X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FPowerPC%2FPPCFastISel.cpp;h=2e524d604789853449a1d1816d31632d156157e6;hb=1bb9c8155ad075e5f78665470a1e7636d3b27d52;hp=a308adebb51479e8739a643d2db1164c7acaf334;hpb=7248968fa529726b44d41bd25403d50c74db4bc4;p=oota-llvm.git

diff --git a/lib/Target/PowerPC/PPCFastISel.cpp b/lib/Target/PowerPC/PPCFastISel.cpp
index a308adebb51..2e524d60478 100644
--- a/lib/Target/PowerPC/PPCFastISel.cpp
+++ b/lib/Target/PowerPC/PPCFastISel.cpp
@@ -13,12 +13,11 @@
 //
 //===----------------------------------------------------------------------===//
 
-#define DEBUG_TYPE "ppcfastisel"
 #include "PPC.h"
+#include "MCTargetDesc/PPCPredicates.h"
 #include "PPCISelLowering.h"
 #include "PPCSubtarget.h"
 #include "PPCTargetMachine.h"
-#include "MCTargetDesc/PPCPredicates.h"
 #include "llvm/ADT/Optional.h"
 #include "llvm/CodeGen/CallingConvLower.h"
 #include "llvm/CodeGen/FastISel.h"
@@ -28,17 +27,38 @@
 #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.
+//   PPCMaterializeGV: Handle TLS.
+//   SelectCall: Handle function pointers.
+//   SelectCall: Handle multi-register return values.
+//   SelectCall: Optimize away nops for local calls.
+//   processCallArgs: Handle bit-converted arguments.
+//   finishCall: Handle multi-register return values.
+//   PPCComputeAddress: Handle parameter references as FrameIndex's.
+//   PPCEmitCmp: Handle immediate as operand 1.
+//   SelectCall: Handle small byval arguments.
+//   SelectIntrinsicCall: Implement.
+//   SelectSelect: Implement.
+//   Consider factoring isTypeLegal into the base class.
+//   Implement switches and jump tables.
+//
+//===----------------------------------------------------------------------===//
 using namespace llvm;
 
+#define DEBUG_TYPE "ppcfastisel"
+
 namespace {
 
 typedef struct Address {
@@ -61,12 +81,12 @@ typedef struct Address {
    }
 } Address;
 
-class PPCFastISel : public FastISel {
+class PPCFastISel final : public FastISel {
 
   const TargetMachine &TM;
   const TargetInstrInfo &TII;
   const TargetLowering &TLI;
-  const PPCSubtarget &PPCSubTarget;
+  const PPCSubtarget *PPCSubTarget;
   LLVMContext *Context;
 
   public:
@@ -76,31 +96,29 @@ class PPCFastISel : public FastISel {
       TM(FuncInfo.MF->getTarget()),
       TII(*TM.getInstrInfo()),
       TLI(*TM.getTargetLowering()),
-      PPCSubTarget(
-       *((static_cast<const PPCTargetMachine *>(&TM))->getSubtargetImpl())
-      ),
+      PPCSubTarget(&TM.getSubtarget<PPCSubtarget>()),
       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 TargetSelectInstruction(const Instruction *I) override;
+    unsigned TargetMaterializeConstant(const Constant *C) override;
+    unsigned TargetMaterializeAlloca(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);
 
   // Instruction selection routines.
   private:
@@ -108,8 +126,14 @@ class PPCFastISel : public FastISel {
     bool SelectStore(const Instruction *I);
     bool SelectBranch(const Instruction *I);
     bool SelectIndirectBr(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);
 
   // Utility routines.
@@ -134,9 +158,23 @@ class PPCFastISel : public FastISel {
                                     const TargetRegisterClass *RC);
     unsigned PPCMaterialize64BitInt(int64_t Imm,
                                     const TargetRegisterClass *RC);
+    unsigned PPCMoveToIntReg(const Instruction *I, MVT VT,
+                             unsigned SrcReg, bool IsSigned);
+    unsigned PPCMoveToFPReg(MVT VT, unsigned SrcReg, bool IsSigned);
 
   // Call handling routines.
   private:
+    bool processCallArgs(SmallVectorImpl<Value*> &Args,
+                         SmallVectorImpl<unsigned> &ArgRegs,
+                         SmallVectorImpl<MVT> &ArgVTs,
+                         SmallVectorImpl<ISD::ArgFlagsTy> &ArgFlags,
+                         SmallVectorImpl<unsigned> &RegArgs,
+                         CallingConv::ID CC,
+                         unsigned &NumBytes,
+                         bool IsVarArg);
+    void finishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs,
+                    const Instruction *I, CallingConv::ID CC,
+                    unsigned &NumBytes, bool IsVarArg);
     CCAssignFn *usePPC32CCs(unsigned Flag);
 
