21 files changed, 700 insertions, 527 deletions

diff --git a/lib/Target/X86/AsmPrinter/CMakeLists.txt b/lib/Target/X86/AsmPrinter/CMakeLists.txt
index 2079a9f..a28c826 100644
--- a/lib/Target/X86/AsmPrinter/CMakeLists.txt
+++ b/lib/Target/X86/AsmPrinter/CMakeLists.txt
@@ -6,3 +6,4 @@ add_llvm_library(LLVMX86AsmPrinter
   X86AsmPrinter.cpp
   X86IntelAsmPrinter.cpp
   )
+add_dependencies(LLVMX86AsmPrinter X86CodeGenTable_gen)
\ No newline at end of file
diff --git a/lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.cpp b/lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.cpp
index 60ed4f0..e75cfc5 100644
--- a/lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.cpp
+++ b/lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.cpp
@@ -23,10 +23,13 @@
 #include "llvm/CallingConv.h"
 #include "llvm/DerivedTypes.h"
 #include "llvm/Module.h"
+#include "llvm/MDNode.h"
 #include "llvm/Type.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringExtras.h"
+#include "llvm/MC/MCContext.h"
 #include "llvm/MC/MCInst.h"
+#include "llvm/MC/MCStreamer.h"
 #include "llvm/CodeGen/DwarfWriter.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
 #include "llvm/Support/CommandLine.h"
@@ -41,18 +44,26 @@ STATISTIC(EmittedInsts, "Number of machine instrs printed");
 static cl::opt<bool> NewAsmPrinter("experimental-asm-printer",
                                    cl::Hidden);
 
-static std::string getPICLabelString(unsigned FnNum,
-                                     const TargetAsmInfo *TAI,
-                                     const X86Subtarget* Subtarget) {
-  std::string label;
+//===----------------------------------------------------------------------===//
+// Primitive Helper Functions.
+//===----------------------------------------------------------------------===//
+
+void X86ATTAsmPrinter::PrintPICBaseSymbol() const {
   if (Subtarget->isTargetDarwin())
-    label =  "\"L" + utostr_32(FnNum) + "$pb\"";
+    O << "\"L" << getFunctionNumber() << "$pb\"";
   else if (Subtarget->isTargetELF())
-    label = ".Lllvm$" + utostr_32(FnNum) + "." "$piclabel";
+    O << ".Lllvm$" << getFunctionNumber() << "." "$piclabel";
   else
     assert(0 && "Don't know how to print PIC label!\n");
+}
 
-  return label;
+/// PrintUnmangledNameSafely - Print out the printable characters in the name.
+/// Don't print things like \\n or \\0.
+static void PrintUnmangledNameSafely(const Value *V, raw_ostream &OS) {
+  for (const char *Name = V->getNameStart(), *E = Name+V->getNameLen();
+       Name != E; ++Name)
+    if (isprint(*Name))
+      OS << *Name;
 }
 
 static X86MachineFunctionInfo calculateFunctionInfo(const Function *F,
@@ -89,15 +100,6 @@ static X86MachineFunctionInfo calculateFunctionInfo(const Function *F,
   return Info;
 }
 
-/// PrintUnmangledNameSafely - Print out the printable characters in the name.
-/// Don't print things like \\n or \\0.
-static void PrintUnmangledNameSafely(const Value *V, raw_ostream &OS) {
-  for (const char *Name = V->getNameStart(), *E = Name+V->getNameLen();
-       Name != E; ++Name)
-    if (isprint(*Name))
-      OS << *Name;
-}
-
 /// decorateName - Query FunctionInfoMap and use this information for various
 /// name decoration.
 void X86ATTAsmPrinter::decorateName(std::string &Name,
@@ -152,6 +154,8 @@ void X86ATTAsmPrinter::decorateName(std::string &Name,
   }
 }
 
+
+
 void X86ATTAsmPrinter::emitFunctionHeader(const MachineFunction &MF) {
   const Function *F = MF.getFunction();
 
@@ -159,9 +163,12 @@ void X86ATTAsmPrinter::emitFunctionHeader(const MachineFunction &MF) {
 
   SwitchToSection(TAI->SectionForGlobal(F));
 
+  // FIXME: A function's alignment should be part of MachineFunction.  There
+  // shouldn't be a policy decision here.
   unsigned FnAlign = 4;
   if (F->hasFnAttr(Attribute::OptimizeForSize))
     FnAlign = 1;
+  
   switch (F->getLinkage()) {
   default: assert(0 && "Unknown linkage type!");
   case Function::InternalLinkage:  // Symbols default to internal.
@@ -283,13 +290,8 @@ bool X86ATTAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
   return false;
 }
 
-static inline bool shouldPrintGOT(TargetMachine &TM, const X86Subtarget* ST) {
-  return ST->isPICStyleGOT() && TM.getRelocationModel() == Reloc::PIC_;
-}
-
 static inline bool shouldPrintPLT(TargetMachine &TM, const X86Subtarget* ST) {
-  return ST->isTargetELF() && TM.getRelocationModel() == Reloc::PIC_ &&
-      (ST->isPICStyleRIPRel() || ST->isPICStyleGOT());
+  return ST->isTargetELF() && TM.getRelocationModel() == Reloc::PIC_;
 }
 
 static inline bool shouldPrintStub(TargetMachine &TM, const X86Subtarget* ST) {
@@ -324,6 +326,8 @@ void X86ATTAsmPrinter::print_pcrel_imm(const MachineInstr *MI, unsigned OpNo) {
     }
     
     if (shouldPrintStub(TM, Subtarget)) {
+      // DARWIN/X86-32 in != static mode.
+      
       // Link-once, declaration, or Weakly-linked global variables need
       // non-lazily-resolved stubs
       if (GV->isDeclaration() || GV->isWeakForLinker()) {
@@ -354,9 +358,8 @@ void X86ATTAsmPrinter::print_pcrel_imm(const MachineInstr *MI, unsigned OpNo) {
         O << Name;
       }
     } else {
-      if (GV->hasDLLImportLinkage()) {
+      if (GV->hasDLLImportLinkage())
         O << "__imp_";
-      }
       O << Name;
       
       if (shouldPrintPLT(TM, Subtarget)) {
@@ -370,9 +373,6 @@ void X86ATTAsmPrinter::print_pcrel_imm(const MachineInstr *MI, unsigned OpNo) {
         FnStubs.insert(Name);
     }
     
-    if (GV->hasExternalWeakLinkage())
-      ExtWeakSymbols.insert(GV);
-    
     printOffset(MO.getOffset());
     
     if (needCloseParen)
@@ -386,7 +386,9 @@ void X86ATTAsmPrinter::print_pcrel_imm(const MachineInstr *MI, unsigned OpNo) {
     Name += MO.getSymbolName();
     // Print function stub suffix unless it's Mac OS X 10.5 and up.
     if (shouldPrintStub(TM, Subtarget) && 
+        // DARWIN/X86-32 in != static mode.
         !(Subtarget->isTargetDarwin() && Subtarget->getDarwinVers() >= 9)) {
+      
       FnStubs.insert(Name);
       printSuffixedName(Name, "$stub");
       return;
@@ -401,23 +403,15 @@ void X86ATTAsmPrinter::print_pcrel_imm(const MachineInstr *MI, unsigned OpNo) {
     
     O << Name;
     
-    if (shouldPrintPLT(TM, Subtarget)) {
-      std::string GOTName(TAI->getGlobalPrefix());
-      GOTName+="_GLOBAL_OFFSET_TABLE_";
-      if (Name == GOTName)
-        // HACK! Emit extra offset to PC during printing GOT offset to
-        // compensate for the size of popl instruction. The resulting code
-        // should look like:
-        //   call .piclabel
-        // piclabel:
-        //   popl %some_register
-        //   addl $_GLOBAL_ADDRESS_TABLE_ + [.-piclabel], %some_register
-        O << " + [.-"
-          << getPICLabelString(getFunctionNumber(), TAI, Subtarget) << ']';
-      
-      O << "@PLT";
+    if (MO.getTargetFlags() == X86II::MO_GOT_ABSOLUTE_ADDRESS) {
+      O << " + [.-";
+      PrintPICBaseSymbol();
+      O << ']';
     }
     
+    if (shouldPrintPLT(TM, Subtarget))
+      O << "@PLT";
+    
     if (needCloseParen)
       O << ')';
     
@@ -427,9 +421,10 @@ void X86ATTAsmPrinter::print_pcrel_imm(const MachineInstr *MI, unsigned OpNo) {
 }
 
 void X86ATTAsmPrinter::printOperand(const MachineInstr *MI, unsigned OpNo,
-                                    const char *Modifier, bool NotRIPRel) {
+                                    const char *Modifier) {
   const MachineOperand &MO = MI->getOperand(OpNo);
   switch (MO.getType()) {
+  default: assert(0 && "unknown operand type!");
   case MachineOperand::MO_Register: {
     assert(TargetRegisterInfo::isPhysicalRegister(MO.getReg()) &&
            "Virtual registers should not make it this far!");
@@ -456,18 +451,7 @@ void X86ATTAsmPrinter::printOperand(const MachineInstr *MI, unsigned OpNo,
     if (!isMemOp) O << '$';
     O << TAI->getPrivateGlobalPrefix() << "JTI" << getFunctionNumber() << '_'
       << MO.getIndex();
-
-    if (TM.getRelocationModel() == Reloc::PIC_) {
-      if (Subtarget->isPICStyleStub())
-        O << "-\"" << TAI->getPrivateGlobalPrefix() << getFunctionNumber()
-          << "$pb\"";
-      else if (Subtarget->isPICStyleGOT())
-        O << "@GOTOFF";
-    }
-
-    if (isMemOp && Subtarget->isPICStyleRIPRel() && !NotRIPRel)
-      O << "(%rip)";
-    return;
+    break;
   }
   case MachineOperand::MO_ConstantPoolIndex: {
     bool isMemOp  = Modifier && !strcmp(Modifier, "mem");
@@ -475,38 +459,17 @@ void X86ATTAsmPrinter::printOperand(const MachineInstr *MI, unsigned OpNo,
     O << TAI->getPrivateGlobalPrefix() << "CPI" << getFunctionNumber() << '_'
       << MO.getIndex();
 
-    if (TM.getRelocationModel() == Reloc::PIC_) {
-      if (Subtarget->isPICStyleStub())
-        O << "-\"" << TAI->getPrivateGlobalPrefix() << getFunctionNumber()
-          << "$pb\"";
-      else if (Subtarget->isPICStyleGOT())
-        O << "@GOTOFF";
-    }
-
     printOffset(MO.getOffset());
-
-    if (isMemOp && Subtarget->isPICStyleRIPRel() && !NotRIPRel)
-      O << "(%rip)";
-    return;
+    break;
   }
   case MachineOperand::MO_GlobalAddress: {
     bool isMemOp = Modifier && !strcmp(Modifier, "mem");
-    bool needCloseParen = false;
 
     const GlobalValue *GV = MO.getGlobal();
-    const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV);
-    if (!GVar) {
-      // If GV is an alias then use the aliasee for determining
-      // thread-localness.
-      if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
-        GVar =dyn_cast_or_null<GlobalVariable>(GA->resolveAliasedGlobal(false));
-    }
-
-    bool isThreadLocal = GVar && GVar->isThreadLocal();
-
     std::string Name = Mang->getValueName(GV);
     decorateName(Name, GV);
 
+    bool needCloseParen = false;
     if (!isMemOp)
       O << '$';
     else if (Name[0] == '$') {
@@ -517,6 +480,8 @@ void X86ATTAsmPrinter::printOperand(const MachineInstr *MI, unsigned OpNo,
     }
 
     if (shouldPrintStub(TM, Subtarget)) {
+      // DARWIN/X86-32 in != static mode.
+
       // Link-once, declaration, or Weakly-linked global variables need
       // non-lazily-resolved stubs
       if (GV->isDeclaration() || GV->isWeakForLinker()) {
@@ -539,118 +504,59 @@ void X86ATTAsmPrinter::printOperand(const MachineInstr *MI, unsigned OpNo,
         O << Name;
       }
 
-      if (TM.getRelocationModel() == Reloc::PIC_)
-        O << '-' << getPICLabelString(getFunctionNumber(), TAI, Subtarget);
+      if (TM.getRelocationModel() == Reloc::PIC_) {
+        O << '-';
+        PrintPICBaseSymbol();
+      }        
     } else {
       if (GV->hasDLLImportLinkage())
         O << "__imp_";
       O << Name;
     }
 
-    if (GV->hasExternalWeakLinkage())
-      ExtWeakSymbols.insert(GV);
-
     printOffset(MO.getOffset());
 
     if (needCloseParen)
       O << ')';
     
