vendor/llvm/llvm-trunk-r256945

7 files changed, 137 insertions, 37 deletions

diff --git a/lib/Target/AArch64/AArch64.td b/lib/Target/AArch64/AArch64.td
index 0bff9b592c15..46ef2c111bae 100644
--- a/lib/Target/AArch64/AArch64.td
+++ b/lib/Target/AArch64/AArch64.td
@@ -124,6 +124,14 @@ def ProcCyclone : SubtargetFeature<"cyclone", "ARMProcFamily", "Cyclone",
                                    FeaturePerfMon,
                                    FeatureZCRegMove, FeatureZCZeroing]>;
 
+def ProcExynosM1 : SubtargetFeature<"exynosm1", "ARMProcFamily", "ExynosM1",
+                                    "Samsung Exynos-M1 processors",
+                                    [FeatureFPARMv8,
+                                    FeatureNEON,
+                                    FeatureCrypto,
+                                    FeatureCRC,
+                                    FeaturePerfMon]>;
+
 def : ProcessorModel<"generic", NoSchedModel, [FeatureFPARMv8,
                                               FeatureNEON,
                                               FeatureCRC,
@@ -136,6 +144,8 @@ def : ProcessorModel<"cortex-a57", CortexA57Model, [ProcA57]>;
 // FIXME: Cortex-A72 is currently modelled as an Cortex-A57.
 def : ProcessorModel<"cortex-a72", CortexA57Model, [ProcA57]>;
 def : ProcessorModel<"cyclone", CycloneModel, [ProcCyclone]>;
+// FIXME: Exynos-M1 is currently modelled without a specific SchedModel.
+def : ProcessorModel<"exynos-m1", NoSchedModel, [ProcExynosM1]>;
 
 //===----------------------------------------------------------------------===//
 // Assembly parser
diff --git a/lib/Target/AArch64/AArch64A57FPLoadBalancing.cpp b/lib/Target/AArch64/AArch64A57FPLoadBalancing.cpp
index 79a84ad8c6c5..3d1ab4e3fc2b 100644
--- a/lib/Target/AArch64/AArch64A57FPLoadBalancing.cpp
+++ b/lib/Target/AArch64/AArch64A57FPLoadBalancing.cpp
@@ -158,7 +158,7 @@ INITIALIZE_PASS_END(AArch64A57FPLoadBalancing, DEBUG_TYPE,
                     "AArch64 A57 FP Load-Balancing", false, false)
 
 namespace {
-/// A Chain is a sequence of instructions that are linked together by 
+/// A Chain is a sequence of instructions that are linked together by
 /// an accumulation operand. For example:
 ///
 ///   fmul d0<def>, ?
@@ -285,7 +285,7 @@ public:
   std::string str() const {
     std::string S;
     raw_string_ostream OS(S);
-    
+
     OS << "{";
     StartInst->print(OS, /* SkipOpers= */true);
     OS << " -> ";
@@ -427,7 +427,7 @@ Chain *AArch64A57FPLoadBalancing::getAndEraseNext(Color PreferredColor,
       return Ch;
     }
   }
-  
+
   // Bailout case - just return the first item.
   Chain *Ch = L.front();
   L.erase(L.begin());
@@ -495,7 +495,7 @@ int AArch64A57FPLoadBalancing::scavengeRegister(Chain *G, Color C,
   RS.enterBasicBlock(&MBB);
   RS.forward(MachineBasicBlock::iterator(G->getStart()));
 
-  // Can we find an appropriate register that is available throughout the life 
+  // Can we find an appropriate register that is available throughout the life
   // of the chain?
   unsigned RegClassID = G->getStart()->getDesc().OpInfo[0].RegClass;
   BitVector AvailableRegs = RS.getRegsAvailable(TRI->getRegClass(RegClassID));
diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index 9f5beff12100..4ecfbe9e2280 100644
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -2426,7 +2426,7 @@ SDValue AArch64TargetLowering::LowerFormalArguments(
 
       continue;
     }
-    
+
     if (VA.isRegLoc()) {
       // Arguments stored in registers.
       EVT RegVT = VA.getLocVT();
@@ -5074,7 +5074,7 @@ static bool isEXTMask(ArrayRef<int> M, EVT VT, bool &ReverseEXT,
 
