-//=======- X86FrameLowering.cpp - X86 Frame Information --------*- C++ -*-====//
+//===-- X86FrameLowering.cpp - X86 Frame Information ----------------------===//
//
// The LLVM Compiler Infrastructure
//
#include "X86MachineFunctionInfo.h"
#include "X86Subtarget.h"
#include "X86TargetMachine.h"
-#include "llvm/Function.h"
+#include "llvm/ADT/SmallSet.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/DataLayout.h"
+#include "llvm/Function.h"
#include "llvm/MC/MCAsmInfo.h"
#include "llvm/MC/MCSymbol.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Target/TargetOptions.h"
#include "llvm/Support/CommandLine.h"
-#include "llvm/ADT/SmallSet.h"
+#include "llvm/Target/TargetOptions.h"
using namespace llvm;
bool X86FrameLowering::hasFP(const MachineFunction &MF) const {
const MachineFrameInfo *MFI = MF.getFrameInfo();
const MachineModuleInfo &MMI = MF.getMMI();
- const TargetRegisterInfo *RI = TM.getRegisterInfo();
+ const TargetRegisterInfo *RegInfo = TM.getRegisterInfo();
- return (DisableFramePointerElim(MF) ||
- RI->needsStackRealignment(MF) ||
+ return (MF.getTarget().Options.DisableFramePointerElim(MF) ||
+ RegInfo->needsStackRealignment(MF) ||
MFI->hasVarSizedObjects() ||
MFI->isFrameAddressTaken() ||
MF.getInfo<X86MachineFunctionInfo>()->getForceFramePointer() ||
- MMI.callsUnwindInit());
+ MMI.callsUnwindInit() || MMI.callsEHReturn());
}
static unsigned getSUBriOpcode(unsigned is64Bit, int64_t Imm) {
}
}
+static unsigned getLEArOpcode(unsigned is64Bit) {
+ return is64Bit ? X86::LEA64r : X86::LEA32r;
+}
+
/// findDeadCallerSavedReg - Return a caller-saved register that isn't live
/// when it reaches the "return" instruction. We can then pop a stack object
/// to this register without worry about clobbering it.
if (!F || MF->getMMI().callsEHReturn())
return 0;
- static const unsigned CallerSavedRegs32Bit[] = {
+ static const uint16_t CallerSavedRegs32Bit[] = {
X86::EAX, X86::EDX, X86::ECX, 0
};
- static const unsigned CallerSavedRegs64Bit[] = {
+ static const uint16_t CallerSavedRegs64Bit[] = {
X86::RAX, X86::RDX, X86::RCX, X86::RSI, X86::RDI,
X86::R8, X86::R9, X86::R10, X86::R11, 0
};
case X86::TCRETURNmi64:
case X86::EH_RETURN:
case X86::EH_RETURN64: {
- SmallSet<unsigned, 8> Uses;
+ SmallSet<uint16_t, 8> Uses;
for (unsigned i = 0, e = MBBI->getNumOperands(); i != e; ++i) {
MachineOperand &MO = MBBI->getOperand(i);
if (!MO.isReg() || MO.isDef())
unsigned Reg = MO.getReg();
if (!Reg)
continue;
- for (const unsigned *AsI = TRI.getOverlaps(Reg); *AsI; ++AsI)
- Uses.insert(*AsI);
+ for (MCRegAliasIterator AI(Reg, &TRI, true); AI.isValid(); ++AI)
+ Uses.insert(*AI);
}
- const unsigned *CS = Is64Bit ? CallerSavedRegs64Bit : CallerSavedRegs32Bit;
+ const uint16_t *CS = Is64Bit ? CallerSavedRegs64Bit : CallerSavedRegs32Bit;
for (; *CS; ++CS)
if (!Uses.count(*CS))
return *CS;
static
void emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
unsigned StackPtr, int64_t NumBytes,
- bool Is64Bit, const TargetInstrInfo &TII,
- const TargetRegisterInfo &TRI) {
+ bool Is64Bit, bool UseLEA,
+ const TargetInstrInfo &TII, const TargetRegisterInfo &TRI) {
bool isSub = NumBytes < 0;
uint64_t Offset = isSub ? -NumBytes : NumBytes;
- unsigned Opc = isSub ?
- getSUBriOpcode(Is64Bit, Offset) :
- getADDriOpcode(Is64Bit, Offset);
+ unsigned Opc;
+ if (UseLEA)
+ Opc = getLEArOpcode(Is64Bit);
+ else
+ Opc = isSub
+ ? getSUBriOpcode(Is64Bit, Offset)
+ : getADDriOpcode(Is64Bit, Offset);
+
uint64_t Chunk = (1LL << 31) - 1;
DebugLoc DL = MBB.findDebugLoc(MBBI);
}
}
- MachineInstr *MI =
- BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr)
- .addReg(StackPtr)
- .addImm(ThisVal);
+ MachineInstr *MI = NULL;
+
+ if (UseLEA) {
+ MI = addRegOffset(BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr),
+ StackPtr, false, isSub ? -ThisVal : ThisVal);
+ } else {
+ MI = BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr)
+ .addReg(StackPtr)
+ .addImm(ThisVal);
+ MI->getOperand(3).setIsDead(); // The EFLAGS implicit def is dead.
+ }
+
if (isSub)
MI->setFlag(MachineInstr::FrameSetup);
- MI->getOperand(3).setIsDead(); // The EFLAGS implicit def is dead.
