Add support for the PowerPC 64-bit SVR4 ABI.

The Link Register is volatile when using the 32-bit SVR4 ABI.
Make it possible to use the 64-bit SVR4 ABI.
Add non-volatile registers for the 64-bit SVR4 ABI.
Make sure r2 is a reserved register when using the 64-bit SVR4 ABI.
Update PPCFrameInfo for the 64-bit SVR4 ABI.
Add FIXME for 64-bit Darwin PPC.
Insert NOP instruction after direct function calls.
Emit official procedure descriptors.
Create TOC entries for GlobalAddress references.
Spill 64-bit non-volatile registers to the correct slots.
Only custom lower VAARG when using the 32-bit SVR4 ABI.
Use simple VASTART lowering for the 64-bit SVR4 ABI.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@79091 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/PowerPC/AsmPrinter/PPCAsmPrinter.cpp

@@ -81,13 +81,14 @@ namespace {
     };
     
     StringMap<FnStubInfo> FnStubs;
-    StringMap<std::string> GVStubs, HiddenGVStubs;
+    StringMap<std::string> GVStubs, HiddenGVStubs, TOC;
     const PPCSubtarget &Subtarget;
+    uint64_t LabelID;
   public:
     explicit PPCAsmPrinter(formatted_raw_ostream &O, TargetMachine &TM,
                            const TargetAsmInfo *T, bool V)
       : AsmPrinter(O, TM, T, V),
-        Subtarget(TM.getSubtarget<PPCSubtarget>()) {}
+        Subtarget(TM.getSubtarget<PPCSubtarget>()), LabelID(0) {}
 
     virtual const char *getPassName() const {
       return "PowerPC Assembly Printer";
@@ -310,6 +311,28 @@ namespace {
       printOperand(MI, OpNo+1);
     }
 
+    void printTOCEntryLabel(const MachineInstr *MI, unsigned OpNo) {
+      const MachineOperand &MO = MI->getOperand(OpNo);
+
+      assert(MO.getType() == MachineOperand::MO_GlobalAddress);
+
+      GlobalValue *GV = MO.getGlobal();
+
+      std::string Name = Mang->getMangledName(GV);
+
+      // Map symbol -> label of TOC entry.
+      if (TOC.count(Name) == 0) {
+        std::string Label;
+        Label += TAI->getPrivateGlobalPrefix();
+        Label += "C";
+        Label += utostr(LabelID++);
+
+        TOC[Name] = Label;
+      }
+
+      O << TOC[Name] << "@toc";
+    }
+
     void printPredicateOperand(const MachineInstr *MI, unsigned OpNo,
                                const char *Modifier);
 
@@ -330,6 +353,7 @@ namespace {
     }
 
     bool runOnMachineFunction(MachineFunction &F);
+    bool doFinalization(Module &M);
 
     void getAnalysisUsage(AnalysisUsage &AU) const {
       AU.setPreservesAll();
@@ -612,7 +636,19 @@ bool PPCLinuxAsmPrinter::runOnMachineFunction(MachineFunction &MF) {
   printVisibility(CurrentFnName, F->getVisibility());
 
   EmitAlignment(MF.getAlignment(), F);
+
+  if (Subtarget.isPPC64()) {
+    // Emit an official procedure descriptor.
+    // FIXME 64-bit SVR4: Use MCSection here?
+    O << "\t.section\t\".opd\",\"aw\"\n";
+    O << "\t.align 3\n";
     O << CurrentFnName << ":\n";
+    O << "\t.quad .L." << CurrentFnName << ",.TOC.@tocbase\n";
+    O << "\t.previous\n";
+    O << ".L." << CurrentFnName << ":\n";
+  } else {
+    O << CurrentFnName << ":\n";
+  }
 
   // Emit pre-function debug information.
   DW->BeginFunction(&MF);
@@ -731,6 +767,24 @@ void PPCLinuxAsmPrinter::PrintGlobalVariable(const GlobalVariable *GVar) {
   O << '\n';
 }
 
+bool PPCLinuxAsmPrinter::doFinalization(Module &M) {
+  const TargetData *TD = TM.getTargetData();
+
+  bool isPPC64 = TD->getPointerSizeInBits() == 64;
+
+  if (isPPC64 && !TOC.empty()) {
+    // FIXME 64-bit SVR4: Use MCSection here?
+    O << "\t.section\t\".toc\",\"aw\"\n";
+
+    for (StringMap<std::string>::iterator I = TOC.begin(), E = TOC.end();
+         I != E; ++I) {
+      O << I->second << ":\n";
+      O << "\t.tc " << I->getKeyData() << "[TC]," << I->getKeyData() << '\n';
+    }
+  }
+
+  return AsmPrinter::doFinalization(M);
+}
 
 /// runOnMachineFunction - This uses the printMachineInstruction()
 /// method to print assembly for each instruction.

lib/Target/PowerPC/PPCFrameInfo.h

@@ -31,33 +31,32 @@ public:
 
   /// getReturnSaveOffset - Return the previous frame offset to save the
   /// return address.
-  static unsigned getReturnSaveOffset(bool LP64, bool isDarwinABI) {
+  static unsigned getReturnSaveOffset(bool isPPC64, bool isDarwinABI) {
     if (isDarwinABI)
-      return LP64 ? 16 : 8;
+      return isPPC64 ? 16 : 8;
     // SVR4 ABI:
-    return 4;
+    return isPPC64 ? 16 : 4;
   }
 
