cmn".

Patch by Manman Ren.
rdar://11245199
PR12545

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@155739 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 65cd899ed7185e2bc9240e4b598001db04005da3..f0dc8e4795c508d3cb35afed2b4ab8256262ddac 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1223,6 +1223,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
   setTargetDAGCombine(ISD::TRUNCATE);
   setTargetDAGCombine(ISD::UINT_TO_FP);
   setTargetDAGCombine(ISD::SINT_TO_FP);
+  setTargetDAGCombine(ISD::SETCC);
   setTargetDAGCombine(ISD::FP_TO_SINT);
   if (Subtarget->is64Bit())
     setTargetDAGCombine(ISD::MUL);
@@ -15000,6 +15001,32 @@ static SDValue PerformZExtCombine(SDNode *N, SelectionDAG &DAG,
   return SDValue();
 }
 
+// Optimize x == -y --> x+y == 0
+//          x != -y --> x+y != 0
+static SDValue PerformISDSETCCCombine(SDNode *N, SelectionDAG &DAG) {
+  ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(2))->get();
+  SDValue LHS = N->getOperand(0);
+  SDValue RHS = N->getOperand(1); 
+
+  if ((CC == ISD::SETNE || CC == ISD::SETEQ) && LHS.getOpcode() == ISD::SUB)
+    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(LHS.getOperand(0)))
+      if (C->getAPIntValue() == 0 && LHS.hasOneUse()) {
+        SDValue addV = DAG.getNode(ISD::ADD, N->getDebugLoc(),
+                                   LHS.getValueType(), RHS, LHS.getOperand(1));
+        return DAG.getSetCC(N->getDebugLoc(), N->getValueType(0),
+                            addV, DAG.getConstant(0, addV.getValueType()), CC);
+      }
+  if ((CC == ISD::SETNE || CC == ISD::SETEQ) && RHS.getOpcode() == ISD::SUB)
+    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(RHS.getOperand(0)))
+      if (C->getAPIntValue() == 0 && RHS.hasOneUse()) {
+        SDValue addV = DAG.getNode(ISD::ADD, N->getDebugLoc(),
+                                   RHS.getValueType(), LHS, RHS.getOperand(1));
+        return DAG.getSetCC(N->getDebugLoc(), N->getValueType(0),
+                            addV, DAG.getConstant(0, addV.getValueType()), CC);
+      }
+  return SDValue();
+}
+
 // Optimize  RES = X86ISD::SETCC CONDCODE, EFLAG_INPUT
 static SDValue PerformSETCCCombine(SDNode *N, SelectionDAG &DAG) {
   unsigned X86CC = N->getConstantOperandVal(0);
@@ -15230,6 +15257,7 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
   case ISD::ZERO_EXTEND:    return PerformZExtCombine(N, DAG, DCI, Subtarget);
   case ISD::SIGN_EXTEND:    return PerformSExtCombine(N, DAG, DCI, Subtarget);
   case ISD::TRUNCATE:       return PerformTruncateCombine(N, DAG, DCI);
+  case ISD::SETCC:          return PerformISDSETCCCombine(N, DAG);
   case X86ISD::SETCC:       return PerformSETCCCombine(N, DAG);
   case X86ISD::SHUFP:       // Handle all target specific shuffles
   case X86ISD::PALIGN:

diff --git a/test/CodeGen/X86/neg_cmp.ll b/test/CodeGen/X86/neg_cmp.ll
new file mode 100644 (file)
index 0000000..866514e
--- /dev/null
+++ b/test/CodeGen/X86/neg_cmp.ll
@@ -0,0 +1,22 @@
+; RUN: llc < %s -march=x86-64 | FileCheck %s
+
+; rdar://11245199
+; PR12545
+define void @f(i32 %x, i32 %y) nounwind uwtable ssp {
+entry:
+; CHECK: f:
+; CHECK-NOT: neg
+; CHECK: add
+  %sub = sub i32 0, %y
+  %cmp = icmp eq i32 %x, %sub
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  tail call void @g() nounwind
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  ret void
+}
+
+declare void @g()