am 2d4629c5: Merge branch \'upstream\' into merge_2013_03_18

* commit '2d4629c5d7dcc6582fa7b85a517744f1a3654eba': (424 commits)
  Change NULL to 0.
  Register the flush function for each compile unit.
  Remove trailing spaces.
  Fix PPC unaligned 64-bit loads and stores
  ARM cost model: Make some vector integer to float casts cheaper
  ARM cost model: Correct cost for some cheap float to integer conversions
  Extend global merge pass to optionally consider global constant variables. Also add some checks to not merge globals used within landing pad instructions or marked as "used".
  Change test cases to handle unaligned references.
  Remove unnecessary leading comment characters in lit-only file
  Add SchedRW annotations to most of X86InstrSSE.td.
  Annotate X86 arithmetic instructions with SchedRW lists.
  Check whether a pointer is non-null (isKnownNonNull) in isKnownNonZero.
  TableGen fix for the new machine model.
  Include '.test' suffix in target specific lit configs that need it
  Make the fields in the diagram match the descriptive text above them.
  Update
  Fix 80-col. violations in PPCCTRLoops
  Fix large count and negative constant count handling in PPCCTRLoops
  Cleanup initial-value constants in PPCCTRLoops
  Fix integer comparison in DIEInteger::BestForm.
  ...

1 files changed, 66 insertions, 36 deletions

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index ef96e56..514971f 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -504,6 +504,7 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::FRINT, MVT::v2f64, Expand);
     setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Expand);
     setOperationAction(ISD::FFLOOR, MVT::v2f64, Expand);
+    setOperationAction(ISD::FMA, MVT::v2f64, Expand);
 
     setOperationAction(ISD::FSQRT, MVT::v4f32, Expand);
     setOperationAction(ISD::FSIN, MVT::v4f32, Expand);
@@ -521,6 +522,23 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Expand);
     setOperationAction(ISD::FFLOOR, MVT::v4f32, Expand);
 
+    // Mark v2f32 intrinsics.
+    setOperationAction(ISD::FSQRT, MVT::v2f32, Expand);
+    setOperationAction(ISD::FSIN, MVT::v2f32, Expand);
+    setOperationAction(ISD::FCOS, MVT::v2f32, Expand);
+    setOperationAction(ISD::FPOWI, MVT::v2f32, Expand);
+    setOperationAction(ISD::FPOW, MVT::v2f32, Expand);
+    setOperationAction(ISD::FLOG, MVT::v2f32, Expand);
+    setOperationAction(ISD::FLOG2, MVT::v2f32, Expand);
+    setOperationAction(ISD::FLOG10, MVT::v2f32, Expand);
+    setOperationAction(ISD::FEXP, MVT::v2f32, Expand);
+    setOperationAction(ISD::FEXP2, MVT::v2f32, Expand);
+    setOperationAction(ISD::FCEIL, MVT::v2f32, Expand);
+    setOperationAction(ISD::FTRUNC, MVT::v2f32, Expand);
+    setOperationAction(ISD::FRINT, MVT::v2f32, Expand);
+    setOperationAction(ISD::FNEARBYINT, MVT::v2f32, Expand);
+    setOperationAction(ISD::FFLOOR, MVT::v2f32, Expand);
+
     // Neon does not support some operations on v1i64 and v2i64 types.
     setOperationAction(ISD::MUL, MVT::v1i64, Expand);
     // Custom handling for some quad-vector types to detect VMULL.
@@ -554,6 +572,12 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
     setOperationAction(ISD::CTPOP,      MVT::v4i16, Custom);
     setOperationAction(ISD::CTPOP,      MVT::v8i16, Custom);
 
+    // NEON only has FMA instructions as of VFP4.
+    if (!Subtarget->hasVFP4()) {
+      setOperationAction(ISD::FMA, MVT::v2f32, Expand);
+      setOperationAction(ISD::FMA, MVT::v4f32, Expand);
+    }
+
     setTargetDAGCombine(ISD::INTRINSIC_VOID);
     setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
     setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
@@ -1581,7 +1605,7 @@ ARMTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
       // On ELF targets for PIC code, direct calls should go through the PLT
       unsigned OpFlags = 0;
       if (Subtarget->isTargetELF() &&
-                  getTargetMachine().getRelocationModel() == Reloc::PIC_)
+          getTargetMachine().getRelocationModel() == Reloc::PIC_)
         OpFlags = ARMII::MO_PLT;
       Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(), 0, OpFlags);
     }
@@ -2247,8 +2271,7 @@ SDValue ARMTargetLowering::LowerGlobalAddressELF(SDValue Op,
   EVT PtrVT = getPointerTy();
   DebugLoc dl = Op.getDebugLoc();
   const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
-  Reloc::Model RelocM = getTargetMachine().getRelocationModel();
-  if (RelocM == Reloc::PIC_) {
+  if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
     bool UseGOTOFF = GV->hasLocalLinkage() || GV->hasHiddenVisibility();
     ARMConstantPoolValue *CPV =
       ARMConstantPoolConstant::Create(GV,
@@ -2292,8 +2315,6 @@ SDValue ARMTargetLowering::LowerGlobalAddressDarwin(SDValue Op,
   DebugLoc dl = Op.getDebugLoc();
   const GlobalValue *GV = cast<GlobalAddressSDNode>(Op)->getGlobal();
   Reloc::Model RelocM = getTargetMachine().getRelocationModel();
-  MachineFunction &MF = DAG.getMachineFunction();
-  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
 
