Move more PEI-related hooks to TFI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120229 91177308-0d34-0410-b5e6-96231b3b80d8

31 files changed, 796 insertions, 793 deletions

diff --git a/include/llvm/Target/TargetFrameInfo.h b/include/llvm/Target/TargetFrameInfo.h
index ebaf221aad..3e502e7d74 100644
--- a/include/llvm/Target/TargetFrameInfo.h
+++ b/include/llvm/Target/TargetFrameInfo.h
@@ -24,6 +24,7 @@ namespace llvm {
   class MachineFunction;
   class MachineBasicBlock;
   class MachineMove;
+  class RegScavenger;
 
 /// Information about stack frame layout on the target.  It holds the direction
 /// of stack growth, the known stack alignment on entry to each function, and
@@ -172,6 +173,22 @@ public:
   /// returned directly, and the base register is returned via FrameReg.
   virtual int getFrameIndexReference(const MachineFunction &MF, int FI,
                                      unsigned &FrameReg) const;
+
+  /// processFunctionBeforeCalleeSavedScan - This method is called immediately
+  /// before PrologEpilogInserter scans the physical registers used to determine
+  /// what callee saved registers should be spilled. This method is optional.
+  virtual void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
+                                                RegScavenger *RS = NULL) const {
+
+  }
+
+  /// processFunctionBeforeFrameFinalized - This method is called immediately
+  /// before the specified function's frame layout (MF.getFrameInfo()) is
+  /// finalized.  Once the frame is finalized, MO_FrameIndex operands are
+  /// replaced with direct constants.  This method is optional.
+  ///
+  virtual void processFunctionBeforeFrameFinalized(MachineFunction &MF) const {
+  }
 };
 
 } // End llvm namespace
diff --git a/include/llvm/Target/TargetRegisterInfo.h b/include/llvm/Target/TargetRegisterInfo.h
index 572bbfb7dd..9caedcb506 100644
--- a/include/llvm/Target/TargetRegisterInfo.h
+++ b/include/llvm/Target/TargetRegisterInfo.h
@@ -678,21 +678,6 @@ public:
     assert(0 && "Call Frame Pseudo Instructions do not exist on this target!");
   }
 
-  /// processFunctionBeforeCalleeSavedScan - This method is called immediately
-  /// before PrologEpilogInserter scans the physical registers used to determine
-  /// what callee saved registers should be spilled. This method is optional.
-  virtual void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
-                                                RegScavenger *RS = NULL) const {
-
-  }
-
-  /// processFunctionBeforeFrameFinalized - This method is called immediately
-  /// before the specified function's frame layout (MF.getFrameInfo()) is
-  /// finalized.  Once the frame is finalized, MO_FrameIndex operands are
-  /// replaced with direct constants.  This method is optional.
-  ///
-  virtual void processFunctionBeforeFrameFinalized(MachineFunction &MF) const {
-  }
 
   /// saveScavengerRegister - Spill the register so it can be used by the
   /// register scavenger. Return true if the register was spilled, false
diff --git a/lib/CodeGen/PrologEpilogInserter.cpp b/lib/CodeGen/PrologEpilogInserter.cpp
index c461623192..5ba7086eef 100644
--- a/lib/CodeGen/PrologEpilogInserter.cpp
+++ b/lib/CodeGen/PrologEpilogInserter.cpp
@@ -65,6 +65,8 @@ FunctionPass *llvm::createPrologEpilogCodeInserter() { return new PEI(); }
 bool PEI::runOnMachineFunction(MachineFunction &Fn) {
   const Function* F = Fn.getFunction();
   const TargetRegisterInfo *TRI = Fn.getTarget().getRegisterInfo();
+  const TargetFrameInfo *TFI = Fn.getTarget().getFrameInfo();
+
   RS = TRI->requiresRegisterScavenging(Fn) ? new RegScavenger() : NULL;
   FrameIndexVirtualScavenging = TRI->requiresFrameIndexScavenging(Fn);
 
