X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTarget%2FPowerPC%2FPPCFastISel.cpp;h=188c8e84c64d6a159bc7a55a875f1c2c12352222;hp=07e3b64f7a837997226eeafe5dd7e07b6dea69c4;hb=18a86c95fc36b5f622e8dc87f71252de37a1ed44;hpb=1836fe5651722c638e32ac94c5b4fe885afa9f15

diff --git a/lib/Target/PowerPC/PPCFastISel.cpp b/lib/Target/PowerPC/PPCFastISel.cpp
index 07e3b64f7a8..188c8e84c64 100644
--- a/lib/Target/PowerPC/PPCFastISel.cpp
+++ b/lib/Target/PowerPC/PPCFastISel.cpp
@@ -13,10 +13,11 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "ppcfastisel"
 #include "PPC.h"
 #include "MCTargetDesc/PPCPredicates.h"
+#include "PPCCallingConv.h"
 #include "PPCISelLowering.h"
+#include "PPCMachineFunctionInfo.h"
 #include "PPCSubtarget.h"
 #include "PPCTargetMachine.h"
 #include "llvm/ADT/Optional.h"
@@ -28,19 +29,19 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/IR/CallingConv.h"
+#include "llvm/IR/GetElementPtrTypeIterator.h"
 #include "llvm/IR/GlobalAlias.h"
 #include "llvm/IR/GlobalVariable.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Operator.h"
 #include "llvm/Support/Debug.h"
-#include "llvm/Support/GetElementPtrTypeIterator.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/TargetMachine.h"
 
 //===----------------------------------------------------------------------===//
 //
 // TBD:
-//   FastLowerArguments: Handle simple cases.
+//   fastLowerArguments: Handle simple cases.
 //   PPCMaterializeGV: Handle TLS.
 //   SelectCall: Handle function pointers.
 //   SelectCall: Handle multi-register return values.
@@ -58,6 +59,8 @@
 //===----------------------------------------------------------------------===//
 using namespace llvm;
 
+#define DEBUG_TYPE "ppcfastisel"
+
 namespace {
 
 typedef struct Address {
@@ -80,46 +83,47 @@ typedef struct Address {
    }
 } Address;
 
-class PPCFastISel : public FastISel {
+class PPCFastISel final : public FastISel {
 
   const TargetMachine &TM;
+  const PPCSubtarget *PPCSubTarget;
+  PPCFunctionInfo *PPCFuncInfo;
   const TargetInstrInfo &TII;
   const TargetLowering &TLI;
-  const PPCSubtarget &PPCSubTarget;
   LLVMContext *Context;
 
   public:
     explicit PPCFastISel(FunctionLoweringInfo &FuncInfo,
                          const TargetLibraryInfo *LibInfo)
-    : FastISel(FuncInfo, LibInfo),
-      TM(FuncInfo.MF->getTarget()),
-      TII(*TM.getInstrInfo()),
-      TLI(*TM.getTargetLowering()),
-      PPCSubTarget(
-       *((static_cast<const PPCTargetMachine *>(&TM))->getSubtargetImpl())
-      ),
-      Context(&FuncInfo.Fn->getContext()) { }
+        : FastISel(FuncInfo, LibInfo), TM(FuncInfo.MF->getTarget()),
+          PPCSubTarget(&FuncInfo.MF->getSubtarget<PPCSubtarget>()),
+          PPCFuncInfo(FuncInfo.MF->getInfo<PPCFunctionInfo>()),
+          TII(*PPCSubTarget->getInstrInfo()),
+          TLI(*PPCSubTarget->getTargetLowering()),
+          Context(&FuncInfo.Fn->getContext()) {}
 
   // Backend specific FastISel code.
   private:
-    virtual bool TargetSelectInstruction(const Instruction *I);
-    virtual unsigned TargetMaterializeConstant(const Constant *C);
-    virtual unsigned TargetMaterializeAlloca(const AllocaInst *AI);
-    virtual bool tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo,
-                                     const LoadInst *LI);
-    virtual bool FastLowerArguments();
-    virtual unsigned FastEmit_i(MVT Ty, MVT RetTy, unsigned Opc, uint64_t Imm);
-    virtual unsigned FastEmitInst_ri(unsigned MachineInstOpcode,
-                                     const TargetRegisterClass *RC,
-                                     unsigned Op0, bool Op0IsKill,
-                                     uint64_t Imm);
-    virtual unsigned FastEmitInst_r(unsigned MachineInstOpcode,
-                                    const TargetRegisterClass *RC,
-                                    unsigned Op0, bool Op0IsKill);
-    virtual unsigned FastEmitInst_rr(unsigned MachineInstOpcode,
-                                     const TargetRegisterClass *RC,
-                                     unsigned Op0, bool Op0IsKill,
-                                     unsigned Op1, bool Op1IsKill);
+    bool fastSelectInstruction(const Instruction *I) override;
+    unsigned fastMaterializeConstant(const Constant *C) override;
+    unsigned fastMaterializeAlloca(const AllocaInst *AI) override;
+    bool tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo,
+                             const LoadInst *LI) override;
+    bool fastLowerArguments() override;
+    unsigned fastEmit_i(MVT Ty, MVT RetTy, unsigned Opc, uint64_t Imm) override;
+    unsigned fastEmitInst_ri(unsigned MachineInstOpcode,
+                             const TargetRegisterClass *RC,
+                             unsigned Op0, bool Op0IsKill,
+                             uint64_t Imm);
+    unsigned fastEmitInst_r(unsigned MachineInstOpcode,
+                            const TargetRegisterClass *RC,
+                            unsigned Op0, bool Op0IsKill);
+    unsigned fastEmitInst_rr(unsigned MachineInstOpcode,
+                             const TargetRegisterClass *RC,
+                             unsigned Op0, bool Op0IsKill,
+                             unsigned Op1, bool Op1IsKill);
+
+    bool fastLowerCall(CallLoweringInfo &CLI) override;
 
   // Instruction selection routines.
   private:
@@ -127,13 +131,11 @@ class PPCFastISel : public FastISel {
     bool SelectStore(const Instruction *I);
     bool SelectBranch(const Instruction *I);
     bool SelectIndirectBr(const Instruction *I);
-    bool SelectCmp(const Instruction *I);
     bool SelectFPExt(const Instruction *I);
     bool SelectFPTrunc(const Instruction *I);
     bool SelectIToFP(const Instruction *I, bool IsSigned);
     bool SelectFPToI(const Instruction *I, bool IsSigned);
     bool SelectBinaryIntOp(const Instruction *I, unsigned ISDOpcode);
-    bool SelectCall(const Instruction *I);
     bool SelectRet(const Instruction *I);
     bool SelectTrunc(const Instruction *I);
     bool SelectIntExt(const Instruction *I);
@@ -142,6 +144,13 @@ class PPCFastISel : public FastISel {
   private:
     bool isTypeLegal(Type *Ty, MVT &VT);
     bool isLoadTypeLegal(Type *Ty, MVT &VT);
+    bool isValueAvailable(const Value *V) const;
+    bool isVSFRCRegister(unsigned Register) const {
+      return MRI.getRegClass(Register)->getID() == PPC::VSFRCRegClassID;
+    }
+    bool isVSSRCRegister(unsigned Register) const {
+      return MRI.getRegClass(Register)->getID() == PPC::VSSRCRegClassID;
+    }
     bool PPCEmitCmp(const Value *Src1Value, const Value *Src2Value,
                     bool isZExt, unsigned DestReg);
     bool PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
@@ -155,7 +164,8 @@ class PPCFastISel : public FastISel {
                            unsigned DestReg, bool IsZExt);
     unsigned PPCMaterializeFP(const ConstantFP *CFP, MVT VT);
     unsigned PPCMaterializeGV(const GlobalValue *GV, MVT VT);
-    unsigned PPCMaterializeInt(const Constant *C, MVT VT);
+    unsigned PPCMaterializeInt(const ConstantInt *CI, MVT VT,
+                               bool UseSExt = true);
     unsigned PPCMaterialize32BitInt(int64_t Imm,
                                     const TargetRegisterClass *RC);
     unsigned PPCMaterialize64BitInt(int64_t Imm,
@@ -174,9 +184,7 @@ class PPCFastISel : public FastISel {
                          CallingConv::ID CC,
                          unsigned &NumBytes,
                          bool IsVarArg);
-    void finishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs,
-                    const Instruction *I, CallingConv::ID CC,
-                    unsigned &NumBytes, bool IsVarArg);
+    bool finishCall(MVT RetVT, CallLoweringInfo &CLI, unsigned &NumBytes);
     CCAssignFn *usePPC32CCs(unsigned Flag);
 
   private:
@@ -255,7 +263,7 @@ static Optional<PPC::Predicate> getComparePred(CmpInst::Predicate Pred) {
 // fast-isel, and return its equivalent machine type in VT.
 // FIXME: Copied directly from ARM -- factor into base class?
 bool PPCFastISel::isTypeLegal(Type *Ty, MVT &VT) {
-  EVT Evt = TLI.getValueType(Ty, true);
+  EVT Evt = TLI.getValueType(DL, Ty, true);
 
