Fold (a | b) == a -> (a & ~b) == 0 when and-not exists

This is especially helpful for AArch64, which simplifies ands cmp to tst.

Author: AZero13

llvm/include/llvm/CodeGen/TargetLowering.h

@@ -5800,6 +5800,8 @@ class LLVM_ABI TargetLowering : public TargetLoweringBase {
 private:
   SDValue foldSetCCWithAnd(EVT VT, SDValue N0, SDValue N1, ISD::CondCode Cond,
                            const SDLoc &DL, DAGCombinerInfo &DCI) const;
+  SDValue foldSetCCWithOr(EVT VT, SDValue N0, SDValue N1, ISD::CondCode Cond,
+                          const SDLoc &DL, DAGCombinerInfo &DCI) const;
   SDValue foldSetCCWithBinOp(EVT VT, SDValue N0, SDValue N1, ISD::CondCode Cond,
                              const SDLoc &DL, DAGCombinerInfo &DCI) const;
 

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

@@ -4212,6 +4212,56 @@ SDValue TargetLowering::foldSetCCWithAnd(EVT VT, SDValue N0, SDValue N1,
   return SDValue();
 }
 
+/// This helper function of SimplifySetCC tries to optimize the comparison when
+/// either operand of the SetCC node is a bitwise-or instruction.
+SDValue TargetLowering::foldSetCCWithOr(EVT VT, SDValue N0, SDValue N1,
+                                         ISD::CondCode Cond, const SDLoc &DL,
+                                         DAGCombinerInfo &DCI) const {
+  if (N1.getOpcode() == ISD::OR && N0.getOpcode() != ISD::OR)
+    std::swap(N0, N1);
+
+  SelectionDAG &DAG = DCI.DAG;
+  EVT OpVT = N0.getValueType();
+  if (N0.getOpcode() != ISD::OR || !OpVT.isInteger() ||
+      (Cond != ISD::SETEQ && Cond != ISD::SETNE))
+    return SDValue();
+
+  // Match these patterns in any of their permutations:
+  // (X | Y) == Y
+  // (X | Y) != Y
+  SDValue X, Y;
+  if (N0.getOperand(0) == N1) {
+    X = N0.getOperand(1);
+    Y = N0.getOperand(0);
+  } else if (N0.getOperand(1) == N1) {
+    X = N0.getOperand(0);
+    Y = N0.getOperand(1);
+  } else {
+    return SDValue();
+  }
+
+  SDValue Zero = DAG.getConstant(0, DL, OpVT);
+  if (N0.hasOneUse() && hasAndNotCompare(Y)) {
+    // If the target supports an 'and-not' or 'and-complement' logic operation,
+    // try to use that to make a comparison operation more efficient.
+    // But don't do this transform if the mask is a single bit because there are
+    // more efficient ways to deal with that case (for example, 'bt' on x86 or
+    // 'rlwinm' on PPC).
+
+    // Bail out if the compare operand that we want to turn into a zero is
+    // already a zero (otherwise, infinite loop).
+    if (isNullConstant(Y))
+      return SDValue();
+
+    // Transform this into: X & ~Y == 0.
+    SDValue NotY = DAG.getNOT(SDLoc(Y), Y, OpVT);
+    SDValue NewAnd = DAG.getNode(ISD::AND, SDLoc(N0), OpVT, X, NotY);
+    return DAG.getSetCC(DL, VT, NewAnd, Zero, Cond);
+  }
+
+  return SDValue();
+}
+
 /// There are multiple IR patterns that could be checking whether certain
 /// truncation of a signed number would be lossy or not. The pattern which is
 /// best at IR level, may not lower optimally. Thus, we want to unfold it.
@@ -5507,6 +5557,9 @@ SDValue TargetLowering::SimplifySetCC(EVT VT, SDValue N0, SDValue N1,
 
     if (SDValue V = foldSetCCWithAnd(VT, N0, N1, Cond, dl, DCI))
       return V;
+
+    if (SDValue V = foldSetCCWithOr(VT, N0, N1, Cond, dl, DCI))
+      return V;
   }
 
   // Fold remainder of division by a constant.