#include "X86.h"
#include "llvm/PassManager.h"
#include "llvm/CodeGen/MachineCodeEmitter.h"
+#include "llvm/CodeGen/JITCodeEmitter.h"
+#include "llvm/CodeGen/ObjectCodeEmitter.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
STATISTIC(NumEmitted, "Number of machine instructions emitted");
namespace {
+template<class CodeEmitter>
class VISIBILITY_HIDDEN Emitter : public MachineFunctionPass {
const X86InstrInfo *II;
const TargetData *TD;
X86TargetMachine &TM;
- MachineCodeEmitter &MCE;
+ CodeEmitter &MCE;
intptr_t PICBaseOffset;
bool Is64BitMode;
bool IsPIC;
public:
static char ID;
- explicit Emitter(X86TargetMachine &tm, MachineCodeEmitter &mce)
+ explicit Emitter(X86TargetMachine &tm, CodeEmitter &mce)
: MachineFunctionPass(&ID), II(0), TD(0), TM(tm),
MCE(mce), PICBaseOffset(0), Is64BitMode(false),
IsPIC(TM.getRelocationModel() == Reloc::PIC_) {}
- Emitter(X86TargetMachine &tm, MachineCodeEmitter &mce,
+ Emitter(X86TargetMachine &tm, CodeEmitter &mce,
const X86InstrInfo &ii, const TargetData &td, bool is64)
: MachineFunctionPass(&ID), II(&ii), TD(&td), TM(tm),
MCE(mce), PICBaseOffset(0), Is64BitMode(is64),
void emitPCRelativeBlockAddress(MachineBasicBlock *MBB);
void emitGlobalAddress(GlobalValue *GV, unsigned Reloc,
intptr_t Disp = 0, intptr_t PCAdj = 0,
- bool NeedStub = false, bool IsNonLazy = false);
+ bool NeedStub = false, bool Indirect = false);
void emitExternalSymbolAddress(const char *ES, unsigned Reloc);
void emitConstPoolAddress(unsigned CPI, unsigned Reloc, intptr_t Disp = 0,
intptr_t PCAdj = 0);
bool gvNeedsNonLazyPtr(const GlobalValue *GV);
};
- char Emitter::ID = 0;
+
+template<class CodeEmitter>
+ char Emitter<CodeEmitter>::ID = 0;
}
/// createX86CodeEmitterPass - Return a pass that emits the collected X86 code
-/// to the specified MCE object.
+/// to the specified templated MachineCodeEmitter object.
+
FunctionPass *llvm::createX86CodeEmitterPass(X86TargetMachine &TM,
MachineCodeEmitter &MCE) {
- return new Emitter(TM, MCE);
+ return new Emitter<MachineCodeEmitter>(TM, MCE);
+}
+FunctionPass *llvm::createX86JITCodeEmitterPass(X86TargetMachine &TM,
+ JITCodeEmitter &JCE) {
+ return new Emitter<JITCodeEmitter>(TM, JCE);
+}
+FunctionPass *llvm::createX86ObjectCodeEmitterPass(X86TargetMachine &TM,
+ ObjectCodeEmitter &OCE) {
+ return new Emitter<ObjectCodeEmitter>(TM, OCE);
}
-bool Emitter::runOnMachineFunction(MachineFunction &MF) {
+template<class CodeEmitter>
+bool Emitter<CodeEmitter>::runOnMachineFunction(MachineFunction &MF) {
MCE.setModuleInfo(&getAnalysis<MachineModuleInfo>());
/// necessary to resolve the address of this block later and emits a dummy
/// value.
///
-void Emitter::emitPCRelativeBlockAddress(MachineBasicBlock *MBB) {
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitPCRelativeBlockAddress(MachineBasicBlock *MBB) {
// Remember where this reference was and where it is to so we can
// deal with it later.
MCE.addRelocation(MachineRelocation::getBB(MCE.getCurrentPCOffset(),
/// emitGlobalAddress - Emit the specified address to the code stream assuming
/// this is part of a "take the address of a global" instruction.