   private:
@@ -243,7 +281,7 @@ bool PPCFastISel::isLoadTypeLegal(Type *Ty, MVT &VT) {
 // 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
@@ -285,24 +323,19 @@ bool PPCFastISel::PPCComputeAddress(const Value *Obj, Address &Addr) {
            II != IE; ++II, ++GTI) {
         const Value *Op = *II;
         if (StructType *STy = dyn_cast<StructType>(*GTI)) {
-          const StructLayout *SL = TD.getStructLayout(STy);
+          const StructLayout *SL = DL.getStructLayout(STy);
           unsigned Idx = cast<ConstantInt>(Op)->getZExtValue();
           TmpOffset += SL->getElementOffset(Idx);
         } else {
-          uint64_t S = TD.getTypeAllocSize(GTI.getIndexedType());
+          uint64_t S = DL.getTypeAllocSize(GTI.getIndexedType());
           for (;;) {
             if (const ConstantInt *CI = dyn_cast<ConstantInt>(Op)) {
               // Constant-offset addressing.
               TmpOffset += CI->getSExtValue() * S;
               break;
             }
-            if (isa<AddOperator>(Op) &&
-                (!isa<Instruction>(Op) ||
-                 FuncInfo.MBBMap[cast<Instruction>(Op)->getParent()]
-                 == FuncInfo.MBB) &&
-                isa<ConstantInt>(cast<AddOperator>(Op)->getOperand(1))) {
-              // An add (in the same block) with a constant operand. Fold the
-              // constant.
+            if (canFoldAddIntoGEP(U, Op)) {
+              // A compatible add with a constant operand. Fold the constant.
               ConstantInt *CI =
               cast<ConstantInt>(cast<AddOperator>(Op)->getOperand(1));
               TmpOffset += CI->getSExtValue() * S;
@@ -372,7 +405,7 @@ void PPCFastISel::PPCSimplifyAddress(Address &Addr, MVT VT, bool &UseOffset,
   // register and continue. This should almost never happen.
   if (!UseOffset && Addr.BaseType == Address::FrameIndexBase) {
     unsigned ResultReg = createResultReg(&PPC::G8RC_and_G8RC_NOX0RegClass);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::ADDI8),
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ADDI8),
             ResultReg).addFrameIndex(Addr.Base.FI).addImm(0);
     Addr.Base.Reg = ResultReg;
     Addr.BaseType = Address::RegBase;
@@ -464,13 +497,13 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
         MachineMemOperand::MOLoad, MFI.getObjectSize(Addr.Base.FI),
         MFI.getObjectAlignment(Addr.Base.FI));
 
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
       .addImm(Addr.Offset).addFrameIndex(Addr.Base.FI).addMemOperand(MMO);
 
   // Base reg with offset in range.
   } else if (UseOffset) {
 
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
       .addImm(Addr.Offset).addReg(Addr.Base.Reg);
 
   // Indexed form.
@@ -494,7 +527,7 @@ bool PPCFastISel::PPCEmitLoad(MVT VT, unsigned &ResultReg, Address &Addr,
       case PPC::LFS:    Opc = PPC::LFSX;    break;
       case PPC::LFD:    Opc = PPC::LFDX;    break;
     }
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
       .addReg(Addr.Base.Reg).addReg(IndexReg);
   }
 
@@ -522,7 +555,7 @@ 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))
@@ -580,18 +613,19 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) {
         MachineMemOperand::MOStore, MFI.getObjectSize(Addr.Base.FI),
         MFI.getObjectAlignment(Addr.Base.FI));
 
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc)).addReg(SrcReg)
-      .addImm(Addr.Offset).addFrameIndex(Addr.Base.FI).addMemOperand(MMO);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc))
+        .addReg(SrcReg)
+        .addImm(Addr.Offset)
+        .addFrameIndex(Addr.Base.FI)
+        .addMemOperand(MMO);
 
   // Base reg with offset in range.
-  } else if (UseOffset) {
-    if (Addr.Offset == 0 && Opc == PPC::STW8)
-      dbgs() << "Possible problem here.\n";
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc))
+  } else if (UseOffset)
+    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.
@@ -607,7 +641,7 @@ bool PPCFastISel::PPCEmitStore(MVT VT, unsigned SrcReg, Address &Addr) {
       case PPC::STFS: Opc = PPC::STFSX; break;
       case PPC::STFD: Opc = PPC::STFDX; break;
     }
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc))
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc))
       .addReg(SrcReg).addReg(Addr.Base.Reg).addReg(IndexReg);
   }
 
@@ -671,9 +705,9 @@ bool PPCFastISel::SelectBranch(const Instruction *I) {
                     CondReg))
       return false;
 
-    BuildMI(*BrBB, FuncInfo.InsertPt, DL, TII.get(PPC::BCC))
+    BuildMI(*BrBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::BCC))
       .addImm(PPCPred).addReg(CondReg).addMBB(TBB);
-    FastEmitBranch(FBB, DL);
+    FastEmitBranch(FBB, DbgLoc);
     FuncInfo.MBB->addSuccessor(TBB);
     return true;
 
@@ -681,7 +715,7 @@ bool PPCFastISel::SelectBranch(const Instruction *I) {
              dyn_cast<ConstantInt>(BI->getCondition())) {
     uint64_t Imm = CI->getZExtValue();
     MachineBasicBlock *Target = (Imm == 0) ? FBB : TBB;
-    FastEmitBranch(Target, DL);
+    FastEmitBranch(Target, DbgLoc);
     return true;
   }
 