   // Only handle simple types.
   if (Evt == MVT::Other || !Evt.isSimple()) return false;
@@ -280,10 +288,18 @@ bool PPCFastISel::isLoadTypeLegal(Type *Ty, MVT &VT) {
   return false;
 }
 
+bool PPCFastISel::isValueAvailable(const Value *V) const {
+  if (!isa<Instruction>(V))
+    return true;
+
+  const auto *I = cast<Instruction>(V);
+  return FuncInfo.MBBMap[I->getParent()] == FuncInfo.MBB;
+}
+
 // Given a value Obj, create an Address object Addr that represents its
 // address.  Return false if we can't handle it.
 bool PPCFastISel::PPCComputeAddress(const Value *Obj, Address &Addr) {
-  const User *U = NULL;
+  const User *U = nullptr;
   unsigned Opcode = Instruction::UserOp1;
   if (const Instruction *I = dyn_cast<Instruction>(Obj)) {
     // Don't walk into other basic blocks unless the object is an alloca from
@@ -306,12 +322,13 @@ bool PPCFastISel::PPCComputeAddress(const Value *Obj, Address &Addr) {
       return PPCComputeAddress(U->getOperand(0), Addr);
     case Instruction::IntToPtr:
       // Look past no-op inttoptrs.
-      if (TLI.getValueType(U->getOperand(0)->getType()) == TLI.getPointerTy())
+      if (TLI.getValueType(DL, U->getOperand(0)->getType()) ==
+          TLI.getPointerTy(DL))
         return PPCComputeAddress(U->getOperand(0), Addr);
       break;
     case Instruction::PtrToInt:
       // Look past no-op ptrtoints.
-      if (TLI.getValueType(U->getType()) == TLI.getPointerTy())
+      if (TLI.getValueType(DL, U->getType()) == TLI.getPointerTy(DL))
         return PPCComputeAddress(U->getOperand(0), Addr);
       break;
     case Instruction::GetElementPtr: {
@@ -485,6 +502,19 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
   // the indexed form.  Also handle stack pointers with special needs.
   unsigned IndexReg = 0;
   PPCSimplifyAddress(Addr, VT, UseOffset, IndexReg);
+
+  // If this is a potential VSX load with an offset of 0, a VSX indexed load can
+  // be used.
+  bool IsVSSRC = (ResultReg != 0) && isVSSRCRegister(ResultReg);
+  bool IsVSFRC = (ResultReg != 0) && isVSFRCRegister(ResultReg);
+  bool Is32VSXLoad = IsVSSRC && Opc == PPC::LFS;
+  bool Is64VSXLoad = IsVSSRC && Opc == PPC::LFD;
+  if ((Is32VSXLoad || Is64VSXLoad) &&
+      (Addr.BaseType != Address::FrameIndexBase) && UseOffset &&
+      (Addr.Offset == 0)) {
+    UseOffset = false;
+  }
+
   if (ResultReg == 0)
     ResultReg = createResultReg(UseRC);
 
@@ -492,10 +522,12 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
   // in range, as otherwise PPCSimplifyAddress would have converted it
   // into a RegBase.
   if (Addr.BaseType == Address::FrameIndexBase) {
+    // VSX only provides an indexed load.
+    if (Is32VSXLoad || Is64VSXLoad) return false;
 
-    MachineMemOperand *MMO =
-      FuncInfo.MF->getMachineMemOperand(
-        MachinePointerInfo::getFixedStack(Addr.Base.FI, Addr.Offset),
+    MachineMemOperand *MMO = FuncInfo.MF->getMachineMemOperand(
+        MachinePointerInfo::getFixedStack(*FuncInfo.MF, Addr.Base.FI,
+                                          Addr.Offset),
         MachineMemOperand::MOLoad, MFI.getObjectSize(Addr.Base.FI),
         MFI.getObjectAlignment(Addr.Base.FI));
 
@@ -504,6 +536,8 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
 
   // Base reg with offset in range.
   } else if (UseOffset) {
+    // VSX only provides an indexed load.
+    if (Is32VSXLoad || Is64VSXLoad) return false;
 
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
       .addImm(Addr.Offset).addReg(Addr.Base.Reg);
@@ -526,8 +560,8 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
       case PPC::LWA:    Opc = PPC::LWAX;    break;
       case PPC::LWA_32: Opc = PPC::LWAX_32; break;
       case PPC::LD:     Opc = PPC::LDX;     break;
-      case PPC::LFS:    Opc = PPC::LFSX;    break;
-      case PPC::LFD:    Opc = PPC::LFDX;    break;
+      case PPC::LFS:    Opc = IsVSSRC ? PPC::LXSSPX : PPC::LFSX; break;
+      case PPC::LFD:    Opc = IsVSFRC ? PPC::LXSDX : PPC::LFDX; break;
     }
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
       .addReg(Addr.Base.Reg).addReg(IndexReg);
@@ -557,12 +591,12 @@ bool PPCFastISel::SelectLoad(const Instruction *I) {
   // to constrain RA from using R0/X0 when this is not legal.
   unsigned AssignedReg = FuncInfo.ValueMap[I];
   const TargetRegisterClass *RC =
-    AssignedReg ? MRI.getRegClass(AssignedReg) : 0;
+    AssignedReg ? MRI.getRegClass(AssignedReg) : nullptr;
 
   unsigned ResultReg = 0;
   if (!PPCEmitLoad(VT, ResultReg, Addr, RC))
     return false;
-  UpdateValueMap(I, ResultReg);
+  updateValueMap(I, ResultReg);
   return true;
 }
 
@@ -605,13 +639,28 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) {
   unsigned IndexReg = 0;
   PPCSimplifyAddress(Addr, VT, UseOffset, IndexReg);
 
+  // If this is a potential VSX store with an offset of 0, a VSX indexed store
+  // can be used.
+  bool IsVSSRC = isVSSRCRegister(SrcReg);
+  bool IsVSFRC = isVSFRCRegister(SrcReg);
+  bool Is32VSXStore = IsVSSRC && Opc == PPC::STFS;
+  bool Is64VSXStore = IsVSFRC && Opc == PPC::STFD;
+  if ((Is32VSXStore || Is64VSXStore) &&
+      (Addr.BaseType != Address::FrameIndexBase) && UseOffset &&
+      (Addr.Offset == 0)) {
+    UseOffset = false;
+  }
+
   // Note: If we still have a frame index here, we know the offset is
   // in range, as otherwise PPCSimplifyAddress would have converted it
   // into a RegBase.
   if (Addr.BaseType == Address::FrameIndexBase) {
-    MachineMemOperand *MMO =
-      FuncInfo.MF->getMachineMemOperand(
-        MachinePointerInfo::getFixedStack(Addr.Base.FI, Addr.Offset),
+    // VSX only provides an indexed store.
+    if (Is32VSXStore || Is64VSXStore) return false;
+
+    MachineMemOperand *MMO = FuncInfo.MF->getMachineMemOperand(
+        MachinePointerInfo::getFixedStack(*FuncInfo.MF, Addr.Base.FI,
+                                          Addr.Offset),
         MachineMemOperand::MOStore, MFI.getObjectSize(Addr.Base.FI),
         MFI.getObjectAlignment(Addr.Base.FI));
 