@@ -75,7 +77,7 @@ bool PEI::runOnMachineFunction(MachineFunction &Fn) {
 
   // Allow the target machine to make some adjustments to the function
   // e.g. UsedPhysRegs before calculateCalleeSavedRegisters.
-  TRI->processFunctionBeforeCalleeSavedScan(Fn, RS);
+  TFI->processFunctionBeforeCalleeSavedScan(Fn, RS);
 
   // Scan the function for modified callee saved registers and insert spill code
   // for any callee saved registers that are modified.
@@ -95,7 +97,7 @@ bool PEI::runOnMachineFunction(MachineFunction &Fn) {
 
   // Allow the target machine to make final modifications to the function
   // before the frame layout is finalized.
-  TRI->processFunctionBeforeFrameFinalized(Fn);
+  TFI->processFunctionBeforeFrameFinalized(Fn);
 
   // Calculate actual frame offsets for all abstract stack objects...
   calculateFrameObjectOffsets(Fn);
diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp
index 42159c9f4f..3e7c1c3075 100644
--- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp
+++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp
@@ -591,328 +591,6 @@ cannotEliminateFrame(const MachineFunction &MF) const {
     || needsStackRealignment(MF);
 }
 
-/// estimateStackSize - Estimate and return the size of the frame.
-static unsigned estimateStackSize(MachineFunction &MF) {
-  const MachineFrameInfo *FFI = MF.getFrameInfo();
-  int Offset = 0;
-  for (int i = FFI->getObjectIndexBegin(); i != 0; ++i) {
-    int FixedOff = -FFI->getObjectOffset(i);
-    if (FixedOff > Offset) Offset = FixedOff;
-  }
-  for (unsigned i = 0, e = FFI->getObjectIndexEnd(); i != e; ++i) {
-    if (FFI->isDeadObjectIndex(i))
-      continue;
-    Offset += FFI->getObjectSize(i);
-    unsigned Align = FFI->getObjectAlignment(i);
-    // Adjust to alignment boundary
-    Offset = (Offset+Align-1)/Align*Align;
-  }
-  return (unsigned)Offset;
-}
-
-/// estimateRSStackSizeLimit - Look at each instruction that references stack
-/// frames and return the stack size limit beyond which some of these
-/// instructions will require a scratch register during their expansion later.
-unsigned
-ARMBaseRegisterInfo::estimateRSStackSizeLimit(MachineFunction &MF) const {
-  const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo();
-  const ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
-  unsigned Limit = (1 << 12) - 1;
-  for (MachineFunction::iterator BB = MF.begin(),E = MF.end(); BB != E; ++BB) {
-    for (MachineBasicBlock::iterator I = BB->begin(), E = BB->end();
-         I != E; ++I) {
-      for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
-        if (!I->getOperand(i).isFI()) continue;
-
-        // When using ADDri to get the address of a stack object, 255 is the
-        // largest offset guaranteed to fit in the immediate offset.
-        if (I->getOpcode() == ARM::ADDri) {
-          Limit = std::min(Limit, (1U << 8) - 1);
-          break;
-        }
-
-        // Otherwise check the addressing mode.
-        switch (I->getDesc().TSFlags & ARMII::AddrModeMask) {
-        case ARMII::AddrMode3:
-        case ARMII::AddrModeT2_i8:
-          Limit = std::min(Limit, (1U << 8) - 1);
-          break;
-        case ARMII::AddrMode5:
-        case ARMII::AddrModeT2_i8s4:
-          Limit = std::min(Limit, ((1U << 8) - 1) * 4);
-          break;
-        case ARMII::AddrModeT2_i12:
-          // i12 supports only positive offset so these will be converted to
-          // i8 opcodes. See llvm::rewriteT2FrameIndex.
-          if (TFI->hasFP(MF) && AFI->hasStackFrame())