///
-void Emitter::emitGlobalAddress(GlobalValue *GV, unsigned Reloc,
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitGlobalAddress(GlobalValue *GV, unsigned Reloc,
intptr_t Disp /* = 0 */,
intptr_t PCAdj /* = 0 */,
bool NeedStub /* = false */,
- bool isNonLazy /* = false */) {
+ bool Indirect /* = false */) {
intptr_t RelocCST = 0;
if (Reloc == X86::reloc_picrel_word)
RelocCST = PICBaseOffset;
else if (Reloc == X86::reloc_pcrel_word)
RelocCST = PCAdj;
- MachineRelocation MR = isNonLazy
- ? MachineRelocation::getGVNonLazyPtr(MCE.getCurrentPCOffset(), Reloc,
- GV, RelocCST, NeedStub)
+ MachineRelocation MR = Indirect
+ ? MachineRelocation::getIndirectSymbol(MCE.getCurrentPCOffset(), Reloc,
+ GV, RelocCST, NeedStub)
: MachineRelocation::getGV(MCE.getCurrentPCOffset(), Reloc,
GV, RelocCST, NeedStub);
MCE.addRelocation(MR);
/// emitExternalSymbolAddress - Arrange for the address of an external symbol to
/// be emitted to the current location in the function, and allow it to be PC
/// relative.
-void Emitter::emitExternalSymbolAddress(const char *ES, unsigned Reloc) {
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitExternalSymbolAddress(const char *ES,
+ unsigned Reloc) {
intptr_t RelocCST = (Reloc == X86::reloc_picrel_word) ? PICBaseOffset : 0;
MCE.addRelocation(MachineRelocation::getExtSym(MCE.getCurrentPCOffset(),
Reloc, ES, RelocCST));
/// emitConstPoolAddress - Arrange for the address of an constant pool
/// to be emitted to the current location in the function, and allow it to be PC
/// relative.
-void Emitter::emitConstPoolAddress(unsigned CPI, unsigned Reloc,
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitConstPoolAddress(unsigned CPI, unsigned Reloc,
intptr_t Disp /* = 0 */,
intptr_t PCAdj /* = 0 */) {
intptr_t RelocCST = 0;
/// emitJumpTableAddress - Arrange for the address of a jump table to
/// be emitted to the current location in the function, and allow it to be PC
/// relative.
-void Emitter::emitJumpTableAddress(unsigned JTI, unsigned Reloc,
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitJumpTableAddress(unsigned JTI, unsigned Reloc,
intptr_t PCAdj /* = 0 */) {
intptr_t RelocCST = 0;
if (Reloc == X86::reloc_picrel_word)
MCE.emitWordLE(0);
}
-unsigned Emitter::getX86RegNum(unsigned RegNo) const {
+template<class CodeEmitter>
+unsigned Emitter<CodeEmitter>::getX86RegNum(unsigned RegNo) const {
return II->getRegisterInfo().getX86RegNum(RegNo);
}
return RM | (RegOpcode << 3) | (Mod << 6);
}
-void Emitter::emitRegModRMByte(unsigned ModRMReg, unsigned RegOpcodeFld){
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitRegModRMByte(unsigned ModRMReg,
+ unsigned RegOpcodeFld){
MCE.emitByte(ModRMByte(3, RegOpcodeFld, getX86RegNum(ModRMReg)));
}
-void Emitter::emitRegModRMByte(unsigned RegOpcodeFld) {
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitRegModRMByte(unsigned RegOpcodeFld) {
MCE.emitByte(ModRMByte(3, RegOpcodeFld, 0));
}
-void Emitter::emitSIBByte(unsigned SS, unsigned Index, unsigned Base) {
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitSIBByte(unsigned SS,
+ unsigned Index,
+ unsigned Base) {
// SIB byte is in the same format as the ModRMByte...