-    bool isRIPRelative = false;
-    if (isThreadLocal) {
-      TLSModel::Model model = getTLSModel(GVar, TM.getRelocationModel());
-      switch (model) {
-      case TLSModel::GeneralDynamic:
-        O << "@TLSGD";
-        break;
-      case TLSModel::LocalDynamic:
-        // O << "@TLSLD"; // local dynamic not implemented
-        O << "@TLSGD";
-        break;
-      case TLSModel::InitialExec:
-        if (Subtarget->is64Bit()) {
-          assert (!NotRIPRel);
-          O << "@GOTTPOFF";
-          isRIPRelative = true;
-        } else {
-          O << "@INDNTPOFF";
-        }
-        break;
-      case TLSModel::LocalExec:
-        if (Subtarget->is64Bit())
-          O << "@TPOFF";
-        else
-          O << "@NTPOFF";
-        break;
-      default:
-        assert (0 && "Unknown TLS model");
-      }
-    } else if (isMemOp) {
-      if (shouldPrintGOT(TM, Subtarget)) {
-        if (Subtarget->GVRequiresExtraLoad(GV, TM, false))
-          O << "@GOT";
-        else
-          O << "@GOTOFF";
-      } else if (Subtarget->isPICStyleRIPRel() &&
-                 !NotRIPRel) {
-        if (TM.getRelocationModel() != Reloc::Static) {
-          if (Subtarget->GVRequiresExtraLoad(GV, TM, false))
-            O << "@GOTPCREL";
-        }
-        
-        isRIPRelative = true;
-      }
-    }
-
-    // Use rip when possible to reduce code size, except when
-    // index or base register are also part of the address. e.g.
-    // foo(%rip)(%rcx,%rax,4) is not legal.
-    if (isRIPRelative)
-      O << "(%rip)";
-    
-    return;
+    break;
   }
-  case MachineOperand::MO_ExternalSymbol: {
-    bool isMemOp  = Modifier && !strcmp(Modifier, "mem");
-    bool needCloseParen = false;
-    std::string Name(TAI->getGlobalPrefix());
-    Name += MO.getSymbolName();
-
-    // Print function stub suffix unless it's Mac OS X 10.5 and up.
-    if (!isMemOp)
-      O << '$';
-    else if (Name[0] == '$') {
-      // The name begins with a dollar-sign. In order to avoid having it look
-      // like an integer immediate to the assembler, enclose it in parens.
-      O << '(';
-      needCloseParen = true;
-    }
-
-    O << Name;
-
-    if (shouldPrintPLT(TM, Subtarget)) {
-      std::string GOTName(TAI->getGlobalPrefix());
-      GOTName+="_GLOBAL_OFFSET_TABLE_";
-      if (Name == GOTName)
-        // HACK! Emit extra offset to PC during printing GOT offset to
-        // compensate for the size of popl instruction. The resulting code
-        // should look like:
-        //   call .piclabel
-        // piclabel:
-        //   popl %some_register
-        //   addl $_GLOBAL_ADDRESS_TABLE_ + [.-piclabel], %some_register
-        O << " + [.-"
-          << getPICLabelString(getFunctionNumber(), TAI, Subtarget) << ']';
-    }
-
-    if (needCloseParen)
-      O << ')';
-
-    if (Subtarget->isPICStyleRIPRel())
-      O << "(%rip)";
-    return;
+  case MachineOperand::MO_ExternalSymbol:
+    /// NOTE: MO_ExternalSymbol in a non-pcrel_imm context is *only* generated
+    /// by _GLOBAL_OFFSET_TABLE_ on X86-32.  All others are call operands, which
+    /// are pcrel_imm's.
+    assert(!Subtarget->is64Bit() && !Subtarget->isPICStyleRIPRel());
+    // These are never used as memory operands.
+    assert(!(Modifier && !strcmp(Modifier, "mem")));
+    
+    O << '$';
+    O << TAI->getGlobalPrefix();
+    O << MO.getSymbolName();
+    break;
   }
+  
+  switch (MO.getTargetFlags()) {
   default:
-    O << "<unknown operand type>"; return;
+    assert(0 && "Unknown target flag on GV operand");
+  case X86II::MO_NO_FLAG:
+    break;
+  case X86II::MO_GOT_ABSOLUTE_ADDRESS:
+    O << " + [.-";
+    PrintPICBaseSymbol();
+    O << ']';
+    break;      
+  case X86II::MO_PIC_BASE_OFFSET:
+    O << '-';
+    PrintPICBaseSymbol();
+    break;
+  case X86II::MO_TLSGD:     O << "@TLSGD";     break;
+  case X86II::MO_GOTTPOFF:  O << "@GOTTPOFF";  break;
+  case X86II::MO_INDNTPOFF: O << "@INDNTPOFF"; break;
+  case X86II::MO_TPOFF:     O << "@TPOFF";     break;
+  case X86II::MO_NTPOFF:    O << "@NTPOFF";    break;
+  case X86II::MO_GOTPCREL:  O << "@GOTPCREL";  break;
+  case X86II::MO_GOT:       O << "@GOT";       break;
+  case X86II::MO_GOTOFF:    O << "@GOTOFF";    break;
   }
 }
 
@@ -670,25 +576,24 @@ void X86ATTAsmPrinter::printSSECC(const MachineInstr *MI, unsigned Op) {
 }
 
 void X86ATTAsmPrinter::printLeaMemReference(const MachineInstr *MI, unsigned Op,
-                                            const char *Modifier,
-                                            bool NotRIPRel) {
+                                            const char *Modifier) {
   MachineOperand BaseReg  = MI->getOperand(Op);
   MachineOperand IndexReg = MI->getOperand(Op+2);
   const MachineOperand &DispSpec = MI->getOperand(Op+3);
 
-  NotRIPRel |= IndexReg.getReg() || BaseReg.getReg();
   if (DispSpec.isGlobal() ||
       DispSpec.isCPI() ||
       DispSpec.isJTI() ||
       DispSpec.isSymbol()) {
-    printOperand(MI, Op+3, "mem", NotRIPRel);
+    printOperand(MI, Op+3, "mem");
   } else {
     int DispVal = DispSpec.getImm();
     if (DispVal || (!IndexReg.getReg() && !BaseReg.getReg()))
       O << DispVal;
   }
 
-  if (IndexReg.getReg() || BaseReg.getReg()) {
+  if ((IndexReg.getReg() || BaseReg.getReg()) &&
+      (Modifier == 0 || strcmp(Modifier, "no-rip"))) {
     unsigned ScaleVal = MI->getOperand(Op+1).getImm();
     unsigned BaseRegOperand = 0, IndexRegOperand = 2;
 
@@ -716,14 +621,14 @@ void X86ATTAsmPrinter::printLeaMemReference(const MachineInstr *MI, unsigned Op,
 }
 
 void X86ATTAsmPrinter::printMemReference(const MachineInstr *MI, unsigned Op,
-                                         const char *Modifier, bool NotRIPRel){
+                                         const char *Modifier) {
   assert(isMem(MI, Op) && "Invalid memory reference!");
   MachineOperand Segment = MI->getOperand(Op+4);
   if (Segment.getReg()) {
       printOperand(MI, Op+4, Modifier);
       O << ':';
     }
-  printLeaMemReference(MI, Op, Modifier, NotRIPRel);
+  printLeaMemReference(MI, Op, Modifier);
 }
 
 void X86ATTAsmPrinter::printPICJumpTableSetLabel(unsigned uid,
@@ -741,13 +646,19 @@ void X86ATTAsmPrinter::printPICJumpTableSetLabel(unsigned uid,
   if (Subtarget->isPICStyleRIPRel())
     O << '-' << TAI->getPrivateGlobalPrefix() << "JTI" << getFunctionNumber()
       << '_' << uid << '\n';
-  else
-    O << '-' << getPICLabelString(getFunctionNumber(), TAI, Subtarget) << '\n';
+  else {
+    O << '-';
+    PrintPICBaseSymbol();
+    O << '\n';
+  }
 }
 
+
 void X86ATTAsmPrinter::printPICLabel(const MachineInstr *MI, unsigned Op) {
-  std::string label = getPICLabelString(getFunctionNumber(), TAI, Subtarget);
-  O << label << '\n' << label << ':';
+  PrintPICBaseSymbol();
+  O << '\n';
+  PrintPICBaseSymbol();
+  O << ':';
 }
 
 
@@ -810,7 +721,7 @@ bool X86ATTAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
     switch (ExtraCode[0]) {
     default: return true;  // Unknown modifier.
     case 'c': // Don't print "$" before a global var name or constant.
-      printOperand(MI, OpNo, "mem", /*NotRIPRel=*/true);
+      printOperand(MI, OpNo, "mem");
       return false;
     case 'b': // Print QImode register
     case 'h': // Print QImode high register
@@ -823,8 +734,19 @@ bool X86ATTAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
       return false;
 
     case 'P': // Don't print @PLT, but do print as memory.
-      printOperand(MI, OpNo, "mem", /*NotRIPRel=*/true);
+      printOperand(MI, OpNo, "mem");
       return false;
+
+      case 'n': { // Negate the immediate or print a '-' before the operand.
+      // Note: this is a temporary solution. It should be handled target
+      // independently as part of the 'MC' work.
+      const MachineOperand &MO = MI->getOperand(OpNo);
+      if (MO.isImm()) {
+        O << -MO.getImm();
+        return false;
+      }
+      O << '-';
+    }
     }
   }
 
@@ -849,7 +771,7 @@ bool X86ATTAsmPrinter::PrintAsmMemoryOperand(const MachineInstr *MI,
       // These only apply to registers, ignore on mem.
       break;
     case 'P': // Don't print @PLT, but do print as memory.
-      printMemReference(MI, OpNo, "mem", /*NotRIPRel=*/true);
+      printMemReference(MI, OpNo, "no-rip");
       return false;
     }
   }
@@ -931,8 +853,13 @@ void X86ATTAsmPrinter::printMachineInstruction(const MachineInstr *MI) {
 
 /// doInitialization
 bool X86ATTAsmPrinter::doInitialization(Module &M) {
-  if (TAI->doesSupportDebugInformation() || TAI->doesSupportExceptionHandling()) 
-    MMI = getAnalysisIfAvailable<MachineModuleInfo>();
+  if (NewAsmPrinter) {
+    Context = new MCContext();
+    // FIXME: Send this to "O" instead of outs().  For now, we force it to
+    // stdout to make it easy to compare.
+    Streamer = createAsmStreamer(*Context, outs());
+  }
+  
   return AsmPrinter::doInitialization(M);
 }
 
@@ -956,6 +883,8 @@ void X86ATTAsmPrinter::printModuleLevelGV(const GlobalVariable* GVar) {
 
   std::string name = Mang->getValueName(GVar);
   Constant *C = GVar->getInitializer();
+  if (isa<MDNode>(C) || isa<MDString>(C))
+    return;
   const Type *Type = C->getType();
   unsigned Size = TD->getTypeAllocSize(Type);
   unsigned Align = TD->getPreferredAlignmentLog(GVar);
@@ -1068,25 +997,6 @@ void X86ATTAsmPrinter::printModuleLevelGV(const GlobalVariable* GVar) {
   EmitGlobalConstant(C);
 }
 
-/// printGVStub - Print stub for a global value.
-///
-void X86ATTAsmPrinter::printGVStub(const char *GV, const char *Prefix) {
-  printSuffixedName(GV, "$non_lazy_ptr", Prefix);
-  O << ":\n\t.indirect_symbol ";
-  if (Prefix) O << Prefix;
-  O << GV << "\n\t.long\t0\n";
-}
-
-/// printHiddenGVStub - Print stub for a hidden global value.
-///
-void X86ATTAsmPrinter::printHiddenGVStub(const char *GV, const char *Prefix) {
-  EmitAlignment(2);
-  printSuffixedName(GV, "$non_lazy_ptr", Prefix);
-  if (Prefix) O << Prefix;
-  O << ":\n" << TAI->getData32bitsDirective() << GV << '\n';
-}
-
-
 bool X86ATTAsmPrinter::doFinalization(Module &M) {
   // Print out module-level global variables here.
   for (Module::const_global_iterator I = M.global_begin(), E = M.global_end();
@@ -1095,100 +1005,62 @@ bool X86ATTAsmPrinter::doFinalization(Module &M) {
 
     if (I->hasDLLExportLinkage())
       DLLExportedGVs.insert(Mang->makeNameProper(I->getName(),""));
-
-    // If the global is a extern weak symbol, remember to emit the weak
-    // reference!
-    // FIXME: This is rather hacky, since we'll emit references to ALL weak
-    // stuff, not used. But currently it's the only way to deal with extern weak
-    // initializers hidden deep inside constant expressions.
-    if (I->hasExternalWeakLinkage())
-      ExtWeakSymbols.insert(I);
-  }
-
-  for (Module::const_iterator I = M.begin(), E = M.end();
-       I != E; ++I) {
-    // If the global is a extern weak symbol, remember to emit the weak
-    // reference!
-    // FIXME: This is rather hacky, since we'll emit references to ALL weak
-    // stuff, not used. But currently it's the only way to deal with extern weak
-    // initializers hidden deep inside constant expressions.
-    if (I->hasExternalWeakLinkage())
-      ExtWeakSymbols.insert(I);
   }
 