   /// getFramePointerSaveOffset - Return the previous frame offset to save the
   /// frame pointer.
-  static unsigned getFramePointerSaveOffset(bool LP64, bool isDarwinABI) {
+  static unsigned getFramePointerSaveOffset(bool isPPC64, bool isDarwinABI) {
     // For the Darwin ABI:
     // Use the TOC save slot in the PowerPC linkage area for saving the frame
     // pointer (if needed.)  LLVM does not generate code that uses the TOC (R2
     // is treated as a caller saved register.)
     if (isDarwinABI)
-      return LP64 ? 40 : 20;
+      return isPPC64 ? 40 : 20;
     
-    // SVR4 ABI:
+    // SVR4 ABI: First slot in the general register save area.
-    // Save it right before the link register
     return -4U;
   }
   
   /// getLinkageSize - Return the size of the PowerPC ABI linkage area.
   ///
-  static unsigned getLinkageSize(bool LP64, bool isDarwinABI) {
+  static unsigned getLinkageSize(bool isPPC64, bool isDarwinABI) {
-    if (isDarwinABI)
+    if (isDarwinABI || isPPC64)
-      return 6 * (LP64 ? 8 : 4);
+      return 6 * (isPPC64 ? 8 : 4);
     
     // SVR4 ABI:
     return 8;
@@ -65,27 +64,27 @@ public:
 
   /// getMinCallArgumentsSize - Return the size of the minium PowerPC ABI
   /// argument area.
-  static unsigned getMinCallArgumentsSize(bool LP64, bool isDarwinABI) {
+  static unsigned getMinCallArgumentsSize(bool isPPC64, bool isDarwinABI) {
-    // For the Darwin ABI:
+    // For the Darwin ABI / 64-bit SVR4 ABI:
     // 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.
-    if (isDarwinABI)
+    if (isDarwinABI || isPPC64)
-      return 8 * (LP64 ? 8 : 4);
+      return 8 * (isPPC64 ? 8 : 4);
     
-    // SVR4 ABI:
+    // 32-bit SVR4 ABI:
     // There is no default stack allocated for the 8 first GPR arguments.
     return 0;
   }
 
   /// getMinCallFrameSize - Return the minimum size a call frame can be using
   /// the PowerPC ABI.
-  static unsigned getMinCallFrameSize(bool LP64, bool isDarwinABI) {
+  static unsigned getMinCallFrameSize(bool isPPC64, bool isDarwinABI) {
     // The call frame needs to be at least big enough for linkage and 8 args.
-    return getLinkageSize(LP64, isDarwinABI) +
+    return getLinkageSize(isPPC64, isDarwinABI) +
-           getMinCallArgumentsSize(LP64, isDarwinABI);
+           getMinCallArgumentsSize(isPPC64, isDarwinABI);
   }
 
   // With the SVR4 ABI, callee-saved registers have fixed offsets on the stack.
@@ -174,10 +173,114 @@ public:
       std::pair<unsigned, int>(PPC::V20, -192)
     };
     
