MFC 261991:

Upgrade our copy of llvm/clang to 3.4 release.  This version supports
all of the features in the current working draft of the upcoming C++
standard, provisionally named C++1y.

The code generator's performance is greatly increased, and the loop
auto-vectorizer is now enabled at -Os and -O2 in addition to -O3.  The
PowerPC backend has made several major improvements to code generation
quality and compile time, and the X86, SPARC, ARM32, Aarch64 and SystemZ
backends have all seen major feature work.

Release notes for llvm and clang can be found here:
<http://llvm.org/releases/3.4/docs/ReleaseNotes.html>
<http://llvm.org/releases/3.4/tools/clang/docs/ReleaseNotes.html>

MFC 262121 (by emaste):

Update lldb for clang/llvm 3.4 import

This commit largely restores the lldb source to the upstream r196259
snapshot with the addition of threaded inferior support and a few bug
fixes.

Specific upstream lldb revisions restored include:
   SVN      git
  181387  779e6ac
  181703  7bef4e2
  182099  b31044e
  182650  f2dcf35
  182683  0d91b80
  183862  15c1774
  183929  99447a6
  184177  0b2934b
  184948  4dc3761
  184954  007e7bc
  186990  eebd175

Sponsored by:	DARPA, AFRL

MFC 262186 (by emaste):

Fix mismerge in r262121

A break statement was lost in the merge.  The error had no functional
impact, but restore it to reduce the diff against upstream.

MFC 262303:

Pull in r197521 from upstream clang trunk (by rdivacky):

  Use the integrated assembler by default on FreeBSD/ppc and ppc64.

Requested by:	jhibbits

MFC 262611:

Pull in r196874 from upstream llvm trunk:

  Fix a crash that occurs when PWD is invalid.

  MCJIT needs to be able to run in hostile environments, even when PWD
  is invalid. There's no need to crash MCJIT in this case.

  The obvious fix is to simply leave MCContext's CompilationDir empty
  when PWD can't be determined. This way, MCJIT clients,
  and other clients that link with LLVM don't need a valid working directory.

  If we do want to guarantee valid CompilationDir, that should be done
  only for clients of getCompilationDir(). This is as simple as checking
  for an empty string.

  The only current use of getCompilationDir is EmitGenDwarfInfo, which
  won't conceivably run with an invalid working dir. However, in the
  purely hypothetically and untestable case that this happens, the
  AT_comp_dir will be omitted from the compilation_unit DIE.

This should help fix assertions occurring with ports-mgmt/tinderbox,
when it is using jails, and sometimes invalidates clang's current
working directory.

Reported by:	decke

MFC 262809:

Pull in r203007 from upstream clang trunk:

  Don't produce an alias between destructors with different calling conventions.

  Fixes pr19007.

(Please note that is an LLVM PR identifier, not a FreeBSD one.)

This should fix Firefox and/or libxul crashes (due to problems with
regparm/stdcall calling conventions) on i386.

Reported by:	multiple users on freebsd-current
PR:		bin/187103

MFC 263048:

Repair recognition of "CC" as an alias for the C++ compiler, since it
was silently broken by upstream for a Windows-specific use-case.

Apparently some versions of CMake still rely on this archaic feature...

Reported by:	rakuco

MFC 263049:

Garbage collect the old way of adding the libstdc++ include directories
in clang's InitHeaderSearch.cpp.  This has been superseded by David
Chisnall's commit in r255321.

Moreover, if libc++ is used, the libstdc++ include directories should
not be in the search path at all.  These directories are now only used
if you pass -stdlib=libstdc++.

1 files changed, 328 insertions, 142 deletions

diff --git a/contrib/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp b/contrib/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
index 6005054..f835a4e 100644
--- a/contrib/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
+++ b/contrib/llvm/lib/Target/ARM/ARMBaseInstrInfo.cpp
@@ -11,10 +11,11 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "ARMBaseInstrInfo.h"
 #include "ARM.h"
+#include "ARMBaseInstrInfo.h"
 #include "ARMBaseRegisterInfo.h"
 #include "ARMConstantPoolValue.h"
+#include "ARMFeatures.h"
 #include "ARMHazardRecognizer.h"
 #include "ARMMachineFunctionInfo.h"
 #include "MCTargetDesc/ARMAddressingModes.h"
@@ -36,7 +37,7 @@
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 
-#define GET_INSTRINFO_CTOR
+#define GET_INSTRINFO_CTOR_DTOR
 #include "ARMGenInstrInfo.inc"
 