+
Offset -= ThisVal;
}
}
MachineBasicBlock::iterator PI = prior(MBBI);
unsigned Opc = PI->getOpcode();
if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
- Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
+ Opc == X86::ADD32ri || Opc == X86::ADD32ri8 ||
+ Opc == X86::LEA32r || Opc == X86::LEA64_32r) &&
PI->getOperand(0).getReg() == StackPtr) {
if (NumBytes)
*NumBytes += PI->getOperand(2).getImm();
void mergeSPUpdatesDown(MachineBasicBlock &MBB,
MachineBasicBlock::iterator &MBBI,
unsigned StackPtr, uint64_t *NumBytes = NULL) {
- // FIXME: THIS ISN'T RUN!!!
+ // FIXME: THIS ISN'T RUN!!!
return;
if (MBBI == MBB.end()) return;
}
/// mergeSPUpdates - Checks the instruction before/after the passed
-/// instruction. If it is an ADD/SUB instruction it is deleted argument and the
-/// stack adjustment is returned as a positive value for ADD and a negative for
+/// instruction. If it is an ADD/SUB/LEA instruction it is deleted argument and the
+/// stack adjustment is returned as a positive value for ADD/LEA and a negative for
/// SUB.
static int mergeSPUpdates(MachineBasicBlock &MBB,
MachineBasicBlock::iterator &MBBI,
int Offset = 0;
if ((Opc == X86::ADD64ri32 || Opc == X86::ADD64ri8 ||
- Opc == X86::ADD32ri || Opc == X86::ADD32ri8) &&
+ Opc == X86::ADD32ri || Opc == X86::ADD32ri8 ||
+ Opc == X86::LEA32r || Opc == X86::LEA64_32r) &&
PI->getOperand(0).getReg() == StackPtr){
Offset += PI->getOperand(2).getImm();
MBB.erase(PI);
if (CSI.empty()) return;
std::vector<MachineMove> &Moves = MMI.getFrameMoves();
- const TargetData *TD = TM.getTargetData();
+ const X86RegisterInfo *RegInfo = TM.getRegisterInfo();
bool HasFP = hasFP(MF);
// Calculate amount of bytes used for return address storing.
- int stackGrowth = -TD->getPointerSize();
+ int stackGrowth = -RegInfo->getSlotSize();
// FIXME: This is dirty hack. The code itself is pretty mess right now.
// It should be rewritten from scratch and generalized sometimes.
/// getCompactUnwindRegNum - Get the compact unwind number for a given
/// register. The number corresponds to the enum lists in
/// compact_unwind_encoding.h.
-static int getCompactUnwindRegNum(const unsigned *CURegs, unsigned Reg) {
- int Idx = 1;
- for (; *CURegs; ++CURegs, ++Idx)
+static int getCompactUnwindRegNum(const uint16_t *CURegs, unsigned Reg) {
+ for (int Idx = 1; *CURegs; ++CURegs, ++Idx)
if (*CURegs == Reg)
return Idx;
return -1;
}
+// Number of registers that can be saved in a compact unwind encoding.
+#define CU_NUM_SAVED_REGS 6
+
/// encodeCompactUnwindRegistersWithoutFrame - Create the permutation encoding
/// used with frameless stacks. It is passed the number of registers to be saved
/// and an array of the registers saved.
-static uint32_t encodeCompactUnwindRegistersWithoutFrame(unsigned SavedRegs[6],
- unsigned RegCount,
- bool Is64Bit) {
+static uint32_t
+encodeCompactUnwindRegistersWithoutFrame(unsigned SavedRegs[CU_NUM_SAVED_REGS],
+ unsigned RegCount, bool Is64Bit) {
// The saved registers are numbered from 1 to 6. In order to encode the order
// in which they were saved, we re-number them according to their place in the
// register order. The re-numbering is relative to the last re-numbered
// 4 3
// 5 3
//
- static const unsigned CU32BitRegs[] = {
+ static const uint16_t CU32BitRegs[] = {
X86::EBX, X86::ECX, X86::EDX, X86::EDI, X86::ESI, X86::EBP, 0
};
- static const unsigned CU64BitRegs[] = {
+ static const uint16_t CU64BitRegs[] = {
X86::RBX, X86::R12, X86::R13, X86::R14, X86::R15, X86::RBP, 0
};
- const unsigned *CURegs = (Is64Bit ? CU64BitRegs : CU32BitRegs);
+ const uint16_t *CURegs = (Is64Bit ? CU64BitRegs : CU32BitRegs);
- uint32_t RenumRegs[6];
- for (unsigned i = 6 - RegCount; i < 6; ++i) {
+ for (unsigned i = 0; i != CU_NUM_SAVED_REGS; ++i) {
int CUReg = getCompactUnwindRegNum(CURegs, SavedRegs[i]);
if (CUReg == -1) return ~0U;
SavedRegs[i] = CUReg;
+ }
+ // Reverse the list.
+ std::swap(SavedRegs[0], SavedRegs[5]);
+ std::swap(SavedRegs[1], SavedRegs[4]);
+ std::swap(SavedRegs[2], SavedRegs[3]);
+
+ uint32_t RenumRegs[CU_NUM_SAVED_REGS];
+ for (unsigned i = CU_NUM_SAVED_REGS - RegCount; i < CU_NUM_SAVED_REGS; ++i) {
unsigned Countless = 0;
- for (unsigned j = 6 - RegCount; j < i; ++j)
+ for (unsigned j = CU_NUM_SAVED_REGS - RegCount; j < i; ++j)
if (SavedRegs[j] < SavedRegs[i])
++Countless;
/// encodeCompactUnwindRegistersWithFrame - Return the registers encoded for a
/// compact encoding with a frame pointer.