-  // Output linker support code for dllexported globals
-  if (!DLLExportedGVs.empty())
-    SwitchToDataSection(".section .drectve");
-
-  for (StringSet<>::iterator i = DLLExportedGVs.begin(),
-         e = DLLExportedGVs.end();
-         i != e; ++i)
-    O << "\t.ascii \" -export:" << i->getKeyData() << ",data\"\n";
-
-  if (!DLLExportedFns.empty()) {
-    SwitchToDataSection(".section .drectve");
-  }
-
-  for (StringSet<>::iterator i = DLLExportedFns.begin(),
-         e = DLLExportedFns.end();
-         i != e; ++i)
-    O << "\t.ascii \" -export:" << i->getKeyData() << "\"\n";
-
   if (Subtarget->isTargetDarwin()) {
     SwitchToDataSection("");
-
-    // Output stubs for dynamically-linked functions
-    for (StringSet<>::iterator i = FnStubs.begin(), e = FnStubs.end();
-         i != e; ++i) {
-      SwitchToDataSection("\t.section __IMPORT,__jump_table,symbol_stubs,"
-                          "self_modifying_code+pure_instructions,5", 0);
-      const char *p = i->getKeyData();
-      printSuffixedName(p, "$stub");
-      O << ":\n"
-           "\t.indirect_symbol " << p << "\n"
-           "\thlt ; hlt ; hlt ; hlt ; hlt\n";
-    }
-
-    O << '\n';
-
-    // Print global value stubs.
-    bool InStubSection = false;
+    
+    // Add the (possibly multiple) personalities to the set of global value
+    // stubs.  Only referenced functions get into the Personalities list.
     if (TAI->doesSupportExceptionHandling() && MMI && !Subtarget->is64Bit()) {
-      // Add the (possibly multiple) personalities to the set of global values.
-      // Only referenced functions get into the Personalities list.
-      const std::vector<Function *>& Personalities = MMI->getPersonalities();
-      for (std::vector<Function *>::const_iterator I = Personalities.begin(),
-             E = Personalities.end(); I != E; ++I) {
-        if (!*I)
+      const std::vector<Function*> &Personalities = MMI->getPersonalities();
+      for (unsigned i = 0, e = Personalities.size(); i != e; ++i) {
+        if (Personalities[i] == 0)
           continue;
-        if (!InStubSection) {
-          SwitchToDataSection(
-                     "\t.section __IMPORT,__pointers,non_lazy_symbol_pointers");
-          InStubSection = true;
-        }
-        printGVStub((*I)->getNameStart(), "_");
+        std::string Name = Mang->getValueName(Personalities[i]);
+        decorateName(Name, Personalities[i]);
+        GVStubs.insert(Name);
       }
     }
 
+    // Output stubs for dynamically-linked functions
+    if (!FnStubs.empty()) {
+      for (StringSet<>::iterator I = FnStubs.begin(), E = FnStubs.end();
+           I != E; ++I) {
+        SwitchToDataSection("\t.section __IMPORT,__jump_table,symbol_stubs,"
+                            "self_modifying_code+pure_instructions,5", 0);
+        const char *Name = I->getKeyData();
+        printSuffixedName(Name, "$stub");
+        O << ":\n"
+             "\t.indirect_symbol " << Name << "\n"
+             "\thlt ; hlt ; hlt ; hlt ; hlt\n";
+      }
+      O << '\n';
+    }
+
     // Output stubs for external and common global variables.
-    if (!InStubSection && !GVStubs.empty())
+    if (!GVStubs.empty()) {
       SwitchToDataSection(
                     "\t.section __IMPORT,__pointers,non_lazy_symbol_pointers");
-    for (StringSet<>::iterator i = GVStubs.begin(), e = GVStubs.end();
-         i != e; ++i)
-      printGVStub(i->getKeyData());
+      for (StringSet<>::iterator I = GVStubs.begin(), E = GVStubs.end();
+           I != E; ++I) {
+        const char *Name = I->getKeyData();
+        printSuffixedName(Name, "$non_lazy_ptr");
+        O << ":\n\t.indirect_symbol " << Name << "\n\t.long\t0\n";
+      }
+    }
 
     if (!HiddenGVStubs.empty()) {
       SwitchToSection(TAI->getDataSection());
-      for (StringSet<>::iterator i = HiddenGVStubs.begin(), e = HiddenGVStubs.end();
-           i != e; ++i)
-        printHiddenGVStub(i->getKeyData());
+      EmitAlignment(2);
+      for (StringSet<>::iterator I = HiddenGVStubs.begin(),
+           E = HiddenGVStubs.end(); I != E; ++I) {
+        const char *Name = I->getKeyData();
+        printSuffixedName(Name, "$non_lazy_ptr");
+        O << ":\n" << TAI->getData32bitsDirective() << Name << '\n';
+      }
     }
 
-    // Emit final debug information.
-    if (TAI->doesSupportDebugInformation() || TAI->doesSupportExceptionHandling())
-      DW->EndModule();
-
     // Funny Darwin hack: This flag tells the linker that no global symbols
     // contain code that falls through to other global symbols (e.g. the obvious
     // implementation of multiple entry points).  If this doesn't occur, the
@@ -1204,17 +1076,40 @@ bool X86ATTAsmPrinter::doFinalization(Module &M) {
         << ";\t.type\t" << (COFF::DT_FCN << COFF::N_BTSHFT)
         << ";\t.endef\n";
     }
-
-    // Emit final debug information.
-    if (TAI->doesSupportDebugInformation() || TAI->doesSupportExceptionHandling())
-      DW->EndModule();
-  } else if (Subtarget->isTargetELF()) {
-    // Emit final debug information.
-    if (TAI->doesSupportDebugInformation() || TAI->doesSupportExceptionHandling())
-      DW->EndModule();
   }
-
-  return AsmPrinter::doFinalization(M);
+  
+  
+  // Output linker support code for dllexported globals on windows.
+  if (!DLLExportedGVs.empty()) {
+    SwitchToDataSection(".section .drectve");
+  
+    for (StringSet<>::iterator i = DLLExportedGVs.begin(),
+         e = DLLExportedGVs.end(); i != e; ++i)
+      O << "\t.ascii \" -export:" << i->getKeyData() << ",data\"\n";
+  }
+  
+  if (!DLLExportedFns.empty()) {
+    SwitchToDataSection(".section .drectve");
+  
+    for (StringSet<>::iterator i = DLLExportedFns.begin(),
+         e = DLLExportedFns.end();
+         i != e; ++i)
+      O << "\t.ascii \" -export:" << i->getKeyData() << "\"\n";
+  }
+  
+  // Do common shutdown.
+  bool Changed = AsmPrinter::doFinalization(M);
+  
+  if (NewAsmPrinter) {
+    Streamer->Finish();
+    
+    delete Streamer;
+    delete Context;
+    Streamer = 0;
+    Context = 0;
+  }
+  
+  return Changed;
 }
 
 // Include the auto-generated portion of the assembly writer.
diff --git a/lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.h b/lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.h
index 68a6bc8..bd96115 100644
--- a/lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.h
+++ b/lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.h
@@ -27,17 +27,23 @@
 namespace llvm {
 
 class MachineJumpTableInfo;
+class MCContext;
 class MCInst;
+class MCStreamer;
 
 class VISIBILITY_HIDDEN X86ATTAsmPrinter : public AsmPrinter {
-  MachineModuleInfo *MMI;
   const X86Subtarget *Subtarget;
+  
+  MCContext *Context;
+  MCStreamer *Streamer;
  public:
   explicit X86ATTAsmPrinter(raw_ostream &O, X86TargetMachine &TM,
                             const TargetAsmInfo *T, CodeGenOpt::Level OL,
                             bool V)
-    : AsmPrinter(O, TM, T, OL, V), MMI(0) {
+    : AsmPrinter(O, TM, T, OL, V) {
     Subtarget = &TM.getSubtarget<X86Subtarget>();
+    Context = 0;
+    Streamer = 0;
   }
 
   virtual const char *getPassName() const {
@@ -69,7 +75,7 @@ class VISIBILITY_HIDDEN X86ATTAsmPrinter : public AsmPrinter {
   bool printInstruction(const MCInst *MI);
 
   void printOperand(const MCInst *MI, unsigned OpNo,
-                    const char *Modifier = 0, bool NotRIPRel = false);
+                    const char *Modifier = 0);
   void printMemReference(const MCInst *MI, unsigned Op);
   void printLeaMemReference(const MCInst *MI, unsigned Op);
   void printSSECC(const MCInst *MI, unsigned Op);
@@ -117,7 +123,7 @@ class VISIBILITY_HIDDEN X86ATTAsmPrinter : public AsmPrinter {
 
   // These methods are used by the tablegen'erated instruction printer.
   void printOperand(const MachineInstr *MI, unsigned OpNo,
-                    const char *Modifier = 0, bool NotRIPRel = false);
+                    const char *Modifier = 0);
   void print_pcrel_imm(const MachineInstr *MI, unsigned OpNo);
   void printi8mem(const MachineInstr *MI, unsigned OpNo) {
     printMemReference(MI, OpNo);
@@ -165,9 +171,9 @@ class VISIBILITY_HIDDEN X86ATTAsmPrinter : public AsmPrinter {
   void printMachineInstruction(const MachineInstr *MI);
   void printSSECC(const MachineInstr *MI, unsigned Op);
   void printMemReference(const MachineInstr *MI, unsigned Op,
-                         const char *Modifier=NULL, bool NotRIPRel = false);
+                         const char *Modifier=NULL);
   void printLeaMemReference(const MachineInstr *MI, unsigned Op,
-                            const char *Modifier=NULL, bool NotRIPRel = false);
+                            const char *Modifier=NULL);
   void printPICJumpTableSetLabel(unsigned uid,
                                  const MachineBasicBlock *MBB) const;
   void printPICJumpTableSetLabel(unsigned uid, unsigned uid2,
@@ -181,9 +187,8 @@ class VISIBILITY_HIDDEN X86ATTAsmPrinter : public AsmPrinter {
   void printPICLabel(const MachineInstr *MI, unsigned Op);
   void printModuleLevelGV(const GlobalVariable* GVar);
 
-  void printGVStub(const char *GV, const char *Prefix = NULL);
-  void printHiddenGVStub(const char *GV, const char *Prefix = NULL);
-
+  void PrintPICBaseSymbol() const;
+  
   bool runOnMachineFunction(MachineFunction &F);
 
   void emitFunctionHeader(const MachineFunction &MF);
diff --git a/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp b/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp
index 9d50edc..fa0ee75 100644
--- a/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp
+++ b/lib/Target/X86/AsmPrinter/X86ATTInstPrinter.cpp
@@ -65,7 +65,7 @@ void X86ATTAsmPrinter::print_pcrel_imm(const MCInst *MI, unsigned OpNo) {
 
 
 void X86ATTAsmPrinter::printOperand(const MCInst *MI, unsigned OpNo,
-                                    const char *Modifier, bool NotRIPRel) {
+                                    const char *Modifier) {
   assert(Modifier == 0 && "Modifiers should not be used");
   
   const MCOperand &Op = MI->getOperand(OpNo);
@@ -93,13 +93,11 @@ void X86ATTAsmPrinter::printOperand(const MCInst *MI, unsigned OpNo,
 }
 
 void X86ATTAsmPrinter::printLeaMemReference(const MCInst *MI, unsigned Op) {
-  bool NotRIPRel = false;
 
   const MCOperand &BaseReg  = MI->getOperand(Op);
   const MCOperand &IndexReg = MI->getOperand(Op+2);
   const MCOperand &DispSpec = MI->getOperand(Op+3);
   
-  NotRIPRel |= IndexReg.getReg() || BaseReg.getReg();
   if (DispSpec.isImm()) {
     int64_t DispVal = DispSpec.getImm();
     if (DispVal || (!IndexReg.getReg() && !BaseReg.getReg()))
@@ -108,7 +106,7 @@ void X86ATTAsmPrinter::printLeaMemReference(const MCInst *MI, unsigned Op) {
     abort();
     //assert(DispSpec.isGlobal() || DispSpec.isCPI() ||
     //       DispSpec.isJTI() || DispSpec.isSymbol());
-    //printOperand(MI, Op+3, "mem", NotRIPRel);
+    //printOperand(MI, Op+3, "mem");
   }
   
   if (IndexReg.getReg() || BaseReg.getReg()) {
diff --git a/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp b/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp
index a39203b..d1623d6 100644
--- a/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp
+++ b/lib/Target/X86/AsmPrinter/X86AsmPrinter.cpp
@@ -47,8 +47,5 @@ namespace {
 extern "C" int X86AsmPrinterForceLink;
 int X86AsmPrinterForceLink = 0;
 