+    static const std::pair<unsigned, int> Offsets64[] = {
+      // Floating-point register save area offsets.
+      std::pair<unsigned, int>(PPC::F31, -8),
+      std::pair<unsigned, int>(PPC::F30, -16),
+      std::pair<unsigned, int>(PPC::F29, -24),
+      std::pair<unsigned, int>(PPC::F28, -32),
+      std::pair<unsigned, int>(PPC::F27, -40),
+      std::pair<unsigned, int>(PPC::F26, -48),
+      std::pair<unsigned, int>(PPC::F25, -56),
+      std::pair<unsigned, int>(PPC::F24, -64),
+      std::pair<unsigned, int>(PPC::F23, -72),
+      std::pair<unsigned, int>(PPC::F22, -80),
+      std::pair<unsigned, int>(PPC::F21, -88),
+      std::pair<unsigned, int>(PPC::F20, -96),
+      std::pair<unsigned, int>(PPC::F19, -104),
+      std::pair<unsigned, int>(PPC::F18, -112),
+      std::pair<unsigned, int>(PPC::F17, -120),
+      std::pair<unsigned, int>(PPC::F16, -128),
+      std::pair<unsigned, int>(PPC::F15, -136),
+      std::pair<unsigned, int>(PPC::F14, -144),
+
+      // General register save area offsets.
+      // FIXME 64-bit SVR4: Are 32-bit registers actually allocated in 64-bit
+      //                    mode?
+      std::pair<unsigned, int>(PPC::R31, -4),
+      std::pair<unsigned, int>(PPC::R30, -12),
+      std::pair<unsigned, int>(PPC::R29, -20),
+      std::pair<unsigned, int>(PPC::R28, -28),
+      std::pair<unsigned, int>(PPC::R27, -36),
+      std::pair<unsigned, int>(PPC::R26, -44),
+      std::pair<unsigned, int>(PPC::R25, -52),
+      std::pair<unsigned, int>(PPC::R24, -60),
+      std::pair<unsigned, int>(PPC::R23, -68),
+      std::pair<unsigned, int>(PPC::R22, -76),
+      std::pair<unsigned, int>(PPC::R21, -84),
+      std::pair<unsigned, int>(PPC::R20, -92),
+      std::pair<unsigned, int>(PPC::R19, -100),
+      std::pair<unsigned, int>(PPC::R18, -108),
+      std::pair<unsigned, int>(PPC::R17, -116),
+      std::pair<unsigned, int>(PPC::R16, -124),
+      std::pair<unsigned, int>(PPC::R15, -132),
+      std::pair<unsigned, int>(PPC::R14, -140),
+
+      std::pair<unsigned, int>(PPC::X31, -8),
+      std::pair<unsigned, int>(PPC::X30, -16),
+      std::pair<unsigned, int>(PPC::X29, -24),
+      std::pair<unsigned, int>(PPC::X28, -32),
+      std::pair<unsigned, int>(PPC::X27, -40),
+      std::pair<unsigned, int>(PPC::X26, -48),
+      std::pair<unsigned, int>(PPC::X25, -56),
+      std::pair<unsigned, int>(PPC::X24, -64),
+      std::pair<unsigned, int>(PPC::X23, -72),
+      std::pair<unsigned, int>(PPC::X22, -80),
+      std::pair<unsigned, int>(PPC::X21, -88),
+      std::pair<unsigned, int>(PPC::X20, -96),
+      std::pair<unsigned, int>(PPC::X19, -104),
+      std::pair<unsigned, int>(PPC::X18, -112),
+      std::pair<unsigned, int>(PPC::X17, -120),
+      std::pair<unsigned, int>(PPC::X16, -128),
+      std::pair<unsigned, int>(PPC::X15, -136),
+      std::pair<unsigned, int>(PPC::X14, -144),
+
+      // CR save area offset.
+      // FIXME SVR4: Disable CR save area for now.
+//      std::pair<unsigned, int>(PPC::CR2, -4),
+//      std::pair<unsigned, int>(PPC::CR3, -4),
+//      std::pair<unsigned, int>(PPC::CR4, -4),
+//      std::pair<unsigned, int>(PPC::CR2LT, -4),
+//      std::pair<unsigned, int>(PPC::CR2GT, -4),
+//      std::pair<unsigned, int>(PPC::CR2EQ, -4),
+//      std::pair<unsigned, int>(PPC::CR2UN, -4),
+//      std::pair<unsigned, int>(PPC::CR3LT, -4),
+//      std::pair<unsigned, int>(PPC::CR3GT, -4),
+//      std::pair<unsigned, int>(PPC::CR3EQ, -4),
+//      std::pair<unsigned, int>(PPC::CR3UN, -4),
+//      std::pair<unsigned, int>(PPC::CR4LT, -4),
+//      std::pair<unsigned, int>(PPC::CR4GT, -4),
+//      std::pair<unsigned, int>(PPC::CR4EQ, -4),
+//      std::pair<unsigned, int>(PPC::CR4UN, -4),
+
+      // VRSAVE save area offset.
+      std::pair<unsigned, int>(PPC::VRSAVE, -4),
+
+      // Vector register save area
+      std::pair<unsigned, int>(PPC::V31, -16),
+      std::pair<unsigned, int>(PPC::V30, -32),
+      std::pair<unsigned, int>(PPC::V29, -48),
+      std::pair<unsigned, int>(PPC::V28, -64),
+      std::pair<unsigned, int>(PPC::V27, -80),
+      std::pair<unsigned, int>(PPC::V26, -96),
+      std::pair<unsigned, int>(PPC::V25, -112),
+      std::pair<unsigned, int>(PPC::V24, -128),
+      std::pair<unsigned, int>(PPC::V23, -144),
+      std::pair<unsigned, int>(PPC::V22, -160),
+      std::pair<unsigned, int>(PPC::V21, -176),
+      std::pair<unsigned, int>(PPC::V20, -192)
+    };
+    
+    if (TM.getSubtarget<PPCSubtarget>().isPPC64()) {
+      NumEntries = array_lengthof(Offsets64);
+
+      return Offsets64;
+    } else {
       NumEntries = array_lengthof(Offsets);
 
       return Offsets;
     }
+  }
 };
 
 } // End llvm namespace

lib/Target/PowerPC/PPCISelLowering.cpp

@@ -212,8 +212,9 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   // VASTART needs to be custom lowered to use the VarArgsFrameIndex
   setOperationAction(ISD::VASTART           , MVT::Other, Custom);
 
-  // VAARG is custom lowered with the SVR4 ABI
+  // VAARG is custom lowered with the 32-bit SVR4 ABI.
-  if (TM.getSubtarget<PPCSubtarget>().isSVR4ABI())
+  if (    TM.getSubtarget<PPCSubtarget>().isSVR4ABI()
+      && !TM.getSubtarget<PPCSubtarget>().isPPC64())
     setOperationAction(ISD::VAARG, MVT::Other, Custom);
   else
     setOperationAction(ISD::VAARG, MVT::Other, Expand);
@@ -419,6 +420,7 @@ const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case PPCISD::VPERM:           return "PPCISD::VPERM";
   case PPCISD::Hi:              return "PPCISD::Hi";
   case PPCISD::Lo:              return "PPCISD::Lo";
+  case PPCISD::TOC_ENTRY:       return "PPCISD::TOC_ENTRY";
   case PPCISD::DYNALLOC:        return "PPCISD::DYNALLOC";
   case PPCISD::GlobalBaseReg:   return "PPCISD::GlobalBaseReg";
   case PPCISD::SRL:             return "PPCISD::SRL";
@@ -428,6 +430,7 @@ const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
   case PPCISD::STD_32:          return "PPCISD::STD_32";
   case PPCISD::CALL_SVR4:       return "PPCISD::CALL_SVR4";
   case PPCISD::CALL_Darwin:     return "PPCISD::CALL_Darwin";
+  case PPCISD::NOP:             return "PPCISD::NOP";
   case PPCISD::MTCTR:           return "PPCISD::MTCTR";
   case PPCISD::BCTRL_Darwin:    return "PPCISD::BCTRL_Darwin";
   case PPCISD::BCTRL_SVR4:      return "PPCISD::BCTRL_SVR4";
@@ -1176,6 +1179,13 @@ SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
 
   const TargetMachine &TM = DAG.getTarget();
 
