[VE] Support for PIC (global data and calls)

[simoll]

Summary

Support for PIC with tests for global variables and function calls.

[arsenm]

llvm/lib/Target/VE/VEAsmPrinter.cpp

210

Typo supporeted repeated a few times

llvm/lib/Target/VE/VEISelDAGToDAG.cpp

135

s/unsigned/Register

llvm/lib/Target/VE/VEISelLowering.cpp

698–700

use the dyn_cast value instead of checking and dyn_casting again

llvm/lib/Target/VE/VEInstrInfo.h

79

s/unsigned/Register

[simoll]

• rebased
• addressed comments

[simoll]

unsigned -> Register

[arsenm]

llvm/lib/Target/VE/VEAsmPrinter.cpp

153

Missing line

[simoll]

• API change EmitInstruction -> emitInstruction
• following coding standard for all new functions (EmitSIC -> emitSIC)

[simoll]

• rebased
• coding standard
• newline

llvm/lib/Target/VE/MCTargetDesc/VEFixupKinds.h

 namespace VE {
 enum Fixups {
   /// fixup_ve_hi32 - 32-bit fixup corresponding to foo@hi
   fixup_ve_hi32 = FirstTargetFixupKind,
 
   /// fixup_ve_lo32 - 32-bit fixup corresponding to foo@lo
   fixup_ve_lo32,
 
+  /// fixup_ve_pc_hi32 - 32-bit fixup corresponding to foo@pc_hi
+  fixup_ve_pc_hi32,
+
+  /// fixup_ve_pc_lo32 - 32-bit fixup corresponding to foo@pc_lo
+  fixup_ve_pc_lo32,
+
+  /// fixup_ve_got_hi32 - 32-bit fixup corresponding to foo@got_hi
+  fixup_ve_got_hi32,
+
+  /// fixup_ve_got_lo32 - 32-bit fixup corresponding to foo@got_lo
+  fixup_ve_got_lo32,
+
+  /// fixup_ve_gotoff_hi32 - 32-bit fixup corresponding to foo@gotoff_hi
+  fixup_ve_gotoff_hi32,
+
+  /// fixup_ve_gotoff_lo32 - 32-bit fixup corresponding to foo@gotoff_lo
+  fixup_ve_gotoff_lo32,
+
+  /// fixup_ve_plt_hi32/lo32
+  fixup_ve_plt_hi32,
+  fixup_ve_plt_lo32,
+
   // Marker
   LastTargetFixupKind,
   NumTargetFixupKinds = LastTargetFixupKind - FirstTargetFixupKind
 };
 } // namespace VE
 } // namespace llvm
 
 #endif

llvm/lib/Target/VE/MCTargetDesc/VEMCExpr.h

 
 class StringRef;
 class VEMCExpr : public MCTargetExpr {
 public:
   enum VariantKind {
     VK_VE_None,
     VK_VE_HI32,
     VK_VE_LO32,
+    VK_VE_PC_HI32,
+    VK_VE_PC_LO32,
+    VK_VE_GOT_HI32,
+    VK_VE_GOT_LO32,
+    VK_VE_GOTOFF_HI32,
+    VK_VE_GOTOFF_LO32,
+    VK_VE_PLT_HI32,
+    VK_VE_PLT_LO32,
   };
 
 private:
   const VariantKind Kind;
   const MCExpr *Expr;
 
   explicit VEMCExpr(VariantKind Kind, const MCExpr *Expr)
       : Kind(Kind), Expr(Expr) {}

llvm/lib/Target/VE/MCTargetDesc/VEMCExpr.cpp

 
 bool VEMCExpr::printVariantKind(raw_ostream &OS, VariantKind Kind) {
   switch (Kind) {
   case VK_VE_None:
     return false;
 
   case VK_VE_HI32:
   case VK_VE_LO32:
+  case VK_VE_PC_HI32:
+  case VK_VE_PC_LO32:
+  case VK_VE_GOT_HI32:
+  case VK_VE_GOT_LO32:
+  case VK_VE_GOTOFF_HI32:
+  case VK_VE_GOTOFF_LO32:
+  case VK_VE_PLT_HI32:
+  case VK_VE_PLT_LO32:
     return false; // OS << "@<text>(";  break;
   }
   return true;
 }
 
 void VEMCExpr::printVariantKindSuffix(raw_ostream &OS, VariantKind Kind) {
   switch (Kind) {
   case VK_VE_None:
     break;
   case VK_VE_HI32:
     OS << "@hi";
     break;
   case VK_VE_LO32:
     OS << "@lo";
     break;
+  case VK_VE_PC_HI32:
+    OS << "@pc_hi";
+    break;
+  case VK_VE_PC_LO32:
+    OS << "@pc_lo";
+    break;
+  case VK_VE_GOT_HI32:
+    OS << "@got_hi";
+    break;
+  case VK_VE_GOT_LO32:
+    OS << "@got_lo";
+    break;
+  case VK_VE_GOTOFF_HI32:
+    OS << "@gotoff_hi";
+    break;
+  case VK_VE_GOTOFF_LO32:
+    OS << "@gotoff_lo";
+    break;
+  case VK_VE_PLT_HI32:
+    OS << "@plt_hi";
+    break;
+  case VK_VE_PLT_LO32:
+    OS << "@plt_lo";
+    break;
   }
 }
 
 VEMCExpr::VariantKind VEMCExpr::parseVariantKind(StringRef name) {
   return StringSwitch<VEMCExpr::VariantKind>(name)
       .Case("hi", VK_VE_HI32)
       .Case("lo", VK_VE_LO32)
+      .Case("pc_hi", VK_VE_PC_HI32)
+      .Case("pc_lo", VK_VE_PC_LO32)
+      .Case("got_hi", VK_VE_GOT_HI32)
+      .Case("got_lo", VK_VE_GOT_LO32)
+      .Case("gotoff_hi", VK_VE_GOTOFF_HI32)
+      .Case("gotoff_lo", VK_VE_GOTOFF_LO32)
+      .Case("plt_hi", VK_VE_PLT_HI32)
+      .Case("plt_lo", VK_VE_PLT_LO32)
       .Default(VK_VE_None);
 }
 