@@ -622,12 +671,15 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) {
         .addMemOperand(MMO);
 
   // Base reg with offset in range.
-  } else if (UseOffset)
+  } else if (UseOffset) {
+    // VSX only provides an indexed store.
+    if (Is32VSXStore || Is64VSXStore) return false;
+    
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc))
       .addReg(SrcReg).addImm(Addr.Offset).addReg(Addr.Base.Reg);
 
   // Indexed form.
-  else {
+  } else {
     // Get the RR opcode corresponding to the RI one.  FIXME: It would be
     // preferable to use the ImmToIdxMap from PPCRegisterInfo.cpp, but it
     // is hard to get at.
@@ -640,11 +692,21 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) {
       case PPC::STH8: Opc = PPC::STHX8; break;
       case PPC::STW8: Opc = PPC::STWX8; break;
       case PPC::STD:  Opc = PPC::STDX;  break;
-      case PPC::STFS: Opc = PPC::STFSX; break;
-      case PPC::STFD: Opc = PPC::STFDX; break;
+      case PPC::STFS: Opc = IsVSSRC ? PPC::STXSSPX : PPC::STFSX; break;
+      case PPC::STFD: Opc = IsVSFRC ? PPC::STXSDX : PPC::STFDX; break;
     }
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc))
-      .addReg(SrcReg).addReg(Addr.Base.Reg).addReg(IndexReg);
+
+    auto MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc))
+        .addReg(SrcReg);
+
+    // If we have an index register defined we use it in the store inst,
+    // otherwise we use X0 as base as it makes the vector instructions to
+    // use zero in the computation of the effective address regardless the
+    // content of the register.
+    if (IndexReg)
+      MIB.addReg(Addr.Base.Reg).addReg(IndexReg);
+    else
+      MIB.addReg(PPC::ZERO8).addReg(Addr.Base.Reg);
   }
 
   return true;
@@ -689,35 +751,35 @@ bool PPCFastISel::SelectBranch(const Instruction *I) {
 
   // For now, just try the simplest case where it's fed by a compare.
   if (const CmpInst *CI = dyn_cast<CmpInst>(BI->getCondition())) {
-    Optional<PPC::Predicate> OptPPCPred = getComparePred(CI->getPredicate());
-    if (!OptPPCPred)
-      return false;
+    if (isValueAvailable(CI)) {
+      Optional<PPC::Predicate> OptPPCPred = getComparePred(CI->getPredicate());
+      if (!OptPPCPred)
+        return false;
 
-    PPC::Predicate PPCPred = OptPPCPred.getValue();
+      PPC::Predicate PPCPred = OptPPCPred.getValue();
 
-    // Take advantage of fall-through opportunities.
-    if (FuncInfo.MBB->isLayoutSuccessor(TBB)) {
-      std::swap(TBB, FBB);
-      PPCPred = PPC::InvertPredicate(PPCPred);
-    }
-
-    unsigned CondReg = createResultReg(&PPC::CRRCRegClass);
+      // Take advantage of fall-through opportunities.
+      if (FuncInfo.MBB->isLayoutSuccessor(TBB)) {
+        std::swap(TBB, FBB);
+        PPCPred = PPC::InvertPredicate(PPCPred);
+      }
 
-    if (!PPCEmitCmp(CI->getOperand(0), CI->getOperand(1), CI->isUnsigned(),
-                    CondReg))
-      return false;
+      unsigned CondReg = createResultReg(&PPC::CRRCRegClass);
 
-    BuildMI(*BrBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::BCC))
-      .addImm(PPCPred).addReg(CondReg).addMBB(TBB);
-    FastEmitBranch(FBB, DbgLoc);
-    FuncInfo.MBB->addSuccessor(TBB);
-    return true;
+      if (!PPCEmitCmp(CI->getOperand(0), CI->getOperand(1), CI->isUnsigned(),
+                      CondReg))
+        return false;
 
+      BuildMI(*BrBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::BCC))
+        .addImm(PPCPred).addReg(CondReg).addMBB(TBB);
+      finishCondBranch(BI->getParent(), TBB, FBB);
+      return true;
+    }
   } else if (const ConstantInt *CI =
              dyn_cast<ConstantInt>(BI->getCondition())) {
     uint64_t Imm = CI->getZExtValue();
     MachineBasicBlock *Target = (Imm == 0) ? FBB : TBB;
-    FastEmitBranch(Target, DbgLoc);
+    fastEmitBranch(Target, DbgLoc);
     return true;
   }
 
@@ -735,11 +797,14 @@ bool PPCFastISel::SelectBranch(const Instruction *I) {
 bool PPCFastISel::PPCEmitCmp(const Value *SrcValue1, const Value *SrcValue2,
                              bool IsZExt, unsigned DestReg) {
   Type *Ty = SrcValue1->getType();
-  EVT SrcEVT = TLI.getValueType(Ty, true);
+  EVT SrcEVT = TLI.getValueType(DL, Ty, true);
   if (!SrcEVT.isSimple())
     return false;
   MVT SrcVT = SrcEVT.getSimpleVT();
 
+  if (SrcVT == MVT::i1 && PPCSubTarget->useCRBits())
+    return false;
+
   // See if operand 2 is an immediate encodeable in the compare.
   // FIXME: Operands are not in canonical order at -O0, so an immediate
   // operand in position 1 is a lost opportunity for now.  We are
@@ -826,8 +891,8 @@ bool PPCFastISel::PPCEmitCmp(const Value *SrcValue1, const Value *SrcValue2,
 // Attempt to fast-select a floating-point extend instruction.
 bool PPCFastISel::SelectFPExt(const Instruction *I) {
   Value *Src  = I->getOperand(0);
-  EVT SrcVT  = TLI.getValueType(Src->getType(), true);
-  EVT DestVT = TLI.getValueType(I->getType(), true);
+  EVT SrcVT = TLI.getValueType(DL, Src->getType(), true);
+  EVT DestVT = TLI.getValueType(DL, I->getType(), true);
 
   if (SrcVT != MVT::f32 || DestVT != MVT::f64)
     return false;
@@ -837,15 +902,15 @@ bool PPCFastISel::SelectFPExt(const Instruction *I) {
     return false;
 
   // No code is generated for a FP extend.
-  UpdateValueMap(I, SrcReg);
+  updateValueMap(I, SrcReg);
   return true;
 }
 
 // Attempt to fast-select a floating-point truncate instruction.
 bool PPCFastISel::SelectFPTrunc(const Instruction *I) {
   Value *Src  = I->getOperand(0);
-  EVT SrcVT  = TLI.getValueType(Src->getType(), true);
-  EVT DestVT = TLI.getValueType(I->getType(), true);
+  EVT SrcVT = TLI.getValueType(DL, Src->getType(), true);
+  EVT DestVT = TLI.getValueType(DL, I->getType(), true);
 
   if (SrcVT != MVT::f64 || DestVT != MVT::f32)
     return false;
@@ -859,12 +924,12 @@ bool PPCFastISel::SelectFPTrunc(const Instruction *I) {
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::FRSP), DestReg)
     .addReg(SrcReg);
 
-  UpdateValueMap(I, DestReg);
+  updateValueMap(I, DestReg);
   return true;
 }
 
 // Move an i32 or i64 value in a GPR to an f64 value in an FPR.
-// FIXME: When direct register moves are implemented (see PowerISA 2.08),
+// FIXME: When direct register moves are implemented (see PowerISA 2.07),
 // those should be used instead of moving via a stack slot when the
 // subtarget permits.
 // FIXME: The code here is sloppy for the 4-byte case.  Can use a 4-byte
@@ -895,11 +960,13 @@ unsigned PPCFastISel::PPCMoveToFPReg(MVT SrcVT, unsigned SrcReg,
   unsigned LoadOpc = PPC::LFD;
 
   if (SrcVT == MVT::i32) {
-    Addr.Offset = 4;
-    if (!IsSigned)
+    if (!IsSigned) {
       LoadOpc = PPC::LFIWZX;
-    else if (PPCSubTarget.hasLFIWAX())
+      Addr.Offset = (PPCSubTarget->isLittleEndian()) ? 0 : 4;
+    } else if (PPCSubTarget->hasLFIWAX()) {
       LoadOpc = PPC::LFIWAX;
+      Addr.Offset = (PPCSubTarget->isLittleEndian()) ? 0 : 4;
+    }
   }
 
   const TargetRegisterClass *RC = &PPC::F8RCRegClass;
@@ -911,6 +978,8 @@ unsigned PPCFastISel::PPCMoveToFPReg(MVT SrcVT, unsigned SrcReg,
 }
 