-static uint32_t encodeCompactUnwindRegistersWithFrame(unsigned SavedRegs[6],
- bool Is64Bit) {
- static const unsigned CU32BitRegs[] = {
+static uint32_t
+encodeCompactUnwindRegistersWithFrame(unsigned SavedRegs[CU_NUM_SAVED_REGS],
+ bool Is64Bit) {
+ static const uint16_t CU32BitRegs[] = {
X86::EBX, X86::ECX, X86::EDX, X86::EDI, X86::ESI, X86::EBP, 0
};
- static const unsigned CU64BitRegs[] = {
+ static const uint16_t CU64BitRegs[] = {
X86::RBX, X86::R12, X86::R13, X86::R14, X86::R15, X86::RBP, 0
};
- const unsigned *CURegs = (Is64Bit ? CU64BitRegs : CU32BitRegs);
+ const uint16_t *CURegs = (Is64Bit ? CU64BitRegs : CU32BitRegs);
// Encode the registers in the order they were saved, 3-bits per register. The
- // registers are numbered from 1 to 6.
+ // registers are numbered from 1 to CU_NUM_SAVED_REGS.
uint32_t RegEnc = 0;
- for (int I = 5; I >= 0; --I) {
+ for (int I = CU_NUM_SAVED_REGS - 1, Idx = 0; I != -1; --I) {
unsigned Reg = SavedRegs[I];
- if (Reg == 0) break;
+ if (Reg == 0) continue;
+
int CURegNum = getCompactUnwindRegNum(CURegs, Reg);
- if (CURegNum == -1)
- return ~0U;
- RegEnc |= (CURegNum & 0x7) << (5 - I);
+ if (CURegNum == -1) return ~0U;
+
+ // Encode the 3-bit register number in order, skipping over 3-bits for each
+ // register.
+ RegEnc |= (CURegNum & 0x7) << (Idx++ * 3);
}
- assert((RegEnc & 0x7FFF) == RegEnc && "Invalid compact register encoding!");
+ assert((RegEnc & 0x3FFFF) == RegEnc && "Invalid compact register encoding!");
return RegEnc;
}
unsigned FramePtr = RegInfo->getFrameRegister(MF);
unsigned StackPtr = RegInfo->getStackRegister();
- X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
- int TailCallReturnAddrDelta = X86FI->getTCReturnAddrDelta();
-
bool Is64Bit = STI.is64Bit();
bool HasFP = hasFP(MF);
- unsigned SavedRegs[6] = { 0, 0, 0, 0, 0, 0 };
- int SavedRegIdx = 6;
+ unsigned SavedRegs[CU_NUM_SAVED_REGS] = { 0, 0, 0, 0, 0, 0 };
+ unsigned SavedRegIdx = 0;
unsigned OffsetSize = (Is64Bit ? 8 : 4);
unsigned PushInstrSize = 1;
unsigned MoveInstr = (Is64Bit ? X86::MOV64rr : X86::MOV32rr);
unsigned MoveInstrSize = (Is64Bit ? 3 : 2);
- unsigned SubtractInstr = getSUBriOpcode(Is64Bit, -TailCallReturnAddrDelta);
unsigned SubtractInstrIdx = (Is64Bit ? 3 : 2);
unsigned StackDivide = (Is64Bit ? 8 : 4);
unsigned InstrOffset = 0;
- unsigned CFAOffset = 0;
unsigned StackAdjust = 0;
+ unsigned StackSize = 0;
MachineBasicBlock &MBB = MF.front(); // Prologue is in entry BB.
bool ExpectEnd = false;
if (Opc == PushInstr) {
// If there are too many saved registers, we cannot use compact encoding.
- if (--SavedRegIdx < 0) return 0;
+ if (SavedRegIdx >= CU_NUM_SAVED_REGS) return 0;
- SavedRegs[SavedRegIdx] = MI.getOperand(0).getReg();
- CFAOffset += OffsetSize;
+ SavedRegs[SavedRegIdx++] = MI.getOperand(0).getReg();
+ StackAdjust += OffsetSize;
InstrOffset += PushInstrSize;
} else if (Opc == MoveInstr) {
unsigned SrcReg = MI.getOperand(1).getReg();
if (DstReg != FramePtr || SrcReg != StackPtr)
return 0;
- CFAOffset = 0;
+ StackAdjust = 0;
memset(SavedRegs, 0, sizeof(SavedRegs));
- SavedRegIdx = 6;
+ SavedRegIdx = 0;
InstrOffset += MoveInstrSize;
- } else if (Opc == SubtractInstr) {
- if (StackAdjust)
- // We all ready have a stack pointer adjustment.
