[llvm] [X86][CodeGen] Support lowering for CCMP/CTEST (PR #91747)

[Shengchen Kan via llvm-commits]

================
@@ -49217,6 +49228,148 @@ static SDValue combineBMILogicOp(SDNode *N, SelectionDAG &DAG,
   return SDValue();
 }
 
+static SDValue combineX86SubCmpForFlags(SDNode *N, SDValue Flag,
+                                        SelectionDAG &DAG,
+                                        TargetLowering::DAGCombinerInfo &DCI,
+                                        const X86Subtarget &ST) {
+  // cmp(setcc(cc, X), 0)
+  // brcond ne
+  //  ->
+  //    X
+  //    brcond cc
+
+  // sub(setcc(cc, X), 1)
+  // brcond ne
+  //  ->
+  //    X
+  //    brcond ~cc
+  //
+  // if only flag has users
+
+  SDValue SetCC = N->getOperand(0);
+
+  // TODO: Remove the check hasCCMP() and update the non-APX tests.
+  if (!ST.hasCCMP() || SetCC.getOpcode() != X86ISD::SETCC || !Flag.hasOneUse())
+    return SDValue();
+
+  // Check the only user of flag is `brcond ne`.
+  SDNode *BrCond = *Flag->uses().begin();
+  if (BrCond->getOpcode() != X86ISD::BRCOND)
+    return SDValue();
+  unsigned CondNo = 2;
+  if (static_cast<X86::CondCode>(BrCond->getConstantOperandVal(CondNo)) !=
+      X86::COND_NE)
+    return SDValue();
+
+  SDValue X = SetCC.getOperand(1);
+  // Replace API is called manually here b/c the number of results may change.
+  DAG.ReplaceAllUsesOfValueWith(Flag, X);
+
+  SDValue CCN = SetCC.getOperand(0);
+  X86::CondCode CC =
+      static_cast<X86::CondCode>(CCN->getAsAPIntVal().getSExtValue());
+  X86::CondCode OppositeCC = X86::GetOppositeBranchCondition(CC);
+  // Update CC for the consumer of the flag.
+  // The old CC is `ne`. Hence, when comparing the result with 0, we are
+  // checking if the second condition evaluates to true. When comparing the
+  // result with 1, we are checking uf the second condition evaluates to false.
+  SmallVector<SDValue> Ops(BrCond->op_values());
+  if (isNullConstant(N->getOperand(1)))
+    Ops[CondNo] = CCN;
+  else if (isOneConstant(N->getOperand(1)))
+    Ops[CondNo] = DAG.getTargetConstant(OppositeCC, SDLoc(BrCond), MVT::i8);
+
+  SDValue NewBrCond =
+      DAG.getNode(X86ISD::BRCOND, SDLoc(BrCond), BrCond->getValueType(0), Ops);
+  // Avoid self-assign error b/c CC1 can be `e/ne`.
+  // Replace API is called manually here b/c we're updating the user of the node
+  // being visited instead of the node itself.
+  if (BrCond != NewBrCond.getNode()) {
+    DAG.ReplaceAllUsesWith(BrCond, &NewBrCond);
+    DCI.recursivelyDeleteUnusedNodes(BrCond);
+  }
+  return X;
+}
+
+static SDValue combineAndOrForCcmpCtest(SDNode *N, SelectionDAG &DAG,
+                                        TargetLowering::DAGCombinerInfo &DCI,
+                                        const X86Subtarget &ST) {
+  // and/or(setcc(cc0, flag0), setcc(cc1, sub (X, Y)))
+  //  ->
+  //    setcc(cc1, ccmp(X, Y, ~cflags/cflags, cc0/~cc0, flag0))
+
+  // and/or(setcc(cc0, flag0), setcc(cc1, cmp (X, 0)))
+  //  ->
+  //    setcc(cc1, ctest(X, X, ~cflags/cflags, cc0/~cc0, flag0))
+  //
+  // where cflags is determined by cc1.
+
+  if (!ST.hasCCMP())
+    return SDValue();
+
+  SDValue SetCC0 = N->getOperand(0);
+  SDValue SetCC1 = N->getOperand(1);
+  if (SetCC0.getOpcode() != X86ISD::SETCC ||
+      SetCC1.getOpcode() != X86ISD::SETCC)
+    return SDValue();
+
+  auto GetCombineToOpc = [&](SDValue V) -> unsigned {
+    SDValue Op = V.getOperand(1);
+    unsigned Opc = Op.getOpcode();
+    if (Opc == X86ISD::SUB)
+      return X86ISD::CCMP;
+    if (Opc == X86ISD::CMP && isNullConstant(Op.getOperand(1)))
+      return X86ISD::CTEST;
+    return 0U;
+  };
+
+  unsigned NewOpc = 0;
+
+  // AND/OR is commutable. Canonicalize the operands to make SETCC with SUB/CMP
+  // appear on the right.
+  if (!(NewOpc = GetCombineToOpc(SetCC1))) {
+    std::swap(SetCC0, SetCC1);
+    if (!(NewOpc = GetCombineToOpc(SetCC1)))
+      return SDValue();
+  }
+
+  X86::CondCode CC0 =
+      static_cast<X86::CondCode>(SetCC0.getConstantOperandVal(0));
+  // CCMP/CTEST is not conditional when the source condition is COND_P/COND_NP.
+  if (CC0 == X86::COND_P || CC0 == X86::COND_NP)
+    return SDValue();
+
+  bool IsOR = N->getOpcode() == ISD::OR;
+
+  // CMP/TEST is executed and updates the EFLAGS normally only when SCC
+  // evaluates to true. So we need to inverse CC0 as SCC when the logic operator
+  // is OR. Similar for CC1.
+  SDValue SCC =
+      IsOR ? DAG.getTargetConstant(X86::GetOppositeBranchCondition(CC0),
+                                   SDLoc(SetCC0.getOperand(0)), MVT::i8)
+           : SetCC0.getOperand(0);
+  SDValue CC1N = SetCC1.getOperand(0);
+  X86::CondCode CC1 =
+      static_cast<X86::CondCode>(CC1N->getAsAPIntVal().getSExtValue());
+  X86::CondCode OppositeCC1 = X86::GetOppositeBranchCondition(CC1);
+  X86::CondCode CFlagsCC = IsOR ? CC1 : OppositeCC1;
+  SDValue CFlags = DAG.getTargetConstant(
+      X86::getCCMPCondFlagsFromCondCode(CFlagsCC), SDLoc(N), MVT::i8);
+  SDValue Sub = SetCC1.getOperand(1);
+
+  // Replace any uses of the old flag produced by SUB/CMP with the new one
+  // produced by CCMP/CTEST.
+  SDValue CCMP = (NewOpc == X86ISD::CCMP)
+                     ? DAG.getNode(X86ISD::CCMP, SDLoc(N), MVT::i32,
+                                   {Sub.getOperand(0), Sub.getOperand(1),
+                                    CFlags, SCC, SetCC0.getOperand(1)})
+                     : DAG.getNode(X86ISD::CTEST, SDLoc(N), MVT::i32,
+                                   {Sub.getOperand(0), Sub.getOperand(0),
+                                    CFlags, SCC, SetCC0.getOperand(1)});
----------------
KanRobert wrote:

I can't. CCMP and CTEST have different parameters.

https://github.com/llvm/llvm-project/pull/91747