 VE::Fixups VEMCExpr::getFixupKind(VEMCExpr::VariantKind Kind) {
   switch (Kind) {
   default:
     llvm_unreachable("Unhandled VEMCExpr::VariantKind");
   case VK_VE_HI32:
     return VE::fixup_ve_hi32;
   case VK_VE_LO32:
     return VE::fixup_ve_lo32;
+  case VK_VE_PC_HI32:
+    return VE::fixup_ve_pc_hi32;
+  case VK_VE_PC_LO32:
+    return VE::fixup_ve_pc_lo32;
+  case VK_VE_GOT_HI32:
+    return VE::fixup_ve_got_hi32;
+  case VK_VE_GOT_LO32:
+    return VE::fixup_ve_got_lo32;
+  case VK_VE_GOTOFF_HI32:
+    return VE::fixup_ve_gotoff_hi32;
+  case VK_VE_GOTOFF_LO32:
+    return VE::fixup_ve_gotoff_lo32;
+  case VK_VE_PLT_HI32:
+    return VE::fixup_ve_plt_hi32;
+  case VK_VE_PLT_LO32:
+    return VE::fixup_ve_plt_lo32;
   }
 }
 
 bool VEMCExpr::evaluateAsRelocatableImpl(MCValue &Res,
                                          const MCAsmLayout *Layout,
                                          const MCFixup *Fixup) const {
   return getSubExpr()->evaluateAsRelocatable(Res, Layout, Fixup);
 }
 
 void VEMCExpr::visitUsedExpr(MCStreamer &Streamer) const {
   Streamer.visitUsedExpr(*getSubExpr());
 }
 
 void VEMCExpr::fixELFSymbolsInTLSFixups(MCAssembler &Asm) const {
   llvm_unreachable("TODO implement");
 }

llvm/lib/Target/VE/VEAsmPrinter.cpp

 //===-- VEAsmPrinter.cpp - VE LLVM assembly writer ------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file contains a printer that converts from our internal representation
 // of machine-dependent LLVM code to GAS-format VE assembly language.
 //
 //===----------------------------------------------------------------------===//
 
 #include "InstPrinter/VEInstPrinter.h"
+#include "MCTargetDesc/VEMCExpr.h"
 #include "MCTargetDesc/VETargetStreamer.h"
 #include "VE.h"
 #include "VEInstrInfo.h"
 #include "VETargetMachine.h"
 #include "llvm/CodeGen/AsmPrinter.h"
 #include "llvm/CodeGen/MachineInstr.h"
 #include "llvm/CodeGen/MachineModuleInfoImpls.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
class VEAsmPrinter : public AsmPrinter {
   }
 
 public:
   explicit VEAsmPrinter(TargetMachine &TM, std::unique_ptr<MCStreamer> Streamer)
       : AsmPrinter(TM, std::move(Streamer)) {}
 
   StringRef getPassName() const override { return "VE Assembly Printer"; }
 
+  void LowerGETGOTAndEmitMCInsts(const MachineInstr *MI,
+                                 const MCSubtargetInfo &STI);
+  void LowerGETFunPLTAndEmitMCInsts(const MachineInstr *MI,
+                                    const MCSubtargetInfo &STI);
+
   void EmitInstruction(const MachineInstr *MI) override;
 
   static const char *getRegisterName(unsigned RegNo) {
     return VEInstPrinter::getRegisterName(RegNo);
   }
 };
 } // end of anonymous namespace
 