MCE.emitByte(ModRMByte(SS, Index, Base));
}
-void Emitter::emitConstant(uint64_t Val, unsigned Size) {
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitConstant(uint64_t Val, unsigned Size) {
// Output the constant in little endian byte order...
for (unsigned i = 0; i != Size; ++i) {
MCE.emitByte(Val & 255);
return Value == (signed char)Value;
}
-bool Emitter::gvNeedsNonLazyPtr(const GlobalValue *GV) {
+template<class CodeEmitter>
+bool Emitter<CodeEmitter>::gvNeedsNonLazyPtr(const GlobalValue *GV) {
// For Darwin, simulate the linktime GOT by using the same non-lazy-pointer
// mechanism as 32-bit mode.
return (!Is64BitMode || TM.getSubtarget<X86Subtarget>().isTargetDarwin()) &&
TM.getSubtarget<X86Subtarget>().GVRequiresExtraLoad(GV, TM, false);
}
-void Emitter::emitDisplacementField(const MachineOperand *RelocOp,
- int DispVal, intptr_t PCAdj) {
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitDisplacementField(const MachineOperand *RelocOp,
+ int DispVal, intptr_t PCAdj) {
// If this is a simple integer displacement that doesn't require a relocation,
// emit it now.
if (!RelocOp) {
unsigned rt = Is64BitMode ? X86::reloc_pcrel_word
: (IsPIC ? X86::reloc_picrel_word : X86::reloc_absolute_word);
bool NeedStub = isa<Function>(RelocOp->getGlobal());
- bool isNonLazy = gvNeedsNonLazyPtr(RelocOp->getGlobal());
+ bool Indirect = gvNeedsNonLazyPtr(RelocOp->getGlobal());
emitGlobalAddress(RelocOp->getGlobal(), rt, RelocOp->getOffset(),
- PCAdj, NeedStub, isNonLazy);
+ PCAdj, NeedStub, Indirect);
} else if (RelocOp->isCPI()) {
unsigned rt = Is64BitMode ? X86::reloc_pcrel_word : X86::reloc_picrel_word;
emitConstPoolAddress(RelocOp->getIndex(), rt,
}
}
-void Emitter::emitMemModRMByte(const MachineInstr &MI,
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitMemModRMByte(const MachineInstr &MI,
unsigned Op, unsigned RegOpcodeField,
intptr_t PCAdj) {
const MachineOperand &Op3 = MI.getOperand(Op+3);
unsigned BaseReg = Base.getReg();
// Is a SIB byte needed?
- if ((!Is64BitMode || DispForReloc) && IndexReg.getReg() == 0 &&
- (BaseReg == 0 || getX86RegNum(BaseReg) != N86::ESP)) {
- if (BaseReg == 0) { // Just a displacement?
+ if ((!Is64BitMode || DispForReloc || BaseReg != 0) &&
+ IndexReg.getReg() == 0 &&
+ (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));
else
IndexRegNo = 4; // For example [ESP+1*<noreg>+4]
emitSIBByte(SS, IndexRegNo, 5);
- } else {
+ } else {
unsigned BaseRegNo = getX86RegNum(BaseReg);
unsigned IndexRegNo;
if (IndexReg.getReg())
}
}
-void Emitter::emitInstruction(const MachineInstr &MI,
+template<class CodeEmitter>
+void Emitter<CodeEmitter>::emitInstruction(
+ const MachineInstr &MI,
const TargetInstrDesc *Desc) {
DOUT << MI;
assert(0 && "psuedo instructions should be removed before code emission");
break;
case TargetInstrInfo::INLINEASM: {
- const char* Value = MI.getOperand(0).getSymbolName();
- /* We allow inline assembler nodes with empty bodies - they can
- implicitly define registers, which is ok for JIT. */
- assert((Value[0] == 0) && "JIT does not support inline asm!\n");
+ // We allow inline assembler nodes with empty bodies - they can
+ // implicitly define registers, which is ok for JIT.
+ if (MI.getOperand(0).getSymbolName()[0]) {
+ assert(0 && "JIT does not support inline asm!\n");
+ abort();
+ }
break;
}
case TargetInstrInfo::DBG_LABEL:
case X86::DWARF_LOC:
case X86::FP_REG_KILL:
break;
- case X86::TLS_tp: {
- MCE.emitByte(BaseOpcode);
- unsigned RegOpcodeField = getX86RegNum(MI.getOperand(0).getReg());
- MCE.emitByte(ModRMByte(0, RegOpcodeField, 5));
- emitConstant(0, 4);
- break;
- }
- case X86::TLS_gs_ri: {
- MCE.emitByte(BaseOpcode);
- unsigned RegOpcodeField = getX86RegNum(MI.getOperand(0).getReg());
- MCE.emitByte(ModRMByte(0, RegOpcodeField, 5));
- GlobalValue* GV = MI.getOperand(1).getGlobal();
- unsigned rt = Is64BitMode ? X86::reloc_pcrel_word
- : (IsPIC ? X86::reloc_picrel_word : X86::reloc_absolute_word);
- emitGlobalAddress(GV, rt);
- break;
- }
case X86::MOVPC32r: {
// This emits the "call" portion of this pseudo instruction.