 // Attempt to fast-select an integer-to-floating-point conversion.
+// FIXME: Once fast-isel has better support for VSX, conversions using
+//        direct moves should be implemented.
 bool PPCFastISel::SelectIToFP(const Instruction *I, bool IsSigned) {
   MVT DstVT;
   Type *DstTy = I->getType();
@@ -921,7 +990,7 @@ bool PPCFastISel::SelectIToFP(const Instruction *I, bool IsSigned) {
     return false;
 
   Value *Src = I->getOperand(0);
-  EVT SrcEVT = TLI.getValueType(Src->getType(), true);
+  EVT SrcEVT = TLI.getValueType(DL, Src->getType(), true);
   if (!SrcEVT.isSimple())
     return false;
 
@@ -937,7 +1006,7 @@ bool PPCFastISel::SelectIToFP(const Instruction *I, bool IsSigned) {
 
   // We can only lower an unsigned convert if we have the newer
   // floating-point conversion operations.
-  if (!IsSigned && !PPCSubTarget.hasFPCVT())
+  if (!IsSigned && !PPCSubTarget->hasFPCVT())
     return false;
 
   // FIXME: For now we require the newer floating-point conversion operations
@@ -945,7 +1014,7 @@ bool PPCFastISel::SelectIToFP(const Instruction *I, bool IsSigned) {
   // to single-precision float.  Otherwise we have to generate a lot of
   // fiddly code to avoid double rounding.  If necessary, the fiddly code
   // can be found in PPCTargetLowering::LowerINT_TO_FP().
-  if (DstVT == MVT::f32 && !PPCSubTarget.hasFPCVT())
+  if (DstVT == MVT::f32 && !PPCSubTarget->hasFPCVT())
     return false;
 
   // Extend the input if necessary.
@@ -976,13 +1045,13 @@ bool PPCFastISel::SelectIToFP(const Instruction *I, bool IsSigned) {
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), DestReg)
     .addReg(FPReg);
 
-  UpdateValueMap(I, DestReg);
+  updateValueMap(I, DestReg);
   return true;
 }
 
 // Move the floating-point value in SrcReg into an integer destination
 // register, and return the register (or zero if we can't handle it).
-// FIXME: When direct register moves are implemented (see PowerISA 2.08),
+// FIXME: When direct register moves are implemented (see PowerISA 2.07),
 // those should be used instead of moving via a stack slot when the
 // subtarget permits.
 unsigned PPCFastISel::PPCMoveToIntReg(const Instruction *I, MVT VT,
@@ -1008,7 +1077,7 @@ unsigned PPCFastISel::PPCMoveToIntReg(const Instruction *I, MVT VT,
   // to determine the required register class.
   unsigned AssignedReg = FuncInfo.ValueMap[I];
   const TargetRegisterClass *RC =
-    AssignedReg ? MRI.getRegClass(AssignedReg) : 0;
+    AssignedReg ? MRI.getRegClass(AssignedReg) : nullptr;
 
   unsigned ResultReg = 0;
   if (!PPCEmitLoad(VT, ResultReg, Addr, RC, !IsSigned))
@@ -1018,6 +1087,8 @@ unsigned PPCFastISel::PPCMoveToIntReg(const Instruction *I, MVT VT,
 }
 
 // Attempt to fast-select a floating-point-to-integer conversion.
+// FIXME: Once fast-isel has better support for VSX, conversions using
+//        direct moves should be implemented.
 bool PPCFastISel::SelectFPToI(const Instruction *I, bool IsSigned) {
   MVT DstVT, SrcVT;
   Type *DstTy = I->getType();
@@ -1027,6 +1098,10 @@ bool PPCFastISel::SelectFPToI(const Instruction *I, bool IsSigned) {
   if (DstVT != MVT::i32 && DstVT != MVT::i64)
     return false;
 
+  // If we don't have FCTIDUZ and we need it, punt to SelectionDAG.
+  if (DstVT == MVT::i64 && !IsSigned && !PPCSubTarget->hasFPCVT())
+    return false;
+
   Value *Src = I->getOperand(0);
   Type *SrcTy = Src->getType();
   if (!isTypeLegal(SrcTy, SrcVT))
@@ -1060,7 +1135,7 @@ bool PPCFastISel::SelectFPToI(const Instruction *I, bool IsSigned) {
     if (IsSigned)
       Opc = PPC::FCTIWZ;
     else
-      Opc = PPCSubTarget.hasFPCVT() ? PPC::FCTIWUZ : PPC::FCTIDZ;
+      Opc = PPCSubTarget->hasFPCVT() ? PPC::FCTIWUZ : PPC::FCTIDZ;
   else
     Opc = IsSigned ? PPC::FCTIDZ : PPC::FCTIDUZ;
 
@@ -1073,14 +1148,14 @@ bool PPCFastISel::SelectFPToI(const Instruction *I, bool IsSigned) {
   if (IntReg == 0)
     return false;
 
-  UpdateValueMap(I, IntReg);
+  updateValueMap(I, IntReg);
   return true;
 }
 
 // Attempt to fast-select a binary integer operation that isn't already
 // handled automatically.
 bool PPCFastISel::SelectBinaryIntOp(const Instruction *I, unsigned ISDOpcode) {
-  EVT DestVT  = TLI.getValueType(I->getType(), true);
+  EVT DestVT = TLI.getValueType(DL, I->getType(), true);
 
   // We can get here in the case when we have a binary operation on a non-legal
   // type and the target independent selector doesn't know how to handle it.
@@ -1162,7 +1237,7 @@ bool PPCFastISel::SelectBinaryIntOp(const Instruction *I, unsigned ISDOpcode) {
                 ResultReg)
             .addReg(SrcReg1)
             .addImm(Imm);
-        UpdateValueMap(I, ResultReg);
+        updateValueMap(I, ResultReg);
         return true;
       }
     }
@@ -1178,7 +1253,7 @@ bool PPCFastISel::SelectBinaryIntOp(const Instruction *I, unsigned ISDOpcode) {
 
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
     .addReg(SrcReg1).addReg(SrcReg2);
-  UpdateValueMap(I, ResultReg);
+  updateValueMap(I, ResultReg);
   return true;
 }
 
@@ -1193,7 +1268,12 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args,
                                   unsigned &NumBytes,
                                   bool IsVarArg) {
   SmallVector<CCValAssign, 16> ArgLocs;
-  CCState CCInfo(CC, IsVarArg, *FuncInfo.MF, TM, ArgLocs, *Context);
+  CCState CCInfo(CC, IsVarArg, *FuncInfo.MF, ArgLocs, *Context);
+
+  // Reserve space for the linkage area on the stack.
+  unsigned LinkageSize = PPCSubTarget->getFrameLowering()->getLinkageSize();
+  CCInfo.AllocateStack(LinkageSize, 8);
+
   CCInfo.AnalyzeCallOperands(ArgVTs, ArgFlags, CC_PPC64_ELF_FIS);
 
   // Bail out if we can't handle any of the arguments.
@@ -1203,7 +1283,7 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args,
 
     // Skip vector arguments for now, as well as long double and
     // uint128_t, and anything that isn't passed in a register.
-    if (ArgVT.isVector() || ArgVT.getSizeInBits() > 64 ||
+    if (ArgVT.isVector() || ArgVT.getSizeInBits() > 64 || ArgVT == MVT::i1 ||
         !VA.isRegLoc() || VA.needsCustom())
       return false;
 