+static MCOperand createVEMCOperand(VEMCExpr::VariantKind Kind, MCSymbol *Sym,
+                                   MCContext &OutContext) {
+  const MCSymbolRefExpr *MCSym = MCSymbolRefExpr::create(Sym, OutContext);
+  const VEMCExpr *expr = VEMCExpr::create(Kind, MCSym, OutContext);
+  return MCOperand::createExpr(expr);
+}
+
+static MCOperand createGOTRelExprOp(VEMCExpr::VariantKind Kind,
+                                    MCSymbol *GOTLabel, MCContext &OutContext) {
+  const MCSymbolRefExpr *GOT = MCSymbolRefExpr::create(GOTLabel, OutContext);
+  const VEMCExpr *expr = VEMCExpr::create(Kind, GOT, OutContext);
+  return MCOperand::createExpr(expr);
+}
+
+static void EmitSIC(MCStreamer &OutStreamer, MCOperand &RD,
+                    const MCSubtargetInfo &STI) {
+  MCInst SICInst;
+  SICInst.setOpcode(VE::SIC);
+  SICInst.addOperand(RD);
+  OutStreamer.EmitInstruction(SICInst, STI);
+}
+
+static void EmitLEAzzi(MCStreamer &OutStreamer, MCOperand &Imm, MCOperand &RD,
+                       const MCSubtargetInfo &STI) {
+  MCInst LEAInst;
+  LEAInst.setOpcode(VE::LEAzzi);
+  LEAInst.addOperand(RD);
+  LEAInst.addOperand(Imm);
+  OutStreamer.EmitInstruction(LEAInst, STI);
+}
+
+static void EmitLEASLzzi(MCStreamer &OutStreamer, MCOperand &Imm, MCOperand &RD,
+                         const MCSubtargetInfo &STI) {
+  MCInst LEASLInst;
+  LEASLInst.setOpcode(VE::LEASLzzi);
+  LEASLInst.addOperand(RD);
+  LEASLInst.addOperand(Imm);
+  OutStreamer.EmitInstruction(LEASLInst, STI);
+}
+
+static void EmitLEAzii(MCStreamer &OutStreamer, MCOperand &RS1, MCOperand &Imm,
+                       MCOperand &RD, const MCSubtargetInfo &STI) {
+  MCInst LEAInst;
+  LEAInst.setOpcode(VE::LEAzii);
+  LEAInst.addOperand(RD);
+  LEAInst.addOperand(RS1);
+  LEAInst.addOperand(Imm);
+  OutStreamer.EmitInstruction(LEAInst, STI);
+}
+
+static void EmitLEASLrri(MCStreamer &OutStreamer, MCOperand &RS1,
+                         MCOperand &RS2, MCOperand &Imm, MCOperand &RD,
+                         const MCSubtargetInfo &STI) {
+  MCInst LEASLInst;
+  LEASLInst.setOpcode(VE::LEASLrri);
+  LEASLInst.addOperand(RS1);
+  LEASLInst.addOperand(RS2);
+  LEASLInst.addOperand(RD);
+  LEASLInst.addOperand(Imm);
+  OutStreamer.EmitInstruction(LEASLInst, STI);
+}
+
+static void EmitBinary(MCStreamer &OutStreamer, unsigned Opcode, MCOperand &RS1,
+                       MCOperand &Src2, MCOperand &RD,
+                       const MCSubtargetInfo &STI) {
+  MCInst Inst;
+  Inst.setOpcode(Opcode);
+  Inst.addOperand(RD);
+  Inst.addOperand(RS1);
+  Inst.addOperand(Src2);
+  OutStreamer.EmitInstruction(Inst, STI);
+}
+
+static void EmitANDrm0(MCStreamer &OutStreamer, MCOperand &RS1, MCOperand &Imm,
+                       MCOperand &RD, const MCSubtargetInfo &STI) {
+  EmitBinary(OutStreamer, VE::ANDrm0, RS1, Imm, RD, STI);
+}
+
+static void EmitHiLo(MCStreamer &OutStreamer, MCSymbol *GOTSym,
+                     VEMCExpr::VariantKind HiKind, VEMCExpr::VariantKind LoKind,
+                     MCOperand &RD, MCContext &OutContext,
+                     const MCSubtargetInfo &STI) {
+
+  MCOperand hi = createVEMCOperand(HiKind, GOTSym, OutContext);
+  MCOperand lo = createVEMCOperand(LoKind, GOTSym, OutContext);
+  MCOperand ci32 = MCOperand::createImm(32);
+  EmitLEAzzi(OutStreamer, lo, RD, STI);
+  EmitANDrm0(OutStreamer, RD, ci32, RD, STI);
+  EmitLEASLzzi(OutStreamer, hi, RD, STI);
+}
+void VEAsmPrinter::LowerGETGOTAndEmitMCInsts(const MachineInstr *MI,
+                                             const MCSubtargetInfo &STI) {
+  MCSymbol *GOTLabel =
+      OutContext.getOrCreateSymbol(Twine("_GLOBAL_OFFSET_TABLE_"));
+
+  const MachineOperand &MO = MI->getOperand(0);
+  MCOperand MCRegOP = MCOperand::createReg(MO.getReg());
+
+  if (!isPositionIndependent()) {
+    // Just load the address of GOT to MCRegOP.
+    switch (TM.getCodeModel()) {
+    default:
+      llvm_unreachable("Unsupported absolute code model");
+    case CodeModel::Small:
+    case CodeModel::Medium:
+    case CodeModel::Large:
+      EmitHiLo(*OutStreamer, GOTLabel, VEMCExpr::VK_VE_HI32,
+               VEMCExpr::VK_VE_LO32, MCRegOP, OutContext, STI);
+      break;
+    }
+    return;
+  }
+
+  MCOperand RegGOT = MCOperand::createReg(VE::SX15); // GOT
+  MCOperand RegPLT = MCOperand::createReg(VE::SX16); // PLT
+
+  // lea %got, _GLOBAL_OFFSET_TABLE_@PC_LO(-24)
+  // and %got, %got, (32)0
+  // sic %plt
+  // lea.sl %got, _GLOBAL_OFFSET_TABLE_@PC_HI(%got, %plt)
+  MCOperand cim24 = MCOperand::createImm(-24);
+  MCOperand loImm =
+      createGOTRelExprOp(VEMCExpr::VK_VE_PC_LO32, GOTLabel, OutContext);
+  EmitLEAzii(*OutStreamer, cim24, loImm, MCRegOP, STI);
+  MCOperand ci32 = MCOperand::createImm(32);
+  EmitANDrm0(*OutStreamer, MCRegOP, ci32, MCRegOP, STI);
+  EmitSIC(*OutStreamer, RegPLT, STI);
+  MCOperand hiImm =
+      createGOTRelExprOp(VEMCExpr::VK_VE_PC_HI32, GOTLabel, OutContext);
+  EmitLEASLrri(*OutStreamer, RegGOT, RegPLT, hiImm, MCRegOP, STI);
+}
+
+void VEAsmPrinter::LowerGETFunPLTAndEmitMCInsts(const MachineInstr *MI,
+                                                const MCSubtargetInfo &STI) {
+  const MachineOperand &MO = MI->getOperand(0);
+  MCOperand MCRegOP = MCOperand::createReg(MO.getReg());
+  const MachineOperand &Addr = MI->getOperand(1);
+  MCSymbol *AddrSym = nullptr;
+
+  switch (Addr.getType()) {
+  default:
+    llvm_unreachable("<unknown operand type>");
+    return;
+  case MachineOperand::MO_MachineBasicBlock:
+    report_fatal_error("MBB is not supported yet");
+    return;
+  case MachineOperand::MO_ConstantPoolIndex:
+    report_fatal_error("ConstantPool is not supported yet");
+    return;
+  case MachineOperand::MO_ExternalSymbol:
+    AddrSym = GetExternalSymbolSymbol(Addr.getSymbolName());
+    break;
+  case MachineOperand::MO_GlobalAddress:
+    AddrSym = getSymbol(Addr.getGlobal());
+    break;
+  }
+
+  if (!isPositionIndependent()) {
+    llvm_unreachable("Unsupported uses of %plt in not PIC code");
+    return;
+  }
+
+  MCOperand RegPLT = MCOperand::createReg(VE::SX16); // PLT
+
+  // lea %dst, %plt_lo(func)(-24)
+  // and %dst, %dst, (32)0
+  // sic %plt                            ; FIXME: is it safe to use %plt here?
+  // lea.sl %dst, %plt_hi(func)(%dst, %plt)
+  MCOperand cim24 = MCOperand::createImm(-24);
+  MCOperand loImm =
+      createGOTRelExprOp(VEMCExpr::VK_VE_PLT_LO32, AddrSym, OutContext);
+  EmitLEAzii(*OutStreamer, cim24, loImm, MCRegOP, STI);
+  MCOperand ci32 = MCOperand::createImm(32);
+  EmitANDrm0(*OutStreamer, MCRegOP, ci32, MCRegOP, STI);
+  EmitSIC(*OutStreamer, RegPLT, STI);
+  MCOperand hiImm =
+      createGOTRelExprOp(VEMCExpr::VK_VE_PLT_HI32, AddrSym, OutContext);
+  EmitLEASLrri(*OutStreamer, MCRegOP, RegPLT, hiImm, MCRegOP, STI);
+}
+
 void VEAsmPrinter::EmitInstruction(const MachineInstr *MI) {
 