 using namespace llvm;
@@ -113,8 +114,7 @@ ScheduleHazardRecognizer *ARMBaseInstrInfo::
 CreateTargetPostRAHazardRecognizer(const InstrItineraryData *II,
                                    const ScheduleDAG *DAG) const {
   if (Subtarget.isThumb2() || Subtarget.hasVFP2())
-    return (ScheduleHazardRecognizer *)
-      new ARMHazardRecognizer(II, *this, getRegisterInfo(), Subtarget, DAG);
+    return (ScheduleHazardRecognizer *)new ARMHazardRecognizer(II, DAG);
   return TargetInstrInfo::CreateTargetPostRAHazardRecognizer(II, DAG);
 }
 
@@ -273,104 +273,90 @@ ARMBaseInstrInfo::AnalyzeBranch(MachineBasicBlock &MBB,MachineBasicBlock *&TBB,
                                 MachineBasicBlock *&FBB,
                                 SmallVectorImpl<MachineOperand> &Cond,
                                 bool AllowModify) const {
-  // If the block has no terminators, it just falls into the block after it.
+  TBB = 0;
+  FBB = 0;
+
   MachineBasicBlock::iterator I = MBB.end();
   if (I == MBB.begin())
-    return false;
+    return false; // Empty blocks are easy.
   --I;
-  while (I->isDebugValue()) {
-    if (I == MBB.begin())
-      return false;
-    --I;
-  }
-
-  // Get the last instruction in the block.
-  MachineInstr *LastInst = I;
-  unsigned LastOpc = LastInst->getOpcode();
 
-  // Check if it's an indirect branch first, this should return 'unanalyzable'
-  // even if it's predicated.
-  if (isIndirectBranchOpcode(LastOpc))
-    return true;
+  // Walk backwards from the end of the basic block until the branch is
+  // analyzed or we give up.
+  while (isPredicated(I) || I->isTerminator()) {
 
-  if (!isUnpredicatedTerminator(I))
-    return false;
+    // Flag to be raised on unanalyzeable instructions. This is useful in cases
+    // where we want to clean up on the end of the basic block before we bail
+    // out.
+    bool CantAnalyze = false;
 
-  // If there is only one terminator instruction, process it.
-  if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
-    if (isUncondBranchOpcode(LastOpc)) {
-      TBB = LastInst->getOperand(0).getMBB();
-      return false;
+    // Skip over DEBUG values and predicated nonterminators.
+    while (I->isDebugValue() || !I->isTerminator()) {
+      if (I == MBB.begin())
+        return false;
+      --I;
     }
-    if (isCondBranchOpcode(LastOpc)) {
-      // Block ends with fall-through condbranch.
-      TBB = LastInst->getOperand(0).getMBB();
-      Cond.push_back(LastInst->getOperand(1));
-      Cond.push_back(LastInst->getOperand(2));
-      return false;
+
+    if (isIndirectBranchOpcode(I->getOpcode()) ||
+        isJumpTableBranchOpcode(I->getOpcode())) {
+      // Indirect branches and jump tables can't be analyzed, but we still want
+      // to clean up any instructions at the tail of the basic block.
+      CantAnalyze = true;
+    } else if (isUncondBranchOpcode(I->getOpcode())) {
+      TBB = I->getOperand(0).getMBB();
+    } else if (isCondBranchOpcode(I->getOpcode())) {
+      // Bail out if we encounter multiple conditional branches.
+      if (!Cond.empty())
+        return true;
+
+      assert(!FBB && "FBB should have been null.");
+      FBB = TBB;
+      TBB = I->getOperand(0).getMBB();
+      Cond.push_back(I->getOperand(1));
+      Cond.push_back(I->getOperand(2));
+    } else if (I->isReturn()) {
+      // Returns can't be analyzed, but we should run cleanup.
+      CantAnalyze = !isPredicated(I);
+    } else {
+      // We encountered other unrecognized terminator. Bail out immediately.
+      return true;
     }
-    return true;  // Can't handle indirect branch.
-  }
 
