[PowerPC] Exploit power9 new instruction setb

Check the expected pattens feeding to SELECT_CC like:
   (select_cc lhs, rhs,  1, (sext (setcc [lr]hs, [lr]hs, cc2)), cc1)
   (select_cc lhs, rhs, -1, (zext (setcc [lr]hs, [lr]hs, cc2)), cc1)
   (select_cc lhs, rhs,  0, (select_cc [lr]hs, [lr]hs,  1, -1, cc2), seteq)
   (select_cc lhs, rhs,  0, (select_cc [lr]hs, [lr]hs, -1,  1, cc2), seteq)
Further transform the sequence to comparison + setb if hits.

Differential Revision: https://reviews.llvm.org/D53275

llvm-svn: 349445

lib/Target/PowerPC/P9InstrResources.td

@@ -111,11 +111,11 @@ def : InstRW<[P9_ALU_3C, IP_EXEC_1C, DISP_1C, DISP_1C],
     (instregex "CNT(L|T)Z(D|W)(8)?(o)?$"),
     (instregex "POPCNT(D|W)$"),
     (instregex "CMPB(8)?$"),
+    (instregex "SETB(8)?$"),
     XSTDIVDP,
     XSTSQRTDP,
     XSXSIGDP,
     XSCVSPDPN,
-    SETB,
     BPERMD
 )>;
 

lib/Target/PowerPC/PPCISelDAGToDAG.cpp

@@ -81,6 +81,8 @@ STATISTIC(NumLogicOpsOnComparison,
           "Number of logical ops on i1 values calculated in GPR.");
 STATISTIC(OmittedForNonExtendUses,
           "Number of compares not eliminated as they have non-extending uses.");
+STATISTIC(NumP9Setb,
+          "Number of compares lowered to setb.");
 
 // FIXME: Remove this once the bug has been fixed!
 cl::opt<bool> ANDIGlueBug("expose-ppc-andi-glue-bug",
@@ -4137,6 +4139,146 @@ void PPCDAGToDAGISel::transferMemOperands(SDNode *N, SDNode *Result) {
   CurDAG->setNodeMemRefs(cast<MachineSDNode>(Result), {MemOp});
 }
 