@@ -1215,6 +1295,14 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args,
   // Get a count of how many bytes are to be pushed onto the stack.
   NumBytes = CCInfo.getNextStackOffset();
 
+  // The prolog code of the callee may store up to 8 GPR argument registers to
+  // the stack, allowing va_start to index over them in memory if its varargs.
+  // Because we cannot tell if this is needed on the caller side, we have to
+  // conservatively assume that it is needed.  As such, make sure we have at
+  // least enough stack space for the caller to store the 8 GPRs.
+  // FIXME: On ELFv2, it may be unnecessary to allocate the parameter area.
+  NumBytes = std::max(NumBytes, LinkageSize + 64);
+
   // Issue CALLSEQ_START.
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
           TII.get(TII.getCallFrameSetupOpcode()))
@@ -1222,7 +1310,7 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args,
 
   // Prepare to assign register arguments.  Every argument uses up a
   // GPR protocol register even if it's passed in a floating-point
-  // register.
+  // register (unless we're using the fast calling convention).
   unsigned NextGPR = PPC::X3;
   unsigned NextFPR = PPC::F1;
 
@@ -1272,7 +1360,8 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args,
     unsigned ArgReg;
     if (ArgVT == MVT::f32 || ArgVT == MVT::f64) {
       ArgReg = NextFPR++;
-      ++NextGPR;
+      if (CC != CallingConv::Fast)
+        ++NextGPR;
     } else
       ArgReg = NextGPR++;
 
@@ -1286,9 +1375,9 @@ bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args,
 
 // For a call that we've determined we can fast-select, finish the
 // call sequence and generate a copy to obtain the return value (if any).
-void PPCFastISel::finishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs,
-                             const Instruction *I, CallingConv::ID CC,
-                             unsigned &NumBytes, bool IsVarArg) {
+bool PPCFastISel::finishCall(MVT RetVT, CallLoweringInfo &CLI, unsigned &NumBytes) {
+  CallingConv::ID CC = CLI.CallConv;
+
   // Issue CallSEQ_END.
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
           TII.get(TII.getCallFrameDestroyOpcode()))
@@ -1299,7 +1388,7 @@ void PPCFastISel::finishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs,
   // any real difficulties there.
   if (RetVT != MVT::isVoid) {
     SmallVector<CCValAssign, 16> RVLocs;
-    CCState CCInfo(CC, IsVarArg, *FuncInfo.MF, TM, RVLocs, *Context);
+    CCState CCInfo(CC, false, *FuncInfo.MF, RVLocs, *Context);
     CCInfo.AnalyzeCallResult(RetVT, RetCC_PPC64_ELF_FIS);
     CCValAssign &VA = RVLocs[0];
     assert(RVLocs.size() == 1 && "No support for multi-reg return values!");
@@ -1344,52 +1433,51 @@ void PPCFastISel::finishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs,
     }
 
     assert(ResultReg && "ResultReg unset!");
-    UsedRegs.push_back(SourcePhysReg);
-    UpdateValueMap(I, ResultReg);
+    CLI.InRegs.push_back(SourcePhysReg);
+    CLI.ResultReg = ResultReg;
+    CLI.NumResultRegs = 1;
   }
+
+  return true;
 }
 
-// Attempt to fast-select a call instruction.
-bool PPCFastISel::SelectCall(const Instruction *I) {
-  const CallInst *CI = cast<CallInst>(I);
-  const Value *Callee = CI->getCalledValue();
+bool PPCFastISel::fastLowerCall(CallLoweringInfo &CLI) {
+  CallingConv::ID CC  = CLI.CallConv;
+  bool IsTailCall     = CLI.IsTailCall;
+  bool IsVarArg       = CLI.IsVarArg;
+  const Value *Callee = CLI.Callee;
+  const MCSymbol *Symbol = CLI.Symbol;
 
-  // Can't handle inline asm.
-  if (isa<InlineAsm>(Callee))
+  if (!Callee && !Symbol)
     return false;
 
   // Allow SelectionDAG isel to handle tail calls.
-  if (CI->isTailCall())
+  if (IsTailCall)
     return false;
 
-  // Obtain calling convention.
-  ImmutableCallSite CS(CI);
-  CallingConv::ID CC = CS.getCallingConv();
-
-  PointerType *PT = cast<PointerType>(CS.getCalledValue()->getType());
-  FunctionType *FTy = cast<FunctionType>(PT->getElementType());
-  bool IsVarArg = FTy->isVarArg();
-
-  // Not ready for varargs yet.
+  // Let SDISel handle vararg functions.
   if (IsVarArg)
     return false;
 
   // Handle simple calls for now, with legal return types and
   // those that can be extended.
-  Type *RetTy = I->getType();
+  Type *RetTy = CLI.RetTy;
   MVT RetVT;
   if (RetTy->isVoidTy())
     RetVT = MVT::isVoid;
   else if (!isTypeLegal(RetTy, RetVT) && RetVT != MVT::i16 &&
            RetVT != MVT::i8)
     return false;
+  else if (RetVT == MVT::i1 && PPCSubTarget->useCRBits())
+    // We can't handle boolean returns when CR bits are in use.
+    return false;
 
   // FIXME: No multi-register return values yet.
   if (RetVT != MVT::isVoid && RetVT != MVT::i8 && RetVT != MVT::i16 &&
       RetVT != MVT::i32 && RetVT != MVT::i64 && RetVT != MVT::f32 &&
       RetVT != MVT::f64) {
     SmallVector<CCValAssign, 16> RVLocs;
-    CCState CCInfo(CC, IsVarArg, *FuncInfo.MF, TM, RVLocs, *Context);
+    CCState CCInfo(CC, IsVarArg, *FuncInfo.MF, RVLocs, *Context);
     CCInfo.AnalyzeCallResult(RetVT, RetCC_PPC64_ELF_FIS);
     if (RVLocs.size() > 1)
       return false;
@@ -1397,7 +1485,7 @@ bool PPCFastISel::SelectCall(const Instruction *I) {
 
   // Bail early if more than 8 arguments, as we only currently
   // handle arguments passed in registers.
-  unsigned NumArgs = CS.arg_size();
+  unsigned NumArgs = CLI.OutVals.size();
   if (NumArgs > 8)
     return false;
 
@@ -1412,28 +1500,16 @@ bool PPCFastISel::SelectCall(const Instruction *I) {
   ArgVTs.reserve(NumArgs);
   ArgFlags.reserve(NumArgs);
 
-  for (ImmutableCallSite::arg_iterator II = CS.arg_begin(), IE = CS.arg_end();
-       II != IE; ++II) {
-    // FIXME: ARM does something for intrinsic calls here, check into that.
-
-    unsigned AttrIdx = II - CS.arg_begin() + 1;
-    
+  for (unsigned i = 0, ie = NumArgs; i != ie; ++i) {
     // Only handle easy calls for now.  It would be reasonably easy
     // to handle <= 8-byte structures passed ByVal in registers, but we
     // have to ensure they are right-justified in the register.
-    if (CS.paramHasAttr(AttrIdx, Attribute::InReg) ||
-        CS.paramHasAttr(AttrIdx, Attribute::StructRet) ||
-        CS.paramHasAttr(AttrIdx, Attribute::Nest) ||
-        CS.paramHasAttr(AttrIdx, Attribute::ByVal))
+    ISD::ArgFlagsTy Flags = CLI.OutFlags[i];
+    if (Flags.isInReg() || Flags.isSRet() || Flags.isNest() || Flags.isByVal())
       return false;
 
-    ISD::ArgFlagsTy Flags;
-    if (CS.paramHasAttr(AttrIdx, Attribute::SExt))
-      Flags.setSExt();
-    if (CS.paramHasAttr(AttrIdx, Attribute::ZExt))
-      Flags.setZExt();
-
-    Type *ArgTy = (*II)->getType();
+    Value *ArgValue = CLI.OutVals[i];
+    Type *ArgTy = ArgValue->getType();
     MVT ArgVT;
     if (!isTypeLegal(ArgTy, ArgVT) && ArgVT != MVT::i16 && ArgVT != MVT::i8)
       return false;
@@ -1441,14 +1517,11 @@ bool PPCFastISel::SelectCall(const Instruction *I) {
     if (ArgVT.isVector())
       return false;
 