-            Limit = std::min(Limit, (1U << 8) - 1);
-          break;
-        case ARMII::AddrMode4:
-        case ARMII::AddrMode6:
-          // Addressing modes 4 & 6 (load/store) instructions can't encode an
-          // immediate offset for stack references.
-          return 0;
-        default:
-          break;
-        }
-        break; // At most one FI per instruction
-      }
-    }
-  }
-
-  return Limit;
-}
-
-static unsigned GetFunctionSizeInBytes(const MachineFunction &MF,
-                                       const ARMBaseInstrInfo &TII) {
-  unsigned FnSize = 0;
-  for (MachineFunction::const_iterator MBBI = MF.begin(), E = MF.end();
-       MBBI != E; ++MBBI) {
-    const MachineBasicBlock &MBB = *MBBI;
-    for (MachineBasicBlock::const_iterator I = MBB.begin(),E = MBB.end();
-         I != E; ++I)
-      FnSize += TII.GetInstSizeInBytes(I);
-  }
-  return FnSize;
-}
-
-void
-ARMBaseRegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
-                                                       RegScavenger *RS) const {
-  // This tells PEI to spill the FP as if it is any other callee-save register
-  // to take advantage the eliminateFrameIndex machinery. This also ensures it
-  // is spilled in the order specified by getCalleeSavedRegs() to make it easier
-  // to combine multiple loads / stores.
-  bool CanEliminateFrame = true;
-  bool CS1Spilled = false;
-  bool LRSpilled = false;
-  unsigned NumGPRSpills = 0;
-  SmallVector<unsigned, 4> UnspilledCS1GPRs;
-  SmallVector<unsigned, 4> UnspilledCS2GPRs;
-  const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo();
-  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
-  MachineFrameInfo *MFI = MF.getFrameInfo();
-
-  // Spill R4 if Thumb2 function requires stack realignment - it will be used as
-  // scratch register. Also spill R4 if Thumb2 function has varsized objects,
-  // since it's always posible to restore sp from fp in a single instruction.
-  // FIXME: It will be better just to find spare register here.
-  if (AFI->isThumb2Function() &&
-      (MFI->hasVarSizedObjects() || needsStackRealignment(MF)))
-    MF.getRegInfo().setPhysRegUsed(ARM::R4);
-
-  // Spill LR if Thumb1 function uses variable length argument lists.
-  if (AFI->isThumb1OnlyFunction() && AFI->getVarArgsRegSaveSize() > 0)
-    MF.getRegInfo().setPhysRegUsed(ARM::LR);
-
-  // Spill the BasePtr if it's used.
-  if (hasBasePointer(MF))
-    MF.getRegInfo().setPhysRegUsed(BasePtr);
-
-  // Don't spill FP if the frame can be eliminated. This is determined
-  // by scanning the callee-save registers to see if any is used.
-  const unsigned *CSRegs = getCalleeSavedRegs();
-  for (unsigned i = 0; CSRegs[i]; ++i) {
-    unsigned Reg = CSRegs[i];
-    bool Spilled = false;
-    if (MF.getRegInfo().isPhysRegUsed(Reg)) {
-      AFI->setCSRegisterIsSpilled(Reg);
-      Spilled = true;
-      CanEliminateFrame = false;
-    } else {
-      // Check alias registers too.
-      for (const unsigned *Aliases = getAliasSet(Reg); *Aliases; ++Aliases) {
-        if (MF.getRegInfo().isPhysRegUsed(*Aliases)) {
-          Spilled = true;
-          CanEliminateFrame = false;
-        }
-      }
-    }
-
-    if (!ARM::GPRRegisterClass->contains(Reg))
-      continue;
-
-    if (Spilled) {
-      NumGPRSpills++;
-
-      if (!STI.isTargetDarwin()) {
-        if (Reg == ARM::LR)
-          LRSpilled = true;
-        CS1Spilled = true;
-        continue;
-      }
-