@@ -704,6 +738,9 @@ bool PPCFastISel::PPCEmitCmp(const Value *SrcValue1, const Value *SrcValue2,
     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
@@ -778,15 +815,275 @@ bool PPCFastISel::PPCEmitCmp(const Value *SrcValue1, const Value *SrcValue2,
   }
 
   if (!UseImm)
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(CmpOpc), DestReg)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CmpOpc), DestReg)
       .addReg(SrcReg1).addReg(SrcReg2);
   else
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(CmpOpc), DestReg)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(CmpOpc), DestReg)
       .addReg(SrcReg1).addImm(Imm);
 
   return true;
 }
 
+// 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);
+
+  if (SrcVT != MVT::f32 || DestVT != MVT::f64)
+    return false;
+
+  unsigned SrcReg = getRegForValue(Src);
+  if (!SrcReg)
+    return false;
+
+  // No code is generated for a FP extend.
+  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);
+
+  if (SrcVT != MVT::f64 || DestVT != MVT::f32)
+    return false;
+
+  unsigned SrcReg = getRegForValue(Src);
+  if (!SrcReg)
+    return false;
+
+  // Round the result to single precision.
+  unsigned DestReg = createResultReg(&PPC::F4RCRegClass);
+  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::FRSP), DestReg)
+    .addReg(SrcReg);
+
+  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),
+// 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
+// stack slot and 4-byte store/load sequence.  Or just sext the 4-byte
+// case to 8 bytes which produces tighter code but wastes stack space.
+unsigned PPCFastISel::PPCMoveToFPReg(MVT SrcVT, unsigned SrcReg,
+                                     bool IsSigned) {
+
+  // If necessary, extend 32-bit int to 64-bit.
+  if (SrcVT == MVT::i32) {
+    unsigned TmpReg = createResultReg(&PPC::G8RCRegClass);
+    if (!PPCEmitIntExt(MVT::i32, SrcReg, MVT::i64, TmpReg, !IsSigned))
+      return 0;
+    SrcReg = TmpReg;
+  }
+
+  // Get a stack slot 8 bytes wide, aligned on an 8-byte boundary.
+  Address Addr;
+  Addr.BaseType = Address::FrameIndexBase;
+  Addr.Base.FI = MFI.CreateStackObject(8, 8, false);
+
+  // Store the value from the GPR.
+  if (!PPCEmitStore(MVT::i64, SrcReg, Addr))
+    return 0;
+
+  // Load the integer value into an FPR.  The kind of load used depends
+  // on a number of conditions.
+  unsigned LoadOpc = PPC::LFD;
+
+  if (SrcVT == MVT::i32) {
+    if (!IsSigned) {
+      LoadOpc = PPC::LFIWZX;
+      Addr.Offset = 4;
+    } else if (PPCSubTarget->hasLFIWAX()) {
+      LoadOpc = PPC::LFIWAX;
+      Addr.Offset = 4;
+    }
+  }
+
+  const TargetRegisterClass *RC = &PPC::F8RCRegClass;
+  unsigned ResultReg = 0;
+  if (!PPCEmitLoad(MVT::f64, ResultReg, Addr, RC, !IsSigned, LoadOpc))
+    return 0;
+
+  return ResultReg;
+}
+
+// Attempt to fast-select an integer-to-floating-point conversion.
+bool PPCFastISel::SelectIToFP(const Instruction *I, bool IsSigned) {
+  MVT DstVT;
+  Type *DstTy = I->getType();
+  if (!isTypeLegal(DstTy, DstVT))
+    return false;
+
+  if (DstVT != MVT::f32 && DstVT != MVT::f64)
+    return false;
+
+  Value *Src = I->getOperand(0);
+  EVT SrcEVT = TLI.getValueType(Src->getType(), true);
+  if (!SrcEVT.isSimple())
+    return false;
+
+  MVT SrcVT = SrcEVT.getSimpleVT();
+
+  if (SrcVT != MVT::i8  && SrcVT != MVT::i16 &&
+      SrcVT != MVT::i32 && SrcVT != MVT::i64)
+    return false;
+
+  unsigned SrcReg = getRegForValue(Src);
+  if (SrcReg == 0)
+    return false;
+
+  // We can only lower an unsigned convert if we have the newer
+  // floating-point conversion operations.
+  if (!IsSigned && !PPCSubTarget->hasFPCVT())
+    return false;
+
+  // FIXME: For now we require the newer floating-point conversion operations
+  // (which are present only on P7 and A2 server models) when converting
+  // 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())
+    return false;
+
+  // Extend the input if necessary.
+  if (SrcVT == MVT::i8 || SrcVT == MVT::i16) {
+    unsigned TmpReg = createResultReg(&PPC::G8RCRegClass);
+    if (!PPCEmitIntExt(SrcVT, SrcReg, MVT::i64, TmpReg, !IsSigned))
+      return false;
+    SrcVT = MVT::i64;
+    SrcReg = TmpReg;
+  }
+
+  // Move the integer value to an FPR.
+  unsigned FPReg = PPCMoveToFPReg(SrcVT, SrcReg, IsSigned);
+  if (FPReg == 0)
+    return false;
+
+  // Determine the opcode for the conversion.
+  const TargetRegisterClass *RC = &PPC::F8RCRegClass;
+  unsigned DestReg = createResultReg(RC);
+  unsigned Opc;
+
+  if (DstVT == MVT::f32)
+    Opc = IsSigned ? PPC::FCFIDS : PPC::FCFIDUS;
+  else
+    Opc = IsSigned ? PPC::FCFID : PPC::FCFIDU;
+
+  // Generate the convert.
+  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), DestReg)
+    .addReg(FPReg);
+
+  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),
+// those should be used instead of moving via a stack slot when the
+// subtarget permits.
+unsigned PPCFastISel::PPCMoveToIntReg(const Instruction *I, MVT VT,
+                                      unsigned SrcReg, bool IsSigned) {
+  // Get a stack slot 8 bytes wide, aligned on an 8-byte boundary.
+  // Note that if have STFIWX available, we could use a 4-byte stack
+  // slot for i32, but this being fast-isel we'll just go with the
+  // easiest code gen possible.
+  Address Addr;
+  Addr.BaseType = Address::FrameIndexBase;
+  Addr.Base.FI = MFI.CreateStackObject(8, 8, false);
+
+  // Store the value from the FPR.
+  if (!PPCEmitStore(MVT::f64, SrcReg, Addr))
+    return 0;
+
+  // Reload it into a GPR.  If we want an i32, modify the address
+  // to have a 4-byte offset so we load from the right place.
+  if (VT == MVT::i32)
+    Addr.Offset = 4;
+
+  // Look at the currently assigned register for this instruction
+  // to determine the required register class.
+  unsigned AssignedReg = FuncInfo.ValueMap[I];
+  const TargetRegisterClass *RC =
+    AssignedReg ? MRI.getRegClass(AssignedReg) : nullptr;
+
+  unsigned ResultReg = 0;
+  if (!PPCEmitLoad(VT, ResultReg, Addr, RC, !IsSigned))
+    return 0;
+
+  return ResultReg;
+}
+
+// Attempt to fast-select a floating-point-to-integer conversion.
+bool PPCFastISel::SelectFPToI(const Instruction *I, bool IsSigned) {
+  MVT DstVT, SrcVT;
+  Type *DstTy = I->getType();
+  if (!isTypeLegal(DstTy, DstVT))
+    return false;
+
+  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))
+    return false;
+
+  if (SrcVT != MVT::f32 && SrcVT != MVT::f64)
+    return false;
+
+  unsigned SrcReg = getRegForValue(Src);
+  if (SrcReg == 0)
+    return false;
+
+  // Convert f32 to f64 if necessary.  This is just a meaningless copy
+  // to get the register class right.  COPY_TO_REGCLASS is needed since
+  // a COPY from F4RC to F8RC is converted to a F4RC-F4RC copy downstream.
+  const TargetRegisterClass *InRC = MRI.getRegClass(SrcReg);
+  if (InRC == &PPC::F4RCRegClass) {
+    unsigned TmpReg = createResultReg(&PPC::F8RCRegClass);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(TargetOpcode::COPY_TO_REGCLASS), TmpReg)
+      .addReg(SrcReg).addImm(PPC::F8RCRegClassID);
+    SrcReg = TmpReg;
+  }
+
+  // Determine the opcode for the conversion, which takes place
+  // entirely within FPRs.
+  unsigned DestReg = createResultReg(&PPC::F8RCRegClass);
+  unsigned Opc;
+
+  if (DstVT == MVT::i32)
+    if (IsSigned)
+      Opc = PPC::FCTIWZ;
+    else
+      Opc = PPCSubTarget->hasFPCVT() ? PPC::FCTIWUZ : PPC::FCTIDZ;
+  else
+    Opc = IsSigned ? PPC::FCTIDZ : PPC::FCTIDUZ;
+
+  // Generate the convert.
+  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), DestReg)
+    .addReg(SrcReg);
+
+  // Now move the integer value from a float register to an integer register.
+  unsigned IntReg = PPCMoveToIntReg(I, DstVT, DestReg, IsSigned);
+  if (IntReg == 0)
+    return false;
+
+  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) {
@@ -868,8 +1165,10 @@ bool PPCFastISel::SelectBinaryIntOp(const Instruction *I, unsigned ISDOpcode) {
       }
 