-  // Get the instruction before it if it is a terminator.
-  MachineInstr *SecondLastInst = I;
-  unsigned SecondLastOpc = SecondLastInst->getOpcode();
-
-  // If AllowModify is true and the block ends with two or more unconditional
-  // branches, delete all but the first unconditional branch.
-  if (AllowModify && isUncondBranchOpcode(LastOpc)) {
-    while (isUncondBranchOpcode(SecondLastOpc)) {
-      LastInst->eraseFromParent();
-      LastInst = SecondLastInst;
-      LastOpc = LastInst->getOpcode();
-      if (I == MBB.begin() || !isUnpredicatedTerminator(--I)) {
-        // Return now the only terminator is an unconditional branch.
-        TBB = LastInst->getOperand(0).getMBB();
-        return false;
-      } else {
-        SecondLastInst = I;
-        SecondLastOpc = SecondLastInst->getOpcode();
+    // Cleanup code - to be run for unpredicated unconditional branches and
+    //                returns.
+    if (!isPredicated(I) &&
+          (isUncondBranchOpcode(I->getOpcode()) ||
+           isIndirectBranchOpcode(I->getOpcode()) ||
+           isJumpTableBranchOpcode(I->getOpcode()) ||
+           I->isReturn())) {
+      // Forget any previous condition branch information - it no longer applies.
+      Cond.clear();
+      FBB = 0;
+
+      // If we can modify the function, delete everything below this
+      // unconditional branch.
+      if (AllowModify) {
+        MachineBasicBlock::iterator DI = llvm::next(I);
+        while (DI != MBB.end()) {
+          MachineInstr *InstToDelete = DI;
+          ++DI;
+          InstToDelete->eraseFromParent();
+        }
       }
     }
-  }
 
-  // If there are three terminators, we don't know what sort of block this is.
-  if (SecondLastInst && I != MBB.begin() && isUnpredicatedTerminator(--I))
-    return true;
-
-  // If the block ends with a B and a Bcc, handle it.
-  if (isCondBranchOpcode(SecondLastOpc) && isUncondBranchOpcode(LastOpc)) {
-    TBB =  SecondLastInst->getOperand(0).getMBB();
-    Cond.push_back(SecondLastInst->getOperand(1));
-    Cond.push_back(SecondLastInst->getOperand(2));
-    FBB = LastInst->getOperand(0).getMBB();
-    return false;
-  }
+    if (CantAnalyze)
+      return true;
 
-  // If the block ends with two unconditional branches, handle it.  The second
-  // one is not executed, so remove it.
-  if (isUncondBranchOpcode(SecondLastOpc) && isUncondBranchOpcode(LastOpc)) {
-    TBB = SecondLastInst->getOperand(0).getMBB();
-    I = LastInst;
-    if (AllowModify)
-      I->eraseFromParent();
-    return false;
-  }
+    if (I == MBB.begin())
+      return false;
 
-  // ...likewise if it ends with a branch table followed by an unconditional
-  // branch. The branch folder can create these, and we must get rid of them for
-  // correctness of Thumb constant islands.
-  if ((isJumpTableBranchOpcode(SecondLastOpc) ||
-       isIndirectBranchOpcode(SecondLastOpc)) &&
-      isUncondBranchOpcode(LastOpc)) {
-    I = LastInst;
-    if (AllowModify)
-      I->eraseFromParent();
-    return true;
+    --I;
   }
 
-  // Otherwise, can't handle this.
-  return true;
+  // We made it past the terminators without bailing out - we must have
+  // analyzed this branch successfully.
+  return false;
 }
 
 
@@ -535,11 +521,17 @@ bool ARMBaseInstrInfo::isPredicable(MachineInstr *MI) const {
   if (!MI->isPredicable())
     return false;
 
-  if ((MI->getDesc().TSFlags & ARMII::DomainMask) == ARMII::DomainNEON) {
-    ARMFunctionInfo *AFI =
-      MI->getParent()->getParent()->getInfo<ARMFunctionInfo>();
-    return AFI->isThumb2Function();
+  ARMFunctionInfo *AFI =
+    MI->getParent()->getParent()->getInfo<ARMFunctionInfo>();
+
+  if (AFI->isThumb2Function()) {
+    if (getSubtarget().restrictIT())
+      return isV8EligibleForIT(MI);
+  } else { // non-Thumb
+    if ((MI->getDesc().TSFlags & ARMII::DomainMask) == ARMII::DomainNEON)
+      return false;
   }
+
   return true;
 }
 