+static bool mayUseP9Setb(SDNode *N, const ISD::CondCode &CC, SelectionDAG *DAG,
+                         bool &NeedSwapOps, bool &IsUnCmp) {
+
+  assert(N->getOpcode() == ISD::SELECT_CC && "Expecting a SELECT_CC here.");
+
+  SDValue LHS = N->getOperand(0);
+  SDValue RHS = N->getOperand(1);
+  SDValue TrueRes = N->getOperand(2);
+  SDValue FalseRes = N->getOperand(3);
+  ConstantSDNode *TrueConst = dyn_cast<ConstantSDNode>(TrueRes);
+  if (!TrueConst)
+    return false;
+
+  assert((N->getSimpleValueType(0) == MVT::i64 ||
+          N->getSimpleValueType(0) == MVT::i32) &&
+         "Expecting either i64 or i32 here.");
+
+  // We are looking for any of:
+  // (select_cc lhs, rhs,  1, (sext (setcc [lr]hs, [lr]hs, cc2)), cc1)
+  // (select_cc lhs, rhs, -1, (zext (setcc [lr]hs, [lr]hs, cc2)), cc1)
+  // (select_cc lhs, rhs,  0, (select_cc [lr]hs, [lr]hs,  1, -1, cc2), seteq)
+  // (select_cc lhs, rhs,  0, (select_cc [lr]hs, [lr]hs, -1,  1, cc2), seteq)
+  int64_t TrueResVal = TrueConst->getSExtValue();
+  if ((TrueResVal < -1 || TrueResVal > 1) ||
+      (TrueResVal == -1 && FalseRes.getOpcode() != ISD::ZERO_EXTEND) ||
+      (TrueResVal == 1 && FalseRes.getOpcode() != ISD::SIGN_EXTEND) ||
+      (TrueResVal == 0 &&
+       (FalseRes.getOpcode() != ISD::SELECT_CC || CC != ISD::SETEQ)))
+    return false;
+
+  bool InnerIsSel = FalseRes.getOpcode() == ISD::SELECT_CC;
+  SDValue SetOrSelCC = InnerIsSel ? FalseRes : FalseRes.getOperand(0);
+  if (SetOrSelCC.getOpcode() != ISD::SETCC &&
+      SetOrSelCC.getOpcode() != ISD::SELECT_CC)
+    return false;
+
+  // Without this setb optimization, the outer SELECT_CC will be manually
+  // selected to SELECT_CC_I4/SELECT_CC_I8 Pseudo, then expand-isel-pseudos pass
+  // transforms pseduo instruction to isel instruction. When there are more than
+  // one use for result like zext/sext, with current optimization we only see
+  // isel is replaced by setb but can't see any significant gain. Since
+  // setb has longer latency than original isel, we should avoid this. Another
+  // point is that setb requires comparison always kept, it can break the
+  // oppotunity to get the comparison away if we have in future.
+  if (!SetOrSelCC.hasOneUse() || (!InnerIsSel && !FalseRes.hasOneUse()))
+    return false;
+
+  SDValue InnerLHS = SetOrSelCC.getOperand(0);
+  SDValue InnerRHS = SetOrSelCC.getOperand(1);
+  ISD::CondCode InnerCC =
+      cast<CondCodeSDNode>(SetOrSelCC.getOperand(InnerIsSel ? 4 : 2))->get();
+  // If the inner comparison is a select_cc, make sure the true/false values are
+  // 1/-1 and canonicalize it if needed.
+  if (InnerIsSel) {
+    ConstantSDNode *SelCCTrueConst =
+        dyn_cast<ConstantSDNode>(SetOrSelCC.getOperand(2));
+    ConstantSDNode *SelCCFalseConst =
+        dyn_cast<ConstantSDNode>(SetOrSelCC.getOperand(3));
+    if (!SelCCTrueConst || !SelCCFalseConst)
+      return false;
+    int64_t SelCCTVal = SelCCTrueConst->getSExtValue();
+    int64_t SelCCFVal = SelCCFalseConst->getSExtValue();
+    // The values must be -1/1 (requiring a swap) or 1/-1.
+    if (SelCCTVal == -1 && SelCCFVal == 1) {
+      std::swap(InnerLHS, InnerRHS);
+    } else if (SelCCTVal != 1 || SelCCFVal != -1)
+      return false;
+  }
+
+  // Canonicalize unsigned case
+  if (InnerCC == ISD::SETULT || InnerCC == ISD::SETUGT) {
+    IsUnCmp = true;
+    InnerCC = (InnerCC == ISD::SETULT) ? ISD::SETLT : ISD::SETGT;
+  }
+
+  bool InnerSwapped = false;
+  if (LHS == InnerRHS && RHS == InnerLHS)
+    InnerSwapped = true;
+  else if (LHS != InnerLHS || RHS != InnerRHS)
+    return false;
+
+  switch (CC) {
+  // (select_cc lhs, rhs,  0, \
+  //     (select_cc [lr]hs, [lr]hs, 1, -1, setlt/setgt), seteq)
+  case ISD::SETEQ:
+    if (!InnerIsSel)
+      return false;
+    if (InnerCC != ISD::SETLT && InnerCC != ISD::SETGT)
+      return false;
+    NeedSwapOps = (InnerCC == ISD::SETGT) ? InnerSwapped : !InnerSwapped;
+    break;
+
+  // (select_cc lhs, rhs, -1, (zext (setcc [lr]hs, [lr]hs, setne)), setu?lt)
+  // (select_cc lhs, rhs, -1, (zext (setcc lhs, rhs, setgt)), setu?lt)
+  // (select_cc lhs, rhs, -1, (zext (setcc rhs, lhs, setlt)), setu?lt)
+  // (select_cc lhs, rhs, 1, (sext (setcc [lr]hs, [lr]hs, setne)), setu?lt)
+  // (select_cc lhs, rhs, 1, (sext (setcc lhs, rhs, setgt)), setu?lt)
+  // (select_cc lhs, rhs, 1, (sext (setcc rhs, lhs, setlt)), setu?lt)
+  case ISD::SETULT:
+    if (!IsUnCmp && InnerCC != ISD::SETNE)
+      return false;
+    IsUnCmp = true;
+    LLVM_FALLTHROUGH;
+  case ISD::SETLT:
+    if (InnerCC == ISD::SETNE || (InnerCC == ISD::SETGT && !InnerSwapped) ||
+        (InnerCC == ISD::SETLT && InnerSwapped))
+      NeedSwapOps = (TrueResVal == 1);
+    else
+      return false;
+    break;
+
+  // (select_cc lhs, rhs, 1, (sext (setcc [lr]hs, [lr]hs, setne)), setu?gt)
+  // (select_cc lhs, rhs, 1, (sext (setcc lhs, rhs, setlt)), setu?gt)
+  // (select_cc lhs, rhs, 1, (sext (setcc rhs, lhs, setgt)), setu?gt)
+  // (select_cc lhs, rhs, -1, (zext (setcc [lr]hs, [lr]hs, setne)), setu?gt)
+  // (select_cc lhs, rhs, -1, (zext (setcc lhs, rhs, setlt)), setu?gt)
+  // (select_cc lhs, rhs, -1, (zext (setcc rhs, lhs, setgt)), setu?gt)
+  case ISD::SETUGT:
+    if (!IsUnCmp && InnerCC != ISD::SETNE)
+      return false;
+    IsUnCmp = true;
+    LLVM_FALLTHROUGH;
+  case ISD::SETGT:
+    if (InnerCC == ISD::SETNE || (InnerCC == ISD::SETLT && !InnerSwapped) ||
+        (InnerCC == ISD::SETGT && InnerSwapped))
+      NeedSwapOps = (TrueResVal == -1);
+    else
+      return false;
+    break;
+
+  default:
+    return false;
+  }
+
+  LLVM_DEBUG(dbgs() << "Found a node that can be lowered to a SETB: ");
+  LLVM_DEBUG(N->dump());
+
+  return true;
+}
+
 // Select - Convert the specified operand from a target-independent to a
 // target-specific node if it hasn't already been changed.
 void PPCDAGToDAGISel::Select(SDNode *N) {
@@ -4599,6 +4741,31 @@ void PPCDAGToDAGISel::Select(SDNode *N) {
         N->getOperand(0).getValueType() == MVT::i1)
       break;
 