+  // 64-bit SVR4 ABI code is always position-independent.
+  // The actual address of the GlobalValue is stored in the TOC.
+  if (PPCSubTarget.isSVR4ABI() && PPCSubTarget.isPPC64()) {
+    return DAG.getNode(PPCISD::TOC_ENTRY, dl, MVT::i64, GA,
+                       DAG.getRegister(PPC::X2, MVT::i64));
+  }
+
   SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, GA, Zero);
   SDValue Lo = DAG.getNode(PPCISD::Lo, dl, PtrVT, GA, Zero);
 
@@ -1308,7 +1318,7 @@ SDValue PPCTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG,
                                         const PPCSubtarget &Subtarget) {
   DebugLoc dl = Op.getDebugLoc();
 
-  if (Subtarget.isDarwinABI()) {
+  if (Subtarget.isDarwinABI() || Subtarget.isPPC64()) {
     // vastart just stores the address of the VarArgsFrameIndex slot into the
     // memory location argument.
     EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
@@ -1317,7 +1327,7 @@ SDValue PPCTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG,
     return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1), SV, 0);
   }
 
-  // For the SVR4 ABI we follow the layout of the va_list struct.
+  // For the 32-bit SVR4 ABI we follow the layout of the va_list struct.
   // We suppose the given va_list is already allocated.
   //
   // typedef struct {
@@ -1450,21 +1460,13 @@ static bool CC_PPC_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, EVT &ValVT,
 }
 
 /// GetFPR - Get the set of FP registers that should be allocated for arguments,
-/// depending on which subtarget is selected.
+/// on Darwin.
-static const unsigned *GetFPR(const PPCSubtarget &Subtarget) {
+static const unsigned *GetFPR() {
-  if (Subtarget.isDarwinABI()) {
   static const unsigned FPR[] = {
     PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
     PPC::F8, PPC::F9, PPC::F10, PPC::F11, PPC::F12, PPC::F13
   };
-    return FPR;
-  }
 
-
-  static const unsigned FPR[] = {
-    PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
-    PPC::F8
-  };
   return FPR;
 }
 
@@ -1487,7 +1489,7 @@ PPCTargetLowering::LowerFormalArguments(SDValue Chain,
                                           &Ins,
                                         DebugLoc dl, SelectionDAG &DAG,
                                         SmallVectorImpl<SDValue> &InVals) {
-  if (PPCSubTarget.isSVR4ABI()) {
+  if (PPCSubTarget.isSVR4ABI() && !PPCSubTarget.isPPC64()) {
     return LowerFormalArguments_SVR4(Chain, CallConv, isVarArg, Ins,
                                      dl, DAG, InVals);
   } else {
@@ -1505,7 +1507,7 @@ PPCTargetLowering::LowerFormalArguments_SVR4(
                                       DebugLoc dl, SelectionDAG &DAG,
                                       SmallVectorImpl<SDValue> &InVals) {
 
-  // SVR4 ABI Stack Frame Layout:
+  // 32-bit SVR4 ABI Stack Frame Layout:
   //              +-----------------------------------+
   //        +-->  |            Back chain             |
   //        |     +-----------------------------------+
@@ -1687,8 +1689,8 @@ PPCTargetLowering::LowerFormalArguments_SVR4(
       FIN = DAG.getNode(ISD::ADD, dl, PtrOff.getValueType(), FIN, PtrOff);
     }
 
-    // FIXME SVR4: We only need to save FP argument registers if CR bit 6 is
+    // FIXME 32-bit SVR4: We only need to save FP argument registers if CR bit 6
-    // set.
+    // is set.
     
     // The double arguments are stored to the VarArgsFrameIndex
     // on the stack.
@@ -1731,7 +1733,6 @@ PPCTargetLowering::LowerFormalArguments_Darwin(
                                         &Ins,
                                       DebugLoc dl, SelectionDAG &DAG,
                                       SmallVectorImpl<SDValue> &InVals) {
-
   // TODO: add description of PPC stack frame format, or at least some docs.
   //
   MachineFunction &MF = DAG.getMachineFunction();
@@ -1756,7 +1757,7 @@ PPCTargetLowering::LowerFormalArguments_Darwin(
     PPC::X7, PPC::X8, PPC::X9, PPC::X10,
   };
 
-  static const unsigned *FPR = GetFPR(PPCSubTarget);
+  static const unsigned *FPR = GetFPR();
 
   static const unsigned VR[] = {
     PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
@@ -1986,7 +1987,7 @@ PPCTargetLowering::LowerFormalArguments_Darwin(
               GPR_idx++;
           }
           ArgOffset += 16;
-          GPR_idx = std::min(GPR_idx+4, Num_GPR_Regs);
+          GPR_idx = std::min(GPR_idx+4, Num_GPR_Regs); // FIXME correct for ppc64?
         }
         ++VR_idx;
       } else {
@@ -2262,8 +2263,8 @@ static SDValue EmitTailCallStoreFPAndRetAddr(SelectionDAG &DAG,
     Chain = DAG.getStore(Chain, dl, OldRetAddr, NewRetAddrFrIdx,
                          PseudoSourceValue::getFixedStack(NewRetAddr), 0);
 
-    // When using the SVR4 ABI there is no need to move the FP stack slot
+    // When using the 32/64-bit SVR4 ABI there is no need to move the FP stack
-    // as the FP is never overwritten.
+    // slot as the FP is never overwritten.
     if (isDarwinABI) {
       int NewFPLoc =
         SPDiff + PPCFrameInfo::getFramePointerSaveOffset(isPPC64, isDarwinABI);
@@ -2311,8 +2312,8 @@ SDValue PPCTargetLowering::EmitTailCallLoadFPAndRetAddr(SelectionDAG & DAG,
     LROpOut = DAG.getLoad(VT, dl, Chain, LROpOut, NULL, 0);
     Chain = SDValue(LROpOut.getNode(), 1);
     