@@ -660,16 +652,16 @@ void ARMBaseInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
                                    unsigned DestReg, unsigned SrcReg,
                                    bool KillSrc) const {
   bool GPRDest = ARM::GPRRegClass.contains(DestReg);
-  bool GPRSrc  = ARM::GPRRegClass.contains(SrcReg);
+  bool GPRSrc = ARM::GPRRegClass.contains(SrcReg);
 
   if (GPRDest && GPRSrc) {
     AddDefaultCC(AddDefaultPred(BuildMI(MBB, I, DL, get(ARM::MOVr), DestReg)
-                                  .addReg(SrcReg, getKillRegState(KillSrc))));
+                                    .addReg(SrcReg, getKillRegState(KillSrc))));
     return;
   }
 
   bool SPRDest = ARM::SPRRegClass.contains(DestReg);
-  bool SPRSrc  = ARM::SPRRegClass.contains(SrcReg);
+  bool SPRSrc = ARM::SPRRegClass.contains(SrcReg);
 
   unsigned Opc = 0;
   if (SPRDest && SPRSrc)
@@ -698,26 +690,47 @@ void ARMBaseInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
   int Spacing = 1;
 
   // Use VORRq when possible.
-  if (ARM::QQPRRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::VORRq, BeginIdx = ARM::qsub_0, SubRegs = 2;
-  else if (ARM::QQQQPRRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::VORRq, BeginIdx = ARM::qsub_0, SubRegs = 4;
+  if (ARM::QQPRRegClass.contains(DestReg, SrcReg)) {
+    Opc = ARM::VORRq;
+    BeginIdx = ARM::qsub_0;
+    SubRegs = 2;
+  } else if (ARM::QQQQPRRegClass.contains(DestReg, SrcReg)) {
+    Opc = ARM::VORRq;
+    BeginIdx = ARM::qsub_0;
+    SubRegs = 4;
   // Fall back to VMOVD.
-  else if (ARM::DPairRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::VMOVD, BeginIdx = ARM::dsub_0, SubRegs = 2;
-  else if (ARM::DTripleRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::VMOVD, BeginIdx = ARM::dsub_0, SubRegs = 3;
-  else if (ARM::DQuadRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::VMOVD, BeginIdx = ARM::dsub_0, SubRegs = 4;
-  else if (ARM::GPRPairRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::MOVr, BeginIdx = ARM::gsub_0, SubRegs = 2;
-
-  else if (ARM::DPairSpcRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::VMOVD, BeginIdx = ARM::dsub_0, SubRegs = 2, Spacing = 2;
-  else if (ARM::DTripleSpcRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::VMOVD, BeginIdx = ARM::dsub_0, SubRegs = 3, Spacing = 2;
-  else if (ARM::DQuadSpcRegClass.contains(DestReg, SrcReg))
-    Opc = ARM::VMOVD, BeginIdx = ARM::dsub_0, SubRegs = 4, Spacing = 2;
+  } else if (ARM::DPairRegClass.contains(DestReg, SrcReg)) {
+    Opc = ARM::VMOVD;
+    BeginIdx = ARM::dsub_0;
+    SubRegs = 2;
+  } else if (ARM::DTripleRegClass.contains(DestReg, SrcReg)) {
+    Opc = ARM::VMOVD;
+    BeginIdx = ARM::dsub_0;
+    SubRegs = 3;
+  } else if (ARM::DQuadRegClass.contains(DestReg, SrcReg)) {
+    Opc = ARM::VMOVD;
+    BeginIdx = ARM::dsub_0;
+    SubRegs = 4;
+  } else if (ARM::GPRPairRegClass.contains(DestReg, SrcReg)) {
+    Opc = Subtarget.isThumb2() ? ARM::tMOVr : ARM::MOVr;
+    BeginIdx = ARM::gsub_0;
+    SubRegs = 2;
+  } else if (ARM::DPairSpcRegClass.contains(DestReg, SrcReg)) {
+    Opc = ARM::VMOVD;
+    BeginIdx = ARM::dsub_0;
+    SubRegs = 2;
+    Spacing = 2;
+  } else if (ARM::DTripleSpcRegClass.contains(DestReg, SrcReg)) {
+    Opc = ARM::VMOVD;
+    BeginIdx = ARM::dsub_0;
+    SubRegs = 3;
+    Spacing = 2;
+  } else if (ARM::DQuadSpcRegClass.contains(DestReg, SrcReg)) {
+    Opc = ARM::VMOVD;
+    BeginIdx = ARM::dsub_0;
+    SubRegs = 4;
+    Spacing = 2;
+  }
 
   assert(Opc && "Impossible reg-to-reg copy");
 
@@ -726,26 +739,28 @@ void ARMBaseInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
 