       if (UseImm) {
-        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg)
-          .addReg(SrcReg1).addImm(Imm);
+        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc),
+                ResultReg)
+            .addReg(SrcReg1)
+            .addImm(Imm);
         UpdateValueMap(I, ResultReg);
         return true;
       }
@@ -884,12 +1183,342 @@ bool PPCFastISel::SelectBinaryIntOp(const Instruction *I, unsigned ISDOpcode) {
   if (ISDOpcode == ISD::SUB)
     std::swap(SrcReg1, SrcReg2);
 
-  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ResultReg)
+  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ResultReg)
     .addReg(SrcReg1).addReg(SrcReg2);
   UpdateValueMap(I, ResultReg);
   return true;
 }
 
+// Handle arguments to a call that we're attempting to fast-select.
+// Return false if the arguments are too complex for us at the moment.
+bool PPCFastISel::processCallArgs(SmallVectorImpl<Value*> &Args,
+                                  SmallVectorImpl<unsigned> &ArgRegs,
+                                  SmallVectorImpl<MVT> &ArgVTs,
+                                  SmallVectorImpl<ISD::ArgFlagsTy> &ArgFlags,
+                                  SmallVectorImpl<unsigned> &RegArgs,
+                                  CallingConv::ID CC,
+                                  unsigned &NumBytes,
+                                  bool IsVarArg) {
+  SmallVector<CCValAssign, 16> ArgLocs;
+  CCState CCInfo(CC, IsVarArg, *FuncInfo.MF, TM, ArgLocs, *Context);
+
+  // Reserve space for the linkage area on the stack.
+  bool isELFv2ABI = PPCSubTarget->isELFv2ABI();
+  unsigned LinkageSize = PPCFrameLowering::getLinkageSize(true, false,
+                                                          isELFv2ABI);
+  CCInfo.AllocateStack(LinkageSize, 8);
+
+  CCInfo.AnalyzeCallOperands(ArgVTs, ArgFlags, CC_PPC64_ELF_FIS);
+
+  // Bail out if we can't handle any of the arguments.
+  for (unsigned I = 0, E = ArgLocs.size(); I != E; ++I) {
+    CCValAssign &VA = ArgLocs[I];
+    MVT ArgVT = ArgVTs[VA.getValNo()];
+
+    // 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 || ArgVT == MVT::i1 ||
+        !VA.isRegLoc() || VA.needsCustom())
+      return false;
+
+    // Skip bit-converted arguments for now.
+    if (VA.getLocInfo() == CCValAssign::BCvt)
+      return false;
+  }
+
+  // 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()))
+    .addImm(NumBytes);
+
+  // Prepare to assign register arguments.  Every argument uses up a
+  // GPR protocol register even if it's passed in a floating-point
+  // register.
+  unsigned NextGPR = PPC::X3;
+  unsigned NextFPR = PPC::F1;
+
+  // Process arguments.
+  for (unsigned I = 0, E = ArgLocs.size(); I != E; ++I) {
+    CCValAssign &VA = ArgLocs[I];
+    unsigned Arg = ArgRegs[VA.getValNo()];
+    MVT ArgVT = ArgVTs[VA.getValNo()];
+
+    // Handle argument promotion and bitcasts.
+    switch (VA.getLocInfo()) {
+      default:
+        llvm_unreachable("Unknown loc info!");
+      case CCValAssign::Full:
+        break;
+      case CCValAssign::SExt: {
+        MVT DestVT = VA.getLocVT();
+        const TargetRegisterClass *RC =
+          (DestVT == MVT::i64) ? &PPC::G8RCRegClass : &PPC::GPRCRegClass;
+        unsigned TmpReg = createResultReg(RC);
+        if (!PPCEmitIntExt(ArgVT, Arg, DestVT, TmpReg, /*IsZExt*/false))
+          llvm_unreachable("Failed to emit a sext!");
+        ArgVT = DestVT;
+        Arg = TmpReg;
+        break;
+      }
+      case CCValAssign::AExt:
+      case CCValAssign::ZExt: {
+        MVT DestVT = VA.getLocVT();
+        const TargetRegisterClass *RC =
+          (DestVT == MVT::i64) ? &PPC::G8RCRegClass : &PPC::GPRCRegClass;
+        unsigned TmpReg = createResultReg(RC);
+        if (!PPCEmitIntExt(ArgVT, Arg, DestVT, TmpReg, /*IsZExt*/true))
+          llvm_unreachable("Failed to emit a zext!");
+        ArgVT = DestVT;
+        Arg = TmpReg;
+        break;
+      }
+      case CCValAssign::BCvt: {
+        // FIXME: Not yet handled.
+        llvm_unreachable("Should have bailed before getting here!");
+        break;
+      }
+    }
+
+    // Copy this argument to the appropriate register.
+    unsigned ArgReg;
+    if (ArgVT == MVT::f32 || ArgVT == MVT::f64) {
+      ArgReg = NextFPR++;
+      ++NextGPR;
+    } else
+      ArgReg = NextGPR++;
+
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(TargetOpcode::COPY), ArgReg).addReg(Arg);
+    RegArgs.push_back(ArgReg);
+  }
+
+  return true;
+}
+
+// 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) {
+  // Issue CallSEQ_END.
+  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+          TII.get(TII.getCallFrameDestroyOpcode()))
+    .addImm(NumBytes).addImm(0);
+
+  // Next, generate a copy to obtain the return value.
+  // FIXME: No multi-register return values yet, though I don't foresee