-    unsigned Arg = getRegForValue(*II);
+    unsigned Arg = getRegForValue(ArgValue);
     if (Arg == 0)
       return false;
 
-    unsigned OriginalAlignment = DL.getABITypeAlignment(ArgTy);
-    Flags.setOrigAlign(OriginalAlignment);
-
-    Args.push_back(*II);
+    Args.push_back(ArgValue);
     ArgRegs.push_back(Arg);
     ArgVTs.push_back(ArgVT);
     ArgFlags.push_back(Flags);
@@ -1462,35 +1535,46 @@ bool PPCFastISel::SelectCall(const Instruction *I) {
                        RegArgs, CC, NumBytes, IsVarArg))
     return false;
 
+  MachineInstrBuilder MIB;
   // FIXME: No handling for function pointers yet.  This requires
   // implementing the function descriptor (OPD) setup.
   const GlobalValue *GV = dyn_cast<GlobalValue>(Callee);
-  if (!GV)
-    return false;
-
-  // Build direct call with NOP for TOC restore.
-  // FIXME: We can and should optimize away the NOP for local calls.
-  MachineInstrBuilder MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
-                                    TII.get(PPC::BL8_NOP));
-  // Add callee.
-  MIB.addGlobalAddress(GV);
+  if (!GV) {
+    // patchpoints are a special case; they always dispatch to a pointer value.
+    // However, we don't actually want to generate the indirect call sequence
+    // here (that will be generated, as necessary, during asm printing), and
+    // the call we generate here will be erased by FastISel::selectPatchpoint,
+    // so don't try very hard...
+    if (CLI.IsPatchPoint)
+      MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::NOP));
+    else
+      return false;
+  } else {
+    // Build direct call with NOP for TOC restore.
+    // FIXME: We can and should optimize away the NOP for local calls.
+    MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+                  TII.get(PPC::BL8_NOP));
+    // Add callee.
+    MIB.addGlobalAddress(GV);
+  }
 
   // Add implicit physical register uses to the call.
   for (unsigned II = 0, IE = RegArgs.size(); II != IE; ++II)
     MIB.addReg(RegArgs[II], RegState::Implicit);
 
+  // Direct calls, in both the ELF V1 and V2 ABIs, need the TOC register live
+  // into the call.
+  PPCFuncInfo->setUsesTOCBasePtr();
+  MIB.addReg(PPC::X2, RegState::Implicit);
+
   // Add a register mask with the call-preserved registers.  Proper
   // defs for return values will be added by setPhysRegsDeadExcept().
-  MIB.addRegMask(TRI.getCallPreservedMask(CC));
+  MIB.addRegMask(TRI.getCallPreservedMask(*FuncInfo.MF, CC));
 
-  // Finish off the call including any return values.
-  SmallVector<unsigned, 4> UsedRegs;
-  finishCall(RetVT, UsedRegs, I, CC, NumBytes, IsVarArg);
-
-  // Set all unused physregs defs as dead.
-  static_cast<MachineInstr *>(MIB)->setPhysRegsDeadExcept(UsedRegs, TRI);
+  CLI.Call = MIB;
 
-  return true;
+  // Finish off the call including any return values.
+  return finishCall(RetVT, CLI, NumBytes);
 }
 
 // Attempt to fast-select a return instruction.
@@ -1508,11 +1592,11 @@ bool PPCFastISel::SelectRet(const Instruction *I) {
 
   if (Ret->getNumOperands() > 0) {
     SmallVector<ISD::OutputArg, 4> Outs;
-    GetReturnInfo(F.getReturnType(), F.getAttributes(), Outs, TLI);
+    GetReturnInfo(F.getReturnType(), F.getAttributes(), Outs, TLI, DL);
 
     // Analyze operands of the call, assigning locations to each operand.
     SmallVector<CCValAssign, 16> ValLocs;
-    CCState CCInfo(CC, F.isVarArg(), *FuncInfo.MF, TM, ValLocs, *Context);
+    CCState CCInfo(CC, F.isVarArg(), *FuncInfo.MF, ValLocs, *Context);
     CCInfo.AnalyzeReturn(Outs, RetCC_PPC64_ELF_FIS);
     const Value *RV = Ret->getOperand(0);
     
@@ -1520,15 +1604,22 @@ bool PPCFastISel::SelectRet(const Instruction *I) {
     if (ValLocs.size() > 1)
       return false;
 
-    // Special case for returning a constant integer of any size.
-    // Materialize the constant as an i64 and copy it to the return
-    // register.  This avoids an unnecessary extend or truncate.
-    if (isa<ConstantInt>(*RV)) {
-      const Constant *C = cast<Constant>(RV);
-      unsigned SrcReg = PPCMaterializeInt(C, MVT::i64);
-      unsigned RetReg = ValLocs[0].getLocReg();
+    // Special case for returning a constant integer of any size - materialize
+    // the constant as an i64 and copy it to the return register.
+    if (const ConstantInt *CI = dyn_cast<ConstantInt>(RV)) {
+      CCValAssign &VA = ValLocs[0];
+
+      unsigned RetReg = VA.getLocReg();
+      // We still need to worry about properly extending the sign. For example,
+      // we could have only a single bit or a constant that needs zero
+      // extension rather than sign extension. Make sure we pass the return
+      // value extension property to integer materialization.
+      unsigned SrcReg =
+          PPCMaterializeInt(CI, MVT::i64, VA.getLocInfo() == CCValAssign::SExt);
+
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
-              TII.get(TargetOpcode::COPY), RetReg).addReg(SrcReg);
+            TII.get(TargetOpcode::COPY), RetReg).addReg(SrcReg);
+
       RetRegs.push_back(RetReg);
 
     } else {
@@ -1545,7 +1636,7 @@ bool PPCFastISel::SelectRet(const Instruction *I) {
         RetRegs.push_back(VA.getLocReg());
         unsigned SrcReg = Reg + VA.getValNo();
 
-        EVT RVEVT = TLI.getValueType(RV->getType());
+        EVT RVEVT = TLI.getValueType(DL, RV->getType());
         if (!RVEVT.isSimple())
           return false;
         MVT RVVT = RVEVT.getSimpleVT();
@@ -1591,7 +1682,7 @@ bool PPCFastISel::SelectRet(const Instruction *I) {
   }
 
   MachineInstrBuilder MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
-                                    TII.get(PPC::BLR));
+                                    TII.get(PPC::BLR8));
 
   for (unsigned i = 0, e = RetRegs.size(); i != e; ++i)
     MIB.addReg(RetRegs[i], RegState::Implicit);
@@ -1664,8 +1755,8 @@ bool PPCFastISel::SelectIndirectBr(const Instruction *I) {
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::BCTR8));
 
   const IndirectBrInst *IB = cast<IndirectBrInst>(I);
-  for (unsigned i = 0, e = IB->getNumSuccessors(); i != e; ++i)
-    FuncInfo.MBB->addSuccessor(FuncInfo.MBBMap[IB->getSuccessor(i)]);
+  for (const BasicBlock *SuccBB : IB->successors())
+    FuncInfo.MBB->addSuccessor(FuncInfo.MBBMap[SuccBB]);
 
   return true;
 }
@@ -1673,8 +1764,8 @@ bool PPCFastISel::SelectIndirectBr(const Instruction *I) {
 // Attempt to fast-select an integer truncate instruction.
 bool PPCFastISel::SelectTrunc(const Instruction *I) {
   Value *Src  = I->getOperand(0);
-  EVT SrcVT  = TLI.getValueType(Src->getType(), true);
-  EVT DestVT = TLI.getValueType(I->getType(), true);
+  EVT SrcVT = TLI.getValueType(DL, Src->getType(), true);
+  EVT DestVT = TLI.getValueType(DL, I->getType(), true);
 
   if (SrcVT != MVT::i64 && SrcVT != MVT::i32 && SrcVT != MVT::i16)
     return false;
@@ -1695,7 +1786,7 @@ bool PPCFastISel::SelectTrunc(const Instruction *I) {
     SrcReg = ResultReg;
   }
 