   // Copy register tuples backward when the first Dest reg overlaps with SrcReg.
   if (TRI->regsOverlap(SrcReg, TRI->getSubReg(DestReg, BeginIdx))) {
-    BeginIdx = BeginIdx + ((SubRegs-1)*Spacing);
+    BeginIdx = BeginIdx + ((SubRegs - 1) * Spacing);
     Spacing = -Spacing;
   }
 #ifndef NDEBUG
   SmallSet<unsigned, 4> DstRegs;
 #endif
   for (unsigned i = 0; i != SubRegs; ++i) {
-    unsigned Dst = TRI->getSubReg(DestReg, BeginIdx + i*Spacing);
-    unsigned Src = TRI->getSubReg(SrcReg,  BeginIdx + i*Spacing);
+    unsigned Dst = TRI->getSubReg(DestReg, BeginIdx + i * Spacing);
+    unsigned Src = TRI->getSubReg(SrcReg, BeginIdx + i * Spacing);
     assert(Dst && Src && "Bad sub-register");
 #ifndef NDEBUG
     assert(!DstRegs.count(Src) && "destructive vector copy");
     DstRegs.insert(Dst);
 #endif
-    Mov = BuildMI(MBB, I, I->getDebugLoc(), get(Opc), Dst)
-      .addReg(Src);
+    Mov = BuildMI(MBB, I, I->getDebugLoc(), get(Opc), Dst).addReg(Src);
     // VORR takes two source operands.
     if (Opc == ARM::VORRq)
       Mov.addReg(Src);
     Mov = AddDefaultPred(Mov);
+    // MOVr can set CC.
+    if (Opc == ARM::MOVr)
+      Mov = AddDefaultCC(Mov);
   }
   // Add implicit super-register defs and kills to the last instruction.
   Mov->addRegisterDefined(DestReg, TRI);
@@ -1214,16 +1229,6 @@ bool ARMBaseInstrInfo::expandPostRAPseudo(MachineBasicBlock::iterator MI) const{
   return true;
 }
 
-MachineInstr*
-ARMBaseInstrInfo::emitFrameIndexDebugValue(MachineFunction &MF,
-                                           int FrameIx, uint64_t Offset,
-                                           const MDNode *MDPtr,
-                                           DebugLoc DL) const {
-  MachineInstrBuilder MIB = BuildMI(MF, DL, get(ARM::DBG_VALUE))
-    .addFrameIndex(FrameIx).addImm(0).addImm(Offset).addMetadata(MDPtr);
-  return &*MIB;
-}
-
 /// Create a copy of a const pool value. Update CPI to the new index and return
 /// the label UID.
 static unsigned duplicateCPV(MachineFunction &MF, unsigned &CPI) {
@@ -1426,9 +1431,11 @@ bool ARMBaseInstrInfo::areLoadsFromSameBasePtr(SDNode *Load1, SDNode *Load2,
   case ARM::VLDRD:
   case ARM::VLDRS:
   case ARM::t2LDRi8:
+  case ARM::t2LDRBi8:
   case ARM::t2LDRDi8:
   case ARM::t2LDRSHi8:
   case ARM::t2LDRi12:
+  case ARM::t2LDRBi12:
   case ARM::t2LDRSHi12:
     break;
   }
@@ -1445,8 +1452,10 @@ bool ARMBaseInstrInfo::areLoadsFromSameBasePtr(SDNode *Load1, SDNode *Load2,
   case ARM::VLDRD:
   case ARM::VLDRS:
   case ARM::t2LDRi8:
+  case ARM::t2LDRBi8:
   case ARM::t2LDRSHi8:
   case ARM::t2LDRi12:
+  case ARM::t2LDRBi12:
   case ARM::t2LDRSHi12:
     break;
   }
@@ -1493,7 +1502,16 @@ bool ARMBaseInstrInfo::shouldScheduleLoadsNear(SDNode *Load1, SDNode *Load2,
   if ((Offset2 - Offset1) / 8 > 64)
     return false;
 
-  if (Load1->getMachineOpcode() != Load2->getMachineOpcode())
+  // Check if the machine opcodes are different. If they are different
+  // then we consider them to not be of the same base address,
+  // EXCEPT in the case of Thumb2 byte loads where one is LDRBi8 and the other LDRBi12.
+  // In this case, they are considered to be the same because they are different
+  // encoding forms of the same basic instruction.
+  if ((Load1->getMachineOpcode() != Load2->getMachineOpcode()) &&
+      !((Load1->getMachineOpcode() == ARM::t2LDRBi8 &&
+         Load2->getMachineOpcode() == ARM::t2LDRBi12) ||
+        (Load1->getMachineOpcode() == ARM::t2LDRBi12 &&
+         Load2->getMachineOpcode() == ARM::t2LDRBi8)))
     return false;  // FIXME: overly conservative?
 