   // The index of an EXT is the first element if it is not UNDEF.
   // Watch out for the beginning UNDEFs. The EXT index should be the expected
-  // value of the first element.  E.g. 
+  // value of the first element.  E.g.
   // <-1, -1, 3, ...> is treated as <1, 2, 3, ...>.
   // <-1, -1, 0, 1, ...> is treated as <2*NumElts-2, 2*NumElts-1, 0, 1, ...>.
   // ExpectedElt is the last mask index plus 1.
@@ -9491,6 +9491,103 @@ static SDValue performBRCONDCombine(SDNode *N,
   return SDValue();
 }
 
+// Optimize some simple tbz/tbnz cases.  Returns the new operand and bit to test
+// as well as whether the test should be inverted.  This code is required to
+// catch these cases (as opposed to standard dag combines) because
+// AArch64ISD::TBZ is matched during legalization.
+static SDValue getTestBitOperand(SDValue Op, unsigned &Bit, bool &Invert,
+                                 SelectionDAG &DAG) {
+
+  if (!Op->hasOneUse())
+    return Op;
+
+  // We don't handle undef/constant-fold cases below, as they should have
+  // already been taken care of (e.g. and of 0, test of undefined shifted bits,
+  // etc.)
+
+  // (tbz (trunc x), b) -> (tbz x, b)
+  // This case is just here to enable more of the below cases to be caught.
+  if (Op->getOpcode() == ISD::TRUNCATE &&
+      Bit < Op->getValueType(0).getSizeInBits()) {
+    return getTestBitOperand(Op->getOperand(0), Bit, Invert, DAG);
+  }
+
+  if (Op->getNumOperands() != 2)
+    return Op;
+
+  auto *C = dyn_cast<ConstantSDNode>(Op->getOperand(1));
+  if (!C)
+    return Op;
+
+  switch (Op->getOpcode()) {
+  default:
+    return Op;
+
+  // (tbz (and x, m), b) -> (tbz x, b)
+  case ISD::AND:
+    if ((C->getZExtValue() >> Bit) & 1)
+      return getTestBitOperand(Op->getOperand(0), Bit, Invert, DAG);
+    return Op;
+
+  // (tbz (shl x, c), b) -> (tbz x, b-c)
+  case ISD::SHL:
+    if (C->getZExtValue() <= Bit &&
+        (Bit - C->getZExtValue()) < Op->getValueType(0).getSizeInBits()) {
+      Bit = Bit - C->getZExtValue();
+      return getTestBitOperand(Op->getOperand(0), Bit, Invert, DAG);
+    }
+    return Op;
+
+  // (tbz (sra x, c), b) -> (tbz x, b+c) or (tbz x, msb) if b+c is > # bits in x
+  case ISD::SRA:
+    Bit = Bit + C->getZExtValue();
+    if (Bit >= Op->getValueType(0).getSizeInBits())
+      Bit = Op->getValueType(0).getSizeInBits() - 1;
+    return getTestBitOperand(Op->getOperand(0), Bit, Invert, DAG);
+
+  // (tbz (srl x, c), b) -> (tbz x, b+c)
+  case ISD::SRL:
+    if ((Bit + C->getZExtValue()) < Op->getValueType(0).getSizeInBits()) {
+      Bit = Bit + C->getZExtValue();
+      return getTestBitOperand(Op->getOperand(0), Bit, Invert, DAG);
+    }
+    return Op;
+
+  // (tbz (xor x, -1), b) -> (tbnz x, b)
+  case ISD::XOR:
+    if ((C->getZExtValue() >> Bit) & 1)
+      Invert = !Invert;
+    return getTestBitOperand(Op->getOperand(0), Bit, Invert, DAG);
+  }
+}
+
+// Optimize test single bit zero/non-zero and branch.
+static SDValue performTBZCombine(SDNode *N,
+                                 TargetLowering::DAGCombinerInfo &DCI,
+                                 SelectionDAG &DAG) {
+  unsigned Bit = cast<ConstantSDNode>(N->getOperand(2))->getZExtValue();
+  bool Invert = false;
+  SDValue TestSrc = N->getOperand(1);
+  SDValue NewTestSrc = getTestBitOperand(TestSrc, Bit, Invert, DAG);
+
+  if (TestSrc == NewTestSrc)
+    return SDValue();
+
+  unsigned NewOpc = N->getOpcode();
+  if (Invert) {
+    if (NewOpc == AArch64ISD::TBZ)
+      NewOpc = AArch64ISD::TBNZ;
+    else {
+      assert(NewOpc == AArch64ISD::TBNZ);
+      NewOpc = AArch64ISD::TBZ;
+    }
+  }
+
+  SDLoc DL(N);
+  return DAG.getNode(NewOpc, DL, MVT::Other, N->getOperand(0), NewTestSrc,
+                     DAG.getConstant(Bit, DL, MVT::i64), N->getOperand(3));
+}
+
 // vselect (v1i1 setcc) ->
 //     vselect (v1iXX setcc)  (XX is the size of the compared operand type)
 // FIXME: Currently the type legalizer can't handle VSELECT having v1i1 as
@@ -9642,6 +9739,9 @@ SDValue AArch64TargetLowering::PerformDAGCombine(SDNode *N,
     return performSTORECombine(N, DCI, DAG, Subtarget);
   case AArch64ISD::BRCOND:
     return performBRCONDCombine(N, DCI, DAG);
+  case AArch64ISD::TBNZ:
+  case AArch64ISD::TBZ:
+    return performTBZCombine(N, DCI, DAG);
   case AArch64ISD::CSEL:
     return performCONDCombine(N, DCI, DAG, 2, 3);
   case AArch64ISD::DUP:
diff --git a/lib/Target/AArch64/AArch64LoadStoreOptimizer.cpp b/lib/Target/AArch64/AArch64LoadStoreOptimizer.cpp
index 566aa2c9a9ba..43664df3b861 100644
--- a/lib/Target/AArch64/AArch64LoadStoreOptimizer.cpp
+++ b/lib/Target/AArch64/AArch64LoadStoreOptimizer.cpp
@@ -613,21 +613,6 @@ static bool isLdOffsetInRangeOfSt(MachineInstr *LoadInst,
          (UnscaledLdOffset + LoadSize <= (UnscaledStOffset + StoreSize));
 }
 