   switch (MI->getOpcode()) {
   default:
     break;
   case TargetOpcode::DBG_VALUE:
     // FIXME: Debug Value.
     return;
+  case VE::GETGOT:
+    LowerGETGOTAndEmitMCInsts(MI, getSubtargetInfo());
+    return;
+  case VE::GETFUNPLT:
+    LowerGETFunPLTAndEmitMCInsts(MI, getSubtargetInfo());
+    return;
   }
+
   MachineBasicBlock::const_instr_iterator I = MI->getIterator();
   MachineBasicBlock::const_instr_iterator E = MI->getParent()->instr_end();
   do {
     MCInst TmpInst;
     LowerVEMachineInstrToMCInst(&*I, TmpInst, *this);
     EmitToStreamer(*OutStreamer, TmpInst);
   } while ((++I != E) && I->isInsideBundle()); // Delay slot check.
 }
 
 // Force static initialization.
 extern "C" void LLVMInitializeVEAsmPrinter() {
   RegisterAsmPrinter<VEAsmPrinter> X(getTheVETarget());
 }

llvm/lib/Target/VE/VEISelDAGToDAG.cpp

public:
   bool SelectADDRri(SDValue N, SDValue &Base, SDValue &Offset);
 
   StringRef getPassName() const override {
     return "VE DAG->DAG Pattern Instruction Selection";
   }
 
   // Include the pieces autogenerated from the target description.
 #include "VEGenDAGISel.inc"
+
+private:
+  SDNode *getGlobalBaseReg();
 };
 } // end anonymous namespace
 
 bool VEDAGToDAGISel::SelectADDRrr(SDValue Addr, SDValue &R1, SDValue &R2) {
   if (Addr.getOpcode() == ISD::FrameIndex)
     return false;
   if (Addr.getOpcode() == ISD::TargetExternalSymbol ||
       Addr.getOpcode() == ISD::TargetGlobalAddress ||

 
 void VEDAGToDAGISel::Select(SDNode *N) {
   SDLoc dl(N);
   if (N->isMachineOpcode()) {
     N->setNodeId(-1);
     return; // Already selected.
   }
 
+  switch (N->getOpcode()) {
+  case VEISD::GLOBAL_BASE_REG:
+    ReplaceNode(N, getGlobalBaseReg());
+    return;
+  }
+
   SelectCode(N);
 }
 
+SDNode *VEDAGToDAGISel::getGlobalBaseReg() {
+  Register GlobalBaseReg = Subtarget->getInstrInfo()->getGlobalBaseReg(MF);
+  return CurDAG
+      ->getRegister(GlobalBaseReg, TLI->getPointerTy(CurDAG->getDataLayout()))
+      .getNode();
+}
+
 /// createVEISelDag - This pass converts a legalized DAG into a
 /// VE-specific DAG, ready for instruction scheduling.
 ///
 FunctionPass *llvm::createVEISelDag(VETargetMachine &TM) {
   return new VEDAGToDAGISel(TM);
 }

llvm/lib/Target/VE/VEISelLowering.h

 
 namespace VEISD {
 enum NodeType : unsigned {
   FIRST_NUMBER = ISD::BUILTIN_OP_END,
 
   Hi,
   Lo, // Hi/Lo operations, typically on a global address.
 
+  GETFUNPLT,       // load function address through %plt insturction
   CALL,            // A call instruction.
-  RET_FLAG, // Return with a flag operand.
+  RET_FLAG,        // Return with a flag operand.
+  GLOBAL_BASE_REG, // Global base reg for PIC.
 };
 }
 
 class VETargetLowering : public TargetLowering {
   const VESubtarget *Subtarget;
 
 public:
   VETargetLowering(const TargetMachine &TM, const VESubtarget &STI);

llvm/lib/Target/VE/VEISelLowering.cpp

SDValue VETargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   // VE needs to get address of callee function in a register
   // So, prepare to copy it to SX12 here.
 
   // If the callee is a GlobalAddress node (quite common, every direct call is)
   // turn it into a TargetGlobalAddress node so that legalize doesn't hack it.
   // Likewise ExternalSymbol -> TargetExternalSymbol.
   SDValue Callee = CLI.Callee;
 
-  assert(!isPositionIndependent() && "TODO PIC");
+  bool IsPICCall = isPositionIndependent();
+
+  // PC-relative references to external symbols should go through $stub.
+  // If so, we need to prepare GlobalBaseReg first.
+  const TargetMachine &TM = DAG.getTarget();
+  const Module *Mod = DAG.getMachineFunction().getFunction().getParent();
+  const GlobalValue *GV = nullptr;
+  auto *CalleeG = dyn_cast<GlobalAddressSDNode>(Callee);
+  if (CalleeG)
+    GV = CalleeG->getGlobal();
+  bool Local = TM.shouldAssumeDSOLocal(*Mod, GV);
+  bool UsePlt = !Local;
+  MachineFunction &MF = DAG.getMachineFunction();
 