-// Force static initialization when called from
-// llvm/InitializeAllAsmPrinters.h
-namespace llvm {
-  void InitializeX86AsmPrinter() { }
-}
+// Force static initialization.
+extern "C" void LLVMInitializeX86AsmPrinter() { }
diff --git a/lib/Target/X86/X86.td b/lib/Target/X86/X86.td
index 8df138d..4d26364 100644
--- a/lib/Target/X86/X86.td
+++ b/lib/Target/X86/X86.td
@@ -55,6 +55,13 @@ def FeatureSlowBTMem : SubtargetFeature<"slow-bt-mem", "IsBTMemSlow", "true",
 def FeatureSSE4A   : SubtargetFeature<"sse4a", "HasSSE4A", "true",
                                       "Support SSE 4a instructions">;
 
+def FeatureAVX     : SubtargetFeature<"avx", "HasAVX", "true",
+                                      "Enable AVX instructions">;
+def FeatureFMA3    : SubtargetFeature<"fma3", "HasFMA3", "true",
+                                      "Enable three-operand fused multiple-add">;
+def FeatureFMA4    : SubtargetFeature<"fma4", "HasFMA4", "true",
+                                      "Enable four-operand fused multiple-add">;
+
 //===----------------------------------------------------------------------===//
 // X86 processors supported.
 //===----------------------------------------------------------------------===//
@@ -82,6 +89,9 @@ def : Proc<"core2",           [FeatureSSSE3,  Feature64Bit, FeatureSlowBTMem]>;
 def : Proc<"penryn",          [FeatureSSE41,  Feature64Bit, FeatureSlowBTMem]>;
 def : Proc<"atom",            [FeatureSSE3,   Feature64Bit, FeatureSlowBTMem]>;
 def : Proc<"corei7",          [FeatureSSE42,  Feature64Bit, FeatureSlowBTMem]>;
+def : Proc<"nehalem",         [FeatureSSE42,  Feature64Bit, FeatureSlowBTMem]>;
+// Sandy Bridge does not have FMA
+def : Proc<"sandybridge",     [FeatureSSE42,  FeatureAVX,   Feature64Bit]>;
 
 def : Proc<"k6",              [FeatureMMX]>;
 def : Proc<"k6-2",            [FeatureMMX,    Feature3DNow]>;
diff --git a/lib/Target/X86/X86CodeEmitter.cpp b/lib/Target/X86/X86CodeEmitter.cpp
index e988a5c..d5846a0 100644
--- a/lib/Target/X86/X86CodeEmitter.cpp
+++ b/lib/Target/X86/X86CodeEmitter.cpp
@@ -301,7 +301,7 @@ bool Emitter<CodeEmitter>::gvNeedsNonLazyPtr(const GlobalValue *GV) {
 
 template<class CodeEmitter>
 void Emitter<CodeEmitter>::emitDisplacementField(const MachineOperand *RelocOp,
-                                    int DispVal, intptr_t PCAdj) {
+                                                 int DispVal, intptr_t PCAdj) {
   // If this is a simple integer displacement that doesn't require a relocation,
   // emit it now.
   if (!RelocOp) {
@@ -371,8 +371,10 @@ void Emitter<CodeEmitter>::emitMemModRMByte(const MachineInstr &MI,
   // Is a SIB byte needed?
   if ((!Is64BitMode || DispForReloc || BaseReg != 0) &&
       IndexReg.getReg() == 0 &&
-      (BaseReg == 0 || getX86RegNum(BaseReg) != N86::ESP)) {
-    if (BaseReg == 0) {  // Just a displacement?
+      (BaseReg == 0 || BaseReg == X86::RIP ||
+       getX86RegNum(BaseReg) != N86::ESP)) {
+    if (BaseReg == 0 ||
+        BaseReg == X86::RIP) {  // Just a displacement?
       // Emit special case [disp32] encoding
       MCE.emitByte(ModRMByte(0, RegOpcodeField, 5));
       
diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index 2bcfd76..8a21b35 100644
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -396,8 +396,7 @@ bool X86FastISel::X86SelectAddress(Value *V, X86AddressMode &AM, bool isCall) {
           // Constant-offset addressing.
           Disp += CI->getSExtValue() * S;
         } else if (IndexReg == 0 &&
-                   (!AM.GV ||
-                    !getTargetMachine()->symbolicAddressesAreRIPRel()) &&
+                   (!AM.GV || !Subtarget->isPICStyleRIPRel()) &&
                    (S == 1 || S == 2 || S == 4 || S == 8)) {
           // Scaled-index addressing.
           Scale = S;
@@ -432,7 +431,7 @@ bool X86FastISel::X86SelectAddress(Value *V, X86AddressMode &AM, bool isCall) {
       return false;
 
     // RIP-relative addresses can't have additional register operands.
-    if (getTargetMachine()->symbolicAddressesAreRIPRel() &&
+    if (Subtarget->isPICStyleRIPRel() &&
         (AM.Base.Reg != 0 || AM.IndexReg != 0))
       return false;
 
@@ -443,6 +442,7 @@ bool X86FastISel::X86SelectAddress(Value *V, X86AddressMode &AM, bool isCall) {
 
     // Set up the basic address.
     AM.GV = GV;
+    
     if (!isCall &&
         TM.getRelocationModel() == Reloc::PIC_ &&
         !Subtarget->is64Bit())
@@ -481,12 +481,16 @@ bool X86FastISel::X86SelectAddress(Value *V, X86AddressMode &AM, bool isCall) {
 
       // Prevent loading GV stub multiple times in same MBB.
       LocalValueMap[V] = AM.Base.Reg;
+    } else if (Subtarget->isPICStyleRIPRel()) {
+      // Use rip-relative addressing if we can.
+      AM.Base.Reg = X86::RIP;
     }
+    
     return true;
   }
 
   // If all else fails, try to materialize the value in a register.
-  if (!AM.GV || !getTargetMachine()->symbolicAddressesAreRIPRel()) {
+  if (!AM.GV || !Subtarget->isPICStyleRIPRel()) {
     if (AM.Base.Reg == 0) {
       AM.Base.Reg = getRegForValue(V);
       return AM.Base.Reg != 0;
@@ -1140,12 +1144,10 @@ bool X86FastISel::X86SelectCall(Instruction *I) {
     return false;
   unsigned CalleeOp = 0;
   GlobalValue *GV = 0;
-  if (CalleeAM.Base.Reg != 0) {
-    assert(CalleeAM.GV == 0);
-    CalleeOp = CalleeAM.Base.Reg;
-  } else if (CalleeAM.GV != 0) {
-    assert(CalleeAM.GV != 0);
+  if (CalleeAM.GV != 0) {
     GV = CalleeAM.GV;
+  } else if (CalleeAM.Base.Reg != 0) {
+    CalleeOp = CalleeAM.Base.Reg;
   } else
     return false;
 
@@ -1493,15 +1495,22 @@ unsigned X86FastISel::TargetMaterializeConstant(Constant *C) {
   
   // x86-32 PIC requires a PIC base register for constant pools.
   unsigned PICBase = 0;
-  if (TM.getRelocationModel() == Reloc::PIC_ &&
-      !Subtarget->is64Bit())
-    PICBase = getInstrInfo()->getGlobalBaseReg(&MF);
+  unsigned char OpFlag = 0;
+  if (TM.getRelocationModel() == Reloc::PIC_) {
+    if (Subtarget->isPICStyleStub()) {
+      OpFlag = X86II::MO_PIC_BASE_OFFSET;
+      PICBase = getInstrInfo()->getGlobalBaseReg(&MF);
+    } else if (Subtarget->isPICStyleGOT()) {
+      OpFlag = X86II::MO_GOTOFF;
+      PICBase = getInstrInfo()->getGlobalBaseReg(&MF);
+    }
+  }
 
   // Create the load from the constant pool.
   unsigned MCPOffset = MCP.getConstantPoolIndex(C, Align);
   unsigned ResultReg = createResultReg(RC);
-  addConstantPoolReference(BuildMI(MBB, DL, TII.get(Opc), ResultReg), MCPOffset,
-                           PICBase);
+  addConstantPoolReference(BuildMI(MBB, DL, TII.get(Opc), ResultReg),
+                           MCPOffset, PICBase, OpFlag);
 
   return ResultReg;
 }
diff --git a/lib/Target/X86/X86ISelDAGToDAG.cpp b/lib/Target/X86/X86ISelDAGToDAG.cpp
index 9cedafc..1336177 100644
--- a/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -65,7 +65,6 @@ namespace {
       int FrameIndex;
     } Base;
 
-    bool isRIPRel;     // RIP as base?
     unsigned Scale;
     SDValue IndexReg; 
     int32_t Disp;
@@ -75,15 +74,35 @@ namespace {
     const char *ES;
     int JT;
     unsigned Align;    // CP alignment.
+    unsigned char SymbolFlags;  // X86II::MO_*
 
     X86ISelAddressMode()
-      : BaseType(RegBase), isRIPRel(false), Scale(1), IndexReg(), Disp(0),
-        Segment(), GV(0), CP(0), ES(0), JT(-1), Align(0) {
+      : BaseType(RegBase), Scale(1), IndexReg(), Disp(0),
+        Segment(), GV(0), CP(0), ES(0), JT(-1), Align(0), SymbolFlags(0) {
     }
 
     bool hasSymbolicDisplacement() const {
       return GV != 0 || CP != 0 || ES != 0 || JT != -1;
     }
+    
+    bool hasBaseOrIndexReg() const {
+      return IndexReg.getNode() != 0 || Base.Reg.getNode() != 0;
+    }
+    
+    /// isRIPRelative - Return true if this addressing mode is already RIP
+    /// relative.
+    bool isRIPRelative() const {
+      if (BaseType != RegBase) return false;
+      if (RegisterSDNode *RegNode =
+            dyn_cast_or_null<RegisterSDNode>(Base.Reg.getNode()))
+        return RegNode->getReg() == X86::RIP;
+      return false;
+    }
+    
+    void setBaseReg(SDValue Reg) {
+      BaseType = RegBase;
+      Base.Reg = Reg;
+    }
 
     void dump() {
       cerr << "X86ISelAddressMode " << this << "\n";
@@ -91,7 +110,7 @@ namespace {
               if (Base.Reg.getNode() != 0) Base.Reg.getNode()->dump(); 
               else cerr << "nul";
       cerr << " Base.FrameIndex " << Base.FrameIndex << "\n";
-      cerr << "isRIPRel " << isRIPRel << " Scale" << Scale << "\n";
+      cerr << " Scale" << Scale << "\n";
       cerr << "IndexReg ";
               if (IndexReg.getNode() != 0) IndexReg.getNode()->dump();
               else cerr << "nul"; 
@@ -200,14 +219,15 @@ namespace {
       // These are 32-bit even in 64-bit mode since RIP relative offset
       // is 32-bit.
       if (AM.GV)
-        Disp = CurDAG->getTargetGlobalAddress(AM.GV, MVT::i32, AM.Disp);
+        Disp = CurDAG->getTargetGlobalAddress(AM.GV, MVT::i32, AM.Disp,
+                                              AM.SymbolFlags);
       else if (AM.CP)
         Disp = CurDAG->getTargetConstantPool(AM.CP, MVT::i32,
-                                             AM.Align, AM.Disp);
+                                             AM.Align, AM.Disp, AM.SymbolFlags);
       else if (AM.ES)
-        Disp = CurDAG->getTargetExternalSymbol(AM.ES, MVT::i32);
+        Disp = CurDAG->getTargetExternalSymbol(AM.ES, MVT::i32, AM.SymbolFlags);
       else if (AM.JT != -1)
-        Disp = CurDAG->getTargetJumpTable(AM.JT, MVT::i32);
+        Disp = CurDAG->getTargetJumpTable(AM.JT, MVT::i32, AM.SymbolFlags);
       else
         Disp = CurDAG->getTargetConstant(AM.Disp, MVT::i32);
 
@@ -683,61 +703,80 @@ bool X86DAGToDAGISel::MatchLoad(SDValue N, X86ISelAddressMode &AM) {
   return true;
 }
 
+/// MatchWrapper - Try to match X86ISD::Wrapper and X86ISD::WrapperRIP nodes
+/// into an addressing mode.  These wrap things that will resolve down into a
+/// symbol reference.  If no match is possible, this returns true, otherwise it
+/// returns false.  
 bool X86DAGToDAGISel::MatchWrapper(SDValue N, X86ISelAddressMode &AM) {
-  bool SymbolicAddressesAreRIPRel =
-    getTargetMachine().symbolicAddressesAreRIPRel();
-  bool is64Bit = Subtarget->is64Bit();
-  DOUT << "Wrapper: 64bit " << is64Bit;
-  DOUT << " AM "; DEBUG(AM.dump()); DOUT << "\n";
-
-  // Under X86-64 non-small code model, GV (and friends) are 64-bits.
-  if (is64Bit && (TM.getCodeModel() != CodeModel::Small))
-    return true;
-
-  // Base and index reg must be 0 in order to use rip as base.
-  bool canUsePICRel = !AM.Base.Reg.getNode() && !AM.IndexReg.getNode();
-  if (is64Bit && !canUsePICRel && SymbolicAddressesAreRIPRel)
-    return true;
-
+  // If the addressing mode already has a symbol as the displacement, we can
+  // never match another symbol.
   if (AM.hasSymbolicDisplacement())
     return true;
-  // If value is available in a register both base and index components have
-  // been picked, we can't fit the result available in the register in the
-  // addressing mode. Duplicate GlobalAddress or ConstantPool as displacement.
 