+ } else if (Opc == X86::SUB64ri32 || Opc == X86::SUB64ri8 ||
+ Opc == X86::SUB32ri || Opc == X86::SUB32ri8) {
+ if (StackSize)
+ // We already have a stack size.
return 0;
if (!MI.getOperand(0).isReg() ||
// %RSP<def> = SUB64ri8 %RSP, 48
return 0;
- StackAdjust = MI.getOperand(2).getImm() / StackDivide;
+ StackSize = MI.getOperand(2).getImm() / StackDivide;
SubtractInstrIdx += InstrOffset;
ExpectEnd = true;
}
// Encode that we are using EBP/RBP as the frame pointer.
uint32_t CompactUnwindEncoding = 0;
- CFAOffset /= StackDivide;
+ StackAdjust /= StackDivide;
if (HasFP) {
- if ((CFAOffset & 0xFF) != CFAOffset)
+ if ((StackAdjust & 0xFF) != StackAdjust)
// Offset was too big for compact encoding.
return 0;
// Get the encoding of the saved registers when we have a frame pointer.
uint32_t RegEnc = encodeCompactUnwindRegistersWithFrame(SavedRegs, Is64Bit);
- if (RegEnc == ~0U)
- return 0;
+ if (RegEnc == ~0U) return 0;
CompactUnwindEncoding |= 0x01000000;
- CompactUnwindEncoding |= (CFAOffset & 0xFF) << 16;
+ CompactUnwindEncoding |= (StackAdjust & 0xFF) << 16;
CompactUnwindEncoding |= RegEnc & 0x7FFF;
} else {
- unsigned FullOffset = CFAOffset + StackAdjust;
- if ((FullOffset & 0xFF) == FullOffset) {
- // Frameless stack.
+ ++StackAdjust;
+ uint32_t TotalStackSize = StackAdjust + StackSize;
+ if ((TotalStackSize & 0xFF) == TotalStackSize) {
+ // Frameless stack with a small stack size.
CompactUnwindEncoding |= 0x02000000;
- CompactUnwindEncoding |= (FullOffset & 0xFF) << 16;
+
+ // Encode the stack size.
+ CompactUnwindEncoding |= (TotalStackSize & 0xFF) << 16;
} else {
- if ((CFAOffset & 0x7) != CFAOffset)
+ if ((StackAdjust & 0x7) != StackAdjust)
// The extra stack adjustments are too big for us to handle.
return 0;
// instruction.
CompactUnwindEncoding |= (SubtractInstrIdx & 0xFF) << 16;
- // Encode any extra stack stack changes (done via push instructions).
- CompactUnwindEncoding |= (CFAOffset & 0x7) << 13;
+ // Encode any extra stack stack adjustments (done via push instructions).
+ CompactUnwindEncoding |= (StackAdjust & 0x7) << 13;
}
+ // Encode the number of registers saved.
+ CompactUnwindEncoding |= (SavedRegIdx & 0x7) << 10;
+
// Get the encoding of the saved registers when we don't have a frame
// pointer.
- uint32_t RegEnc = encodeCompactUnwindRegistersWithoutFrame(SavedRegs,
- 6 - SavedRegIdx,
- Is64Bit);
+ uint32_t RegEnc =
+ encodeCompactUnwindRegistersWithoutFrame(SavedRegs, SavedRegIdx,
+ Is64Bit);
if (RegEnc == ~0U) return 0;
+
+ // Encode the register encoding.
CompactUnwindEncoding |= RegEnc & 0x3FF;
}
return CompactUnwindEncoding;
}
+/// usesTheStack - This function checks if any of the users of EFLAGS
+/// copies the EFLAGS. We know that the code that lowers COPY of EFLAGS has
+/// to use the stack, and if we don't adjust the stack we clobber the first
+/// frame index.
+/// See X86InstrInfo::copyPhysReg.
+static bool usesTheStack(MachineFunction &MF) {
+ MachineRegisterInfo &MRI = MF.getRegInfo();
+
+ for (MachineRegisterInfo::reg_iterator ri = MRI.reg_begin(X86::EFLAGS),
+ re = MRI.reg_end(); ri != re; ++ri)
+ if (ri->isCopy())
+ return true;
+
+ return false;
+}
+
/// emitPrologue - Push callee-saved registers onto the stack, which
/// automatically adjust the stack pointer. Adjust the stack pointer to allocate
/// space for local variables. Also emit labels used by the exception handler to
bool HasFP = hasFP(MF);
bool Is64Bit = STI.is64Bit();
bool IsWin64 = STI.isTargetWin64();
+ bool UseLEA = STI.useLeaForSP();
unsigned StackAlign = getStackAlignment();
unsigned SlotSize = RegInfo->getSlotSize();
unsigned FramePtr = RegInfo->getFrameRegister(MF);
unsigned StackPtr = RegInfo->getStackRegister();
+ unsigned BasePtr = RegInfo->getBaseRegister();
DebugLoc DL;
// If we're forcing a stack realignment we can't rely on just the frame
// If this is x86-64 and the Red Zone is not disabled, if we are a leaf
// function, and use up to 128 bytes of stack space, 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).
- if (Is64Bit && !Fn->hasFnAttr(Attribute::NoRedZone) &&
+ // stack pointer (we fit in the Red Zone). We also check that we don't
+ // push and pop from the stack.
+ if (Is64Bit && !Fn->getFnAttributes().hasAttribute(Attribute::NoRedZone) &&
!RegInfo->needsStackRealignment(MF) &&
- !MFI->hasVarSizedObjects() && // No dynamic alloca.
- !MFI->adjustsStack() && // No calls.
- !IsWin64 && // Win64 has no Red Zone
- !EnableSegmentedStacks) { // Regular stack
+ !MFI->hasVarSizedObjects() && // No dynamic alloca.
+ !MFI->adjustsStack() && // No calls.
+ !IsWin64 && // Win64 has no Red Zone
+ !usesTheStack(MF) && // Don't push and pop.