-// Copy MachineMemOperands from Op0 and Op1 to a new array assigned to MI.
-static void concatenateMemOperands(MachineInstr *MI, MachineInstr *Op0,
-                                   MachineInstr *Op1) {
-  assert(MI->memoperands_empty() && "expected a new machineinstr");
-  size_t numMemRefs = (Op0->memoperands_end() - Op0->memoperands_begin()) +
-                      (Op1->memoperands_end() - Op1->memoperands_begin());
-
-  MachineFunction *MF = MI->getParent()->getParent();
-  MachineSDNode::mmo_iterator MemBegin = MF->allocateMemRefsArray(numMemRefs);
-  MachineSDNode::mmo_iterator MemEnd =
-      std::copy(Op0->memoperands_begin(), Op0->memoperands_end(), MemBegin);
-  MemEnd = std::copy(Op1->memoperands_begin(), Op1->memoperands_end(), MemEnd);
-  MI->setMemRefs(MemBegin, MemEnd);
-}
-
 MachineBasicBlock::iterator
 AArch64LoadStoreOpt::mergePairedInsns(MachineBasicBlock::iterator I,
                                       MachineBasicBlock::iterator Paired,
@@ -692,10 +677,8 @@ AArch64LoadStoreOpt::mergePairedInsns(MachineBasicBlock::iterator I,
                        TII->get(NewOpc))
                    .addOperand(getLdStRegOp(RtNewDest))
                    .addOperand(BaseRegOp)
-                   .addImm(OffsetImm);
-
-    // Copy MachineMemOperands from the original loads.
-    concatenateMemOperands(NewMemMI, I, Paired);
+                   .addImm(OffsetImm)
+                   .setMemRefs(I->mergeMemRefsWith(*Paired));
 
     DEBUG(
         dbgs()
@@ -786,9 +769,8 @@ AArch64LoadStoreOpt::mergePairedInsns(MachineBasicBlock::iterator I,
                   TII->get(NewOpc))
               .addOperand(getLdStRegOp(I))
               .addOperand(BaseRegOp)
-              .addImm(OffsetImm);
-    // Copy MachineMemOperands from the original stores.
-    concatenateMemOperands(MIB, I, Paired);
+              .addImm(OffsetImm)
+              .setMemRefs(I->mergeMemRefsWith(*Paired));
   } else {
     // Handle Unscaled
     if (IsUnscaled)
diff --git a/lib/Target/AArch64/AArch64Subtarget.h b/lib/Target/AArch64/AArch64Subtarget.h
index 1b8b9b27719c..151133b2f32c 100644
--- a/lib/Target/AArch64/AArch64Subtarget.h
+++ b/lib/Target/AArch64/AArch64Subtarget.h
@@ -33,7 +33,14 @@ class Triple;
 
 class AArch64Subtarget : public AArch64GenSubtargetInfo {
 protected:
-  enum ARMProcFamilyEnum {Others, CortexA35, CortexA53, CortexA57, Cyclone};
+  enum ARMProcFamilyEnum {
+    Others,
+    CortexA35,
+    CortexA53,
+    CortexA57,
+    Cyclone,
+    ExynosM1
+  };
 