   // Four loads in a row should be sufficient.
@@ -1708,7 +1726,7 @@ MachineInstr *ARMBaseInstrInfo::optimizeSelect(MachineInstr *MI,
                                                bool PreferFalse) const {
   assert((MI->getOpcode() == ARM::MOVCCr || MI->getOpcode() == ARM::t2MOVCCr) &&
          "Unknown select instruction");
-  const MachineRegisterInfo &MRI = MI->getParent()->getParent()->getRegInfo();
+  MachineRegisterInfo &MRI = MI->getParent()->getParent()->getRegInfo();
   MachineInstr *DefMI = canFoldIntoMOVCC(MI->getOperand(2).getReg(), MRI, this);
   bool Invert = !DefMI;
   if (!DefMI)
@@ -1716,11 +1734,17 @@ MachineInstr *ARMBaseInstrInfo::optimizeSelect(MachineInstr *MI,
   if (!DefMI)
     return 0;
 
+  // Find new register class to use.
+  MachineOperand FalseReg = MI->getOperand(Invert ? 2 : 1);
+  unsigned       DestReg  = MI->getOperand(0).getReg();
+  const TargetRegisterClass *PreviousClass = MRI.getRegClass(FalseReg.getReg());
+  if (!MRI.constrainRegClass(DestReg, PreviousClass))
+    return 0;
+
   // Create a new predicated version of DefMI.
   // Rfalse is the first use.
   MachineInstrBuilder NewMI = BuildMI(*MI->getParent(), MI, MI->getDebugLoc(),
-                                      DefMI->getDesc(),
-                                      MI->getOperand(0).getReg());
+                                      DefMI->getDesc(), DestReg);
 
   // Copy all the DefMI operands, excluding its (null) predicate.
   const MCInstrDesc &DefDesc = DefMI->getDesc();
@@ -1743,7 +1767,6 @@ MachineInstr *ARMBaseInstrInfo::optimizeSelect(MachineInstr *MI,
   // register operand tied to the first def.
   // The tie makes the register allocator ensure the FalseReg is allocated the
   // same register as operand 0.
-  MachineOperand FalseReg = MI->getOperand(Invert ? 2 : 1);
   FalseReg.setImplicit();
   NewMI.addOperand(FalseReg);
   NewMI->tieOperands(0, NewMI->getNumOperands() - 1);
@@ -1803,6 +1826,14 @@ void llvm::emitARMRegPlusImmediate(MachineBasicBlock &MBB,
                                unsigned DestReg, unsigned BaseReg, int NumBytes,
                                ARMCC::CondCodes Pred, unsigned PredReg,
                                const ARMBaseInstrInfo &TII, unsigned MIFlags) {
+  if (NumBytes == 0 && DestReg != BaseReg) {
+    BuildMI(MBB, MBBI, dl, TII.get(ARM::MOVr), DestReg)
+      .addReg(BaseReg, RegState::Kill)
+      .addImm((unsigned)Pred).addReg(PredReg).addReg(0)
+      .setMIFlags(MIFlags);
+    return;
+  }
+
   bool isSub = NumBytes < 0;
   if (isSub) NumBytes = -NumBytes;
 
@@ -1826,6 +1857,115 @@ void llvm::emitARMRegPlusImmediate(MachineBasicBlock &MBB,
   }
 }
 