-      // Keep track if LR and any of R4, R5, R6, and R7 is spilled.
-      switch (Reg) {
-      case ARM::LR:
-        LRSpilled = true;
-        // Fallthrough
-      case ARM::R4:
-      case ARM::R5:
-      case ARM::R6:
-      case ARM::R7:
-        CS1Spilled = true;
-        break;
-      default:
-        break;
-      }
-    } else {
-      if (!STI.isTargetDarwin()) {
-        UnspilledCS1GPRs.push_back(Reg);
-        continue;
-      }
-
-      switch (Reg) {
-      case ARM::R4:
-      case ARM::R5:
-      case ARM::R6:
-      case ARM::R7:
-      case ARM::LR:
-        UnspilledCS1GPRs.push_back(Reg);
-        break;
-      default:
-        UnspilledCS2GPRs.push_back(Reg);
-        break;
-      }
-    }
-  }
-
-  bool ForceLRSpill = false;
-  if (!LRSpilled && AFI->isThumb1OnlyFunction()) {
-    unsigned FnSize = GetFunctionSizeInBytes(MF, TII);
-    // Force LR to be spilled if the Thumb function size is > 2048. This enables
-    // use of BL to implement far jump. If it turns out that it's not needed
-    // then the branch fix up path will undo it.
-    if (FnSize >= (1 << 11)) {
-      CanEliminateFrame = false;
-      ForceLRSpill = true;
-    }
-  }
-
-  // If any of the stack slot references may be out of range of an immediate
-  // offset, make sure a register (or a spill slot) is available for the
-  // register scavenger. Note that if we're indexing off the frame pointer, the
-  // effective stack size is 4 bytes larger since the FP points to the stack
-  // slot of the previous FP. Also, if we have variable sized objects in the
-  // function, stack slot references will often be negative, and some of
-  // our instructions are positive-offset only, so conservatively consider
-  // that case to want a spill slot (or register) as well. Similarly, if
-  // the function adjusts the stack pointer during execution and the
-  // adjustments aren't already part of our stack size estimate, our offset
-  // calculations may be off, so be conservative.
-  // FIXME: We could add logic to be more precise about negative offsets
-  //        and which instructions will need a scratch register for them. Is it
-  //        worth the effort and added fragility?
-  bool BigStack =
-    (RS &&
-     (estimateStackSize(MF) + ((TFI->hasFP(MF) && AFI->hasStackFrame()) ? 4:0) >=
-      estimateRSStackSizeLimit(MF)))
-    || MFI->hasVarSizedObjects()
-    || (MFI->adjustsStack() && !TFI->canSimplifyCallFramePseudos(MF));
-
-  bool ExtraCSSpill = false;
-  if (BigStack || !CanEliminateFrame || cannotEliminateFrame(MF)) {
-    AFI->setHasStackFrame(true);
-
-    // If LR is not spilled, but at least one of R4, R5, R6, and R7 is spilled.
-    // Spill LR as well so we can fold BX_RET to the registers restore (LDM).
-    if (!LRSpilled && CS1Spilled) {
-      MF.getRegInfo().setPhysRegUsed(ARM::LR);
-      AFI->setCSRegisterIsSpilled(ARM::LR);
-      NumGPRSpills++;
-      UnspilledCS1GPRs.erase(std::find(UnspilledCS1GPRs.begin(),
-                                    UnspilledCS1GPRs.end(), (unsigned)ARM::LR));
-      ForceLRSpill = false;
-      ExtraCSSpill = true;
-    }
-
-    if (TFI->hasFP(MF)) {
-      MF.getRegInfo().setPhysRegUsed(FramePtr);
-      NumGPRSpills++;
-    }
-
-    // If stack and double are 8-byte aligned and we are spilling an odd number
-    // of GPRs, spill one extra callee save GPR so we won't have to pad between
-    // the integer and double callee save areas.
-    unsigned TargetAlign = MF.getTarget().getFrameInfo()->getStackAlignment();