+  // any real difficulties there.
+  if (RetVT != MVT::isVoid) {
+    SmallVector<CCValAssign, 16> RVLocs;
+    CCState CCInfo(CC, IsVarArg, *FuncInfo.MF, TM, RVLocs, *Context);
+    CCInfo.AnalyzeCallResult(RetVT, RetCC_PPC64_ELF_FIS);
+    CCValAssign &VA = RVLocs[0];
+    assert(RVLocs.size() == 1 && "No support for multi-reg return values!");
+    assert(VA.isRegLoc() && "Can only return in registers!");
+
+    MVT DestVT = VA.getValVT();
+    MVT CopyVT = DestVT;
+
+    // Ints smaller than a register still arrive in a full 64-bit
+    // register, so make sure we recognize this.
+    if (RetVT == MVT::i8 || RetVT == MVT::i16 || RetVT == MVT::i32)
+      CopyVT = MVT::i64;
+
+    unsigned SourcePhysReg = VA.getLocReg();
+    unsigned ResultReg = 0;
+
+    if (RetVT == CopyVT) {
+      const TargetRegisterClass *CpyRC = TLI.getRegClassFor(CopyVT);
+      ResultReg = createResultReg(CpyRC);
+
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+              TII.get(TargetOpcode::COPY), ResultReg)
+        .addReg(SourcePhysReg);
+
+    // If necessary, round the floating result to single precision.
+    } else if (CopyVT == MVT::f64) {
+      ResultReg = createResultReg(TLI.getRegClassFor(RetVT));
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::FRSP),
+              ResultReg).addReg(SourcePhysReg);
+
+    // If only the low half of a general register is needed, generate
+    // a GPRC copy instead of a G8RC copy.  (EXTRACT_SUBREG can't be
+    // used along the fast-isel path (not lowered), and downstream logic
+    // also doesn't like a direct subreg copy on a physical reg.)
+    } else if (RetVT == MVT::i8 || RetVT == MVT::i16 || RetVT == MVT::i32) {
+      ResultReg = createResultReg(&PPC::GPRCRegClass);
+      // Convert physical register from G8RC to GPRC.
+      SourcePhysReg -= PPC::X0 - PPC::R0;
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+              TII.get(TargetOpcode::COPY), ResultReg)
+        .addReg(SourcePhysReg);
+    }
+
+    assert(ResultReg && "ResultReg unset!");
+    UsedRegs.push_back(SourcePhysReg);
+    UpdateValueMap(I, ResultReg);
+  }
+}
+
+// Attempt to fast-select a call instruction.
+bool PPCFastISel::SelectCall(const Instruction *I) {
+  const CallInst *CI = cast<CallInst>(I);
+  const Value *Callee = CI->getCalledValue();
+
+  // Can't handle inline asm.
+  if (isa<InlineAsm>(Callee))
+    return false;
+
+  // Allow SelectionDAG isel to handle tail calls.
+  if (CI->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.
+  if (IsVarArg)
+    return false;
+
+  // Handle simple calls for now, with legal return types and
+  // those that can be extended.
+  Type *RetTy = I->getType();
+  MVT RetVT;
+  if (RetTy->isVoidTy())
+    RetVT = MVT::isVoid;
+  else if (!isTypeLegal(RetTy, RetVT) && RetVT != MVT::i16 &&
+           RetVT != MVT::i8)
+    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);
+    CCInfo.AnalyzeCallResult(RetVT, RetCC_PPC64_ELF_FIS);
+    if (RVLocs.size() > 1)
+      return false;
+  }
+
+  // Bail early if more than 8 arguments, as we only currently
+  // handle arguments passed in registers.
+  unsigned NumArgs = CS.arg_size();
+  if (NumArgs > 8)
+    return false;
+
+  // Set up the argument vectors.
+  SmallVector<Value*, 8> Args;
+  SmallVector<unsigned, 8> ArgRegs;
+  SmallVector<MVT, 8> ArgVTs;
+  SmallVector<ISD::ArgFlagsTy, 8> ArgFlags;
+
+  Args.reserve(NumArgs);
+  ArgRegs.reserve(NumArgs);
+  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;
+    
+    // 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))
+      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();
+    MVT ArgVT;
+    if (!isTypeLegal(ArgTy, ArgVT) && ArgVT != MVT::i16 && ArgVT != MVT::i8)
+      return false;
+
+    if (ArgVT.isVector())
+      return false;
+
+    unsigned Arg = getRegForValue(*II);
+    if (Arg == 0)
+      return false;
+
+    unsigned OriginalAlignment = DL.getABITypeAlignment(ArgTy);
+    Flags.setOrigAlign(OriginalAlignment);
+
+    Args.push_back(*II);
+    ArgRegs.push_back(Arg);
+    ArgVTs.push_back(ArgVT);
+    ArgFlags.push_back(Flags);
+  }
+
+  // Process the arguments.
+  SmallVector<unsigned, 8> RegArgs;
+  unsigned NumBytes;
+
+  if (!processCallArgs(Args, ArgRegs, ArgVTs, ArgFlags,
+                       RegArgs, CC, NumBytes, IsVarArg))
+    return false;
+
+  // 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);
+
+  // 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 the ELFv2 ABI need the TOC register live into the call.
+  if (PPCSubTarget->isELFv2ABI())
+    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));
+
+  // 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);
+
+  return true;
+}
+
 // Attempt to fast-select a return instruction.
 bool PPCFastISel::SelectRet(const Instruction *I) {
 