   // Turn GlobalAddress/ExternalSymbol node into a value node
   // containing the address of them here.
-  if (isa<GlobalAddressSDNode>(Callee)) {
+  if (CalleeG) {
+    if (IsPICCall) {
+      if (UsePlt)
+        Subtarget->getInstrInfo()->getGlobalBaseReg(&MF);
+      Callee = DAG.getTargetGlobalAddress(GV, DL, PtrVT, 0, 0);
+      Callee = DAG.getNode(VEISD::GETFUNPLT, DL, PtrVT, Callee);
+    } else {
       Callee =
           makeHiLoPair(Callee, VEMCExpr::VK_VE_HI32, VEMCExpr::VK_VE_LO32, DAG);
-  } else if (isa<ExternalSymbolSDNode>(Callee)) {
+    }
+  } else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee)) {
+    if (IsPICCall) {
+      if (UsePlt)
+        Subtarget->getInstrInfo()->getGlobalBaseReg(&MF);
+      Callee = DAG.getTargetExternalSymbol(E->getSymbol(), PtrVT, 0);
+      Callee = DAG.getNode(VEISD::GETFUNPLT, DL, PtrVT, Callee);
+    } else {
       Callee =
           makeHiLoPair(Callee, VEMCExpr::VK_VE_HI32, VEMCExpr::VK_VE_LO32, DAG);
     }
+  }
 
   RegsToPass.push_back(std::make_pair(VE::SX12, Callee));
 
   for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
     CCValAssign &VA = ArgLocs[i];
     SDValue Arg = CLI.OutVals[i];
 
     // Promote the value if needed.

 #define TARGET_NODE_CASE(NAME)                                                 \
   case VEISD::NAME:                                                            \
     return "VEISD::" #NAME;
   switch ((VEISD::NodeType)Opcode) {
   case VEISD::FIRST_NUMBER:
     break;
     TARGET_NODE_CASE(Lo)
     TARGET_NODE_CASE(Hi)
+    TARGET_NODE_CASE(GETFUNPLT)
     TARGET_NODE_CASE(CALL)
     TARGET_NODE_CASE(RET_FLAG)
+    TARGET_NODE_CASE(GLOBAL_BASE_REG)
   }
 #undef TARGET_NODE_CASE
   return nullptr;
 }
 
 EVT VETargetLowering::getSetCCResultType(const DataLayout &, LLVMContext &,
                                          EVT VT) const {
   return MVT::i32;
SDValue VETargetLowering::makeHiLoPair(SDValue Op, unsigned HiTF, unsigned LoTF,
   SDValue Lo = DAG.getNode(VEISD::Lo, DL, VT, withTargetFlags(Op, LoTF, DAG));
   return DAG.getNode(ISD::ADD, DL, VT, Hi, Lo);
 }
 
 // Build SDNodes for producing an address from a GlobalAddress, ConstantPool,
 // or ExternalSymbol SDNode.
 SDValue VETargetLowering::makeAddress(SDValue Op, SelectionDAG &DAG) const {
   SDLoc DL(Op);
+  EVT PtrVT = Op.getValueType();
 
-  assert(!isPositionIndependent() && "TODO implement PIC");
+  // Handle PIC mode first. VE needs a got load for every variable!
+  if (isPositionIndependent()) {
+    // GLOBAL_BASE_REG codegen'ed with call. Inform MFI that this
+    // function has calls.
+    MachineFrameInfo &MFI = DAG.getMachineFunction().getFrameInfo();
+    MFI.setHasCalls(true);
+    auto GlobalN = dyn_cast<GlobalAddressSDNode>(Op);
+
+    if (isa<ConstantPoolSDNode>(Op) ||
+        (GlobalN && GlobalN->getGlobal()->hasLocalLinkage())) {
+      // Create following instructions for local linkage PIC code.
+      //     lea %s35, %gotoff_lo(.LCPI0_0)
+      //     and %s35, %s35, (32)0
+      //     lea.sl %s35, %gotoff_hi(.LCPI0_0)(%s35)
+      //     adds.l %s35, %s15, %s35                  ; %s15 is GOT
+      // FIXME: use lea.sl %s35, %gotoff_hi(.LCPI0_0)(%s35, %s15)
+      SDValue HiLo = makeHiLoPair(Op, VEMCExpr::VK_VE_GOTOFF_HI32,
+                                  VEMCExpr::VK_VE_GOTOFF_LO32, DAG);
+      SDValue GlobalBase = DAG.getNode(VEISD::GLOBAL_BASE_REG, DL, PtrVT);
+      return DAG.getNode(ISD::ADD, DL, PtrVT, GlobalBase, HiLo);
+    }
+    // Create following instructions for not local linkage PIC code.
+    //     lea %s35, %got_lo(.LCPI0_0)
+    //     and %s35, %s35, (32)0
+    //     lea.sl %s35, %got_hi(.LCPI0_0)(%s35)
+    //     adds.l %s35, %s15, %s35                  ; %s15 is GOT
+    //     ld     %s35, (,%s35)
+    // FIXME: use lea.sl %s35, %gotoff_hi(.LCPI0_0)(%s35, %s15)
+    SDValue HiLo = makeHiLoPair(Op, VEMCExpr::VK_VE_GOT_HI32,
+                                VEMCExpr::VK_VE_GOT_LO32, DAG);
+    SDValue GlobalBase = DAG.getNode(VEISD::GLOBAL_BASE_REG, DL, PtrVT);
+    SDValue AbsAddr = DAG.getNode(ISD::ADD, DL, PtrVT, GlobalBase, HiLo);
+    return DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), AbsAddr,
+                       MachinePointerInfo::getGOT(DAG.getMachineFunction()));
+  }
 
   // This is one of the absolute code models.
   switch (getTargetMachine().getCodeModel()) {
   default:
     llvm_unreachable("Unsupported absolute code model");
   case CodeModel::Small:
   case CodeModel::Medium:
   case CodeModel::Large:

llvm/lib/Target/VE/VEInstrInfo.h

void storeRegToStackSlot(MachineBasicBlock &MBB,
                            const TargetRegisterInfo *TRI) const override;
 
   void loadRegFromStackSlot(MachineBasicBlock &MBB,
                             MachineBasicBlock::iterator MBBI, Register DestReg,
                             int FrameIndex, const TargetRegisterClass *RC,
                             const TargetRegisterInfo *TRI) const override;
   /// } Stack Spill & Reload
 