-    if (TargetAlign == 8 && (NumGPRSpills & 1)) {
-      if (CS1Spilled && !UnspilledCS1GPRs.empty()) {
-        for (unsigned i = 0, e = UnspilledCS1GPRs.size(); i != e; ++i) {
-          unsigned Reg = UnspilledCS1GPRs[i];
-          // Don't spill high register if the function is thumb1
-          if (!AFI->isThumb1OnlyFunction() ||
-              isARMLowRegister(Reg) || Reg == ARM::LR) {
-            MF.getRegInfo().setPhysRegUsed(Reg);
-            AFI->setCSRegisterIsSpilled(Reg);
-            if (!isReservedReg(MF, Reg))
-              ExtraCSSpill = true;
-            break;
-          }
-        }
-      } else if (!UnspilledCS2GPRs.empty() &&
-                 !AFI->isThumb1OnlyFunction()) {
-        unsigned Reg = UnspilledCS2GPRs.front();
-        MF.getRegInfo().setPhysRegUsed(Reg);
-        AFI->setCSRegisterIsSpilled(Reg);
-        if (!isReservedReg(MF, Reg))
-          ExtraCSSpill = true;
-      }
-    }
-
-    // Estimate if we might need to scavenge a register at some point in order
-    // to materialize a stack offset. If so, either spill one additional
-    // callee-saved register or reserve a special spill slot to facilitate
-    // register scavenging. Thumb1 needs a spill slot for stack pointer
-    // adjustments also, even when the frame itself is small.
-    if (BigStack && !ExtraCSSpill) {
-      // If any non-reserved CS register isn't spilled, just spill one or two
-      // extra. That should take care of it!
-      unsigned NumExtras = TargetAlign / 4;
-      SmallVector<unsigned, 2> Extras;
-      while (NumExtras && !UnspilledCS1GPRs.empty()) {
-        unsigned Reg = UnspilledCS1GPRs.back();
-        UnspilledCS1GPRs.pop_back();
-        if (!isReservedReg(MF, Reg) &&
-            (!AFI->isThumb1OnlyFunction() || isARMLowRegister(Reg) ||
-             Reg == ARM::LR)) {
-          Extras.push_back(Reg);
-          NumExtras--;
-        }
-      }
-      // For non-Thumb1 functions, also check for hi-reg CS registers
-      if (!AFI->isThumb1OnlyFunction()) {
-        while (NumExtras && !UnspilledCS2GPRs.empty()) {
-          unsigned Reg = UnspilledCS2GPRs.back();
-          UnspilledCS2GPRs.pop_back();
-          if (!isReservedReg(MF, Reg)) {
-            Extras.push_back(Reg);
-            NumExtras--;
-          }
-        }
-      }
-      if (Extras.size() && NumExtras == 0) {
-        for (unsigned i = 0, e = Extras.size(); i != e; ++i) {
-          MF.getRegInfo().setPhysRegUsed(Extras[i]);
-          AFI->setCSRegisterIsSpilled(Extras[i]);
-        }
-      } else if (!AFI->isThumb1OnlyFunction()) {
-        // note: Thumb1 functions spill to R12, not the stack.  Reserve a slot
-        // closest to SP or frame pointer.
-        const TargetRegisterClass *RC = ARM::GPRRegisterClass;
-        RS->setScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
-                                                           RC->getAlignment(),
-                                                           false));
-      }
-    }
-  }
-
-  if (ForceLRSpill) {
-    MF.getRegInfo().setPhysRegUsed(ARM::LR);
-    AFI->setCSRegisterIsSpilled(ARM::LR);
-    AFI->setLRIsSpilledForFarJump(true);
-  }
-}
-
 unsigned ARMBaseRegisterInfo::getRARegister() const {
   return ARM::LR;
 }
diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.h b/lib/Target/ARM/ARMBaseRegisterInfo.h
index 718f484296..23a466fad9 100644
--- a/lib/Target/ARM/ARMBaseRegisterInfo.h
+++ b/lib/Target/ARM/ARMBaseRegisterInfo.h
@@ -154,9 +154,6 @@ public:
 