   // FIXME: Enable this for static codegen when tool issues are fixed.  Also
   // update ARMFastISel::ARMMaterializeGV.
@@ -2321,6 +2342,7 @@ SDValue ARMTargetLowering::LowerGlobalAddressDarwin(SDValue Op,
   if (RelocM == Reloc::Static) {
     CPAddr = DAG.getTargetConstantPool(GV, PtrVT, 4);
   } else {
+    ARMFunctionInfo *AFI = DAG.getMachineFunction().getInfo<ARMFunctionInfo>();
     ARMPCLabelIndex = AFI->createPICLabelUId();
     unsigned PCAdj = (RelocM != Reloc::PIC_) ? 0 : (Subtarget->isThumb()?4:8);
     ARMConstantPoolValue *CPV =
@@ -2401,7 +2423,6 @@ ARMTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG,
     ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
     unsigned ARMPCLabelIndex = AFI->createPICLabelUId();
     EVT PtrVT = getPointerTy();
-    DebugLoc dl = Op.getDebugLoc();
     Reloc::Model RelocM = getTargetMachine().getRelocationModel();
     SDValue CPAddr;
     unsigned PCAdj = (RelocM != Reloc::PIC_)
@@ -2661,7 +2682,7 @@ ARMTargetLowering::LowerFormalArguments(SDValue Chain,
   CCInfo.AnalyzeFormalArguments(Ins,
                                 CCAssignFnForNode(CallConv, /* Return*/ false,
                                                   isVarArg));
-  
+
   SmallVector<SDValue, 16> ArgValues;
   int lastInsIndex = -1;
   SDValue ArgValue;
@@ -2776,7 +2797,7 @@ ARMTargetLowering::LowerFormalArguments(SDValue Chain,
             } else {
               int FI = MFI->CreateFixedObject(Flags.getByValSize(),
                                               VA.getLocMemOffset(), false);
-              InVals.push_back(DAG.getFrameIndex(FI, getPointerTy()));              
+              InVals.push_back(DAG.getFrameIndex(FI, getPointerTy()));
             }
           } else {
             int FI = MFI->CreateFixedObject(VA.getLocVT().getSizeInBits()/8,
@@ -3573,7 +3594,7 @@ static SDValue LowerCTTZ(SDNode *N, SelectionDAG &DAG,
 /// input    = [v0    v1    v2    v3   ] (vi 16-bit element)
 /// cast: N0 = [w0 w1 w2 w3 w4 w5 w6 w7] (v0 = [w0 w1], wi 8-bit element)
 /// vcnt: N1 = [b0 b1 b2 b3 b4 b5 b6 b7] (bi = bit-count of 8-bit element wi)
-/// vrev: N2 = [b1 b0 b3 b2 b5 b4 b7 b6] 
+/// vrev: N2 = [b1 b0 b3 b2 b5 b4 b7 b6]
 ///            [b0 b1 b2 b3 b4 b5 b6 b7]
 ///           +[b1 b0 b3 b2 b5 b4 b7 b6]
 /// N3=N1+N2 = [k0 k0 k1 k1 k2 k2 k3 k3] (k0 = b0+b1 = bit-count of 16-bit v0,
@@ -3594,7 +3615,7 @@ static SDValue getCTPOP16BitCounts(SDNode *N, SelectionDAG &DAG) {
 /// bit-count for each 16-bit element from the operand.  We need slightly
 /// different sequencing for v4i16 and v8i16 to stay within NEON's available
 /// 64/128-bit registers.
-/// 
+///
 /// Trace for v4i16:
 /// input           = [v0    v1    v2    v3    ] (vi 16-bit element)
 /// v8i8: BitCounts = [k0 k1 k2 k3 k0 k1 k2 k3 ] (ki is the bit-count of vi)
@@ -3625,7 +3646,7 @@ static SDValue lowerCTPOP16BitElements(SDNode *N, SelectionDAG &DAG) {
 /// input    = [v0    v1    ] (vi: 32-bit elements)
 /// Bitcast  = [w0 w1 w2 w3 ] (wi: 16-bit elements, v0 = [w0 w1])
 /// Counts16 = [k0 k1 k2 k3 ] (ki: 16-bit elements, bit-count of wi)
-/// vrev: N0 = [k1 k0 k3 k2 ] 
+/// vrev: N0 = [k1 k0 k3 k2 ]
 ///            [k0 k1 k2 k3 ]
 ///       N1 =+[k1 k0 k3 k2 ]
 ///            [k0 k2 k1 k3 ]
@@ -4403,7 +4424,7 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
 