-  UpdateValueMap(I, SrcReg);
+  updateValueMap(I, SrcReg);
   return true;
 }
 
@@ -1710,8 +1801,8 @@ bool PPCFastISel::SelectIntExt(const Instruction *I) {
   if (!SrcReg) return false;
 
   EVT SrcEVT, DestEVT;
-  SrcEVT = TLI.getValueType(SrcTy, true);
-  DestEVT = TLI.getValueType(DestTy, true);
+  SrcEVT = TLI.getValueType(DL, SrcTy, true);
+  DestEVT = TLI.getValueType(DL, DestTy, true);
   if (!SrcEVT.isSimple())
     return false;
   if (!DestEVT.isSimple())
@@ -1734,13 +1825,13 @@ bool PPCFastISel::SelectIntExt(const Instruction *I) {
   if (!PPCEmitIntExt(SrcVT, SrcReg, DestVT, ResultReg, IsZExt))
     return false;
 
-  UpdateValueMap(I, ResultReg);
+  updateValueMap(I, ResultReg);
   return true;
 }
 
 // Attempt to fast-select an instruction that wasn't handled by
 // the table-generated machinery.
-bool PPCFastISel::TargetSelectInstruction(const Instruction *I) {
+bool PPCFastISel::fastSelectInstruction(const Instruction *I) {
 
   switch (I->getOpcode()) {
     case Instruction::Load:
@@ -1770,9 +1861,7 @@ bool PPCFastISel::TargetSelectInstruction(const Instruction *I) {
     case Instruction::Sub:
       return SelectBinaryIntOp(I, ISD::SUB);
     case Instruction::Call:
-      if (dyn_cast<IntrinsicInst>(I))
-        return false;
-      return SelectCall(I);
+      return selectCall(I);
     case Instruction::Ret:
       return SelectRet(I);
     case Instruction::Trunc:
@@ -1803,14 +1892,14 @@ unsigned PPCFastISel::PPCMaterializeFP(const ConstantFP *CFP, MVT VT) {
   unsigned DestReg = createResultReg(TLI.getRegClassFor(VT));
   CodeModel::Model CModel = TM.getCodeModel();
 
-  MachineMemOperand *MMO =
-    FuncInfo.MF->getMachineMemOperand(
-      MachinePointerInfo::getConstantPool(), MachineMemOperand::MOLoad,
-      (VT == MVT::f32) ? 4 : 8, Align);
+  MachineMemOperand *MMO = FuncInfo.MF->getMachineMemOperand(
+      MachinePointerInfo::getConstantPool(*FuncInfo.MF),
+      MachineMemOperand::MOLoad, (VT == MVT::f32) ? 4 : 8, Align);
 
   unsigned Opc = (VT == MVT::f32) ? PPC::LFS : PPC::LFD;
   unsigned TmpReg = createResultReg(&PPC::G8RC_and_G8RC_NOX0RegClass);
 
+  PPCFuncInfo->setUsesTOCBasePtr();
   // For small code model, generate a LF[SD](0, LDtocCPT(Idx, X2)).
   if (CModel == CodeModel::Small || CModel == CodeModel::JITDefault) {
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtocCPT),
@@ -1855,18 +1944,12 @@ unsigned PPCFastISel::PPCMaterializeGV(const GlobalValue *GV, MVT VT) {
   // FIXME: Jump tables are not yet required because fast-isel doesn't
   // handle switches; if that changes, we need them as well.  For now,
   // what follows assumes everything's a generic (or TLS) global address.
-  const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV);
-  if (!GVar) {
-    // If GV is an alias, use the aliasee for determining thread-locality.
-    if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
-      GVar = dyn_cast_or_null<GlobalVariable>(GA->resolveAliasedGlobal(false));
-  }
 
   // FIXME: We don't yet handle the complexity of TLS.
-  bool IsTLS = GVar && GVar->isThreadLocal();
-  if (IsTLS)
+  if (GV->isThreadLocal())
     return 0;
 
+  PPCFuncInfo->setUsesTOCBasePtr();
   // For small code model, generate a simple TOC load.
   if (CModel == CodeModel::Small || CModel == CodeModel::JITDefault)
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtoc),
@@ -1874,8 +1957,8 @@ unsigned PPCFastISel::PPCMaterializeGV(const GlobalValue *GV, MVT VT) {
         .addGlobalAddress(GV)
         .addReg(PPC::X2);
   else {
-    // If the address is an externally defined symbol, a symbol with
-    // common or externally available linkage, a function address, or a
+    // If the address is an externally defined symbol, a symbol with common
+    // or externally available linkage, a non-local function address, or a
     // jump table address (not yet needed), or if we are generating code
     // for large code model, we generate:
     //       LDtocL(GV, ADDIStocHA(%X2, GV))
@@ -1886,18 +1969,15 @@ unsigned PPCFastISel::PPCMaterializeGV(const GlobalValue *GV, MVT VT) {
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ADDIStocHA),
             HighPartReg).addReg(PPC::X2).addGlobalAddress(GV);
 
-    // !GVar implies a function address.  An external variable is one
-    // without an initializer.
-    // If/when switches are implemented, jump tables should be handled
-    // on the "if" path here.
-    if (CModel == CodeModel::Large || !GVar || !GVar->hasInitializer() ||
-        GVar->hasCommonLinkage() || GVar->hasAvailableExternallyLinkage())
+    unsigned char GVFlags = PPCSubTarget->classifyGlobalReference(GV);
+    if (GVFlags & PPCII::MO_NLP_FLAG) {
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtocL),
               DestReg).addGlobalAddress(GV).addReg(HighPartReg);
-    else
+    } else {
       // Otherwise generate the ADDItocL.
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ADDItocL),
               DestReg).addReg(HighPartReg).addGlobalAddress(GV);
+    }
   }
 
   return DestReg;
@@ -1994,7 +2074,16 @@ unsigned PPCFastISel::PPCMaterialize64BitInt(int64_t Imm,
 
 // Materialize an integer constant into a register, and return
 // the register number (or zero if we failed to handle it).
-unsigned PPCFastISel::PPCMaterializeInt(const Constant *C, MVT VT) {
+unsigned PPCFastISel::PPCMaterializeInt(const ConstantInt *CI, MVT VT,
+                                        bool UseSExt) {
+  // If we're using CR bit registers for i1 values, handle that as a special
+  // case first.
+  if (VT == MVT::i1 && PPCSubTarget->useCRBits()) {
+    unsigned ImmReg = createResultReg(&PPC::CRBITRCRegClass);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(CI->isZero() ? PPC::CRUNSET : PPC::CRSET), ImmReg);
+    return ImmReg;
+  }
 
   if (VT != MVT::i64 && VT != MVT::i32 && VT != MVT::i16 &&
       VT != MVT::i8 && VT != MVT::i1) 
@@ -2004,17 +2093,22 @@ unsigned PPCFastISel::PPCMaterializeInt(const Constant *C, MVT VT) {
                                    &PPC::GPRCRegClass);
 
   // If the constant is in range, use a load-immediate.
-  const ConstantInt *CI = cast<ConstantInt>(C);
-  if (isInt<16>(CI->getSExtValue())) {
+  if (UseSExt && isInt<16>(CI->getSExtValue())) {
     unsigned Opc = (VT == MVT::i64) ? PPC::LI8 : PPC::LI;
     unsigned ImmReg = createResultReg(RC);
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ImmReg)
-      .addImm(CI->getSExtValue());
+        .addImm(CI->getSExtValue());
+    return ImmReg;
+  } else if (!UseSExt && isUInt<16>(CI->getZExtValue())) {
+    unsigned Opc = (VT == MVT::i64) ? PPC::LI8 : PPC::LI;
+    unsigned ImmReg = createResultReg(RC);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ImmReg)
+        .addImm(CI->getZExtValue());
     return ImmReg;
   }
 