+bool llvm::tryFoldSPUpdateIntoPushPop(MachineFunction &MF,
+                                      MachineInstr *MI,
+                                      unsigned NumBytes) {
+  // This optimisation potentially adds lots of load and store
+  // micro-operations, it's only really a great benefit to code-size.
+  if (!MF.getFunction()->hasFnAttribute(Attribute::MinSize))
+    return false;
+
+  // If only one register is pushed/popped, LLVM can use an LDR/STR
+  // instead. We can't modify those so make sure we're dealing with an
+  // instruction we understand.
+  bool IsPop = isPopOpcode(MI->getOpcode());
+  bool IsPush = isPushOpcode(MI->getOpcode());
+  if (!IsPush && !IsPop)
+    return false;
+
+  bool IsVFPPushPop = MI->getOpcode() == ARM::VSTMDDB_UPD ||
+                      MI->getOpcode() == ARM::VLDMDIA_UPD;
+  bool IsT1PushPop = MI->getOpcode() == ARM::tPUSH ||
+                     MI->getOpcode() == ARM::tPOP ||
+                     MI->getOpcode() == ARM::tPOP_RET;
+
+  assert((IsT1PushPop || (MI->getOperand(0).getReg() == ARM::SP &&
+                          MI->getOperand(1).getReg() == ARM::SP)) &&
+         "trying to fold sp update into non-sp-updating push/pop");
+
+  // The VFP push & pop act on D-registers, so we can only fold an adjustment
+  // by a multiple of 8 bytes in correctly. Similarly rN is 4-bytes. Don't try
+  // if this is violated.
+  if (NumBytes % (IsVFPPushPop ? 8 : 4) != 0)
+    return false;
+
+  // ARM and Thumb2 push/pop insts have explicit "sp, sp" operands (+
+  // pred) so the list starts at 4. Thumb1 starts after the predicate.
+  int RegListIdx = IsT1PushPop ? 2 : 4;
+
+  // Calculate the space we'll need in terms of registers.
+  unsigned FirstReg = MI->getOperand(RegListIdx).getReg();
+  unsigned RD0Reg, RegsNeeded;
+  if (IsVFPPushPop) {
+    RD0Reg = ARM::D0;
+    RegsNeeded = NumBytes / 8;
+  } else {
+    RD0Reg = ARM::R0;
+    RegsNeeded = NumBytes / 4;
+  }
+
+  // We're going to have to strip all list operands off before
+  // re-adding them since the order matters, so save the existing ones
+  // for later.
+  SmallVector<MachineOperand, 4> RegList;
+  for (int i = MI->getNumOperands() - 1; i >= RegListIdx; --i)
+    RegList.push_back(MI->getOperand(i));
+
+  MachineBasicBlock *MBB = MI->getParent();
+  const TargetRegisterInfo *TRI = MF.getRegInfo().getTargetRegisterInfo();
+  const MCPhysReg *CSRegs = TRI->getCalleeSavedRegs(&MF);
+
+  // Now try to find enough space in the reglist to allocate NumBytes.
+  for (unsigned CurReg = FirstReg - 1; CurReg >= RD0Reg && RegsNeeded;
+       --CurReg) {
+    if (!IsPop) {
+      // Pushing any register is completely harmless, mark the
+      // register involved as undef since we don't care about it in
+      // the slightest.
+      RegList.push_back(MachineOperand::CreateReg(CurReg, false, false,
+                                                  false, false, true));
+      --RegsNeeded;
+      continue;
+    }
+
+    // However, we can only pop an extra register if it's not live. For
+    // registers live within the function we might clobber a return value
+    // register; the other way a register can be live here is if it's
+    // callee-saved.
+    if (isCalleeSavedRegister(CurReg, CSRegs) ||
+        MBB->computeRegisterLiveness(TRI, CurReg, MI) !=
+            MachineBasicBlock::LQR_Dead) {
+      // VFP pops don't allow holes in the register list, so any skip is fatal
+      // for our transformation. GPR pops do, so we should just keep looking.
+      if (IsVFPPushPop)
+        return false;
+      else
+        continue;
+    }
+
+    // Mark the unimportant registers as <def,dead> in the POP.
+    RegList.push_back(MachineOperand::CreateReg(CurReg, true, false, false,
+                                                true));
+    --RegsNeeded;
+  }
+
+  if (RegsNeeded > 0)
+    return false;
+
+  // Finally we know we can profitably perform the optimisation so go
+  // ahead: strip all existing registers off and add them back again
+  // in the right order.
+  for (int i = MI->getNumOperands() - 1; i >= RegListIdx; --i)
+    MI->RemoveOperand(i);
+
+  // Add the complete list back in.
+  MachineInstrBuilder MIB(MF, &*MI);
+  for (int i = RegList.size() - 1; i >= 0; --i)
+    MIB.addOperand(RegList[i]);
+
+  return true;
+}
+
 bool llvm::rewriteARMFrameIndex(MachineInstr &MI, unsigned FrameRegIdx,
                                 unsigned FrameReg, int &Offset,
                                 const ARMBaseInstrInfo &TII) {
@@ -2232,8 +2372,32 @@ optimizeCompareInstr(MachineInstr *CmpInstr, unsigned SrcReg, unsigned SrcReg2,
           isSafe = true;
           break;
         }
-        // Condition code is after the operand before CPSR.
-        ARMCC::CondCodes CC = (ARMCC::CondCodes)Instr.getOperand(IO-1).getImm();
+        // Condition code is after the operand before CPSR except for VSELs.
+        ARMCC::CondCodes CC;
+        bool IsInstrVSel = true;
+        switch (Instr.getOpcode()) {
+        default:
+          IsInstrVSel = false;
+          CC = (ARMCC::CondCodes)Instr.getOperand(IO - 1).getImm();
+          break;
+        case ARM::VSELEQD:
+        case ARM::VSELEQS:
+          CC = ARMCC::EQ;
+          break;
+        case ARM::VSELGTD:
+        case ARM::VSELGTS:
+          CC = ARMCC::GT;
+          break;
+        case ARM::VSELGED:
+        case ARM::VSELGES:
+          CC = ARMCC::GE;
+          break;
+        case ARM::VSELVSS:
+        case ARM::VSELVSD:
+          CC = ARMCC::VS;
+          break;
+        }
+
         if (Sub) {
           ARMCC::CondCodes NewCC = getSwappedCondition(CC);
           if (NewCC == ARMCC::AL)
@@ -2244,11 +2408,14 @@ optimizeCompareInstr(MachineInstr *CmpInstr, unsigned SrcReg, unsigned SrcReg2,
           // If it is safe to remove CmpInstr, the condition code of these
           // operands will be modified.
           if (SrcReg2 != 0 && Sub->getOperand(1).getReg() == SrcReg2 &&
-              Sub->getOperand(2).getReg() == SrcReg)
-            OperandsToUpdate.push_back(std::make_pair(&((*I).getOperand(IO-1)),
-                                                      NewCC));
-        }
-        else
+              Sub->getOperand(2).getReg() == SrcReg) {
+            // VSel doesn't support condition code update.
+            if (IsInstrVSel)
+              return false;
+            OperandsToUpdate.push_back(
+                std::make_pair(&((*I).getOperand(IO - 1)), NewCC));
+          }
+        } else
           switch (CC) {
           default:
             // CPSR can be used multiple times, we should continue.
@@ -3604,6 +3771,24 @@ ARMBaseInstrInfo::getOperandLatency(const InstrItineraryData *ItinData,
   return Latency;
 }
 