   SDValue N0 = N.getOperand(0);
-  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(N0)) {
-    uint64_t Offset = G->getOffset();
-    if (!is64Bit || isInt32(AM.Disp + Offset)) {
-      GlobalValue *GV = G->getGlobal();
-      bool isRIPRel = SymbolicAddressesAreRIPRel;
-      if (N0.getOpcode() == llvm::ISD::TargetGlobalTLSAddress) {
-        TLSModel::Model model =
-          getTLSModel (GV, TM.getRelocationModel());
-        if (is64Bit && model == TLSModel::InitialExec)
-          isRIPRel = true;
-      }
-      AM.GV = GV;
-      AM.Disp += Offset;
-      AM.isRIPRel = isRIPRel;
-      return false;
-    }
-  } else if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(N0)) {
-    uint64_t Offset = CP->getOffset();
-    if (!is64Bit || isInt32(AM.Disp + Offset)) {
+  
+  // Handle X86-64 rip-relative addresses.  We check this before checking direct
+  // folding because RIP is preferable to non-RIP accesses.
+  if (Subtarget->is64Bit() &&
+      // Under X86-64 non-small code model, GV (and friends) are 64-bits, so
+      // they cannot be folded into immediate fields.
+      // FIXME: This can be improved for kernel and other models?
+      TM.getCodeModel() == CodeModel::Small &&
+      
+      // Base and index reg must be 0 in order to use %rip as base and lowering
+      // must allow RIP.
+      !AM.hasBaseOrIndexReg() && N.getOpcode() == X86ISD::WrapperRIP) {
+  
+    if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(N0)) {
+      int64_t Offset = AM.Disp + G->getOffset();
+      if (!isInt32(Offset)) return true;
+      AM.GV = G->getGlobal();
+      AM.Disp = Offset;
+      AM.SymbolFlags = G->getTargetFlags();
+    } else if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(N0)) {
+      int64_t Offset = AM.Disp + CP->getOffset();
+      if (!isInt32(Offset)) return true;
       AM.CP = CP->getConstVal();
       AM.Align = CP->getAlignment();
-      AM.Disp += Offset;
-      AM.isRIPRel = SymbolicAddressesAreRIPRel;
-      return false;
+      AM.Disp = Offset;
+      AM.SymbolFlags = CP->getTargetFlags();
+    } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(N0)) {
+      AM.ES = S->getSymbol();
+      AM.SymbolFlags = S->getTargetFlags();
+    } else {
+      JumpTableSDNode *J = cast<JumpTableSDNode>(N0);
+      AM.JT = J->getIndex();
+      AM.SymbolFlags = J->getTargetFlags();
     }
-  } else if (ExternalSymbolSDNode *S =dyn_cast<ExternalSymbolSDNode>(N0)) {
-    AM.ES = S->getSymbol();
-    AM.isRIPRel = SymbolicAddressesAreRIPRel;
+  
+    if (N.getOpcode() == X86ISD::WrapperRIP)
+      AM.setBaseReg(CurDAG->getRegister(X86::RIP, MVT::i64));
     return false;
-  } else if (JumpTableSDNode *J = dyn_cast<JumpTableSDNode>(N0)) {
-    AM.JT = J->getIndex();
-    AM.isRIPRel = SymbolicAddressesAreRIPRel;
+  }
+
+  // Handle the case when globals fit in our immediate field: This is true for
+  // X86-32 always and X86-64 when in -static -mcmodel=small mode.  In 64-bit
+  // mode, this results in a non-RIP-relative computation.
+  if (!Subtarget->is64Bit() ||
+      (TM.getCodeModel() == CodeModel::Small &&
+       TM.getRelocationModel() == Reloc::Static)) {
+    if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(N0)) {
+      AM.GV = G->getGlobal();
+      AM.Disp += G->getOffset();
+      AM.SymbolFlags = G->getTargetFlags();
+    } else if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(N0)) {
+      AM.CP = CP->getConstVal();
+      AM.Align = CP->getAlignment();
+      AM.Disp += CP->getOffset();
+      AM.SymbolFlags = CP->getTargetFlags();
+    } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(N0)) {
+      AM.ES = S->getSymbol();
+      AM.SymbolFlags = S->getTargetFlags();
+    } else {
+      JumpTableSDNode *J = cast<JumpTableSDNode>(N0);
+      AM.JT = J->getIndex();
+      AM.SymbolFlags = J->getTargetFlags();
+    }
     return false;
   }
 
@@ -756,12 +795,19 @@ bool X86DAGToDAGISel::MatchAddress(SDValue N, X86ISelAddressMode &AM,
   if (Depth > 5)
     return MatchAddressBase(N, AM);
   
+  // If this is already a %rip relative address, we can only merge immediates
+  // into it.  Instead of handling this in every case, we handle it here.
   // RIP relative addressing: %rip + 32-bit displacement!
-  if (AM.isRIPRel) {
-    if (!AM.ES && AM.JT != -1 && N.getOpcode() == ISD::Constant) {
-      uint64_t Val = cast<ConstantSDNode>(N)->getSExtValue();
-      if (!is64Bit || isInt32(AM.Disp + Val)) {
-        AM.Disp += Val;
+  if (AM.isRIPRelative()) {
+    // FIXME: JumpTable and ExternalSymbol address currently don't like
+    // displacements.  It isn't very important, but this should be fixed for
+    // consistency.
+    if (!AM.ES && AM.JT != -1) return true;
+    
+    if (ConstantSDNode *Cst = dyn_cast<ConstantSDNode>(N)) {
+      int64_t Val = AM.Disp + Cst->getSExtValue();
+      if (isInt32(Val)) {
+        AM.Disp = Val;
         return false;
       }
     }
@@ -785,6 +831,7 @@ bool X86DAGToDAGISel::MatchAddress(SDValue N, X86ISelAddressMode &AM,
     break;
 
   case X86ISD::Wrapper:
+  case X86ISD::WrapperRIP:
     if (!MatchWrapper(N, AM))
       return false;
     break;
@@ -804,7 +851,7 @@ bool X86DAGToDAGISel::MatchAddress(SDValue N, X86ISelAddressMode &AM,
     break;
 
   case ISD::SHL:
-    if (AM.IndexReg.getNode() != 0 || AM.Scale != 1 || AM.isRIPRel)
+    if (AM.IndexReg.getNode() != 0 || AM.Scale != 1)
       break;
       
     if (ConstantSDNode
@@ -845,8 +892,7 @@ bool X86DAGToDAGISel::MatchAddress(SDValue N, X86ISelAddressMode &AM,
     // X*[3,5,9] -> X+X*[2,4,8]
     if (AM.BaseType == X86ISelAddressMode::RegBase &&
         AM.Base.Reg.getNode() == 0 &&
-        AM.IndexReg.getNode() == 0 &&
-        !AM.isRIPRel) {
+        AM.IndexReg.getNode() == 0) {
       if (ConstantSDNode
             *CN = dyn_cast<ConstantSDNode>(N.getNode()->getOperand(1)))
         if (CN->getZExtValue() == 3 || CN->getZExtValue() == 5 ||
@@ -895,7 +941,7 @@ bool X86DAGToDAGISel::MatchAddress(SDValue N, X86ISelAddressMode &AM,
       break;
     }
     // Test if the index field is free for use.
-    if (AM.IndexReg.getNode() || AM.isRIPRel) {
+    if (AM.IndexReg.getNode() || AM.isRIPRelative()) {
       AM = Backup;
       break;
     }
@@ -966,8 +1012,7 @@ bool X86DAGToDAGISel::MatchAddress(SDValue N, X86ISelAddressMode &AM,
     // the add.
     if (AM.BaseType == X86ISelAddressMode::RegBase &&
         !AM.Base.Reg.getNode() &&
-        !AM.IndexReg.getNode() &&
-        !AM.isRIPRel) {
+        !AM.IndexReg.getNode()) {
       AM.Base.Reg = N.getNode()->getOperand(0);
       AM.IndexReg = N.getNode()->getOperand(1);
       AM.Scale = 1;
@@ -1006,9 +1051,6 @@ bool X86DAGToDAGISel::MatchAddress(SDValue N, X86ISelAddressMode &AM,
     // Scale must not be used already.
     if (AM.IndexReg.getNode() != 0 || AM.Scale != 1) break;
 
-    // Not when RIP is used as the base.
-    if (AM.isRIPRel) break;
-
     SDValue X = Shift.getOperand(0);
     ConstantSDNode *C2 = dyn_cast<ConstantSDNode>(N.getOperand(1));
     ConstantSDNode *C1 = dyn_cast<ConstantSDNode>(Shift.getOperand(1));
@@ -1130,7 +1172,7 @@ bool X86DAGToDAGISel::MatchAddressBase(SDValue N, X86ISelAddressMode &AM) {
   // Is the base register already occupied?
   if (AM.BaseType != X86ISelAddressMode::RegBase || AM.Base.Reg.getNode()) {
     // If so, check to see if the scale index register is set.
-    if (AM.IndexReg.getNode() == 0 && !AM.isRIPRel) {
+    if (AM.IndexReg.getNode() == 0) {
       AM.IndexReg = N;
       AM.Scale = 1;
       return false;
@@ -1157,7 +1199,7 @@ bool X86DAGToDAGISel::SelectAddr(SDValue Op, SDValue N, SDValue &Base,
   if (AvoidDupAddrCompute && !N.hasOneUse()) {
     unsigned Opcode = N.getOpcode();
     if (Opcode != ISD::Constant && Opcode != ISD::FrameIndex &&
-        Opcode != X86ISD::Wrapper) {
+        Opcode != X86ISD::Wrapper && Opcode != X86ISD::WrapperRIP) {
       // If we are able to fold N into addressing mode, then we'll allow it even
       // if N has multiple uses. In general, addressing computation is used as
       // addresses by all of its uses. But watch out for CopyToReg uses, that
@@ -1307,7 +1349,8 @@ bool X86DAGToDAGISel::SelectTLSADDRAddr(SDValue Op, SDValue N, SDValue &Base,
   AM.GV = GA->getGlobal();
   AM.Disp += GA->getOffset();
   AM.Base.Reg = CurDAG->getRegister(0, N.getValueType());
-  
+  AM.SymbolFlags = GA->getTargetFlags();
+
   if (N.getValueType() == MVT::i32) {
     AM.Scale = 1;
     AM.IndexReg = CurDAG->getRegister(X86::EBX, MVT::i32);
@@ -1687,7 +1730,8 @@ SDNode *X86DAGToDAGISel::Select(SDValue N) {
       