   bool cannotEliminateFrame(const MachineFunction &MF) const;
 
-  void processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
-                                            RegScavenger *RS = NULL) const;
-
   // Debug information queries.
   unsigned getRARegister() const;
   unsigned getFrameRegister(const MachineFunction &MF) const;
@@ -198,8 +195,6 @@ public:
                                    int SPAdj, RegScavenger *RS = NULL) const;
 
 private:
-  unsigned estimateRSStackSizeLimit(MachineFunction &MF) const;
-
   unsigned getRegisterPairEven(unsigned Reg, const MachineFunction &MF) const;
 
   unsigned getRegisterPairOdd(unsigned Reg, const MachineFunction &MF) const;
diff --git a/lib/Target/ARM/ARMFrameInfo.cpp b/lib/Target/ARM/ARMFrameInfo.cpp
index d379bf0e9c..e2531d04d0 100644
--- a/lib/Target/ARM/ARMFrameInfo.cpp
+++ b/lib/Target/ARM/ARMFrameInfo.cpp
@@ -18,6 +18,7 @@
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
 #include "llvm/Target/TargetOptions.h"
 
 using namespace llvm;
@@ -614,3 +615,327 @@ bool ARMFrameInfo::restoreCalleeSavedRegisters(MachineBasicBlock &MBB,
 