-    // When using the SVR4 ABI there is no need to load the FP stack slot
+    // When using the 32/64-bit SVR4 ABI there is no need to load the FP stack
-    // as the FP is never overwritten.
+    // slot as the FP is never overwritten.
     if (isDarwinABI) {
       FPOpOut = getFramePointerFrameIndex(DAG);
       FPOpOut = DAG.getLoad(VT, dl, Chain, FPOpOut, NULL, 0);
@@ -2487,7 +2488,6 @@ PPCTargetLowering::FinishCall(unsigned CallConv, DebugLoc dl, bool isTailCall,
                               int SPDiff, unsigned NumBytes,
                               const SmallVectorImpl<ISD::InputArg> &Ins,
                               SmallVectorImpl<SDValue> &InVals) {
-
   std::vector<EVT> NodeTys;
   SmallVector<SDValue, 8> Ops;
   unsigned CallOpc = PrepareCall(DAG, Callee, InFlag, Chain, dl, SPDiff,
@@ -2529,6 +2529,19 @@ PPCTargetLowering::FinishCall(unsigned CallConv, DebugLoc dl, bool isTailCall,
   Chain = DAG.getNode(CallOpc, dl, NodeTys, &Ops[0], Ops.size());
   InFlag = Chain.getValue(1);
 
+  // Add a NOP immediately after the branch instruction when using the 64-bit
+  // SVR4 ABI. At link time, if caller and callee are in a different module and
+  // thus have a different TOC, the call will be replaced with a call to a stub
+  // function which saves the current TOC, loads the TOC of the callee and
+  // branches to the callee. The NOP will be replaced with a load instruction
+  // which restores the TOC of the caller from the TOC save slot of the current
+  // stack frame. If caller and callee belong to the same module (and have the
+  // same TOC), the NOP will remain unchanged.
+  if (!isTailCall && PPCSubTarget.isSVR4ABI()&& PPCSubTarget.isPPC64()) {
+    // Insert NOP.
+    InFlag = DAG.getNode(PPCISD::NOP, dl, MVT::Flag, InFlag);
+  }
+
   Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
                              DAG.getIntPtrConstant(BytesCalleePops, true),
                              InFlag);
@@ -2547,7 +2560,7 @@ PPCTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
                              const SmallVectorImpl<ISD::InputArg> &Ins,
                              DebugLoc dl, SelectionDAG &DAG,
                              SmallVectorImpl<SDValue> &InVals) {
-  if (PPCSubTarget.isSVR4ABI()) {
+  if (PPCSubTarget.isSVR4ABI() && !PPCSubTarget.isPPC64()) {
     return LowerCall_SVR4(Chain, Callee, CallConv, isVarArg,
                           isTailCall, Outs, Ins,
                           dl, DAG, InVals);
@@ -2567,7 +2580,7 @@ PPCTargetLowering::LowerCall_SVR4(SDValue Chain, SDValue Callee,
                                   DebugLoc dl, SelectionDAG &DAG,
                                   SmallVectorImpl<SDValue> &InVals) {
   // See PPCTargetLowering::LowerFormalArguments_SVR4() for a description
-  // of the SVR4 ABI stack frame layout.
+  // of the 32-bit SVR4 ABI stack frame layout.
 
   assert((!isTailCall ||
           (CallConv == CallingConv::Fast && PerformTailCallOpt)) &&
@@ -2846,7 +2859,7 @@ PPCTargetLowering::LowerCall_Darwin(SDValue Chain, SDValue Callee,
     PPC::X3, PPC::X4, PPC::X5, PPC::X6,
     PPC::X7, PPC::X8, PPC::X9, PPC::X10,
   };
-  static const unsigned *FPR = GetFPR(PPCSubTarget);
+  static const unsigned *FPR = GetFPR();
 
   static const unsigned VR[] = {
     PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,

lib/Target/PowerPC/PPCISelLowering.h

@@ -60,6 +60,8 @@ namespace llvm {
       /// though these are usually folded into other nodes.
       Hi, Lo,
       
+      TOC_ENTRY,
+
       /// OPRC, CHAIN = DYNALLOC(CHAIN, NEGSIZE, FRAME_INDEX)
       /// This instruction is lowered in PPCRegisterInfo::eliminateFrameIndex to
       /// compute an allocation on the stack.
@@ -84,6 +86,9 @@ namespace llvm {
       /// CALL - A direct function call.
       CALL_Darwin, CALL_SVR4,
       
+      /// NOP - Special NOP which follows 64-bit SVR4 calls.
+      NOP,
+
       /// CHAIN,FLAG = MTCTR(VAL, CHAIN[, INFLAG]) - Directly corresponds to a
       /// MTCTR instruction.
       MTCTR,

lib/Target/PowerPC/PPCInstr64Bit.td

@@ -123,6 +123,8 @@ def : Pat<(PPCcall_SVR4 (i64 tglobaladdr:$dst)),
           (BL8_ELF tglobaladdr:$dst)>;
 def : Pat<(PPCcall_SVR4 (i64 texternalsym:$dst)),
           (BL8_ELF texternalsym:$dst)>;
+def : Pat<(PPCnop),
+          (NOP)>;
 