@@ -924,8 +1553,8 @@ bool PPCFastISel::SelectRet(const Instruction *I) {
       const Constant *C = cast<Constant>(RV);
       unsigned SrcReg = PPCMaterializeInt(C, MVT::i64);
       unsigned RetReg = ValLocs[0].getLocReg();
-      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(TargetOpcode::COPY),
-              RetReg).addReg(SrcReg);
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+              TII.get(TargetOpcode::COPY), RetReg).addReg(SrcReg);
       RetRegs.push_back(RetReg);
 
     } else {
@@ -980,14 +1609,14 @@ bool PPCFastISel::SelectRet(const Instruction *I) {
           }
         }
 
-        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+        BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                 TII.get(TargetOpcode::COPY), RetRegs[i])
           .addReg(SrcReg);
       }
     }
   }
 
-  MachineInstrBuilder MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+  MachineInstrBuilder MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                                     TII.get(PPC::BLR));
 
   for (unsigned i = 0, e = RetRegs.size(); i != e; ++i)
@@ -1017,7 +1646,7 @@ bool PPCFastISel::PPCEmitIntExt(MVT SrcVT, unsigned SrcReg, MVT DestVT,
       assert(DestVT == MVT::i64 && "Signed extend from i32 to i32??");
       Opc = PPC::EXTSW_32_64;
     }
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), DestReg)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), DestReg)
       .addReg(SrcReg);
 
   // Unsigned 32-bit extensions use RLWINM.
@@ -1029,7 +1658,7 @@ bool PPCFastISel::PPCEmitIntExt(MVT SrcVT, unsigned SrcReg, MVT DestVT,
       assert(SrcVT == MVT::i16 && "Unsigned extend from i32 to i32??");
       MB = 16;
     }
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::RLWINM),
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::RLWINM),
             DestReg)
       .addReg(SrcReg).addImm(/*SH=*/0).addImm(MB).addImm(/*ME=*/31);
 
@@ -1042,7 +1671,7 @@ bool PPCFastISel::PPCEmitIntExt(MVT SrcVT, unsigned SrcReg, MVT DestVT,
       MB = 48;
     else
       MB = 32;
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
             TII.get(PPC::RLDICL_32_64), DestReg)
       .addReg(SrcReg).addImm(/*SH=*/0).addImm(MB);
   }
@@ -1056,9 +1685,9 @@ bool PPCFastISel::SelectIndirectBr(const Instruction *I) {
   if (AddrReg == 0)
     return false;
 