+ !MF.getTarget().Options.EnableSegmentedStacks) { // Regular stack
uint64_t MinSize = X86FI->getCalleeSavedFrameSize();
if (HasFP) MinSize += SlotSize;
StackSize = std::max(MinSize, StackSize > 128 ? StackSize - 128 : 0);
// ELSE => DW_CFA_offset_extended
std::vector<MachineMove> &Moves = MMI.getFrameMoves();
- const TargetData *TD = MF.getTarget().getTargetData();
uint64_t NumBytes = 0;
- int stackGrowth = -TD->getPointerSize();
+ int stackGrowth = -SlotSize;
if (HasFP) {
// Calculate required stack adjustment.
uint64_t FrameSize = StackSize - SlotSize;
- if (RegInfo->needsStackRealignment(MF))
- FrameSize = (FrameSize + MaxAlign - 1) / MaxAlign * MaxAlign;
-
- NumBytes = FrameSize - X86FI->getCalleeSavedFrameSize();
+ if (RegInfo->needsStackRealignment(MF)) {
+ // Callee-saved registers are pushed on stack before the stack
+ // is realigned.
+ FrameSize -= X86FI->getCalleeSavedFrameSize();
+ NumBytes = (FrameSize + MaxAlign - 1) / MaxAlign * MaxAlign;
+ } else {
+ NumBytes = FrameSize - X86FI->getCalleeSavedFrameSize();
+ }
// Get the offset of the stack slot for the EBP register, which is
// guaranteed to be the last slot by processFunctionBeforeFrameFinalized.
for (MachineFunction::iterator I = llvm::next(MF.begin()), E = MF.end();
I != E; ++I)
I->addLiveIn(FramePtr);
-
- // Realign stack
- if (RegInfo->needsStackRealignment(MF)) {
- MachineInstr *MI =
- BuildMI(MBB, MBBI, DL,
- TII.get(Is64Bit ? X86::AND64ri32 : X86::AND32ri), StackPtr)
- .addReg(StackPtr)
- .addImm(-MaxAlign)
- .setMIFlag(MachineInstr::FrameSetup);
-
- // The EFLAGS implicit def is dead.
- MI->getOperand(3).setIsDead();
- }
} else {
NumBytes = StackSize - X86FI->getCalleeSavedFrameSize();
}
}
}
- DL = MBB.findDebugLoc(MBBI);
+ // Realign stack after we pushed callee-saved registers (so that we'll be
+ // able to calculate their offsets from the frame pointer).
+
+ // NOTE: We push the registers before realigning the stack, so
+ // vector callee-saved (xmm) registers may be saved w/o proper
+ // alignment in this way. However, currently these regs are saved in
+ // stack slots (see X86FrameLowering::spillCalleeSavedRegisters()), so
+ // this shouldn't be a problem.
+ if (RegInfo->needsStackRealignment(MF)) {
+ assert(HasFP && "There should be a frame pointer if stack is realigned.");
+ MachineInstr *MI =
+ BuildMI(MBB, MBBI, DL,
+ TII.get(Is64Bit ? X86::AND64ri32 : X86::AND32ri), StackPtr)
+ .addReg(StackPtr)
+ .addImm(-MaxAlign)
+ .setMIFlag(MachineInstr::FrameSetup);
+
+ // The EFLAGS implicit def is dead.
+ MI->getOperand(3).setIsDead();
+ }
// If there is an SUB32ri of ESP immediately before this instruction, merge
// the two. This can be the case when tail call elimination is enabled and
// FIXME: %rax preserves the offset and should be available.
if (isSPUpdateNeeded)
emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit,
- TII, *RegInfo);
+ UseLEA, TII, *RegInfo);
if (isEAXAlive) {
// Restore EAX
}
} else if (NumBytes)
emitSPUpdate(MBB, MBBI, StackPtr, -(int64_t)NumBytes, Is64Bit,
- TII, *RegInfo);
+ UseLEA, TII, *RegInfo);
+
+ // If we need a base pointer, set it up here. It's whatever the value
+ // of the stack pointer is at this point. Any variable size objects
+ // will be allocated after this, so we can still use the base pointer
+ // to reference locals.
+ if (RegInfo->hasBasePointer(MF)) {
+ // Update the frame pointer with the current stack pointer.
+ unsigned Opc = Is64Bit ? X86::MOV64rr : X86::MOV32rr;
+ BuildMI(MBB, MBBI, DL, TII.get(Opc), BasePtr)
+ .addReg(StackPtr)
+ .setMIFlag(MachineInstr::FrameSetup);
+ }
if (( (!HasFP && NumBytes) || PushedRegs) && needsFrameMoves) {
// Mark end of stack pointer adjustment.
unsigned RetOpcode = MBBI->getOpcode();
DebugLoc DL = MBBI->getDebugLoc();
bool Is64Bit = STI.is64Bit();
+ bool UseLEA = STI.useLeaForSP();
unsigned StackAlign = getStackAlignment();
unsigned SlotSize = RegInfo->getSlotSize();
unsigned FramePtr = RegInfo->getFrameRegister(MF);
if (hasFP(MF)) {
// Calculate required stack adjustment.
uint64_t FrameSize = StackSize - SlotSize;
- if (RegInfo->needsStackRealignment(MF))
- FrameSize = (FrameSize + MaxAlign - 1)/MaxAlign*MaxAlign;
-
- NumBytes = FrameSize - CSSize;
+ if (RegInfo->needsStackRealignment(MF)) {
+ // Callee-saved registers were pushed on stack before the stack
+ // was realigned.