 // Atomic operations
 let usesCustomDAGSchedInserter = 1 in {
@@ -543,6 +545,10 @@ let canFoldAsLoad = 1, PPC970_Unit = 2 in {
 def LD   : DSForm_1<58, 0, (outs G8RC:$rD), (ins memrix:$src),
                     "ld $rD, $src", LdStLD,
                     [(set G8RC:$rD, (load ixaddr:$src))]>, isPPC64;
+def LDtoc: DSForm_1<58, 0, (outs G8RC:$rD), (ins tocentry:$disp, G8RC:$reg),
+                    "ld $rD, $disp($reg)", LdStLD,
+                    [(set G8RC:$rD,
+                     (PPCtoc_entry tglobaladdr:$disp, G8RC:$reg))]>, isPPC64;
 def LDX  : XForm_1<31,  21, (outs G8RC:$rD), (ins memrr:$src),
                    "ldx $rD, $src", LdStLD,
                    [(set G8RC:$rD, (load xaddr:$src))]>, isPPC64;

lib/Target/PowerPC/PPCInstrInfo.td

@@ -53,6 +53,8 @@ def SDT_PPCTC_ret : SDTypeProfile<0, 2, [
   SDTCisPtrTy<0>, SDTCisVT<1, i32>
 ]>;
 
+def SDT_PPCnop : SDTypeProfile<0, 0, []>;
+
 //===----------------------------------------------------------------------===//
 // PowerPC specific DAG Nodes.
 //
@@ -85,6 +87,7 @@ def PPCfsel   : SDNode<"PPCISD::FSEL",
 
 def PPChi       : SDNode<"PPCISD::Hi", SDTIntBinOp, []>;
 def PPClo       : SDNode<"PPCISD::Lo", SDTIntBinOp, []>;
+def PPCtoc_entry: SDNode<"PPCISD::TOC_ENTRY", SDTIntBinOp, [SDNPMayLoad]>;
 def PPCvmaddfp  : SDNode<"PPCISD::VMADDFP", SDTFPTernaryOp, []>;
 def PPCvnmsubfp : SDNode<"PPCISD::VNMSUBFP", SDTFPTernaryOp, []>;
 
@@ -111,6 +114,7 @@ def PPCcall_Darwin : SDNode<"PPCISD::CALL_Darwin", SDT_PPCCall,
                             [SDNPHasChain, SDNPOptInFlag, SDNPOutFlag]>;
 def PPCcall_SVR4  : SDNode<"PPCISD::CALL_SVR4", SDT_PPCCall,
                            [SDNPHasChain, SDNPOptInFlag, SDNPOutFlag]>;
+def PPCnop : SDNode<"PPCISD::NOP", SDT_PPCnop, [SDNPInFlag, SDNPOutFlag]>;
 def PPCmtctr      : SDNode<"PPCISD::MTCTR", SDT_PPCCall,
                            [SDNPHasChain, SDNPOptInFlag, SDNPOutFlag]>;
 def PPCbctrl_Darwin  : SDNode<"PPCISD::BCTRL_Darwin", SDTNone,
@@ -306,6 +310,10 @@ def memrix : Operand<iPTR> {   // memri where the imm is shifted 2 bits.
   let PrintMethod = "printMemRegImmShifted";
   let MIOperandInfo = (ops i32imm:$imm, ptr_rc:$reg);
 }
+def tocentry : Operand<iPTR> {
+  let PrintMethod = "printTOCEntryLabel";
+  let MIOperandInfo = (ops i32imm:$imm);
+}
 
 // PowerPC Predicate operand.  20 = (0<<5)|20 = always, CR0 is a dummy reg
 // that doesn't matter.

lib/Target/PowerPC/PPCRegisterInfo.cpp

@@ -175,6 +175,7 @@ PPCRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
     PPC::LR,  0
   };
 
+  // 32-bit SVR4 calling convention.
   static const unsigned SVR4_CalleeSavedRegs[] = {
                         PPC::R14, PPC::R15,
     PPC::R16, PPC::R17, PPC::R18, PPC::R19,
@@ -200,7 +201,7 @@ PPCRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
     PPC::CR3LT, PPC::CR3GT, PPC::CR3EQ, PPC::CR3UN,
     PPC::CR4LT, PPC::CR4GT, PPC::CR4EQ, PPC::CR4UN,
     
-    PPC::LR,  0
+    0
   };
   // 64-bit Darwin calling convention. 
   static const unsigned Darwin64_CalleeSavedRegs[] = {
@@ -228,11 +229,40 @@ PPCRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
     PPC::LR8,  0
   };
 
+  // 64-bit SVR4 calling convention.
+  static const unsigned SVR4_64_CalleeSavedRegs[] = {
+    PPC::X14, PPC::X15,
+    PPC::X16, PPC::X17, PPC::X18, PPC::X19,
+    PPC::X20, PPC::X21, PPC::X22, PPC::X23,
+    PPC::X24, PPC::X25, PPC::X26, PPC::X27,
+    PPC::X28, PPC::X29, PPC::X30, PPC::X31,
+
+    PPC::F14, PPC::F15, PPC::F16, PPC::F17,
+    PPC::F18, PPC::F19, PPC::F20, PPC::F21,
+    PPC::F22, PPC::F23, PPC::F24, PPC::F25,
+    PPC::F26, PPC::F27, PPC::F28, PPC::F29,
+    PPC::F30, PPC::F31,
+
+    PPC::CR2, PPC::CR3, PPC::CR4,
+
+    PPC::VRSAVE,
+
+    PPC::V20, PPC::V21, PPC::V22, PPC::V23,
+    PPC::V24, PPC::V25, PPC::V26, PPC::V27,
+    PPC::V28, PPC::V29, PPC::V30, PPC::V31,
+
+    PPC::CR2LT, PPC::CR2GT, PPC::CR2EQ, PPC::CR2UN,
+    PPC::CR3LT, PPC::CR3GT, PPC::CR3EQ, PPC::CR3UN,
+    PPC::CR4LT, PPC::CR4GT, PPC::CR4EQ, PPC::CR4UN,
+
+    0
+  };
+  
   if (Subtarget.isDarwinABI())
     return Subtarget.isPPC64() ? Darwin64_CalleeSavedRegs :
                                  Darwin32_CalleeSavedRegs;
 