       // If N2 is not Wrapper(decriptor) then the llvm.declare is mangled
       // somehow, just ignore it.
-      if (N2.getOpcode() != X86ISD::Wrapper) {
+      if (N2.getOpcode() != X86ISD::Wrapper &&
+          N2.getOpcode() != X86ISD::WrapperRIP) {
         ReplaceUses(N.getValue(0), Chain);
         return NULL;
       }
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 8d0ea66..9614e69 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -19,6 +19,7 @@
 #include "llvm/CallingConv.h"
 #include "llvm/Constants.h"
 #include "llvm/DerivedTypes.h"
+#include "llvm/GlobalAlias.h"
 #include "llvm/GlobalVariable.h"
 #include "llvm/Function.h"
 #include "llvm/Intrinsics.h"
@@ -4311,21 +4312,102 @@ X86TargetLowering::LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) {
 SDValue
 X86TargetLowering::LowerConstantPool(SDValue Op, SelectionDAG &DAG) {
   ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
-  // FIXME there isn't really any debug info here, should come from the parent
-  DebugLoc dl = CP->getDebugLoc();
+  
+  // In PIC mode (unless we're in RIPRel PIC mode) we add an offset to the
+  // global base reg.
+  unsigned char OpFlag = 0;
+  unsigned WrapperKind = X86ISD::Wrapper;
+  if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
+    if (Subtarget->isPICStyleStub())
+      OpFlag = X86II::MO_PIC_BASE_OFFSET;
+    else if (Subtarget->isPICStyleGOT())
+      OpFlag = X86II::MO_GOTOFF;
+    else if (Subtarget->isPICStyleRIPRel() &&
+             getTargetMachine().getCodeModel() == CodeModel::Small)
+      WrapperKind = X86ISD::WrapperRIP;
+  }
+  
   SDValue Result = DAG.getTargetConstantPool(CP->getConstVal(), getPointerTy(),
-                                             CP->getAlignment());
-  Result = DAG.getNode(X86ISD::Wrapper, dl, getPointerTy(), Result);
+                                             CP->getAlignment(),
+                                             CP->getOffset(), OpFlag);
+  DebugLoc DL = CP->getDebugLoc();
+  Result = DAG.getNode(WrapperKind, DL, getPointerTy(), Result);
+  // With PIC, the address is actually $g + Offset.
+  if (OpFlag) {
+    Result = DAG.getNode(ISD::ADD, DL, getPointerTy(),
+                         DAG.getNode(X86ISD::GlobalBaseReg,
+                                     DebugLoc::getUnknownLoc(), getPointerTy()),
+                         Result);
+  }
+
+  return Result;
+}
+
+SDValue X86TargetLowering::LowerJumpTable(SDValue Op, SelectionDAG &DAG) {
+  JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
+  
+  // In PIC mode (unless we're in RIPRel PIC mode) we add an offset to the
+  // global base reg.
+  unsigned char OpFlag = 0;
+  unsigned WrapperKind = X86ISD::Wrapper;
+  if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
+    if (Subtarget->isPICStyleStub())
+      OpFlag = X86II::MO_PIC_BASE_OFFSET;
+    else if (Subtarget->isPICStyleGOT())
+      OpFlag = X86II::MO_GOTOFF;
+    else if (Subtarget->isPICStyleRIPRel())
+      WrapperKind = X86ISD::WrapperRIP;
+  }
+  
+  SDValue Result = DAG.getTargetJumpTable(JT->getIndex(), getPointerTy(),
+                                          OpFlag);
+  DebugLoc DL = JT->getDebugLoc();
+  Result = DAG.getNode(WrapperKind, DL, getPointerTy(), Result);
+  
+  // With PIC, the address is actually $g + Offset.
+  if (OpFlag) {
+    Result = DAG.getNode(ISD::ADD, DL, getPointerTy(),
+                         DAG.getNode(X86ISD::GlobalBaseReg,
+                                     DebugLoc::getUnknownLoc(), getPointerTy()),
+                         Result);
+  }
+  
+  return Result;
+}
+
+SDValue
+X86TargetLowering::LowerExternalSymbol(SDValue Op, SelectionDAG &DAG) {
+  const char *Sym = cast<ExternalSymbolSDNode>(Op)->getSymbol();
+  
+  // In PIC mode (unless we're in RIPRel PIC mode) we add an offset to the
+  // global base reg.
+  unsigned char OpFlag = 0;
+  unsigned WrapperKind = X86ISD::Wrapper;
+  if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
+    if (Subtarget->isPICStyleStub())
+      OpFlag = X86II::MO_PIC_BASE_OFFSET;
+    else if (Subtarget->isPICStyleGOT())
+      OpFlag = X86II::MO_GOTOFF;
+    else if (Subtarget->isPICStyleRIPRel())
+      WrapperKind = X86ISD::WrapperRIP;
+  }
+  
+  SDValue Result = DAG.getTargetExternalSymbol(Sym, getPointerTy(), OpFlag);
+  
+  DebugLoc DL = Op.getDebugLoc();
+  Result = DAG.getNode(WrapperKind, DL, getPointerTy(), Result);
+  
+  
   // With PIC, the address is actually $g + Offset.
   if (getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
       !Subtarget->isPICStyleRIPRel()) {
-    Result = DAG.getNode(ISD::ADD, dl, getPointerTy(),
+    Result = DAG.getNode(ISD::ADD, DL, getPointerTy(),
                          DAG.getNode(X86ISD::GlobalBaseReg,
                                      DebugLoc::getUnknownLoc(),
                                      getPointerTy()),
                          Result);
   }
-
+  
   return Result;
 }
 
@@ -4343,9 +4425,29 @@ X86TargetLowering::LowerGlobalAddress(const GlobalValue *GV, DebugLoc dl,
   if (!IsPic && !ExtraLoadRequired && isInt32(Offset)) {
     Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), Offset);
     Offset = 0;
-  } else
-    Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), 0);
-  Result = DAG.getNode(X86ISD::Wrapper, dl, getPointerTy(), Result);
+  } else {
+    unsigned char OpFlags = 0;
+    
+    if (Subtarget->isPICStyleRIPRel() &&
+        getTargetMachine().getRelocationModel() != Reloc::Static) {
+      if (ExtraLoadRequired)
+        OpFlags = X86II::MO_GOTPCREL;
+    } else if (Subtarget->isPICStyleGOT() &&
+               getTargetMachine().getRelocationModel() == Reloc::PIC_) {
+      if (ExtraLoadRequired)
+        OpFlags = X86II::MO_GOT;
+      else
+        OpFlags = X86II::MO_GOTOFF;
+    }
+    
+    Result = DAG.getTargetGlobalAddress(GV, getPointerTy(), 0, OpFlags);
+  }
+  
+  if (Subtarget->isPICStyleRIPRel() &&
+      getTargetMachine().getCodeModel() == CodeModel::Small)
+    Result = DAG.getNode(X86ISD::WrapperRIP, dl, getPointerTy(), Result);
+  else
+    Result = DAG.getNode(X86ISD::Wrapper, dl, getPointerTy(), Result);
 
   // With PIC, the address is actually $g + Offset.
   if (IsPic && !Subtarget->isPICStyleRIPRel()) {
@@ -4381,12 +4483,14 @@ X86TargetLowering::LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) {
 
 static SDValue
 GetTLSADDR(SelectionDAG &DAG, SDValue Chain, GlobalAddressSDNode *GA,
-           SDValue *InFlag, const MVT PtrVT, unsigned ReturnReg) {
+           SDValue *InFlag, const MVT PtrVT, unsigned ReturnReg,
+           unsigned char OperandFlags) {
   SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
   DebugLoc dl = GA->getDebugLoc();
   SDValue TGA = DAG.getTargetGlobalAddress(GA->getGlobal(),
                                            GA->getValueType(0),
-                                           GA->getOffset());
+                                           GA->getOffset(),
+                                           OperandFlags);
   if (InFlag) {
     SDValue Ops[] = { Chain,  TGA, *InFlag };
     Chain = DAG.getNode(X86ISD::TLSADDR, dl, NodeTys, Ops, 3);
@@ -4410,14 +4514,15 @@ LowerToTLSGeneralDynamicModel32(GlobalAddressSDNode *GA, SelectionDAG &DAG,
                                                  PtrVT), InFlag);
   InFlag = Chain.getValue(1);
 
-  return GetTLSADDR(DAG, Chain, GA, &InFlag, PtrVT, X86::EAX);
+  return GetTLSADDR(DAG, Chain, GA, &InFlag, PtrVT, X86::EAX, X86II::MO_TLSGD);
 }
 
 // Lower ISD::GlobalTLSAddress using the "general dynamic" model, 64 bit
 static SDValue
 LowerToTLSGeneralDynamicModel64(GlobalAddressSDNode *GA, SelectionDAG &DAG,
                                 const MVT PtrVT) {
-  return GetTLSADDR(DAG, DAG.getEntryNode(), GA, NULL, PtrVT, X86::RAX);
+  return GetTLSADDR(DAG, DAG.getEntryNode(), GA, NULL, PtrVT,
+                    X86::RAX, X86II::MO_TLSGD);
 }
 
 // Lower ISD::GlobalTLSAddress using the "initial exec" (for no-pic) or
@@ -4435,11 +4540,26 @@ static SDValue LowerToTLSExecModel(GlobalAddressSDNode *GA, SelectionDAG &DAG,
   SDValue ThreadPointer = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Base,
                                       NULL, 0);
 
+  unsigned char OperandFlags = 0;
+  // Most TLS accesses are not RIP relative, even on x86-64.  One exception is
+  // initialexec.
+  unsigned WrapperKind = X86ISD::Wrapper;
+  if (model == TLSModel::LocalExec) {
+    OperandFlags = is64Bit ? X86II::MO_TPOFF : X86II::MO_NTPOFF;
+  } else if (is64Bit) {
+    assert(model == TLSModel::InitialExec);
+    OperandFlags = X86II::MO_GOTTPOFF;
+    WrapperKind = X86ISD::WrapperRIP;
+  } else {
+    assert(model == TLSModel::InitialExec);
+    OperandFlags = X86II::MO_INDNTPOFF;
+  }
+  
   // emit "addl x@ntpoff,%eax" (local exec) or "addl x@indntpoff,%eax" (initial
   // exec)
   SDValue TGA = DAG.getTargetGlobalAddress(GA->getGlobal(), GA->getValueType(0),
-                                           GA->getOffset());
-  SDValue Offset = DAG.getNode(X86ISD::Wrapper, dl, PtrVT, TGA);
+                                           GA->getOffset(), OperandFlags);
+  SDValue Offset = DAG.getNode(WrapperKind, dl, PtrVT, TGA);
 
   if (model == TLSModel::InitialExec)
     Offset = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Offset,
@@ -4457,72 +4577,33 @@ X86TargetLowering::LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) {
   assert(Subtarget->isTargetELF() &&
          "TLS not implemented for non-ELF targets");
   GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
-  GlobalValue *GV = GA->getGlobal();
-  TLSModel::Model model =
-    getTLSModel (GV, getTargetMachine().getRelocationModel());
-  if (Subtarget->is64Bit()) {
-    switch (model) {
-    case TLSModel::GeneralDynamic:
-    case TLSModel::LocalDynamic: // not implemented
+  const GlobalValue *GV = GA->getGlobal();
+  
+  // If GV is an alias then use the aliasee for determining
+  // thread-localness.
+  if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
+    GV = GA->resolveAliasedGlobal(false);
+  
+  TLSModel::Model model = getTLSModel(GV,
+                                      getTargetMachine().getRelocationModel());
+  
+  switch (model) {
+  case TLSModel::GeneralDynamic:
+  case TLSModel::LocalDynamic: // not implemented
+    if (Subtarget->is64Bit())
       return LowerToTLSGeneralDynamicModel64(GA, DAG, getPointerTy());
-
-    case TLSModel::InitialExec:
-    case TLSModel::LocalExec:
-      return LowerToTLSExecModel(GA, DAG, getPointerTy(), model, true);
-    }
-  } else {
-    switch (model) {
-    case TLSModel::GeneralDynamic:
-    case TLSModel::LocalDynamic: // not implemented
-      return LowerToTLSGeneralDynamicModel32(GA, DAG, getPointerTy());
-
-    case TLSModel::InitialExec:
-    case TLSModel::LocalExec:
-      return LowerToTLSExecModel(GA, DAG, getPointerTy(), model, false);
-    }
+    return LowerToTLSGeneralDynamicModel32(GA, DAG, getPointerTy());
+    
+  case TLSModel::InitialExec:
+  case TLSModel::LocalExec:
+    return LowerToTLSExecModel(GA, DAG, getPointerTy(), model,
+                               Subtarget->is64Bit());
   }
+  
   assert(0 && "Unreachable");
   return SDValue();
 }
 
-SDValue
-X86TargetLowering::LowerExternalSymbol(SDValue Op, SelectionDAG &DAG) {
-  // FIXME there isn't really any debug info here
-  DebugLoc dl = Op.getDebugLoc();
-  const char *Sym = cast<ExternalSymbolSDNode>(Op)->getSymbol();
-  SDValue Result = DAG.getTargetExternalSymbol(Sym, getPointerTy());
-  Result = DAG.getNode(X86ISD::Wrapper, dl, getPointerTy(), Result);
-  // With PIC, the address is actually $g + Offset.
-  if (getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
-      !Subtarget->isPICStyleRIPRel()) {
-    Result = DAG.getNode(ISD::ADD, dl, getPointerTy(),
-                         DAG.getNode(X86ISD::GlobalBaseReg,
-                                     DebugLoc::getUnknownLoc(),
-                                     getPointerTy()),
-                         Result);
-  }
-
-  return Result;
-}
-
-SDValue X86TargetLowering::LowerJumpTable(SDValue Op, SelectionDAG &DAG) {
-  JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
-  // FIXME there isn't really any debug into here
-  DebugLoc dl = JT->getDebugLoc();
-  SDValue Result = DAG.getTargetJumpTable(JT->getIndex(), getPointerTy());
-  Result = DAG.getNode(X86ISD::Wrapper, dl, getPointerTy(), Result);
-  // With PIC, the address is actually $g + Offset.
-  if (getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
-      !Subtarget->isPICStyleRIPRel()) {
-    Result = DAG.getNode(ISD::ADD, dl, getPointerTy(),
-                         DAG.getNode(X86ISD::GlobalBaseReg,
-                                     DebugLoc::getUnknownLoc(),
-                                     getPointerTy()),
-                         Result);
-  }
-
-  return Result;
-}
 