+    if (PPCSubTarget->isISA3_0() && PPCSubTarget->isPPC64()) {
+      bool NeedSwapOps = false;
+      bool IsUnCmp = false;
+      if (mayUseP9Setb(N, CC, CurDAG, NeedSwapOps, IsUnCmp)) {
+        SDValue LHS = N->getOperand(0);
+        SDValue RHS = N->getOperand(1);
+        if (NeedSwapOps)
+          std::swap(LHS, RHS);
+
+        // Make use of SelectCC to generate the comparison to set CR bits, for
+        // equality comparisons having one literal operand, SelectCC probably
+        // doesn't need to materialize the whole literal and just use xoris to
+        // check it first, it leads the following comparison result can't
+        // exactly represent GT/LT relationship. So to avoid this we specify
+        // SETGT/SETUGT here instead of SETEQ.
+        SDValue GenCC =
+            SelectCC(LHS, RHS, IsUnCmp ? ISD::SETUGT : ISD::SETGT, dl);
+        CurDAG->SelectNodeTo(
+            N, N->getSimpleValueType(0) == MVT::i64 ? PPC::SETB8 : PPC::SETB,
+            N->getValueType(0), GenCC);
+        NumP9Setb++;
+        return;
+      }
+    }
+
     // Handle the setcc cases here.  select_cc lhs, 0, 1, 0, cc
     if (!isPPC64)
       if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N->getOperand(1)))

lib/Target/PowerPC/PPCInstr64Bit.td

@@ -777,8 +777,12 @@ def MADDHDU : VAForm_1a<49, (outs g8rc :$RT), (ins g8rc:$RA, g8rc:$RB, g8rc:$RC)
                        "maddhdu $RT, $RA, $RB, $RC", IIC_IntMulHD, []>, isPPC64;
 def MADDLD : VAForm_1a<51, (outs g8rc :$RT), (ins g8rc:$RA, g8rc:$RB, g8rc:$RC),
                        "maddld $RT, $RA, $RB, $RC", IIC_IntMulHD, []>, isPPC64;
-def SETB : XForm_44<31, 128, (outs g8rc:$RT), (ins crrc:$BFA),
+def SETB : XForm_44<31, 128, (outs gprc:$RT), (ins crrc:$BFA),
                        "setb $RT, $BFA", IIC_IntGeneral>, isPPC64;
+let Interpretation64Bit = 1, isCodeGenOnly = 1 in {
+  def SETB8 : XForm_44<31, 128, (outs g8rc:$RT), (ins crrc:$BFA),
+                       "setb $RT, $BFA", IIC_IntGeneral>, isPPC64;
+}
 def DARN : XForm_45<31, 755, (outs g8rc:$RT), (ins i32imm:$L),
                      "darn $RT, $L", IIC_LdStLD>, isPPC64;
 def ADDPCIS : DXForm<19, 2, (outs g8rc:$RT), (ins i32imm:$D),

lib/Target/PowerPC/PPCInstrInfo.cpp

@@ -3488,6 +3488,7 @@ static bool isSignExtendingOp(const MachineInstr &MI) {
       Opcode == PPC::EXTSH  || Opcode == PPC::EXTSHo  ||
       Opcode == PPC::EXTSB8 || Opcode == PPC::EXTSH8  ||
       Opcode == PPC::EXTSW  || Opcode == PPC::EXTSWo  ||
+      Opcode == PPC::SETB   || Opcode == PPC::SETB8   ||
       Opcode == PPC::EXTSH8_32_64 || Opcode == PPC::EXTSW_32_64 ||
       Opcode == PPC::EXTSB8_32_64)
     return true;