MCE.emitByte(BaseOpcode);
emitConstant(0, X86InstrInfo::sizeOfImm(Desc));
// Remember PIC base.
- PICBaseOffset = MCE.getCurrentPCOffset();
+ PICBaseOffset = (intptr_t) MCE.getCurrentPCOffset();
X86JITInfo *JTI = TM.getJITInfo();
JTI->setPICBase(MCE.getCurrentPCValue());
break;
} else if (MO.isSymbol()) {
emitExternalSymbolAddress(MO.getSymbolName(), X86::reloc_pcrel_word);
} else if (MO.isImm()) {
- emitConstant(MO.getImm(), X86InstrInfo::sizeOfImm(Desc));
+ if (Opcode == X86::CALLpcrel32 || Opcode == X86::CALL64pcrel32) {
+ // Fix up immediate operand for pc relative calls.
+ intptr_t Imm = (intptr_t)MO.getImm();
+ Imm = Imm - MCE.getCurrentPCValue() - 4;
+ emitConstant(Imm, X86InstrInfo::sizeOfImm(Desc));
+ } else
+ emitConstant(MO.getImm(), X86InstrInfo::sizeOfImm(Desc));
} else {
assert(0 && "Unknown RawFrm operand!");
}
rt = X86::reloc_absolute_dword; // FIXME: add X86II flag?
if (MO1.isGlobal()) {
bool NeedStub = isa<Function>(MO1.getGlobal());
- bool isNonLazy = gvNeedsNonLazyPtr(MO1.getGlobal());
+ bool Indirect = gvNeedsNonLazyPtr(MO1.getGlobal());
emitGlobalAddress(MO1.getGlobal(), rt, MO1.getOffset(), 0,
- NeedStub, isNonLazy);
+ NeedStub, Indirect);
} else if (MO1.isSymbol())
emitExternalSymbolAddress(MO1.getSymbolName(), rt);
else if (MO1.isCPI())
}
case X86II::MRMDestMem: {
MCE.emitByte(BaseOpcode);
- emitMemModRMByte(MI, CurOp, getX86RegNum(MI.getOperand(CurOp+4).getReg()));
- CurOp += 5;
+ emitMemModRMByte(MI, CurOp,
+ getX86RegNum(MI.getOperand(CurOp + X86AddrNumOperands)
+ .getReg()));
+ CurOp += X86AddrNumOperands + 1;
if (CurOp != NumOps)
emitConstant(MI.getOperand(CurOp++).getImm(), X86InstrInfo::sizeOfImm(Desc));
break;
getX86RegNum(MI.getOperand(CurOp).getReg()));
CurOp += 2;
if (CurOp != NumOps)
- emitConstant(MI.getOperand(CurOp++).getImm(), X86InstrInfo::sizeOfImm(Desc));
+ emitConstant(MI.getOperand(CurOp++).getImm(),
+ X86InstrInfo::sizeOfImm(Desc));
break;
case X86II::MRMSrcMem: {
- intptr_t PCAdj = (CurOp+5 != NumOps) ? X86InstrInfo::sizeOfImm(Desc) : 0;
+ // FIXME: Maybe lea should have its own form?
+ int AddrOperands;
+ if (Opcode == X86::LEA64r || Opcode == X86::LEA64_32r ||
+ Opcode == X86::LEA16r || Opcode == X86::LEA32r)
+ AddrOperands = X86AddrNumOperands - 1; // No segment register
+ else
+ AddrOperands = X86AddrNumOperands;
+
+ intptr_t PCAdj = (CurOp + AddrOperands + 1 != NumOps) ?