 /// LowerShift - Lower SRA_PARTS and friends, which return two i32 values and
 /// take a 2 x i32 value to shift plus a shift amount.
@@ -6779,6 +6860,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::REP_MOVS:           return "X86ISD::REP_MOVS";
   case X86ISD::GlobalBaseReg:      return "X86ISD::GlobalBaseReg";
   case X86ISD::Wrapper:            return "X86ISD::Wrapper";
+  case X86ISD::WrapperRIP:         return "X86ISD::WrapperRIP";
   case X86ISD::PEXTRB:             return "X86ISD::PEXTRB";
   case X86ISD::PEXTRW:             return "X86ISD::PEXTRW";
   case X86ISD::INSERTPS:           return "X86ISD::INSERTPS";
diff --git a/lib/Target/X86/X86Instr64bit.td b/lib/Target/X86/X86Instr64bit.td
index 063913f..472ba4c 100644
--- a/lib/Target/X86/X86Instr64bit.td
+++ b/lib/Target/X86/X86Instr64bit.td
@@ -45,7 +45,8 @@ def lea64_32mem : Operand<i32> {
 // Complex Pattern Definitions.
 //
 def lea64addr : ComplexPattern<i64, 4, "SelectLEAAddr",
-                        [add, mul, X86mul_imm, shl, or, frameindex, X86Wrapper],
+                        [add, mul, X86mul_imm, shl, or, frameindex, X86Wrapper,
+                         X86WrapperRIP],
                         []>;
 
 def tls64addr : ComplexPattern<i64, 4, "SelectTLSADDRAddr",
@@ -1418,6 +1419,9 @@ def : Pat<(i64 (X86Wrapper tglobaladdr :$dst)),
 def : Pat<(i64 (X86Wrapper texternalsym:$dst)),
           (MOV64ri texternalsym:$dst)>, Requires<[NotSmallCode]>;
 
+// If we have small model and -static mode, it is safe to store global addresses
+// directly as immediates.  FIXME: This is really a hack, the 'imm' predicate
+// should handle this sort of thing.
 def : Pat<(store (i64 (X86Wrapper tconstpool:$src)), addr:$dst),
           (MOV64mi32 addr:$dst, tconstpool:$src)>,
           Requires<[SmallCode, IsStatic]>;
@@ -1431,6 +1435,23 @@ def : Pat<(store (i64 (X86Wrapper texternalsym:$src)), addr:$dst),
           (MOV64mi32 addr:$dst, texternalsym:$src)>,
           Requires<[SmallCode, IsStatic]>;
 
+// If we have small model and -static mode, it is safe to store global addresses
+// directly as immediates.  FIXME: This is really a hack, the 'imm' predicate
+// should handle this sort of thing.
+def : Pat<(store (i64 (X86WrapperRIP tconstpool:$src)), addr:$dst),
+          (MOV64mi32 addr:$dst, tconstpool:$src)>,
+          Requires<[SmallCode, IsStatic]>;
+def : Pat<(store (i64 (X86WrapperRIP tjumptable:$src)), addr:$dst),
+          (MOV64mi32 addr:$dst, tjumptable:$src)>,
+          Requires<[SmallCode, IsStatic]>;
+def : Pat<(store (i64 (X86WrapperRIP tglobaladdr:$src)), addr:$dst),
+          (MOV64mi32 addr:$dst, tglobaladdr:$src)>,
+          Requires<[SmallCode, IsStatic]>;
+def : Pat<(store (i64 (X86WrapperRIP texternalsym:$src)), addr:$dst),
+          (MOV64mi32 addr:$dst, texternalsym:$src)>,
+          Requires<[SmallCode, IsStatic]>;
+
+
 // Calls
 // Direct PC relative function call for small code model. 32-bit displacement
 // sign extended to 64-bit.
diff --git a/lib/Target/X86/X86InstrBuilder.h b/lib/Target/X86/X86InstrBuilder.h
index 39504cd..b50dd65 100644
--- a/lib/Target/X86/X86InstrBuilder.h
+++ b/lib/Target/X86/X86InstrBuilder.h
@@ -157,10 +157,10 @@ addFrameReference(const MachineInstrBuilder &MIB, int FI, int Offset = 0) {
 ///
 inline const MachineInstrBuilder &
 addConstantPoolReference(const MachineInstrBuilder &MIB, unsigned CPI,
-                         unsigned GlobalBaseReg = 0) {
+                         unsigned GlobalBaseReg, unsigned char OpFlags) {
   //FIXME: factor this
   return MIB.addReg(GlobalBaseReg).addImm(1).addReg(0)
-    .addConstantPoolIndex(CPI).addReg(0);
+    .addConstantPoolIndex(CPI, 0, OpFlags).addReg(0);
 }
 
 } // End llvm namespace
diff --git a/lib/Target/X86/X86InstrInfo.cpp b/lib/Target/X86/X86InstrInfo.cpp
index 8a9b7c9..21f71ec 100644
--- a/lib/Target/X86/X86InstrInfo.cpp
+++ b/lib/Target/X86/X86InstrInfo.cpp
@@ -18,6 +18,7 @@
 #include "X86MachineFunctionInfo.h"
 #include "X86Subtarget.h"
 #include "X86TargetMachine.h"
+#include "llvm/GlobalVariable.h"
 #include "llvm/DerivedTypes.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
@@ -28,7 +29,6 @@
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/Target/TargetAsmInfo.h"
-
 using namespace llvm;
 
 namespace {
@@ -781,6 +781,29 @@ static bool regIsPICBase(unsigned BaseReg, const MachineRegisterInfo &MRI) {
 static inline bool isGVStub(GlobalValue *GV, X86TargetMachine &TM) {
   return TM.getSubtarget<X86Subtarget>().GVRequiresExtraLoad(GV, TM, false);
 }
+
+/// CanRematLoadWithDispOperand - Return true if a load with the specified
+/// operand is a candidate for remat: for this to be true we need to know that
+/// the load will always return the same value, even if moved.
+static bool CanRematLoadWithDispOperand(const MachineOperand &MO,
+                                        X86TargetMachine &TM) {
+  // Loads from constant pool entries can be remat'd.
+  if (MO.isCPI()) return true;
+  
+  // We can remat globals in some cases.
+  if (MO.isGlobal()) {
+    // If this is a load of a stub, not of the global, we can remat it.  This
+    // access will always return the address of the global.
+    if (isGVStub(MO.getGlobal(), TM))
+      return true;
+    
+    // If the global itself is constant, we can remat the load.
+    if (GlobalVariable *GV = dyn_cast<GlobalVariable>(MO.getGlobal()))
+      if (GV->isConstant())
+        return true;
+  }
+  return false;
+}
  
 bool
 X86InstrInfo::isReallyTriviallyReMaterializable(const MachineInstr *MI) const {
@@ -802,11 +825,9 @@ X86InstrInfo::isReallyTriviallyReMaterializable(const MachineInstr *MI) const {
       if (MI->getOperand(1).isReg() &&
           MI->getOperand(2).isImm() &&
           MI->getOperand(3).isReg() && MI->getOperand(3).getReg() == 0 &&
-          (MI->getOperand(4).isCPI() ||
-           (MI->getOperand(4).isGlobal() &&
-            isGVStub(MI->getOperand(4).getGlobal(), TM)))) {
+          CanRematLoadWithDispOperand(MI->getOperand(4), TM)) {
         unsigned BaseReg = MI->getOperand(1).getReg();
-        if (BaseReg == 0)
+        if (BaseReg == 0 || BaseReg == X86::RIP)
           return true;
         // Allow re-materialization of PIC load.
         if (!ReMatPICStubLoad && MI->getOperand(4).isGlobal())
@@ -3190,9 +3211,8 @@ unsigned X86InstrInfo::GetInstSizeInBytes(const MachineInstr *MI) const {
   bool IsPIC = (TM.getRelocationModel() == Reloc::PIC_);
   bool Is64BitMode = TM.getSubtargetImpl()->is64Bit();
   unsigned Size = GetInstSizeWithDesc(*MI, &Desc, IsPIC, Is64BitMode);
-  if (Desc.getOpcode() == X86::MOVPC32r) {
+  if (Desc.getOpcode() == X86::MOVPC32r)
     Size += GetInstSizeWithDesc(*MI, &get(X86::POP32r), IsPIC, Is64BitMode);
-  }
   return Size;
 }
 
@@ -3220,17 +3240,17 @@ unsigned X86InstrInfo::getGlobalBaseReg(MachineFunction *MF) const {
   const TargetInstrInfo *TII = TM.getInstrInfo();
   // Operand of MovePCtoStack is completely ignored by asm printer. It's
   // only used in JIT code emission as displacement to pc.
-  BuildMI(FirstMBB, MBBI, DL, TII->get(X86::MOVPC32r), PC)
-    .addImm(0);
+  BuildMI(FirstMBB, MBBI, DL, TII->get(X86::MOVPC32r), PC).addImm(0);
   
   // If we're using vanilla 'GOT' PIC style, we should use relative addressing
-  // not to pc, but to _GLOBAL_ADDRESS_TABLE_ external
+  // not to pc, but to _GLOBAL_OFFSET_TABLE_ external.
   if (TM.getRelocationModel() == Reloc::PIC_ &&
       TM.getSubtarget<X86Subtarget>().isPICStyleGOT()) {
-    GlobalBaseReg =
-      RegInfo.createVirtualRegister(X86::GR32RegisterClass);
+    GlobalBaseReg = RegInfo.createVirtualRegister(X86::GR32RegisterClass);
+    // Generate addl $__GLOBAL_OFFSET_TABLE_ + [.-piclabel], %some_register
     BuildMI(FirstMBB, MBBI, DL, TII->get(X86::ADD32ri), GlobalBaseReg)
-      .addReg(PC).addExternalSymbol("_GLOBAL_OFFSET_TABLE_");
+      .addReg(PC).addExternalSymbol("_GLOBAL_OFFSET_TABLE_", 0,
+                                    X86II::MO_GOT_ABSOLUTE_ADDRESS);
   } else {
     GlobalBaseReg = PC;
   }
diff --git a/lib/Target/X86/X86InstrInfo.h b/lib/Target/X86/X86InstrInfo.h
index e09769e..83f0194 100644
--- a/lib/Target/X86/X86InstrInfo.h
+++ b/lib/Target/X86/X86InstrInfo.h
@@ -71,7 +71,86 @@ namespace X86 {
 namespace X86II {
   enum {
     //===------------------------------------------------------------------===//
-    // Instruction types.  These are the standard/most common forms for X86
+    // X86 Specific MachineOperand flags.
+    
+    MO_NO_FLAG = 0,
+    
+    /// MO_GOT_ABSOLUTE_ADDRESS - On a symbol operand, this represents a
+    /// relocation of:
+    ///    SYMBOL_LABEL + [. - PICBASELABEL]
+    MO_GOT_ABSOLUTE_ADDRESS = 1,
+    
+    /// MO_PIC_BASE_OFFSET - On a symbol operand this indicates that the
+    /// immediate should get the value of the symbol minus the PIC base label:
+    ///    SYMBOL_LABEL - PICBASELABEL
+    MO_PIC_BASE_OFFSET = 2,
+
+    /// MO_GOT - On a symbol operand this indicates that the immediate is the
+    /// offset to the GOT entry for the symbol name from the base of the GOT.
+    ///
+    /// See the X86-64 ELF ABI supplement for more details. 
+    ///    SYMBOL_LABEL @GOT
+    MO_GOT = 3,
+    
+    /// MO_GOTOFF - On a symbol operand this indicates that the immediate is
+    /// the offset to the location of the symbol name from the base of the GOT. 
+    ///
+    /// See the X86-64 ELF ABI supplement for more details. 
+    ///    SYMBOL_LABEL @GOTOFF
+    MO_GOTOFF = 4,
+    
+    /// MO_GOTPCREL - On a symbol operand this indicates that the immediate is
+    /// offset to the GOT entry for the symbol name from the current code
+    /// location. 
+    ///
+    /// See the X86-64 ELF ABI supplement for more details. 
+    ///    SYMBOL_LABEL @GOTPCREL
+    MO_GOTPCREL = 5,
+    
+    /// MO_PLT - On a symbol operand this indicates that the immediate is
+    /// offset to the PLT entry of symbol name from the current code location. 
+    ///
+    /// See the X86-64 ELF ABI supplement for more details. 
+    ///    SYMBOL_LABEL @PLT
+    MO_PLT = 6,
+    
+    /// MO_TLSGD - On a symbol operand this indicates that the immediate is
+    /// some TLS offset.
+    ///
+    /// See 'ELF Handling for Thread-Local Storage' for more details. 
+    ///    SYMBOL_LABEL @TLSGD
+    MO_TLSGD = 7,
+    
+    /// MO_GOTTPOFF - On a symbol operand this indicates that the immediate is
+    /// some TLS offset.
+    ///
+    /// See 'ELF Handling for Thread-Local Storage' for more details. 
+    ///    SYMBOL_LABEL @GOTTPOFF
+    MO_GOTTPOFF = 8,
+   
+    /// MO_INDNTPOFF - On a symbol operand this indicates that the immediate is
+    /// some TLS offset.
+    ///
+    /// See 'ELF Handling for Thread-Local Storage' for more details. 
+    ///    SYMBOL_LABEL @INDNTPOFF
+    MO_INDNTPOFF = 9,
+    
+    /// MO_TPOFF - On a symbol operand this indicates that the immediate is
+    /// some TLS offset.
+    ///
+    /// See 'ELF Handling for Thread-Local Storage' for more details. 
+    ///    SYMBOL_LABEL @TPOFF
+    MO_TPOFF = 10,
+    
+    /// MO_NTPOFF - On a symbol operand this indicates that the immediate is
+    /// some TLS offset.
+    ///
+    /// See 'ELF Handling for Thread-Local Storage' for more details. 
+    ///    SYMBOL_LABEL @NTPOFF
+    MO_NTPOFF = 11,
+    
+    //===------------------------------------------------------------------===//
+    // Instruction encodings.  These are the standard/most common forms for X86
     // instructions.
     //
 
diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td
index 2d8f55f..a6b0880 100644
--- a/lib/Target/X86/X86InstrInfo.td
+++ b/lib/Target/X86/X86InstrInfo.td
@@ -50,9 +50,9 @@ def SDTX86atomicBinary : SDTypeProfile<2, 3, [SDTCisInt<0>, SDTCisInt<1>,
                                 SDTCisPtrTy<2>, SDTCisInt<3>,SDTCisInt<4>]>;
 def SDTX86Ret     : SDTypeProfile<0, -1, [SDTCisVT<0, i16>]>;
 
-def SDT_X86CallSeqStart : SDCallSeqStart<[ SDTCisVT<0, i32> ]>;
-def SDT_X86CallSeqEnd   : SDCallSeqEnd<[ SDTCisVT<0, i32>,
-                                         SDTCisVT<1, i32> ]>;
+def SDT_X86CallSeqStart : SDCallSeqStart<[SDTCisVT<0, i32>]>;
+def SDT_X86CallSeqEnd   : SDCallSeqEnd<[SDTCisVT<0, i32>,
+                                        SDTCisVT<1, i32>]>;
 
 def SDT_X86Call   : SDTypeProfile<0, -1, [SDTCisVT<0, iPTR>]>;
 
@@ -236,6 +236,10 @@ def HasSSE3      : Predicate<"Subtarget->hasSSE3()">;
 def HasSSSE3     : Predicate<"Subtarget->hasSSSE3()">;
 def HasSSE41     : Predicate<"Subtarget->hasSSE41()">;
 def HasSSE42     : Predicate<"Subtarget->hasSSE42()">;
+def HasSSE4A     : Predicate<"Subtarget->hasSSE4A()">;
+def HasAVX       : Predicate<"Subtarget->hasAVX()">;
+def HasFMA3      : Predicate<"Subtarget->hasFMA3()">;
+def HasFMA4      : Predicate<"Subtarget->hasFMA4()">;
 def FPStackf32   : Predicate<"!Subtarget->hasSSE1()">;
 def FPStackf64   : Predicate<"!Subtarget->hasSSE2()">;
 def In32BitMode  : Predicate<"!Subtarget->is64Bit()">;
diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index 43fadc2..b79a006 100644
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -481,11 +481,11 @@ def MMX_PMOVMSKBrr : MMXI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR64:$src),
 
 // Misc.
 let Uses = [EDI] in
-def MMX_MASKMOVQ : MMXI<0xF7, MRMDestMem, (outs), (ins VR64:$src, VR64:$mask),
+def MMX_MASKMOVQ : MMXI<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
                         "maskmovq\t{$mask, $src|$src, $mask}",
                         [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, EDI)]>;
 let Uses = [RDI] in
-def MMX_MASKMOVQ64: MMXI64<0xF7, MRMDestMem, (outs), (ins VR64:$src, VR64:$mask),
+def MMX_MASKMOVQ64: MMXI64<0xF7, MRMSrcReg, (outs), (ins VR64:$src, VR64:$mask),
                            "maskmovq\t{$mask, $src|$src, $mask}",
                            [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)]>;
 
diff --git a/lib/Target/X86/X86RegisterInfo.td b/lib/Target/X86/X86RegisterInfo.td
index d552cb3..996baa0 100644
--- a/lib/Target/X86/X86RegisterInfo.td
+++ b/lib/Target/X86/X86RegisterInfo.td
@@ -439,7 +439,7 @@ def GR32 : RegisterClass<"X86", [i32], 32,
 
 def GR64 : RegisterClass<"X86", [i64], 64, 
                          [RAX, RCX, RDX, RSI, RDI, R8, R9, R10, R11,
-                          RBX, R14, R15, R12, R13, RBP, RSP]> {
+                          RBX, R14, R15, R12, R13, RBP, RSP, RIP]> {
   let SubRegClassList = [GR8, GR8, GR16, GR32];
   let MethodProtos = [{
     iterator allocation_order_end(const MachineFunction &MF) const;
@@ -453,9 +453,9 @@ def GR64 : RegisterClass<"X86", [i64], 64,
       if (!Subtarget.is64Bit())
         return begin();  // None of these are allocatable in 32-bit.
       if (RI->hasFP(MF)) // Does the function dedicate RBP to being a frame ptr?
-        return end()-2;  // If so, don't allocate RSP or RBP
+        return end()-3;  // If so, don't allocate RIP, RSP or RBP
       else
-        return end()-1;  // If not, just don't allocate RSP
+        return end()-2;  // If not, just don't allocate RIP or RSP
     }
   }];
 }
diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp
index 56983ce..8506fa6 100644
--- a/lib/Target/X86/X86Subtarget.cpp
+++ b/lib/Target/X86/X86Subtarget.cpp
@@ -207,6 +207,10 @@ void X86Subtarget::AutoDetectSubtargetFeatures() {
 
   bool IsIntel = memcmp(text.c, "GenuineIntel", 12) == 0;
   bool IsAMD   = !IsIntel && memcmp(text.c, "AuthenticAMD", 12) == 0;
+
+  HasFMA3 = IsIntel && ((ECX >> 12) & 0x1);
+  HasAVX = ((ECX >> 28) & 0x1);
+
   if (IsIntel || IsAMD) {
     // Determine if bit test memory instructions are slow.
     unsigned Family = 0;
@@ -217,6 +221,7 @@ void X86Subtarget::AutoDetectSubtargetFeatures() {
     X86::GetCpuIDAndInfo(0x80000001, &EAX, &EBX, &ECX, &EDX);
     HasX86_64 = (EDX >> 29) & 0x1;
     HasSSE4A = IsAMD && ((ECX >> 6) & 0x1);
+    HasFMA4 = IsAMD && ((ECX >> 16) & 0x1);
   }
 }
 
@@ -342,6 +347,10 @@ X86Subtarget::X86Subtarget(const Module &M, const std::string &FS, bool is64Bit)
   , X86SSELevel(NoMMXSSE)
   , X863DNowLevel(NoThreeDNow)
   , HasX86_64(false)
+  , HasSSE4A(false)
+  , HasAVX(false)
+  , HasFMA3(false)
+  , HasFMA4(false)
   , IsBTMemSlow(false)
   , DarwinVers(0)
   , IsLinux(false)
diff --git a/lib/Target/X86/X86Subtarget.h b/lib/Target/X86/X86Subtarget.h
index 694b0eb..f4f6cce 100644
--- a/lib/Target/X86/X86Subtarget.h
+++ b/lib/Target/X86/X86Subtarget.h
@@ -64,12 +64,21 @@ protected:
   ///
   bool HasX86_64;
 
-  /// IsBTMemSlow - True if BT (bit test) of memory instructions are slow.
-  bool IsBTMemSlow;
-  
   /// HasSSE4A - True if the processor supports SSE4A instructions.
   bool HasSSE4A;
 
+  /// HasAVX - Target has AVX instructions
+  bool HasAVX;
+
+  /// HasFMA3 - Target has 3-operand fused multiply-add
+  bool HasFMA3;
+
+  /// HasFMA4 - Target has 4-operand fused multiply-add
+  bool HasFMA4;
+
+  /// IsBTMemSlow - True if BT (bit test) of memory instructions are slow.
+  bool IsBTMemSlow;
+  
   /// DarwinVers - Nonzero if this is a darwin platform: the numeric
   /// version of the platform, e.g. 8 = 10.4 (Tiger), 9 = 10.5 (Leopard), etc.
   unsigned char DarwinVers; // Is any darwin-x86 platform.
@@ -133,6 +142,9 @@ public:
   bool hasSSE4A() const { return HasSSE4A; }
   bool has3DNow() const { return X863DNowLevel >= ThreeDNow; }
   bool has3DNowA() const { return X863DNowLevel >= ThreeDNowA; }
+  bool hasAVX() const { return hasAVX(); }
+  bool hasFMA3() const { return HasFMA3; }
+  bool hasFMA4() const { return HasFMA4; }
 
   bool isBTMemSlow() const { return IsBTMemSlow; }
 
diff --git a/lib/Target/X86/X86TargetMachine.cpp b/lib/Target/X86/X86TargetMachine.cpp
index 53c46c3..67dcd01 100644
--- a/lib/Target/X86/X86TargetMachine.cpp
+++ b/lib/Target/X86/X86TargetMachine.cpp
@@ -36,10 +36,8 @@ X("x86",    "32-bit X86: Pentium-Pro and above");
 static RegisterTarget<X86_64TargetMachine>
 Y("x86-64", "64-bit X86: EM64T and AMD64");
 
-// Force static initialization when called from llvm/InitializeAllTargets.h
-namespace llvm {
-  void InitializeX86Target() { }
-}
+// Force static initialization.
+extern "C" void LLVMInitializeX86Target() { }
 
 // No assembler printer by default
 X86TargetMachine::AsmPrinterCtorFn X86TargetMachine::AsmPrinterCtor = 0;
@@ -222,7 +220,8 @@ bool X86TargetMachine::addAssemblyEmitter(PassManagerBase &PM,
   // On Darwin, override 64-bit static relocation to pic_ since the
   // assembler doesn't support it.
   if (DefRelocModel == Reloc::Static &&
-      Subtarget.isTargetDarwin() && Subtarget.is64Bit())
+      Subtarget.isTargetDarwin() && Subtarget.is64Bit() &&
+      getCodeModel() == CodeModel::Small)
     setRelocationModel(Reloc::PIC_);
 
   assert(AsmPrinterCtor && "AsmPrinter was not linked in");
@@ -319,11 +318,3 @@ bool X86TargetMachine::addSimpleCodeEmitter(PassManagerBase &PM,
   return false;
 }
 
-/// symbolicAddressesAreRIPRel - Return true if symbolic addresses are
-/// RIP-relative on this machine, taking into consideration the relocation
-/// model and subtarget. RIP-relative addresses cannot have a separate
-/// base or index register.
-bool X86TargetMachine::symbolicAddressesAreRIPRel() const {
-  return getRelocationModel() != Reloc::Static &&
-         Subtarget.isPICStyleRIPRel();
-}
diff --git a/lib/Target/X86/X86TargetMachine.h b/lib/Target/X86/X86TargetMachine.h
index ecc1d39..ba73ca8 100644
--- a/lib/Target/X86/X86TargetMachine.h
+++ b/lib/Target/X86/X86TargetMachine.h
@@ -91,12 +91,6 @@ public:
   virtual bool addSimpleCodeEmitter(PassManagerBase &PM,
                                     CodeGenOpt::Level OptLevel,
                                     bool DumpAsm, JITCodeEmitter &JCE);
-
-  /// symbolicAddressesAreRIPRel - Return true if symbolic addresses are
-  /// RIP-relative on this machine, taking into consideration the relocation
-  /// model and subtarget. RIP-relative addresses cannot have a separate
-  /// base or index register.
-  bool symbolicAddressesAreRIPRel() const;
 };
 
 /// X86_32TargetMachine - X86 32-bit target machine.