+ FrameSize -= CSSize;
+ NumBytes = (FrameSize + MaxAlign - 1) / MaxAlign * MaxAlign;
+ } else {
+ NumBytes = FrameSize - CSSize;
+ }
// Pop EBP.
BuildMI(MBB, MBBI, DL,
}
// Skip the callee-saved pop instructions.
- MachineBasicBlock::iterator LastCSPop = MBBI;
while (MBBI != MBB.begin()) {
MachineBasicBlock::iterator PI = prior(MBBI);
unsigned Opc = PI->getOpcode();
if (Opc != X86::POP32r && Opc != X86::POP64r && Opc != X86::DBG_VALUE &&
- !PI->getDesc().isTerminator())
+ !PI->isTerminator())
break;
--MBBI;
}
+ MachineBasicBlock::iterator FirstCSPop = MBBI;
DL = MBBI->getDebugLoc();
// If dynamic alloca is used, then reset esp to point to the last callee-saved
// slot before popping them off! Same applies for the case, when stack was
// realigned.
- if (RegInfo->needsStackRealignment(MF)) {
- // We cannot use LEA here, because stack pointer was realigned. We need to
- // deallocate local frame back.
- if (CSSize) {
- emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, TII, *RegInfo);
- MBBI = prior(LastCSPop);
- }
-
- BuildMI(MBB, MBBI, DL,
- TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr),
- StackPtr).addReg(FramePtr);
- } else if (MFI->hasVarSizedObjects()) {
- if (CSSize) {
- unsigned Opc = Is64Bit ? X86::LEA64r : X86::LEA32r;
- MachineInstr *MI =
- addRegOffset(BuildMI(MF, DL, TII.get(Opc), StackPtr),
- FramePtr, false, -CSSize);
- MBB.insert(MBBI, MI);
+ if (RegInfo->needsStackRealignment(MF) || MFI->hasVarSizedObjects()) {
+ if (RegInfo->needsStackRealignment(MF))
+ MBBI = FirstCSPop;
+ if (CSSize != 0) {
+ unsigned Opc = getLEArOpcode(Is64Bit);
+ addRegOffset(BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr),
+ FramePtr, false, -CSSize);
} else {
- BuildMI(MBB, MBBI, DL,
- TII.get(Is64Bit ? X86::MOV64rr : X86::MOV32rr), StackPtr)
+ unsigned Opc = (Is64Bit ? X86::MOV64rr : X86::MOV32rr);
+ BuildMI(MBB, MBBI, DL, TII.get(Opc), StackPtr)
.addReg(FramePtr);
}
} else if (NumBytes) {
// Adjust stack pointer back: ESP += numbytes.
- emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, TII, *RegInfo);
+ emitSPUpdate(MBB, MBBI, StackPtr, NumBytes, Is64Bit, UseLEA, TII, *RegInfo);
}
// We're returning from function via eh_return.
if (Offset) {
// Check for possible merge with preceding ADD instruction.
Offset += mergeSPUpdates(MBB, MBBI, StackPtr, true);
- emitSPUpdate(MBB, MBBI, StackPtr, Offset, Is64Bit, TII, *RegInfo);
+ emitSPUpdate(MBB, MBBI, StackPtr, Offset, Is64Bit, UseLEA, TII, *RegInfo);
}
// Jump to label or value in register.
}
MachineInstr *NewMI = prior(MBBI);
- for (unsigned i = 2, e = MBBI->getNumOperands(); i != e; ++i)
- NewMI->addOperand(MBBI->getOperand(i));
+ NewMI->copyImplicitOps(MF, MBBI);
// Delete the pseudo instruction TCRETURN.
MBB.erase(MBBI);
// Check for possible merge with preceding ADD instruction.
delta += mergeSPUpdates(MBB, MBBI, StackPtr, true);
- emitSPUpdate(MBB, MBBI, StackPtr, delta, Is64Bit, TII, *RegInfo);
+ emitSPUpdate(MBB, MBBI, StackPtr, delta, Is64Bit, UseLEA, TII, *RegInfo);
}
}
int X86FrameLowering::getFrameIndexOffset(const MachineFunction &MF, int FI) const {
- const X86RegisterInfo *RI =
+ const X86RegisterInfo *RegInfo =
static_cast<const X86RegisterInfo*>(MF.getTarget().getRegisterInfo());
const MachineFrameInfo *MFI = MF.getFrameInfo();
int Offset = MFI->getObjectOffset(FI) - getOffsetOfLocalArea();
uint64_t StackSize = MFI->getStackSize();
- if (RI->needsStackRealignment(MF)) {
+ if (RegInfo->hasBasePointer(MF)) {
+ assert (hasFP(MF) && "VLAs and dynamic stack realign, but no FP?!");
+ if (FI < 0) {
+ // Skip the saved EBP.
+ return Offset + RegInfo->getSlotSize();
+ } else {
+ assert((-(Offset + StackSize)) % MFI->getObjectAlignment(FI) == 0);
+ return Offset + StackSize;
+ }
+ } else if (RegInfo->needsStackRealignment(MF)) {
if (FI < 0) {
// Skip the saved EBP.
- Offset += RI->getSlotSize();
+ return Offset + RegInfo->getSlotSize();
} else {
assert((-(Offset + StackSize)) % MFI->getObjectAlignment(FI) == 0);
return Offset + StackSize;
return Offset + StackSize;
// Skip the saved EBP.