-  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::MTCTR8))
+  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::MTCTR8))
     .addReg(AddrReg);
-  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::BCTR8));
+  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)
@@ -1067,6 +1696,35 @@ bool PPCFastISel::SelectIndirectBr(const Instruction *I) {
   return true;
 }
 
+// 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);
+
+  if (SrcVT != MVT::i64 && SrcVT != MVT::i32 && SrcVT != MVT::i16)
+    return false;
+
+  if (DestVT != MVT::i32 && DestVT != MVT::i16 && DestVT != MVT::i8)
+    return false;
+
+  unsigned SrcReg = getRegForValue(Src);
+  if (!SrcReg)
+    return false;
+
+  // The only interesting case is when we need to switch register classes.
+  if (SrcVT == MVT::i64) {
+    unsigned ResultReg = createResultReg(&PPC::GPRCRegClass);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(TargetOpcode::COPY),
+            ResultReg).addReg(SrcReg, 0, PPC::sub_32);
+    SrcReg = ResultReg;
+  }
+
+  UpdateValueMap(I, SrcReg);
+  return true;
+}
+
 // Attempt to fast-select an integer extend instruction.
 bool PPCFastISel::SelectIntExt(const Instruction *I) {
   Type *DestTy = I->getType();
@@ -1119,14 +1777,32 @@ bool PPCFastISel::TargetSelectInstruction(const Instruction *I) {
       return SelectBranch(I);
     case Instruction::IndirectBr:
       return SelectIndirectBr(I);
+    case Instruction::FPExt:
+      return SelectFPExt(I);
+    case Instruction::FPTrunc:
+      return SelectFPTrunc(I);
+    case Instruction::SIToFP:
+      return SelectIToFP(I, /*IsSigned*/ true);
+    case Instruction::UIToFP:
+      return SelectIToFP(I, /*IsSigned*/ false);
+    case Instruction::FPToSI:
+      return SelectFPToI(I, /*IsSigned*/ true);
+    case Instruction::FPToUI:
+      return SelectFPToI(I, /*IsSigned*/ false);
     case Instruction::Add:
       return SelectBinaryIntOp(I, ISD::ADD);
     case Instruction::Or:
       return SelectBinaryIntOp(I, ISD::OR);
     case Instruction::Sub:
       return SelectBinaryIntOp(I, ISD::SUB);
+    case Instruction::Call:
+      if (dyn_cast<IntrinsicInst>(I))
+        return false;
+      return SelectCall(I);
     case Instruction::Ret:
       return SelectRet(I);
+    case Instruction::Trunc:
+      return SelectTrunc(I);
     case Instruction::ZExt:
     case Instruction::SExt:
       return SelectIntExt(I);
@@ -1147,7 +1823,7 @@ unsigned PPCFastISel::PPCMaterializeFP(const ConstantFP *CFP, MVT VT) {
     return 0;
 
   // All FP constants are loaded from the constant pool.
-  unsigned Align = TD.getPrefTypeAlignment(CFP->getType());
+  unsigned Align = DL.getPrefTypeAlignment(CFP->getType());
   assert(Align > 0 && "Unexpectedly missing alignment information!");
   unsigned Idx = MCP.getConstantPoolIndex(cast<Constant>(CFP), Align);
   unsigned DestReg = createResultReg(TLI.getRegClassFor(VT));
@@ -1163,19 +1839,28 @@ unsigned PPCFastISel::PPCMaterializeFP(const ConstantFP *CFP, MVT VT) {
 
   // For small code model, generate a LF[SD](0, LDtocCPT(Idx, X2)).
   if (CModel == CodeModel::Small || CModel == CodeModel::JITDefault) {
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::LDtocCPT),
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtocCPT),
             TmpReg)
       .addConstantPoolIndex(Idx).addReg(PPC::X2);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), DestReg)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), DestReg)
       .addImm(0).addReg(TmpReg).addMemOperand(MMO);
   } else {
     // Otherwise we generate LF[SD](Idx[lo], ADDIStocHA(X2, Idx)).
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::ADDIStocHA),
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ADDIStocHA),
             TmpReg).addReg(PPC::X2).addConstantPoolIndex(Idx);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), DestReg)
-      .addConstantPoolIndex(Idx, 0, PPCII::MO_TOC_LO)
-      .addReg(TmpReg)
-      .addMemOperand(MMO);
+    // But for large code model, we must generate a LDtocL followed
+    // by the LF[SD].
+    if (CModel == CodeModel::Large) {
+      unsigned TmpReg2 = createResultReg(&PPC::G8RC_and_G8RC_NOX0RegClass);
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtocL),
+              TmpReg2).addConstantPoolIndex(Idx).addReg(TmpReg);
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), DestReg)
+        .addImm(0).addReg(TmpReg2);
+    } else 
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), DestReg)
+        .addConstantPoolIndex(Idx, 0, PPCII::MO_TOC_LO)
+        .addReg(TmpReg)
+        .addMemOperand(MMO);
   }
 
   return DestReg;
@@ -1196,26 +1881,20 @@ 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));
-    assert((GVar || isa<Function>(GV)) && "Unexpected GV subclass!");
-  }
 
   // FIXME: We don't yet handle the complexity of TLS.
-  bool IsTLS = GVar && GVar->isThreadLocal();
-  if (IsTLS)
+  if (GV->isThreadLocal())
     return 0;
 