+  Register getGlobalBaseReg(MachineFunction *MF) const;
+
   // Lower pseudo instructions after register allocation.
   bool expandPostRAPseudo(MachineInstr &MI) const override;
 
   bool expandExtendStackPseudo(MachineInstr &MI) const;
 };
 
 } // namespace llvm
 
 #endif

llvm/lib/Target/VE/VEInstrInfo.cpp

 //===-- VEInstrInfo.cpp - VE Instruction Information ----------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file contains the VE implementation of the TargetInstrInfo class.
 //
 //===----------------------------------------------------------------------===//
 
 #include "VEInstrInfo.h"
 #include "VE.h"
+#include "VEMachineFunctionInfo.h"
 #include "VESubtarget.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineMemOperand.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/CommandLine.h"
if (RC == &VE::I64RegClass) {
     BuildMI(MBB, I, DL, get(VE::LDUri), DestReg)
         .addFrameIndex(FI)
         .addImm(0)
         .addMemOperand(MMO);
   } else
     report_fatal_error("Can't load this register from stack slot");
 }
 
+Register VEInstrInfo::getGlobalBaseReg(MachineFunction *MF) const {
+  VEMachineFunctionInfo *VEFI = MF->getInfo<VEMachineFunctionInfo>();
+  Register GlobalBaseReg = VEFI->getGlobalBaseReg();
+  if (GlobalBaseReg != 0)
+    return GlobalBaseReg;
+
+  // We use %s15 (%got) as a global base register
+  GlobalBaseReg = VE::SX15;
+
+  // Insert a pseudo instruction to set the GlobalBaseReg into the first
+  // MBB of the function
+  MachineBasicBlock &FirstMBB = MF->front();
+  MachineBasicBlock::iterator MBBI = FirstMBB.begin();
+  DebugLoc dl;
+  BuildMI(FirstMBB, MBBI, dl, get(VE::GETGOT), GlobalBaseReg);
+  VEFI->setGlobalBaseReg(GlobalBaseReg);
+  return GlobalBaseReg;
+}
+
 bool VEInstrInfo::expandPostRAPseudo(MachineInstr &MI) const {
   switch (MI.getOpcode()) {
   case VE::EXTEND_STACK: {
     return expandExtendStackPseudo(MI);
   }
   case VE::EXTEND_STACK_GUARD: {
     MI.eraseFromParent(); // The pseudo instruction is gone now.
     return true;

llvm/lib/Target/VE/VEInstrInfo.td

 
 def SDT_SPCall    : SDTypeProfile<0, -1, [SDTCisVT<0, i64>]>;
 def call          : SDNode<"VEISD::CALL", SDT_SPCall,
                            [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue,
                             SDNPVariadic]>;
 
 def retflag       : SDNode<"VEISD::RET_FLAG", SDTNone,
                            [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
+
+def getGOT        : Operand<iPTR>;
+
+// GETFUNPLT for PIC
+def GetFunPLT : SDNode<"VEISD::GETFUNPLT", SDTIntUnaryOp>;
+
+
 //===----------------------------------------------------------------------===//
 // VE Flag Conditions
 //===----------------------------------------------------------------------===//
 
 // Note that these values must be kept in sync with the CCOp::CondCode enum
 // values.
 class CC_VAL<int N> : PatLeaf<(i32 N)>;
 def CC_IG    : CC_VAL< 0>;  // Greater
def rzi : RM<
     opc, (outs RC:$sx), (ins RC:$sz, immOp2:$imm32),
     !strconcat(opcStr, " $sx, ${imm32}(${sz})"),
     [(set Ty:$sx, (OpNode Ty:$sz, (Ty simm32:$imm32)))]> {
     let cy = 0;
     let sy = 0;
     let cz = 1;
     let hasSideEffects = 0;
   }
+  def zii : RM<
+    opc, (outs RC:$sx), (ins immOp:$sy, immOp2:$imm32),
+    !strconcat(opcStr, " $sx, ${imm32}(${sy})"),
+    [/* Not define DAG pattern here to avoid llvm uses LEAzii for all add
+        instructions.
+        (set Ty:$sx, (OpNode (Ty simm7:$sy), (Ty simm32:$imm32))) */]> {
+    let cy = 0;
+    let cz = 0;
+    let sz = 0;
+    let hasSideEffects = 0;
+  }
   def zzi : RM<
     opc, (outs RC:$sx), (ins immOp2:$imm32),
     !strconcat(opcStr, " $sx, $imm32")> {
     let cy = 0;
     let sy = 0;
     let cz = 0;
     let sz = 0;
     let hasSideEffects = 0;
def SHMri : RM<
     "shm.l $sx, $addr">;
 }
 
 let cx = 0, sx = 0, cy = 0, sy = 0, cz = 0, sz = 0, hasSideEffects = 0 in
 def MONC : RR<
     0x3F, (outs), (ins),
     "monc">;
 
+// Save Instruction Counter
+
+let cx = 0, cy = 0, sy = 0, cz = 0, sz = 0, hasSideEffects = 0 /* , Uses = [IC] */ in
+def SIC : RR<0x28, (outs I32:$sx), (ins), "sic $sx">;
+
 //===----------------------------------------------------------------------===//
 // Instructions for CodeGenOnly
 //===----------------------------------------------------------------------===//
 
 let isCodeGenOnly = 1 in {
 
 // Call instruction
 let Defs = [SX10], Uses = [SX11], hasDelaySlot = 1, isCall = 1, hasSideEffects = 0 in {
def : Pat<(brcc cond:$cond, f32:$l, f32:$r, bb:$addr),
           (BCRSrr (fcond2cc $cond), $l, $r, bb:$addr)>;
 def : Pat<(brcc cond:$cond, f64:$l, f64:$r, bb:$addr),
           (BCRDrr (fcond2cc $cond), $l, $r, bb:$addr)>;
 
 //===----------------------------------------------------------------------===//
 // Pseudo Instructions
 //===----------------------------------------------------------------------===//
 