-  return SVR4_CalleeSavedRegs;
+  return Subtarget.isPPC64() ? SVR4_64_CalleeSavedRegs : SVR4_CalleeSavedRegs;
 }
 
 const TargetRegisterClass* const*
@@ -267,6 +297,7 @@ PPCRegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
     &PPC::GPRCRegClass, 0
   };
   
+  // 32-bit SVR4 calling convention.
   static const TargetRegisterClass * const SVR4_CalleeSavedRegClasses[] = {
                                           &PPC::GPRCRegClass,&PPC::GPRCRegClass,
     &PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,
@@ -295,7 +326,7 @@ PPCRegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
     &PPC::CRBITRCRegClass,&PPC::CRBITRCRegClass,&PPC::CRBITRCRegClass,
     &PPC::CRBITRCRegClass, 
     
-    &PPC::GPRCRegClass, 0
+    0
   };
   
   // 64-bit Darwin calling convention.
@@ -328,11 +359,44 @@ PPCRegisterInfo::getCalleeSavedRegClasses(const MachineFunction *MF) const {
     &PPC::G8RCRegClass, 0
   };
 
+  // 64-bit SVR4 calling convention.
+  static const TargetRegisterClass * const SVR4_64_CalleeSavedRegClasses[] = {
+    &PPC::G8RCRegClass,&PPC::G8RCRegClass,
+    &PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
+    &PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
+    &PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
+    &PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
+
+    &PPC::F8RCRegClass,&PPC::F8RCRegClass,&PPC::F8RCRegClass,&PPC::F8RCRegClass,
+    &PPC::F8RCRegClass,&PPC::F8RCRegClass,&PPC::F8RCRegClass,&PPC::F8RCRegClass,
+    &PPC::F8RCRegClass,&PPC::F8RCRegClass,&PPC::F8RCRegClass,&PPC::F8RCRegClass,
+    &PPC::F8RCRegClass,&PPC::F8RCRegClass,&PPC::F8RCRegClass,&PPC::F8RCRegClass,
+    &PPC::F8RCRegClass,&PPC::F8RCRegClass,
+
+    &PPC::CRRCRegClass,&PPC::CRRCRegClass,&PPC::CRRCRegClass,
+
+    &PPC::VRSAVERCRegClass,
+
+    &PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,
+    &PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,
+    &PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,
+
+    &PPC::CRBITRCRegClass,&PPC::CRBITRCRegClass,&PPC::CRBITRCRegClass,
+    &PPC::CRBITRCRegClass,
+    &PPC::CRBITRCRegClass,&PPC::CRBITRCRegClass,&PPC::CRBITRCRegClass,
+    &PPC::CRBITRCRegClass,
+    &PPC::CRBITRCRegClass,&PPC::CRBITRCRegClass,&PPC::CRBITRCRegClass,
+    &PPC::CRBITRCRegClass,
+
+    0
+  };
+  
   if (Subtarget.isDarwinABI())
     return Subtarget.isPPC64() ? Darwin64_CalleeSavedRegClasses :
                                  Darwin32_CalleeSavedRegClasses;
   
-  return SVR4_CalleeSavedRegClasses;
+  return Subtarget.isPPC64() ? SVR4_64_CalleeSavedRegClasses
+                             : SVR4_CalleeSavedRegClasses;
 }
 
 // needsFP - Return true if the specified function should have a dedicated frame
@@ -364,9 +428,9 @@ BitVector PPCRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
     Reserved.set(PPC::R13); // Small Data Area pointer register
   }
   
-  // On PPC64, r13 is the thread pointer. Never allocate this register. Note
+  // On PPC64, r13 is the thread pointer. Never allocate this register.
-  // that this is over conservative, as it also prevents allocation of R31 when
+  // Note that this is over conservative, as it also prevents allocation of R31
-  // the FP is not needed.
+  // when the FP is not needed.
   if (Subtarget.isPPC64()) {
     Reserved.set(PPC::R13);
     Reserved.set(PPC::R31);
@@ -378,6 +442,11 @@ BitVector PPCRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
     Reserved.set(PPC::X1);
     Reserved.set(PPC::X13);
     Reserved.set(PPC::X31);
+
+    // The 64-bit SVR4 ABI reserves r2 for the TOC pointer.
+    if (Subtarget.isSVR4ABI()) {
+      Reserved.set(PPC::X2);
+    }
   }
 
   if (needsFP(MF))
@@ -911,7 +980,7 @@ void PPCRegisterInfo::determineFrameLayout(MachineFunction &MF) const {
   // don't have a frame pointer, calls, or dynamic alloca then we do not need
   // to adjust the stack pointer (we fit in the Red Zone).
   bool DisableRedZone = MF.getFunction()->hasFnAttr(Attribute::NoRedZone);
-  // FIXME SVR4 The SVR4 ABI has no red zone.
+  // FIXME SVR4 The 32-bit SVR4 ABI has no red zone.
   if (!DisableRedZone &&
       FrameSize <= 224 &&                          // Fits in red zone.
       !MFI->hasVarSizedObjects() &&                // No dynamic alloca.
@@ -1017,16 +1086,19 @@ PPCRegisterInfo::processFunctionBeforeFrameFinalized(MachineFunction &MF)
   }
   
   unsigned MinGPR = PPC::R31;
+  unsigned MinG8R = PPC::X31;
   unsigned MinFPR = PPC::F31;
   unsigned MinVR = PPC::V31;
   
   bool HasGPSaveArea = false;
+  bool HasG8SaveArea = false;
   bool HasFPSaveArea = false;
   bool HasCRSaveArea = false;
   bool HasVRSAVESaveArea = false;
   bool HasVRSaveArea = false;
   