   // For small code model, generate a simple TOC load.
   if (CModel == CodeModel::Small || CModel == CodeModel::JITDefault)
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::LDtoc), DestReg)
-      .addGlobalAddress(GV).addReg(PPC::X2);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtoc),
+            DestReg)
+        .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))
@@ -1223,20 +1902,21 @@ unsigned PPCFastISel::PPCMaterializeGV(const GlobalValue *GV, MVT VT) {
     //       ADDItocL(ADDIStocHA(%X2, GV), GV)
     // Either way, start with the ADDIStocHA:
     unsigned HighPartReg = createResultReg(RC);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::ADDIStocHA),
+    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())
-      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::LDtocL),
+    if (CModel == CodeModel::Large ||
+        (GV->getType()->getElementType()->isFunctionTy() &&
+         (GV->isDeclaration() || GV->isWeakForLinker())) ||
+        GV->isDeclaration() || GV->hasCommonLinkage() ||
+        GV->hasAvailableExternallyLinkage())
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtocL),
               DestReg).addGlobalAddress(GV).addReg(HighPartReg);
     else
       // Otherwise generate the ADDItocL.
-      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::ADDItocL),
+      BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ADDItocL),
               DestReg).addReg(HighPartReg).addGlobalAddress(GV);
   }
 
@@ -1254,21 +1934,21 @@ unsigned PPCFastISel::PPCMaterialize32BitInt(int64_t Imm,
   bool IsGPRC = RC->hasSuperClassEq(&PPC::GPRCRegClass);
 
   if (isInt<16>(Imm))
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
             TII.get(IsGPRC ? PPC::LI : PPC::LI8), ResultReg)
       .addImm(Imm);
   else if (Lo) {
     // Both Lo and Hi have nonzero bits.
     unsigned TmpReg = createResultReg(RC);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
             TII.get(IsGPRC ? PPC::LIS : PPC::LIS8), TmpReg)
       .addImm(Hi);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
             TII.get(IsGPRC ? PPC::ORI : PPC::ORI8), ResultReg)
       .addReg(TmpReg).addImm(Lo);
   } else
     // Just Hi bits.
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
             TII.get(IsGPRC ? PPC::LIS : PPC::LIS8), ResultReg)
       .addImm(Hi);
   
@@ -1308,7 +1988,7 @@ unsigned PPCFastISel::PPCMaterialize64BitInt(int64_t Imm,
   unsigned TmpReg2;
   if (Imm) {
     TmpReg2 = createResultReg(RC);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::RLDICR),
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::RLDICR),
             TmpReg2).addReg(TmpReg1).addImm(Shift).addImm(63 - Shift);
   } else
     TmpReg2 = TmpReg1;
@@ -1316,14 +1996,14 @@ unsigned PPCFastISel::PPCMaterialize64BitInt(int64_t Imm,
   unsigned TmpReg3, Hi, Lo;
   if ((Hi = (Remainder >> 16) & 0xFFFF)) {
     TmpReg3 = createResultReg(RC);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::ORIS8),
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ORIS8),
             TmpReg3).addReg(TmpReg2).addImm(Hi);
   } else
     TmpReg3 = TmpReg2;
 
   if ((Lo = Remainder & 0xFFFF)) {
     unsigned ResultReg = createResultReg(RC);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(PPC::ORI8),
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ORI8),
             ResultReg).addReg(TmpReg3).addImm(Lo);
     return ResultReg;
   }
@@ -1335,6 +2015,15 @@ 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) {
+  // If we're using CR bit registers for i1 values, handle that as a special
+  // case first.
+  if (VT == MVT::i1 && PPCSubTarget->useCRBits()) {
+    const ConstantInt *CI = cast<ConstantInt>(C);
+    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) 
@@ -1348,7 +2037,7 @@ unsigned PPCFastISel::PPCMaterializeInt(const Constant *C, MVT VT) {
   if (isInt<16>(CI->getSExtValue())) {
     unsigned Opc = (VT == MVT::i64) ? PPC::LI8 : PPC::LI;
     unsigned ImmReg = createResultReg(RC);
-    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(Opc), ImmReg)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(Opc), ImmReg)
       .addImm(CI->getSExtValue());
     return ImmReg;
   }
@@ -1379,15 +2068,30 @@ unsigned PPCFastISel::TargetMaterializeConstant(const Constant *C) {
     return PPCMaterializeGV(GV, VT);
   else if (isa<ConstantInt>(C))
     return PPCMaterializeInt(C, VT);
-  // TBD: Global values.
 
   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).  TBD.
+// return the register number (or zero if we failed to handle it).
 unsigned PPCFastISel::TargetMaterializeAlloca(const AllocaInst *AI) {
-  return AI && 0;
+  // Don't handle dynamic allocas.
+  if (!FuncInfo.StaticAllocaMap.count(AI)) return 0;
+
+  MVT VT;
+  if (!isLoadTypeLegal(AI->getType(), VT)) return 0;
+
+  DenseMap<const AllocaInst*, int>::iterator SI =
+    FuncInfo.StaticAllocaMap.find(AI);
+
+  if (SI != FuncInfo.StaticAllocaMap.end()) {
+    unsigned ResultReg = createResultReg(&PPC::G8RC_and_G8RC_NOX0RegClass);
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ADDI8),
+            ResultReg).addFrameIndex(SI->second).addImm(0);
+    return ResultReg;
+  }
+
+  return 0;
 }
 
 // Fold loads into extends when possible.
@@ -1461,7 +2165,7 @@ 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();
@@ -1485,6 +2189,15 @@ 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;
@@ -1564,6 +2277,6 @@ namespace llvm {
     if (Subtarget->isPPC64() && Subtarget->isSVR4ABI())
       return new PPCFastISel(FuncInfo, LibInfo);
 
-    return 0;
+    return nullptr;
   }
 }