+ X86InstrInfo::sizeOfImm(Desc) : 0;
MCE.emitByte(BaseOpcode);
emitMemModRMByte(MI, CurOp+1, getX86RegNum(MI.getOperand(CurOp).getReg()),
PCAdj);
- CurOp += 5;
+ CurOp += AddrOperands + 1;
if (CurOp != NumOps)
- emitConstant(MI.getOperand(CurOp++).getImm(), X86InstrInfo::sizeOfImm(Desc));
+ emitConstant(MI.getOperand(CurOp++).getImm(),
+ X86InstrInfo::sizeOfImm(Desc));
break;
}
case X86II::MRM6r: case X86II::MRM7r: {
MCE.emitByte(BaseOpcode);
- // Special handling of lfence and mfence.
+ // Special handling of lfence, mfence, monitor, and mwait.
if (Desc->getOpcode() == X86::LFENCE ||
- Desc->getOpcode() == X86::MFENCE)
+ Desc->getOpcode() == X86::MFENCE ||
+ Desc->getOpcode() == X86::MONITOR ||
+ Desc->getOpcode() == X86::MWAIT) {
emitRegModRMByte((Desc->TSFlags & X86II::FormMask)-X86II::MRM0r);
- else
+
+ switch (Desc->getOpcode()) {
+ default: break;
+ case X86::MONITOR:
+ MCE.emitByte(0xC8);
+ break;
+ case X86::MWAIT:
+ MCE.emitByte(0xC9);
+ break;
+ }
+ } else {
emitRegModRMByte(MI.getOperand(CurOp++).getReg(),
(Desc->TSFlags & X86II::FormMask)-X86II::MRM0r);
+ }
if (CurOp != NumOps) {
const MachineOperand &MO1 = MI.getOperand(CurOp++);
rt = X86::reloc_absolute_word; // FIXME: add X86II flag?
if (MO1.isGlobal()) {
bool NeedStub = isa<Function>(MO1.getGlobal());
- bool isNonLazy = gvNeedsNonLazyPtr(MO1.getGlobal());
+ bool Indirect = gvNeedsNonLazyPtr(MO1.getGlobal());
emitGlobalAddress(MO1.getGlobal(), rt, MO1.getOffset(), 0,
- NeedStub, isNonLazy);
+ NeedStub, Indirect);
} else if (MO1.isSymbol())
emitExternalSymbolAddress(MO1.getSymbolName(), rt);
else if (MO1.isCPI())
case X86II::MRM2m: case X86II::MRM3m:
case X86II::MRM4m: case X86II::MRM5m:
case X86II::MRM6m: case X86II::MRM7m: {
- intptr_t PCAdj = (CurOp+4 != NumOps) ?
- (MI.getOperand(CurOp+4).isImm() ? X86InstrInfo::sizeOfImm(Desc) : 4) : 0;
+ intptr_t PCAdj = (CurOp + X86AddrNumOperands != NumOps) ?
+ (MI.getOperand(CurOp+X86AddrNumOperands).isImm() ?
+ X86InstrInfo::sizeOfImm(Desc) : 4) : 0;
MCE.emitByte(BaseOpcode);
emitMemModRMByte(MI, CurOp, (Desc->TSFlags & X86II::FormMask)-X86II::MRM0m,
PCAdj);
- CurOp += 4;
+ CurOp += X86AddrNumOperands;
if (CurOp != NumOps) {
const MachineOperand &MO = MI.getOperand(CurOp++);
rt = X86::reloc_absolute_word; // FIXME: add X86II flag?
if (MO.isGlobal()) {
bool NeedStub = isa<Function>(MO.getGlobal());
- bool isNonLazy = gvNeedsNonLazyPtr(MO.getGlobal());
+ bool Indirect = gvNeedsNonLazyPtr(MO.getGlobal());
emitGlobalAddress(MO.getGlobal(), rt, MO.getOffset(), 0,
- NeedStub, isNonLazy);
+ NeedStub, Indirect);
} else if (MO.isSymbol())
emitExternalSymbolAddress(MO.getSymbolName(), rt);
else if (MO.isCPI())
abort();
}
}
+