   return true;
 }
+
+// FIXME: Make generic?
+static unsigned GetFunctionSizeInBytes(const MachineFunction &MF,
+                                       const ARMBaseInstrInfo &TII) {
+  unsigned FnSize = 0;
+  for (MachineFunction::const_iterator MBBI = MF.begin(), E = MF.end();
+       MBBI != E; ++MBBI) {
+    const MachineBasicBlock &MBB = *MBBI;
+    for (MachineBasicBlock::const_iterator I = MBB.begin(),E = MBB.end();
+         I != E; ++I)
+      FnSize += TII.GetInstSizeInBytes(I);
+  }
+  return FnSize;
+}
+
+/// estimateStackSize - Estimate and return the size of the frame.
+/// FIXME: Make generic?
+static unsigned estimateStackSize(MachineFunction &MF) {
+  const MachineFrameInfo *FFI = MF.getFrameInfo();
+  int Offset = 0;
+  for (int i = FFI->getObjectIndexBegin(); i != 0; ++i) {
+    int FixedOff = -FFI->getObjectOffset(i);
+    if (FixedOff > Offset) Offset = FixedOff;
+  }
+  for (unsigned i = 0, e = FFI->getObjectIndexEnd(); i != e; ++i) {
+    if (FFI->isDeadObjectIndex(i))
+      continue;
+    Offset += FFI->getObjectSize(i);
+    unsigned Align = FFI->getObjectAlignment(i);
+    // Adjust to alignment boundary
+    Offset = (Offset+Align-1)/Align*Align;
+  }
+  return (unsigned)Offset;
+}
+
+/// estimateRSStackSizeLimit - Look at each instruction that references stack
+/// frames and return the stack size limit beyond which some of these
+/// instructions will require a scratch register during their expansion later.
+// FIXME: Move to TII?
+static unsigned estimateRSStackSizeLimit(MachineFunction &MF,
+                                         const TargetFrameInfo *TFI) {
+  const ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+  unsigned Limit = (1 << 12) - 1;
+  for (MachineFunction::iterator BB = MF.begin(),E = MF.end(); BB != E; ++BB) {
+    for (MachineBasicBlock::iterator I = BB->begin(), E = BB->end();
+         I != E; ++I) {
+      for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) {
+        if (!I->getOperand(i).isFI()) continue;
+
+        // When using ADDri to get the address of a stack object, 255 is the
+        // largest offset guaranteed to fit in the immediate offset.
+        if (I->getOpcode() == ARM::ADDri) {
+          Limit = std::min(Limit, (1U << 8) - 1);
+          break;
+        }
+
+        // Otherwise check the addressing mode.
+        switch (I->getDesc().TSFlags & ARMII::AddrModeMask) {
+        case ARMII::AddrMode3:
+        case ARMII::AddrModeT2_i8:
+          Limit = std::min(Limit, (1U << 8) - 1);
+          break;
+        case ARMII::AddrMode5:
+        case ARMII::AddrModeT2_i8s4:
+          Limit = std::min(Limit, ((1U << 8) - 1) * 4);
+          break;
+        case ARMII::AddrModeT2_i12:
+          // i12 supports only positive offset so these will be converted to
+          // i8 opcodes. See llvm::rewriteT2FrameIndex.
+          if (TFI->hasFP(MF) && AFI->hasStackFrame())
+            Limit = std::min(Limit, (1U << 8) - 1);
+          break;
+        case ARMII::AddrMode4:
+        case ARMII::AddrMode6:
+          // Addressing modes 4 & 6 (load/store) instructions can't encode an
+          // immediate offset for stack references.
+          return 0;
+        default:
+          break;
+        }
+        break; // At most one FI per instruction
+      }
+    }
+  }
+
+  return Limit;
+}
+
+void
+ARMFrameInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
+                                                   RegScavenger *RS) const {
+  // This tells PEI to spill the FP as if it is any other callee-save register
+  // to take advantage the eliminateFrameIndex machinery. This also ensures it
+  // is spilled in the order specified by getCalleeSavedRegs() to make it easier
+  // to combine multiple loads / stores.
+  bool CanEliminateFrame = true;
+  bool CS1Spilled = false;
+  bool LRSpilled = false;
+  unsigned NumGPRSpills = 0;
+  SmallVector<unsigned, 4> UnspilledCS1GPRs;
+  SmallVector<unsigned, 4> UnspilledCS2GPRs;
+  const ARMBaseRegisterInfo *RegInfo =
+    static_cast<const ARMBaseRegisterInfo*>(MF.getTarget().getRegisterInfo());
+  const ARMBaseInstrInfo &TII =
+    *static_cast<const ARMBaseInstrInfo*>(MF.getTarget().getInstrInfo());
+  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+  MachineFrameInfo *MFI = MF.getFrameInfo();
+  unsigned FramePtr = RegInfo->getFrameRegister(MF);
+
+  // Spill R4 if Thumb2 function requires stack realignment - it will be used as
+  // scratch register. Also spill R4 if Thumb2 function has varsized objects,
+  // since it's always posible to restore sp from fp in a single instruction.
+  // FIXME: It will be better just to find spare register here.