+// GETGOT for PIC
+let Defs = [SX15 /* %got */, SX16 /* %plt */], hasSideEffects = 0 in {
+  def GETGOT : Pseudo<(outs getGOT:$getpcseq), (ins), "$getpcseq">;
+}
+
+// GETFUNPLT for PIC
+let hasSideEffects = 0 in
+def GETFUNPLT : Pseudo<(outs I64:$dst), (ins i64imm:$addr),
+                       "$dst, $addr",
+                       [(set iPTR:$dst, (GetFunPLT tglobaladdr:$addr))] >;
+
+def : Pat<(GetFunPLT tglobaladdr:$dst),
+          (GETFUNPLT tglobaladdr:$dst)>;
+def : Pat<(GetFunPLT texternalsym:$dst),
+          (GETFUNPLT texternalsym:$dst)>;
+
+
 let Defs = [SX11], Uses = [SX11], hasSideEffects = 0 in {
 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i64imm:$amt, i64imm:$amt2),
                               "# ADJCALLSTACKDOWN $amt, $amt2",
                               [(callseq_start timm:$amt, timm:$amt2)]>;
 def ADJCALLSTACKUP : Pseudo<(outs), (ins i64imm:$amt1, i64imm:$amt2),
                             "# ADJCALLSTACKUP $amt1",
                             [(callseq_end timm:$amt1, timm:$amt2)]>;
 }

llvm/lib/Target/VE/VEMachineFunctionInfo.h

 #include "llvm/CodeGen/MachineFunction.h"
 
 namespace llvm {
 
 class VEMachineFunctionInfo : public MachineFunctionInfo {
   virtual void anchor();
 
 private:
+  unsigned GlobalBaseReg;
+
   /// VarArgsFrameOffset - Frame offset to start of varargs area.
   int VarArgsFrameOffset;
 
   /// IsLeafProc - True if the function is a leaf procedure.
   bool IsLeafProc;
 
 public:
-  VEMachineFunctionInfo() : VarArgsFrameOffset(0), IsLeafProc(false) {}
+  VEMachineFunctionInfo()
+      : GlobalBaseReg(0), VarArgsFrameOffset(0), IsLeafProc(false) {}
   explicit VEMachineFunctionInfo(MachineFunction &MF)
-      : VarArgsFrameOffset(0), IsLeafProc(false) {}
+      : GlobalBaseReg(0), VarArgsFrameOffset(0), IsLeafProc(false) {}
+
+  unsigned getGlobalBaseReg() const { return GlobalBaseReg; }
+  void setGlobalBaseReg(unsigned Reg) { GlobalBaseReg = Reg; }
 
   int getVarArgsFrameOffset() const { return VarArgsFrameOffset; }
   void setVarArgsFrameOffset(int Offset) { VarArgsFrameOffset = Offset; }
 
   void setLeafProc(bool rhs) { IsLeafProc = rhs; }
   bool isLeafProc() const { return IsLeafProc; }
 };
 } // namespace llvm
 
 #endif

llvm/test/CodeGen/VE/pic_access_data.ll

+; RUN: llc -relocation-model=pic < %s -mtriple=ve-unknown-unknown | FileCheck %s
+
+@dst = external global i32, align 4
+@ptr = external global i32*, align 8
+@src = external global i32, align 4
+
+define i32 @func() {
+; CHECK-LABEL: func:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    lea %s15, _GLOBAL_OFFSET_TABLE_@pc_lo(-24)
+; CHECK-NEXT:    and %s15, %s15, (32)0
+; CHECK-NEXT:    sic %s16
+; CHECK-NEXT:    lea.sl %s15, _GLOBAL_OFFSET_TABLE_@pc_hi(%s16, %s15)
+; CHECK-NEXT:    lea %s0, dst@got_lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, dst@got_hi(%s0)
+; CHECK-NEXT:    adds.l %s0, %s15, %s0
+; CHECK-NEXT:    ld %s1, (,%s0)
+; CHECK-NEXT:    lea %s0, ptr@got_lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, ptr@got_hi(%s0)
+; CHECK-NEXT:    lea %s2, src@got_lo
+; CHECK-NEXT:    and %s2, %s2, (32)0
+; CHECK-NEXT:    lea.sl %s2, src@got_hi(%s2)
+; CHECK-NEXT:    adds.l %s2, %s15, %s2
+; CHECK-NEXT:    ld %s2, (,%s2)
+; CHECK-NEXT:    adds.l %s0, %s15, %s0
+; CHECK-NEXT:    ld %s0, (,%s0)
+; CHECK-NEXT:    ldl.sx %s2, (,%s2)
+; CHECK-NEXT:    st %s1, (,%s0)
+; CHECK-NEXT:    or %s0, 1, (0)1
+; CHECK-NEXT:    stl %s2, (,%s1)
+; CHECK-NEXT:    or %s11, 0, %s9
+
+  store i32* @dst, i32** @ptr, align 8
+  %1 = load i32, i32* @src, align 4
+  store i32 %1, i32* @dst, align 4
+  ret i32 1
+}