   // Construct the constant piecewise.
-  int64_t Imm = CI->getZExtValue();
+  int64_t Imm = UseSExt ? CI->getSExtValue() : CI->getZExtValue();
 
   if (VT == MVT::i64)
     return PPCMaterialize64BitInt(Imm, RC);
@@ -2026,8 +2120,8 @@ unsigned PPCFastISel::PPCMaterializeInt(const Constant *C, MVT VT) {
 
 // Materialize a constant into a register, and return the register
 // number (or zero if we failed to handle it).
-unsigned PPCFastISel::TargetMaterializeConstant(const Constant *C) {
-  EVT CEVT = TLI.getValueType(C->getType(), true);
+unsigned PPCFastISel::fastMaterializeConstant(const Constant *C) {
+  EVT CEVT = TLI.getValueType(DL, C->getType(), true);
 
   // Only handle simple types.
   if (!CEVT.isSimple()) return 0;
@@ -2037,15 +2131,15 @@ unsigned PPCFastISel::TargetMaterializeConstant(const Constant *C) {
     return PPCMaterializeFP(CFP, VT);
   else if (const GlobalValue *GV = dyn_cast<GlobalValue>(C))
     return PPCMaterializeGV(GV, VT);
-  else if (isa<ConstantInt>(C))
-    return PPCMaterializeInt(C, VT);
+  else if (const ConstantInt *CI = dyn_cast<ConstantInt>(C))
+    return PPCMaterializeInt(CI, VT, VT != MVT::i1);
 
   return 0;
 }
 
 // Materialize the address created by an alloca into a register, and
 // return the register number (or zero if we failed to handle it).
-unsigned PPCFastISel::TargetMaterializeAlloca(const AllocaInst *AI) {
+unsigned PPCFastISel::fastMaterializeAlloca(const AllocaInst *AI) {
   // Don't handle dynamic allocas.
   if (!FuncInfo.StaticAllocaMap.count(AI)) return 0;
 
@@ -2136,7 +2230,7 @@ bool PPCFastISel::tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo,
 
   unsigned ResultReg = MI->getOperand(0).getReg();
 
-  if (!PPCEmitLoad(VT, ResultReg, Addr, 0, IsZExt))
+  if (!PPCEmitLoad(VT, ResultReg, Addr, nullptr, IsZExt))
     return false;
 
   MI->eraseFromParent();
@@ -2145,7 +2239,7 @@ bool PPCFastISel::tryToFoldLoadIntoMI(MachineInstr *MI, unsigned OpNo,
 
 // Attempt to lower call arguments in a faster way than done by
 // the selection DAG code.
-bool PPCFastISel::FastLowerArguments() {
+bool PPCFastISel::fastLowerArguments() {
   // Defer to normal argument lowering for now.  It's reasonably
   // efficient.  Consider doing something like ARM to handle the
   // case where all args fit in registers, no varargs, no float
@@ -2155,11 +2249,20 @@ bool PPCFastISel::FastLowerArguments() {
 
 // Handle materializing integer constants into a register.  This is not
 // automatically generated for PowerPC, so must be explicitly created here.
-unsigned PPCFastISel::FastEmit_i(MVT Ty, MVT VT, unsigned Opc, uint64_t Imm) {
+unsigned PPCFastISel::fastEmit_i(MVT Ty, MVT VT, unsigned Opc, uint64_t Imm) {
   
   if (Opc != ISD::Constant)
     return 0;
 
+  // If we're using CR bit registers for i1 values, handle that as a special
+  // case first.
+  if (VT == MVT::i1 && PPCSubTarget->useCRBits()) {
+    unsigned ImmReg = createResultReg(&PPC::CRBITRCRegClass);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(Imm == 0 ? PPC::CRUNSET : PPC::CRSET), ImmReg);
+    return ImmReg;
+  }
+
   if (VT != MVT::i64 && VT != MVT::i32 && VT != MVT::i16 &&
       VT != MVT::i8 && VT != MVT::i1) 
     return 0;
@@ -2183,7 +2286,7 @@ unsigned PPCFastISel::FastEmit_i(MVT Ty, MVT VT, unsigned Opc, uint64_t Imm) {
 // assigning R0 or X0 to the output register for GPRC and G8RC
 // register classes, as any such result could be used in ADDI, etc.,
 // where those regs have another meaning.
-unsigned PPCFastISel::FastEmitInst_ri(unsigned MachineInstOpcode,
+unsigned PPCFastISel::fastEmitInst_ri(unsigned MachineInstOpcode,
                                       const TargetRegisterClass *RC,
                                       unsigned Op0, bool Op0IsKill,
                                       uint64_t Imm) {
@@ -2196,27 +2299,27 @@ unsigned PPCFastISel::FastEmitInst_ri(unsigned MachineInstOpcode,
     (RC == &PPC::GPRCRegClass ? &PPC::GPRC_and_GPRC_NOR0RegClass :
      (RC == &PPC::G8RCRegClass ? &PPC::G8RC_and_G8RC_NOX0RegClass : RC));
 
-  return FastISel::FastEmitInst_ri(MachineInstOpcode, UseRC,
+  return FastISel::fastEmitInst_ri(MachineInstOpcode, UseRC,
                                    Op0, Op0IsKill, Imm);
 }
 
 // Override for instructions with one register operand to avoid use of
 // R0/X0.  The automatic infrastructure isn't aware of the context so
 // we must be conservative.
-unsigned PPCFastISel::FastEmitInst_r(unsigned MachineInstOpcode,
+unsigned PPCFastISel::fastEmitInst_r(unsigned MachineInstOpcode,
                                      const TargetRegisterClass* RC,
                                      unsigned Op0, bool Op0IsKill) {
   const TargetRegisterClass *UseRC =
     (RC == &PPC::GPRCRegClass ? &PPC::GPRC_and_GPRC_NOR0RegClass :
      (RC == &PPC::G8RCRegClass ? &PPC::G8RC_and_G8RC_NOX0RegClass : RC));
 
-  return FastISel::FastEmitInst_r(MachineInstOpcode, UseRC, Op0, Op0IsKill);
+  return FastISel::fastEmitInst_r(MachineInstOpcode, UseRC, Op0, Op0IsKill);
 }
 
 // Override for instructions with two register operands to avoid use
 // of R0/X0.  The automatic infrastructure isn't aware of the context
 // so we must be conservative.
-unsigned PPCFastISel::FastEmitInst_rr(unsigned MachineInstOpcode,
+unsigned PPCFastISel::fastEmitInst_rr(unsigned MachineInstOpcode,
                                       const TargetRegisterClass* RC,
                                       unsigned Op0, bool Op0IsKill,
                                       unsigned Op1, bool Op1IsKill) {
@@ -2224,7 +2327,7 @@ unsigned PPCFastISel::FastEmitInst_rr(unsigned MachineInstOpcode,
     (RC == &PPC::GPRCRegClass ? &PPC::GPRC_and_GPRC_NOR0RegClass :
      (RC == &PPC::G8RCRegClass ? &PPC::G8RC_and_G8RC_NOX0RegClass : RC));
 
-  return FastISel::FastEmitInst_rr(MachineInstOpcode, UseRC, Op0, Op0IsKill,
+  return FastISel::fastEmitInst_rr(MachineInstOpcode, UseRC, Op0, Op0IsKill,
                                    Op1, Op1IsKill);
 }
 
@@ -2232,13 +2335,10 @@ namespace llvm {
   // Create the fast instruction selector for PowerPC64 ELF.
   FastISel *PPC::createFastISel(FunctionLoweringInfo &FuncInfo,
                                 const TargetLibraryInfo *LibInfo) {
-    const TargetMachine &TM = FuncInfo.MF->getTarget();
-
     // Only available on 64-bit ELF for now.
-    const PPCSubtarget *Subtarget = &TM.getSubtarget<PPCSubtarget>();
-    if (Subtarget->isPPC64() && Subtarget->isSVR4ABI())
+    const PPCSubtarget &Subtarget = FuncInfo.MF->getSubtarget<PPCSubtarget>();
+    if (Subtarget.isPPC64() && Subtarget.isSVR4ABI())
       return new PPCFastISel(FuncInfo, LibInfo);
-
-    return 0;
+    return nullptr;
   }
 }