     ValueCounts.insert(std::make_pair(V, 0));
     unsigned &Count = ValueCounts[V];
-    
+
     // Is this value dominant? (takes up more than half of the lanes)
     if (++Count > (NumElts / 2)) {
       hasDominantValue = true;
@@ -4431,8 +4452,11 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
 
       // If we are VDUPing a value that comes directly from a vector, that will
       // cause an unnecessary move to and from a GPR, where instead we could
-      // just use VDUPLANE.
-      if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
+      // just use VDUPLANE. We can only do this if the lane being extracted
+      // is at a constant index, as the VDUP from lane instructions only have
+      // constant-index forms.
+      if (Value->getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
+          isa<ConstantSDNode>(Value->getOperand(1))) {
         // We need to create a new undef vector to use for the VDUPLANE if the
         // size of the vector from which we get the value is different than the
         // size of the vector that we need to create. We will insert the element
@@ -4447,12 +4471,10 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
                  DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, DAG.getUNDEF(VT),
                         Value, DAG.getConstant(index, MVT::i32)),
                            DAG.getConstant(index, MVT::i32));
-        } else {
+        } else
           N = DAG.getNode(ARMISD::VDUPLANE, dl, VT,
                         Value->getOperand(0), Value->getOperand(1));
-        }
-      }
-      else
+      } else
         N = DAG.getNode(ARMISD::VDUP, dl, VT, Value);
 
       if (!usesOnlyOneValue) {
@@ -4484,7 +4506,7 @@ SDValue ARMTargetLowering::LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
     if (usesOnlyOneValue) {
       SDValue Val = IsSingleInstrConstant(Value, DAG, ST, dl);
       if (isConstant && Val.getNode())
-        return DAG.getNode(ARMISD::VDUP, dl, VT, Val); 
+        return DAG.getNode(ARMISD::VDUP, dl, VT, Val);
     }
   }
 
@@ -6329,6 +6351,7 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
     MF->getOrCreateJumpTableInfo(MachineJumpTableInfo::EK_Inline);
   unsigned MJTI = JTI->createJumpTableIndex(LPadList);
   unsigned UId = AFI->createJumpTableUId();
+  Reloc::Model RelocM = getTargetMachine().getRelocationModel();
 
   // Create the MBBs for the dispatch code.
 
@@ -6338,14 +6361,11 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
 
   MachineBasicBlock *TrapBB = MF->CreateMachineBasicBlock();
   unsigned trap_opcode;
-  if (Subtarget->isThumb()) {
+  if (Subtarget->isThumb())
     trap_opcode = ARM::tTRAP;
-  } else {
-    if (Subtarget->useNaClTrap())
-      trap_opcode = ARM::TRAPNaCl;
-    else
-      trap_opcode = ARM::TRAP;
-  }
+  else
+    trap_opcode = Subtarget->useNaClTrap() ? ARM::TRAPNaCl : ARM::TRAP;
+
   BuildMI(TrapBB, dl, TII->get(trap_opcode));
   DispatchBB->addSuccessor(TrapBB);
 
@@ -6492,11 +6512,14 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
                    .addImm(0)
                    .addMemOperand(JTMMOLd));
 
-    unsigned NewVReg6 = MRI->createVirtualRegister(TRC);
-    AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg6)
-                   .addReg(ARM::CPSR, RegState::Define)
-                   .addReg(NewVReg5, RegState::Kill)
-                   .addReg(NewVReg3));
+    unsigned NewVReg6 = NewVReg5;
+    if (RelocM == Reloc::PIC_) {
+      NewVReg6 = MRI->createVirtualRegister(TRC);
+      AddDefaultPred(BuildMI(DispContBB, dl, TII->get(ARM::tADDrr), NewVReg6)
+                     .addReg(ARM::CPSR, RegState::Define)
+                     .addReg(NewVReg5, RegState::Kill)
+                     .addReg(NewVReg3));
+    }
 
     BuildMI(DispContBB, dl, TII->get(ARM::tBR_JTr))
       .addReg(NewVReg6, RegState::Kill)
@@ -6576,11 +6599,18 @@ EmitSjLjDispatchBlock(MachineInstr *MI, MachineBasicBlock *MBB) const {
       .addImm(0)
       .addMemOperand(JTMMOLd));
 
-    BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd))
-      .addReg(NewVReg5, RegState::Kill)
-      .addReg(NewVReg4)
-      .addJumpTableIndex(MJTI)
-      .addImm(UId);
+    if (RelocM == Reloc::PIC_) {
+      BuildMI(DispContBB, dl, TII->get(ARM::BR_JTadd))
+        .addReg(NewVReg5, RegState::Kill)
+        .addReg(NewVReg4)
+        .addJumpTableIndex(MJTI)
+        .addImm(UId);
+    } else {
+      BuildMI(DispContBB, dl, TII->get(ARM::BR_JTr))
+        .addReg(NewVReg5, RegState::Kill)
+        .addJumpTableIndex(MJTI)
+        .addImm(UId);
+    }
   }
 
   // Add the jump table entries as successors to the MBB.