- Offset += RI->getSlotSize();
+ Offset += RegInfo->getSlotSize();
// Skip the RETADDR move area
const X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
return Offset;
}
+int X86FrameLowering::getFrameIndexReference(const MachineFunction &MF, int FI,
+ unsigned &FrameReg) const {
+ const X86RegisterInfo *RegInfo =
+ static_cast<const X86RegisterInfo*>(MF.getTarget().getRegisterInfo());
+ // We can't calculate offset from frame pointer if the stack is realigned,
+ // so enforce usage of stack/base pointer. The base pointer is used when we
+ // have dynamic allocas in addition to dynamic realignment.
+ if (RegInfo->hasBasePointer(MF))
+ FrameReg = RegInfo->getBaseRegister();
+ else if (RegInfo->needsStackRealignment(MF))
+ FrameReg = RegInfo->getStackRegister();
+ else
+ FrameReg = RegInfo->getFrameRegister(MF);
+ return getFrameIndexOffset(MF, FI);
+}
+
bool X86FrameLowering::spillCalleeSavedRegisters(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
const std::vector<CalleeSavedInfo> &CSI,
"Slot for EBP register must be last in order to be found!");
(void)FrameIdx;
}
+
+ // Spill the BasePtr if it's used.
+ if (RegInfo->hasBasePointer(MF))
+ MF.getRegInfo().setPhysRegUsed(RegInfo->getBaseRegister());
}
static bool
return false;
}
+
+/// GetScratchRegister - Get a register for performing work in the segmented
+/// stack prologue. Depending on platform and the properties of the function
+/// either one or two registers will be needed. Set primary to true for
+/// the first register, false for the second.
static unsigned
-GetScratchRegister(bool Is64Bit, const MachineFunction &MF) {
- if (Is64Bit) {
- return X86::R11;
- } else {
- CallingConv::ID CallingConvention = MF.getFunction()->getCallingConv();
- bool IsNested = HasNestArgument(&MF);
-
- if (CallingConvention == CallingConv::X86_FastCall) {
- if (IsNested) {
- report_fatal_error("Segmented stacks does not support fastcall with "
- "nested function.");
- return -1;
- } else {
- return X86::EAX;
- }
- } else {
- if (IsNested)
- return X86::EDX;
- else
- return X86::ECX;
- }
+GetScratchRegister(bool Is64Bit, const MachineFunction &MF, bool Primary) {
+ if (Is64Bit)
+ return Primary ? X86::R11 : X86::R12;
+
+ CallingConv::ID CallingConvention = MF.getFunction()->getCallingConv();
+ bool IsNested = HasNestArgument(&MF);
+
+ if (CallingConvention == CallingConv::X86_FastCall ||
+ CallingConvention == CallingConv::Fast) {
+ if (IsNested)
+ report_fatal_error("Segmented stacks does not support fastcall with "
+ "nested function.");
+ return Primary ? X86::EAX : X86::ECX;
}
+ if (IsNested)
+ return Primary ? X86::EDX : X86::EAX;
+ return Primary ? X86::ECX : X86::EAX;
}
+// The stack limit in the TCB is set to this many bytes above the actual stack
+// limit.
+static const uint64_t kSplitStackAvailable = 256;
+
void
X86FrameLowering::adjustForSegmentedStacks(MachineFunction &MF) const {
MachineBasicBlock &prologueMBB = MF.front();
DebugLoc DL;
const X86Subtarget *ST = &MF.getTarget().getSubtarget<X86Subtarget>();
- unsigned ScratchReg = GetScratchRegister(Is64Bit, MF);
+ unsigned ScratchReg = GetScratchRegister(Is64Bit, MF, true);
assert(!MF.getRegInfo().isLiveIn(ScratchReg) &&
"Scratch register is live-in");
if (MF.getFunction()->isVarArg())
report_fatal_error("Segmented stacks do not support vararg functions.");
- if (!ST->isTargetLinux())
- report_fatal_error("Segmented stacks supported only on linux.");
+ if (!ST->isTargetLinux() && !ST->isTargetDarwin() &&
+ !ST->isTargetWin32() && !ST->isTargetFreeBSD())
+ report_fatal_error("Segmented stacks not supported on this platform.");
MachineBasicBlock *allocMBB = MF.CreateMachineBasicBlock();
MachineBasicBlock *checkMBB = MF.CreateMachineBasicBlock();
// prologue.
StackSize = MFI->getStackSize();
+ // When the frame size is less than 256 we just compare the stack
+ // boundary directly to the value of the stack pointer, per gcc.
+ bool CompareStackPointer = StackSize < kSplitStackAvailable;
+
// Read the limit off the current stacklet off the stack_guard location.
if (Is64Bit) {
- TlsReg = X86::FS;
- TlsOffset = 0x70;
+ if (ST->isTargetLinux()) {
+ TlsReg = X86::FS;
+ TlsOffset = 0x70;
+ } else if (ST->isTargetDarwin()) {
+ TlsReg = X86::GS;
+ TlsOffset = 0x60 + 90*8; // See pthread_machdep.h. Steal TLS slot 90.