+  if (AFI->isThumb2Function() &&
+      (MFI->hasVarSizedObjects() || RegInfo->needsStackRealignment(MF)))
+    MF.getRegInfo().setPhysRegUsed(ARM::R4);
+
+  // Spill LR if Thumb1 function uses variable length argument lists.
+  if (AFI->isThumb1OnlyFunction() && AFI->getVarArgsRegSaveSize() > 0)
+    MF.getRegInfo().setPhysRegUsed(ARM::LR);
+
+  // Spill the BasePtr if it's used.
+  if (RegInfo->hasBasePointer(MF))
+    MF.getRegInfo().setPhysRegUsed(RegInfo->getBaseRegister());
+
+  // Don't spill FP if the frame can be eliminated. This is determined
+  // by scanning the callee-save registers to see if any is used.
+  const unsigned *CSRegs = RegInfo->getCalleeSavedRegs();
+  for (unsigned i = 0; CSRegs[i]; ++i) {
+    unsigned Reg = CSRegs[i];
+    bool Spilled = false;
+    if (MF.getRegInfo().isPhysRegUsed(Reg)) {
+      AFI->setCSRegisterIsSpilled(Reg);
+      Spilled = true;
+      CanEliminateFrame = false;
+    } else {
+      // Check alias registers too.
+      for (const unsigned *Aliases =
+             RegInfo->getAliasSet(Reg); *Aliases; ++Aliases) {
+        if (MF.getRegInfo().isPhysRegUsed(*Aliases)) {
+          Spilled = true;
+          CanEliminateFrame = false;
+        }
+      }
+    }
+
+    if (!ARM::GPRRegisterClass->contains(Reg))
+      continue;
+
+    if (Spilled) {
+      NumGPRSpills++;
+
+      if (!STI.isTargetDarwin()) {
+        if (Reg == ARM::LR)
+          LRSpilled = true;
+        CS1Spilled = true;
+        continue;
+      }
+
+      // Keep track if LR and any of R4, R5, R6, and R7 is spilled.
+      switch (Reg) {
+      case ARM::LR:
+        LRSpilled = true;
+        // Fallthrough
+      case ARM::R4: case ARM::R5:
+      case ARM::R6: case ARM::R7:
+        CS1Spilled = true;
+        break;
+      default:
+        break;
+      }
+    } else {
+      if (!STI.isTargetDarwin()) {
+        UnspilledCS1GPRs.push_back(Reg);
+        continue;
+      }
+
+      switch (Reg) {
+      case ARM::R4: case ARM::R5:
+      case ARM::R6: case ARM::R7:
+      case ARM::LR:
+        UnspilledCS1GPRs.push_back(Reg);
+        break;
+      default:
+        UnspilledCS2GPRs.push_back(Reg);
+        break;
+      }
+    }
+  }
+
+  bool ForceLRSpill = false;
+  if (!LRSpilled && AFI->isThumb1OnlyFunction()) {
+    unsigned FnSize = GetFunctionSizeInBytes(MF, TII);
+    // Force LR to be spilled if the Thumb function size is > 2048. This enables
+    // use of BL to implement far jump. If it turns out that it's not needed
+    // then the branch fix up path will undo it.
+    if (FnSize >= (1 << 11)) {
+      CanEliminateFrame = false;
+      ForceLRSpill = true;
+    }
+  }
+
+  // If any of the stack slot references may be out of range of an immediate
+  // offset, make sure a register (or a spill slot) is available for the
+  // register scavenger. Note that if we're indexing off the frame pointer, the
+  // effective stack size is 4 bytes larger since the FP points to the stack
+  // slot of the previous FP. Also, if we have variable sized objects in the
+  // function, stack slot references will often be negative, and some of
+  // our instructions are positive-offset only, so conservatively consider
+  // that case to want a spill slot (or register) as well. Similarly, if
+  // the function adjusts the stack pointer during execution and the
+  // adjustments aren't already part of our stack size estimate, our offset
+  // calculations may be off, so be conservative.
+  // FIXME: We could add logic to be more precise about negative offsets
+  //        and which instructions will need a scratch register for them. Is it
+  //        worth the effort and added fragility?
+  bool BigStack =
+    (RS &&
+     (estimateStackSize(MF) + ((hasFP(MF) && AFI->hasStackFrame()) ? 4:0) >=
+      estimateRSStackSizeLimit(MF, this)))
+    || MFI->hasVarSizedObjects()
+    || (MFI->adjustsStack() && !canSimplifyCallFramePseudos(MF));
+
+  bool ExtraCSSpill = false;
+  if (BigStack || !CanEliminateFrame || RegInfo->cannotEliminateFrame(MF)) {
+    AFI->setHasStackFrame(true);
+
+    // If LR is not spilled, but at least one of R4, R5, R6, and R7 is spilled.
+    // Spill LR as well so we can fold BX_RET to the registers restore (LDM).
+    if (!LRSpilled && CS1Spilled) {
+      MF.getRegInfo().setPhysRegUsed(ARM::LR);
+      AFI->setCSRegisterIsSpilled(ARM::LR);
+      NumGPRSpills++;
+      UnspilledCS1GPRs.erase(std::find(UnspilledCS1GPRs.begin(),
+                                    UnspilledCS1GPRs.end(), (unsigned)ARM::LR));
+      ForceLRSpill = false;
+      ExtraCSSpill = true;
+    }
+
+    if (hasFP(MF)) {
+      MF.getRegInfo().setPhysRegUsed(FramePtr);