llvm/test/CodeGen/VE/pic_access_static_data.ll

+; RUN: llc -relocation-model=pic < %s -mtriple=ve-unknown-unknown | FileCheck %s
+
+@dst = internal unnamed_addr global i32 0, align 4
+@src = internal unnamed_addr global i1 false, align 4
+@.str = private unnamed_addr constant [3 x i8] c"%d\00", align 1
+
+define void @func() {
+; CHECK-LABEL: func:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    lea %s15, _GLOBAL_OFFSET_TABLE_@pc_lo(-24)
+; CHECK-NEXT:    and %s15, %s15, (32)0
+; CHECK-NEXT:    sic %s16
+; CHECK-NEXT:    lea.sl %s15, _GLOBAL_OFFSET_TABLE_@pc_hi(%s16, %s15)
+; CHECK-NEXT:    lea %s0, src@gotoff_lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, src@gotoff_hi(%s0)
+; CHECK-NEXT:    adds.l %s0, %s15, %s0
+; CHECK-NEXT:    ld1b.zx %s0, (,%s0)
+; CHECK-NEXT:    or %s1, 0, (0)1
+; CHECK-NEXT:    lea %s2, 100
+; CHECK-NEXT:    cmov.w.ne %s1, %s2, %s0
+; CHECK-NEXT:    lea %s0, dst@gotoff_lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, dst@gotoff_hi(%s0)
+; CHECK-NEXT:    adds.l %s0, %s15, %s0
+; CHECK-NEXT:    stl %s1, (,%s0)
+; CHECK-NEXT:    or %s11, 0, %s9
+
+  %1 = load i1, i1* @src, align 4
+  %2 = select i1 %1, i32 100, i32 0
+  store i32 %2, i32* @dst, align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define i32 @main() {
+; CHECK-LABEL: main:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    lea %s15, _GLOBAL_OFFSET_TABLE_@pc_lo(-24)
+; CHECK-NEXT:    and %s15, %s15, (32)0
+; CHECK-NEXT:    sic %s16
+; CHECK-NEXT:    lea.sl %s15, _GLOBAL_OFFSET_TABLE_@pc_hi(%s16, %s15)
+; CHECK-NEXT:    lea %s0, src@gotoff_lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, src@gotoff_hi(%s0)
+; CHECK-NEXT:    adds.l %s0, %s15, %s0
+; CHECK-NEXT:    or %s1, 1, (0)1
+; CHECK-NEXT:    st1b %s1, (,%s0)
+; CHECK-NEXT:    lea %s12, func@plt_lo(-24)
+; CHECK-NEXT:    and %s12, %s12, (32)0
+; CHECK-NEXT:    sic %s16
+; CHECK-NEXT:    lea.sl %s12, func@plt_hi(%s16, %s12)
+; CHECK-NEXT:    bsic %lr, (,%s12)
+; CHECK-NEXT:    lea %s0, dst@gotoff_lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, dst@gotoff_hi(%s0)
+; CHECK-NEXT:    adds.l %s0, %s15, %s0
+; CHECK-NEXT:    ldl.sx %s1, (,%s0)
+; CHECK-NEXT:    stl %s1, 184(,%s11)
+; CHECK-NEXT:    lea %s0, .L.str@gotoff_lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, .L.str@gotoff_hi(%s0)
+; CHECK-NEXT:    adds.l %s0, %s15, %s0
+; CHECK-NEXT:    lea %s12, printf@plt_lo(-24)
+; CHECK-NEXT:    and %s12, %s12, (32)0
+; CHECK-NEXT:    sic %s16
+; CHECK-NEXT:    lea.sl %s12, printf@plt_hi(%s16, %s12)
+; CHECK-NEXT:    st %s0, 176(,%s11)
+; CHECK-NEXT:    bsic %lr, (,%s12)
+; CHECK-NEXT:    or %s0, 0, (0)1
+; CHECK-NEXT:    or %s11, 0, %s9
+  store i1 true, i1* @src, align 4
+  tail call void @func()
+  %1 = load i32, i32* @dst, align 4
+  %2 = tail call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @.str, i64 0, i64 0), i32 %1)
+  ret i32 0
+}
+
+declare i32 @printf(i8* nocapture readonly, ...)

llvm/test/CodeGen/VE/pic_func_call.ll

+; RUN: llc -relocation-model=pic < %s -mtriple=ve-unknown-unknown | FileCheck %s
+
+define void @func() {
+; CHECK-LABEL: func:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    lea %s15, _GLOBAL_OFFSET_TABLE_@pc_lo(-24)
+; CHECK-NEXT:    and %s15, %s15, (32)0
+; CHECK-NEXT:    sic %s16
+; CHECK-NEXT:    lea.sl %s15, _GLOBAL_OFFSET_TABLE_@pc_hi(%s16, %s15)
+; CHECK-NEXT:    lea %s12, function@plt_lo(-24)
+; CHECK-NEXT:    and %s12, %s12, (32)0
+; CHECK-NEXT:    sic %s16
+; CHECK-NEXT:    lea.sl %s12, function@plt_hi(%s16, %s12)
+; CHECK-NEXT:    bsic %lr, (,%s12)
+; CHECK-NEXT:    or %s11, 0, %s9
+
+  call void bitcast (void (...)* @function to void ()*)()
+  ret void
+}
+
+declare void @function(...)

llvm/test/CodeGen/VE/pic_indirect_func_call.ll

+; RUN: llc -relocation-model=pic < %s -mtriple=ve-unknown-unknown | FileCheck %s
+
+@ptr = external global void (...)*, align 8
+
+define void @func() {
+; CHECK-LABEL: func:
+; CHECK:       .LBB{{[0-9]+}}_2:
+; CHECK-NEXT:    lea %s15, _GLOBAL_OFFSET_TABLE_@pc_lo(-24)
+; CHECK-NEXT:    and %s15, %s15, (32)0
+; CHECK-NEXT:    sic %s16
+; CHECK-NEXT:    lea.sl %s15, _GLOBAL_OFFSET_TABLE_@pc_hi(%s16, %s15)
+; CHECK-NEXT:    lea %s0, function@got_lo
+; CHECK-NEXT:    and %s0, %s0, (32)0
+; CHECK-NEXT:    lea.sl %s0, function@got_hi(%s0)
+; CHECK-NEXT:    adds.l %s0, %s15, %s0
+; CHECK-NEXT:    ld %s0, (,%s0)
+; CHECK-NEXT:    lea %s1, ptr@got_lo
+; CHECK-NEXT:    and %s1, %s1, (32)0
+; CHECK-NEXT:    lea.sl %s1, ptr@got_hi(%s1)
+; CHECK-NEXT:    adds.l %s1, %s15, %s1
+; CHECK-NEXT:    ld %s1, (,%s1)
+; CHECK-NEXT:    st %s0, (,%s1)
+; CHECK-NEXT:    or %s12, 0, %s0
+; CHECK-NEXT:    bsic %lr, (,%s12)
+; CHECK-NEXT:    or %s11, 0, %s9
+
+  store void (...)* @function, void (...)** @ptr, align 8
+  %1 = load void (...)*, void (...)** @ptr, align 8
+  %2 = bitcast void (...)* %1 to void ()*
+  call void %2()
+  ret void
+}
+
+declare void @function(...)