+ } else if (ST->isTargetFreeBSD()) {
+ TlsReg = X86::FS;
+ TlsOffset = 0x18;
+ } else {
+ report_fatal_error("Segmented stacks not supported on this platform.");
+ }
+
+ if (CompareStackPointer)
+ ScratchReg = X86::RSP;
+ else
+ BuildMI(checkMBB, DL, TII.get(X86::LEA64r), ScratchReg).addReg(X86::RSP)
+ .addImm(1).addReg(0).addImm(-StackSize).addReg(0);
- BuildMI(checkMBB, DL, TII.get(X86::LEA64r), ScratchReg).addReg(X86::RSP)
- .addImm(0).addReg(0).addImm(-StackSize).addReg(0);
BuildMI(checkMBB, DL, TII.get(X86::CMP64rm)).addReg(ScratchReg)
- .addReg(0).addImm(0).addReg(0).addImm(TlsOffset).addReg(TlsReg);
+ .addReg(0).addImm(1).addReg(0).addImm(TlsOffset).addReg(TlsReg);
} else {
- TlsReg = X86::GS;
- TlsOffset = 0x30;
+ if (ST->isTargetLinux()) {
+ TlsReg = X86::GS;
+ TlsOffset = 0x30;
+ } else if (ST->isTargetDarwin()) {
+ TlsReg = X86::GS;
+ TlsOffset = 0x48 + 90*4;
+ } else if (ST->isTargetWin32()) {
+ TlsReg = X86::FS;
+ TlsOffset = 0x14; // pvArbitrary, reserved for application use
+ } else if (ST->isTargetFreeBSD()) {
+ report_fatal_error("Segmented stacks not supported on FreeBSD i386.");
+ } else {
+ report_fatal_error("Segmented stacks not supported on this platform.");
+ }
+
+ if (CompareStackPointer)
+ ScratchReg = X86::ESP;
+ else
+ BuildMI(checkMBB, DL, TII.get(X86::LEA32r), ScratchReg).addReg(X86::ESP)
+ .addImm(1).addReg(0).addImm(-StackSize).addReg(0);
+
+ if (ST->isTargetLinux() || ST->isTargetWin32()) {
+ BuildMI(checkMBB, DL, TII.get(X86::CMP32rm)).addReg(ScratchReg)
+ .addReg(0).addImm(0).addReg(0).addImm(TlsOffset).addReg(TlsReg);
+ } else if (ST->isTargetDarwin()) {
+
+ // TlsOffset doesn't fit into a mod r/m byte so we need an extra register
+ unsigned ScratchReg2;
+ bool SaveScratch2;
+ if (CompareStackPointer) {
+ // The primary scratch register is available for holding the TLS offset
+ ScratchReg2 = GetScratchRegister(Is64Bit, MF, true);
+ SaveScratch2 = false;
+ } else {
+ // Need to use a second register to hold the TLS offset
+ ScratchReg2 = GetScratchRegister(Is64Bit, MF, false);
+
+ // Unfortunately, with fastcc the second scratch register may hold an arg
+ SaveScratch2 = MF.getRegInfo().isLiveIn(ScratchReg2);
+ }
+
+ // If Scratch2 is live-in then it needs to be saved
+ assert((!MF.getRegInfo().isLiveIn(ScratchReg2) || SaveScratch2) &&
+ "Scratch register is live-in and not saved");
- BuildMI(checkMBB, DL, TII.get(X86::LEA32r), ScratchReg).addReg(X86::ESP)
- .addImm(0).addReg(0).addImm(-StackSize).addReg(0);
- BuildMI(checkMBB, DL, TII.get(X86::CMP32rm)).addReg(ScratchReg)
- .addReg(0).addImm(0).addReg(0).addImm(TlsOffset).addReg(TlsReg);
+ if (SaveScratch2)
+ BuildMI(checkMBB, DL, TII.get(X86::PUSH32r))
+ .addReg(ScratchReg2, RegState::Kill);
+
+ BuildMI(checkMBB, DL, TII.get(X86::MOV32ri), ScratchReg2)
+ .addImm(TlsOffset);
+ BuildMI(checkMBB, DL, TII.get(X86::CMP32rm))
+ .addReg(ScratchReg)
+ .addReg(ScratchReg2).addImm(1).addReg(0)
+ .addImm(0)
+ .addReg(TlsReg);
+
+ if (SaveScratch2)
+ BuildMI(checkMBB, DL, TII.get(X86::POP32r), ScratchReg2);
+ }
}
// This jump is taken if SP >= (Stacklet Limit + Stack Space required).
// It jumps to normal execution of the function body.
- BuildMI(checkMBB, DL, TII.get(X86::JG_4)).addMBB(&prologueMBB);
+ BuildMI(checkMBB, DL, TII.get(X86::JA_4)).addMBB(&prologueMBB);
// On 32 bit we first push the arguments size and then the frame size. On 64
// bit, we pass the stack frame size in r10 and the argument size in r11.
MF.getRegInfo().setPhysRegUsed(X86::R10);
MF.getRegInfo().setPhysRegUsed(X86::R11);
} else {
- // Since we'll call __morestack, stack alignment needs to be preserved.
- BuildMI(allocMBB, DL, TII.get(X86::SUB32ri), X86::ESP).addReg(X86::ESP)
- .addImm(8);
BuildMI(allocMBB, DL, TII.get(X86::PUSHi32))
.addImm(X86FI->getArgumentStackSize());
BuildMI(allocMBB, DL, TII.get(X86::PUSHi32))
BuildMI(allocMBB, DL, TII.get(X86::CALLpcrel32))
.addExternalSymbol("__morestack");
- // __morestack only seems to remove 8 bytes off the stack. Add back the
- // additional 8 bytes we added before pushing the arguments.
- if (!Is64Bit)
- BuildMI(allocMBB, DL, TII.get(X86::ADD32ri), X86::ESP).addReg(X86::ESP)
- .addImm(8);
if (IsNested)
BuildMI(allocMBB, DL, TII.get(X86::MORESTACK_RET_RESTORE_R10));
else