   SmallVector<CalleeSavedInfo, 18> GPRegs;
+  SmallVector<CalleeSavedInfo, 18> G8Regs;
   SmallVector<CalleeSavedInfo, 18> FPRegs;
   SmallVector<CalleeSavedInfo, 18> VRegs;
   
@@ -1042,6 +1114,14 @@ PPCRegisterInfo::processFunctionBeforeFrameFinalized(MachineFunction &MF)
       if (Reg < MinGPR) {
         MinGPR = Reg;
       }
+    } else if (RC == PPC::G8RCRegisterClass) {
+      HasG8SaveArea = true;
+
+      G8Regs.push_back(CSI[i]);
+
+      if (Reg < MinG8R) {
+        MinG8R = Reg;
+      }
     } else if (RC == PPC::F8RCRegisterClass) {
       HasFPSaveArea = true;
       
@@ -1104,7 +1184,7 @@ PPCRegisterInfo::processFunctionBeforeFrameFinalized(MachineFunction &MF)
   
   // General register save area starts right below the Floating-point
   // register save area.
-  if (HasGPSaveArea) {
+  if (HasGPSaveArea || HasG8SaveArea) {
     // Move general register save area spill slots down, taking into account
     // the size of the Floating-point register save area.
     for (unsigned i = 0, e = GPRegs.size(); i != e; ++i) {
@@ -1113,7 +1193,22 @@ PPCRegisterInfo::processFunctionBeforeFrameFinalized(MachineFunction &MF)
       FFI->setObjectOffset(FI, LowerBound + FFI->getObjectOffset(FI));
     }
     
-    LowerBound -= (31 - getRegisterNumbering(MinGPR) + 1) * 4;
+    // Move general register save area spill slots down, taking into account
+    // the size of the Floating-point register save area.
+    for (unsigned i = 0, e = G8Regs.size(); i != e; ++i) {
+      int FI = G8Regs[i].getFrameIdx();
+
+      FFI->setObjectOffset(FI, LowerBound + FFI->getObjectOffset(FI));
+    }
+
+    unsigned MinReg = std::min<unsigned>(getRegisterNumbering(MinGPR),
+                                         getRegisterNumbering(MinG8R));
+
+    if (Subtarget.isPPC64()) {
+      LowerBound -= (31 - MinReg + 1) * 8;
+    } else {
+      LowerBound -= (31 - MinReg + 1) * 4;
+    }
   }
   
   // The CR save area is below the general register save area.

lib/Target/PowerPC/PPCRegisterInfo.td

@@ -280,7 +280,8 @@ def GPRC : RegisterClass<"PPC", [i32], 32,
   let MethodBodies = [{
     GPRCClass::iterator
     GPRCClass::allocation_order_begin(const MachineFunction &MF) const {
-      // In Linux, r2 is reserved for the OS.
+      // 32-bit SVR4 ABI: r2 is reserved for the OS.
+      // 64-bit SVR4 ABI: r2 is reserved for the TOC pointer.
       if (!MF.getTarget().getSubtarget<PPCSubtarget>().isDarwin())
         return begin()+1;
 
@@ -291,7 +292,7 @@ def GPRC : RegisterClass<"PPC", [i32], 32,
       // On PPC64, r13 is the thread pointer.  Never allocate this register.
       // Note that this is overconservative, as it also prevents allocation of
       // R31 when the FP is not needed.
-      // When using the SVR4 ABI, r13 is reserved for the Small Data Area
+      // When using the 32-bit SVR4 ABI, r13 is reserved for the Small Data Area
       // pointer.
       const PPCSubtarget &Subtarget
         = MF.getTarget().getSubtarget<PPCSubtarget>();
@@ -318,6 +319,10 @@ def G8RC : RegisterClass<"PPC", [i64], 64,
   let MethodBodies = [{
     G8RCClass::iterator
     G8RCClass::allocation_order_begin(const MachineFunction &MF) const {
+      // 64-bit SVR4 ABI: r2 is reserved for the TOC pointer.
+      if (!MF.getTarget().getSubtarget<PPCSubtarget>().isDarwin())
+        return begin()+1;
+
       return begin();
     }
     G8RCClass::iterator
@@ -372,4 +377,3 @@ def CRRC : RegisterClass<"PPC", [i32], 32, [CR0, CR1, CR5, CR6, CR7, CR2,
 def CTRRC : RegisterClass<"PPC", [i32], 32, [CTR]>;
 def CTRRC8 : RegisterClass<"PPC", [i64], 64, [CTR8]>;
 def VRSAVERC : RegisterClass<"PPC", [i32], 32, [VRSAVE]>;
-

lib/Target/PowerPC/PPCSubtarget.h

@@ -138,8 +138,9 @@ public:
   /// getDarwinVers - Return the darwin version number, 8 = tiger, 9 = leopard.
   unsigned getDarwinVers() const { return DarwinVers; }
 
-  bool isDarwinABI() const { return isDarwin() || IsPPC64; }
+  bool isDarwinABI() const { return isDarwin(); }
-  bool isSVR4ABI() const { return !isDarwin() && !IsPPC64; }
+  bool isSVR4ABI() const { return !isDarwin(); }
+
 };
 } // End llvm namespace