X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTarget%2FX86%2FX86RegisterInfo.cpp;h=06545bc209622a1997dbce268b51c715559b6d9f;hp=0fa38f453706dbec450b32a30d2c62052ef85e10;hb=c1896335867238b01457db7e5a2a1a66d81a39d9;hpb=acd5f13c889cc39e72bee7b0426f8ef3426a09c3 diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp index 0fa38f45370..06545bc2096 100644 --- a/lib/Target/X86/X86RegisterInfo.cpp +++ b/lib/Target/X86/X86RegisterInfo.cpp @@ -14,6 +14,7 @@ //===----------------------------------------------------------------------===// #include "X86RegisterInfo.h" +#include "X86FrameLowering.h" #include "X86InstrBuilder.h" #include "X86MachineFunctionInfo.h" #include "X86Subtarget.h" @@ -53,34 +54,35 @@ static cl::opt EnableBasePointer("x86-use-base-pointer", cl::Hidden, cl::init(true), cl::desc("Enable use of a base pointer for complex stack frames")); -X86RegisterInfo::X86RegisterInfo(const X86Subtarget &STI) - : X86GenRegisterInfo( - (STI.is64Bit() ? X86::RIP : X86::EIP), - X86_MC::getDwarfRegFlavour(STI.getTargetTriple(), false), - X86_MC::getDwarfRegFlavour(STI.getTargetTriple(), true), - (STI.is64Bit() ? X86::RIP : X86::EIP)), - Subtarget(STI) { +X86RegisterInfo::X86RegisterInfo(const Triple &TT) + : X86GenRegisterInfo((TT.isArch64Bit() ? X86::RIP : X86::EIP), + X86_MC::getDwarfRegFlavour(TT, false), + X86_MC::getDwarfRegFlavour(TT, true), + (TT.isArch64Bit() ? X86::RIP : X86::EIP)) { X86_MC::InitLLVM2SEHRegisterMapping(this); // Cache some information. - Is64Bit = Subtarget.is64Bit(); - IsWin64 = Subtarget.isTargetWin64(); + Is64Bit = TT.isArch64Bit(); + IsWin64 = Is64Bit && TT.isOSWindows(); + // Use a callee-saved register as the base pointer. These registers must + // not conflict with any ABI requirements. For example, in 32-bit mode PIC + // requires GOT in the EBX register before function calls via PLT GOT pointer. if (Is64Bit) { SlotSize = 8; - StackPtr = (Subtarget.isTarget64BitLP64() || Subtarget.isTargetNaCl64()) ? - X86::RSP : X86::ESP; - FramePtr = (Subtarget.isTarget64BitLP64() || Subtarget.isTargetNaCl64()) ? - X86::RBP : X86::EBP; + // This matches the simplified 32-bit pointer code in the data layout + // computation. + // FIXME: Should use the data layout? + bool Use64BitReg = TT.getEnvironment() != Triple::GNUX32; + StackPtr = Use64BitReg ? X86::RSP : X86::ESP; + FramePtr = Use64BitReg ? X86::RBP : X86::EBP; + BasePtr = Use64BitReg ? X86::RBX : X86::EBX; } else { SlotSize = 4; StackPtr = X86::ESP; FramePtr = X86::EBP; + BasePtr = X86::ESI; } - // Use a callee-saved register as the base pointer. These registers must - // not conflict with any ABI requirements. For example, in 32-bit mode PIC - // requires GOT in the EBX register before function calls via PLT GOT pointer. - BasePtr = Is64Bit ? X86::RBX : X86::ESI; } bool @@ -119,8 +121,9 @@ X86RegisterInfo::getMatchingSuperRegClass(const TargetRegisterClass *A, return X86GenRegisterInfo::getMatchingSuperRegClass(A, B, SubIdx); } -const TargetRegisterClass* -X86RegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC) const{ +const TargetRegisterClass * +X86RegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC, + const MachineFunction &MF) const { // Don't allow super-classes of GR8_NOREX. This class is only used after // extracting sub_8bit_hi sub-registers. The H sub-registers cannot be copied // to the full GR8 register class in 64-bit mode, so we cannot allow the @@ -160,6 +163,7 @@ X86RegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC) const{ const TargetRegisterClass * X86RegisterInfo::getPointerRegClass(const MachineFunction &MF, unsigned Kind) const { + const X86Subtarget &Subtarget = MF.getSubtarget(); switch (Kind) { default: llvm_unreachable("Unexpected Kind in getPointerRegClass!"); case 0: // Normal GPRs. @@ -171,9 +175,9 @@ X86RegisterInfo::getPointerRegClass(const MachineFunction &MF, return &X86::GR64_NOSPRegClass; return &X86::GR32_NOSPRegClass; case 2: // Available for tailcall (not callee-saved GPRs). - if (Subtarget.isTargetWin64()) + if (IsWin64) return &X86::GR64_TCW64RegClass; - else if (Subtarget.is64Bit()) + else if (Is64Bit) return &X86::GR64_TCRegClass; const Function *F = MF.getFunction(); @@ -209,7 +213,7 @@ X86RegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, case X86::GR64RegClassID: return 12 - FPDiff; case X86::VR128RegClassID: - return Subtarget.is64Bit() ? 10 : 4; + return Is64Bit ? 