   /// ARMProcFamily - ARM processor family: Cortex-A53, Cortex-A57, and others.
   ARMProcFamilyEnum ARMProcFamily;
@@ -143,6 +150,7 @@ public:
   bool isCyclone() const { return CPUString == "cyclone"; }
   bool isCortexA57() const { return CPUString == "cortex-a57"; }
   bool isCortexA53() const { return CPUString == "cortex-a53"; }
+  bool isExynosM1() const { return CPUString == "exynos-m1"; }
 
   bool useAA() const override { return isCortexA53(); }
 
diff --git a/lib/Target/AArch64/Utils/AArch64BaseInfo.cpp b/lib/Target/AArch64/Utils/AArch64BaseInfo.cpp
index 78f5289ec26d..cde1c6df2608 100644
--- a/lib/Target/AArch64/Utils/AArch64BaseInfo.cpp
+++ b/lib/Target/AArch64/Utils/AArch64BaseInfo.cpp
@@ -834,7 +834,7 @@ const AArch64NamedImmMapper::Mapping AArch64SysReg::SysRegMapper::SysRegMappings
 };
 
 uint32_t
-AArch64SysReg::SysRegMapper::fromString(StringRef Name, 
+AArch64SysReg::SysRegMapper::fromString(StringRef Name,
     const FeatureBitset& FeatureBits, bool &Valid) const {
   std::string NameLower = Name.lower();
 
@@ -878,7 +878,7 @@ AArch64SysReg::SysRegMapper::fromString(StringRef Name,
 }
 
 std::string
-AArch64SysReg::SysRegMapper::toString(uint32_t Bits, 
+AArch64SysReg::SysRegMapper::toString(uint32_t Bits,
                                       const FeatureBitset& FeatureBits) const {
   // First search the registers shared by all
   for (unsigned i = 0; i < array_lengthof(SysRegMappings); ++i) {
diff --git a/lib/Target/AArch64/Utils/AArch64BaseInfo.h b/lib/Target/AArch64/Utils/AArch64BaseInfo.h
index f649cb9b8a8d..e63627eae123 100644
--- a/lib/Target/AArch64/Utils/AArch64BaseInfo.h
+++ b/lib/Target/AArch64/Utils/AArch64BaseInfo.h
@@ -285,17 +285,17 @@ struct AArch64NamedImmMapper {
     // Zero value of FeatureBitSet means the mapping is always available
     FeatureBitset FeatureBitSet;
 
-    bool isNameEqual(std::string Other, 
+    bool isNameEqual(std::string Other,
                      const FeatureBitset& FeatureBits) const {
-      if (FeatureBitSet.any() && 
+      if (FeatureBitSet.any() &&
           (FeatureBitSet & FeatureBits).none())
         return false;
       return Name == Other;
     }
 
-    bool isValueEqual(uint32_t Other, 
+    bool isValueEqual(uint32_t Other,
                       const FeatureBitset& FeatureBits) const {
-      if (FeatureBitSet.any() && 
+      if (FeatureBitSet.any() &&
           (FeatureBitSet & FeatureBits).none())
         return false;
       return Value == Other;
@@ -310,7 +310,7 @@ struct AArch64NamedImmMapper {
   StringRef toString(uint32_t Value, const FeatureBitset& FeatureBits,
                      bool &Valid) const;
   // Maps string to value, depending on availability for FeatureBits given
-  uint32_t fromString(StringRef Name, const FeatureBitset& FeatureBits, 
+  uint32_t fromString(StringRef Name, const FeatureBitset& FeatureBits,
                      bool &Valid) const;
 
   /// Many of the instructions allow an alternative assembly form consisting of
@@ -1322,7 +1322,7 @@ namespace AArch64TLBI {
       return true;
     }
   }
-} 
+}
 
 namespace AArch64II {
   /// Target Operand Flag enum.