+unsigned ARMBaseInstrInfo::getPredicationCost(const MachineInstr *MI) const {
+   if (MI->isCopyLike() || MI->isInsertSubreg() ||
+      MI->isRegSequence() || MI->isImplicitDef())
+    return 0;
+
+  if (MI->isBundle())
+    return 0;
+
+  const MCInstrDesc &MCID = MI->getDesc();
+
+  if (MCID.isCall() || MCID.hasImplicitDefOfPhysReg(ARM::CPSR)) {
+    // When predicated, CPSR is an additional source operand for CPSR updating
+    // instructions, this apparently increases their latencies.
+    return 1;
+  }
+  return 0;
+}
+
 unsigned ARMBaseInstrInfo::getInstrLatency(const InstrItineraryData *ItinData,
                                            const MachineInstr *MI,
                                            unsigned *PredCost) const {
@@ -3685,8 +3870,7 @@ hasHighOperandLatency(const InstrItineraryData *ItinData,
     return true;
 
   // Hoist VFP / NEON instructions with 4 or higher latency.
-  int Latency = computeOperandLatency(ItinData, DefMI, DefIdx, UseMI, UseIdx,
-                                      /*FindMin=*/false);
+  int Latency = computeOperandLatency(ItinData, DefMI, DefIdx, UseMI, UseIdx);
   if (Latency < 0)
     Latency = getInstrLatency(ItinData, DefMI);
   if (Latency <= 3)
@@ -4137,7 +4321,7 @@ breakPartialRegDependency(MachineBasicBlock::iterator MI,
   // FIXME: In some cases, VLDRS can be changed to a VLD1DUPd32 which defines
   // the full D-register by loading the same value to both lanes.  The
   // instruction is micro-coded with 2 uops, so don't do this until we can
-  // properly schedule micro-coded instuctions.  The dispatcher stalls cause
+  // properly schedule micro-coded instructions.  The dispatcher stalls cause
   // too big regressions.
 
   // Insert the dependency-breaking FCONSTD before MI.
@@ -4152,6 +4336,8 @@ bool ARMBaseInstrInfo::hasNOP() const {
 }
 
 bool ARMBaseInstrInfo::isSwiftFastImmShift(const MachineInstr *MI) const {
+  if (MI->getNumOperands() < 4)
+    return true;
   unsigned ShOpVal = MI->getOperand(3).getImm();
   unsigned ShImm = ARM_AM::getSORegOffset(ShOpVal);
   // Swift supports faster shifts for: lsl 2, lsl 1, and lsr 1.