10 : 4; case X86::VR64RegClassID: return 4; } @@ -217,8 +221,10 @@ X86RegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC, const MCPhysReg * X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const { + const X86Subtarget &Subtarget = MF->getSubtarget(); bool HasAVX = Subtarget.hasAVX(); bool HasAVX512 = Subtarget.hasAVX512(); + bool CallsEHReturn = MF->getMMI().callsEHReturn(); assert(MF && "MachineFunction required"); switch (MF->getFunction()->getCallingConv()) { @@ -252,11 +258,16 @@ X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const { if (Is64Bit) return CSR_64_MostRegs_SaveList; break; + case CallingConv::X86_64_Win64: + return CSR_Win64_SaveList; + case CallingConv::X86_64_SysV: + if (CallsEHReturn) + return CSR_64EHRet_SaveList; + return CSR_64_SaveList; default: break; } - bool CallsEHReturn = MF->getMMI().callsEHReturn(); if (Is64Bit) { if (IsWin64) return CSR_Win64_SaveList; @@ -269,8 +280,10 @@ X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const { return CSR_32_SaveList; } -const uint32_t* -X86RegisterInfo::getCallPreservedMask(CallingConv::ID CC) const { +const uint32_t * +X86RegisterInfo::getCallPreservedMask(const MachineFunction &MF, + CallingConv::ID CC) const { + const X86Subtarget &Subtarget = MF.getSubtarget(); bool HasAVX = Subtarget.hasAVX(); bool HasAVX512 = Subtarget.hasAVX512(); @@ -307,6 +320,10 @@ X86RegisterInfo::getCallPreservedMask(CallingConv::ID CC) const { break; default: break; + case CallingConv::X86_64_Win64: + return CSR_Win64_RegMask; + case CallingConv::X86_64_SysV: + return CSR_64_RegMask; } // Unlike getCalleeSavedRegs(), we don't have MMI so we can't check @@ -348,13 +365,15 @@ BitVector X86RegisterInfo::getReservedRegs(const MachineFunction &MF) const { // Set the base-pointer register and its aliases as reserved if needed. if (hasBasePointer(MF)) { CallingConv::ID CC = MF.getFunction()->getCallingConv(); - const uint32_t* RegMask = getCallPreservedMask(CC); + const uint32_t *RegMask = getCallPreservedMask(MF, CC); if (MachineOperand::clobbersPhysReg(RegMask, getBaseRegister())) report_fatal_error( "Stack realignment in presence of dynamic allocas is not supported with" "this calling convention."); - for (MCSubRegIterator I(getBaseRegister(), this, /*IncludeSelf=*/true); + unsigned BasePtr = getX86SubSuperRegister(getBaseRegister(), MVT::i64, + false); + for (MCSubRegIterator I(BasePtr, this, /*IncludeSelf=*/true); I.isValid(); ++I) Reserved.set(*I); } @@ -390,7 +409,7 @@ BitVector X86RegisterInfo::getReservedRegs(const MachineFunction &MF) const { Reserved.set(*AI); } } - if (!Is64Bit || !Subtarget.hasAVX512()) { + if (!Is64Bit || !MF.getSubtarget().hasAVX512()) { for (unsigned n = 16; n != 32; ++n) { for (MCRegAliasIterator AI(X86::XMM0 + n, this, true); AI.isValid(); ++AI) Reserved.set(*AI); @@ -445,10 +464,8 @@ bool X86RegisterInfo::needsStackRealignment(const MachineFunction &MF) const { const Function *F = MF.getFunction(); unsigned StackAlign = MF.getSubtarget().getFrameLowering()->getStackAlignment(); - bool requiresRealignment = - ((MFI->getMaxAlignment() > StackAlign) || - F->getAttributes().hasAttribute(AttributeSet::FunctionIndex, - Attribute::StackAlignment)); + bool requiresRealignment = ((MFI->getMaxAlignment() > StackAlign) || + F->hasFnAttribute(Attribute::StackAlignment)); // If we've requested that we force align the stack do so now. if (ForceStackAlign) @@ -485,6 +502,24 @@ X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, else BasePtr = (TFI->hasFP(MF) ? FramePtr : StackPtr); + // FRAME_ALLOC uses a single offset, with no register. It only works in the + // simple FP case, and doesn't work with stack realignment. On 32-bit, the + // offset is from the traditional base pointer location. On 64-bit, the + // offset is from the SP at the end of the prologue, not the FP location. This + // matches the behavior of llvm.frameaddress. + if (Opc == TargetOpcode::FRAME_ALLOC) { + MachineOperand &FI = MI.getOperand(FIOperandNum); + bool IsWinEH = MF.getTarget().getMCAsmInfo()->usesWindowsCFI(); + int Offset; + if (IsWinEH) + Offset = static_cast(TFI) + ->getFrameIndexOffsetFromSP(MF, FrameIndex); + else + Offset = TFI->getFrameIndexOffset(MF, FrameIndex); + FI.ChangeToImmediate(Offset); + return; + } + // For LEA64_32r when BasePtr is 32-bits (X32) we can use full-size 64-bit // register as source operand, semantic is the same and destination is // 32-bits. It saves one byte per lea in code since 0x67 prefix is avoided. @@ -536,8 +571,9 @@ unsigned X86RegisterInfo::getFrameRegister(const MachineFunction &MF) const { return TFI->hasFP(MF) ? FramePtr : StackPtr; } -unsigned X86RegisterInfo::getPtrSizedFrameRegister( - const MachineFunction &MF) const { +unsigned +X86RegisterInfo::getPtrSizedFrameRegister(const MachineFunction &MF) const { + const X86Subtarget &Subtarget = MF.getSubtarget(); unsigned FrameReg = getFrameRegister(MF); if (Subtarget.isTarget64BitILP32()) FrameReg = getX86SubSuperRegister(FrameReg, MVT::i32, false);