#include "PPCRegisterInfo.h"
#include "PPCFrameInfo.h"
#include "PPCSubtarget.h"
+#include "llvm/CallingConv.h"
#include "llvm/Constants.h"
+#include "llvm/Function.h"
#include "llvm/Type.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineLocation.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/RegisterScavenging.h"
-#include "llvm/CodeGen/SelectionDAGNodes.h"
#include "llvm/Target/TargetFrameInfo.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetMachine.h"
ImmToIdxMap[PPC::ADDI8] = PPC::ADD8; ImmToIdxMap[PPC::STD_32] = PPC::STDX_32;
}
-void PPCRegisterInfo::reMaterialize(MachineBasicBlock &MBB,
- MachineBasicBlock::iterator I,
- unsigned DestReg,
- const MachineInstr *Orig) const {
- MachineInstr *MI = Orig->clone();
- MI->getOperand(0).setReg(DestReg);
- MBB.insert(I, MI);
+/// getPointerRegClass - Return the register class to use to hold pointers.
+/// This is used for addressing modes.
+const TargetRegisterClass *PPCRegisterInfo::getPointerRegClass() const {
+ if (Subtarget.isPPC64())
+ return &PPC::G8RCRegClass;
+ else
+ return &PPC::GPRCRegClass;
}
const unsigned*
//
static bool needsFP(const MachineFunction &MF) {
const MachineFrameInfo *MFI = MF.getFrameInfo();
- return NoFramePointerElim || MFI->hasVarSizedObjects();
+ return NoFramePointerElim || MFI->hasVarSizedObjects() ||
+ (PerformTailCallOpt && MF.getInfo<PPCFunctionInfo>()->hasFastCall());
}
static bool spillsCR(const MachineFunction &MF) {
Reserved.set(PPC::R1);
Reserved.set(PPC::LR);
Reserved.set(PPC::LR8);
+ Reserved.set(PPC::RM);
// In Linux, r2 is reserved for the OS.
if (!Subtarget.isDarwin())
/// MustSaveLR - Return true if this function requires that we save the LR
/// register onto the stack in the prolog and restore it in the epilog of the
/// function.
-static bool MustSaveLR(const MachineFunction &MF) {
+static bool MustSaveLR(const MachineFunction &MF, unsigned LR) {
const PPCFunctionInfo *MFI = MF.getInfo<PPCFunctionInfo>();
- // We need an save/restore of LR if there is any use/def of LR explicitly, or
- // if there is some use of the LR stack slot (e.g. for builtin_return_address.
- return MFI->usesLR() || MFI->isLRStoreRequired() ||
- // FIXME: Anything that has a call should clobber the LR register,
- // isn't this redundant??
- MF.getFrameInfo()->hasCalls();
+ // We need a save/restore of LR if there is any def of LR (which is
+ // defined by calls, including the PIC setup sequence), or if there is
+ // some use of the LR stack slot (e.g. for builtin_return_address).
+ // (LR comes in 32 and 64 bit versions.)
+ MachineRegisterInfo::def_iterator RI = MF.getRegInfo().def_begin(LR);
+ return RI !=MF.getRegInfo().def_end() || MFI->isLRStoreRequired();
}
+
+
void PPCRegisterInfo::
eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
MachineBasicBlock::iterator I) const {
+ if (PerformTailCallOpt && I->getOpcode() == PPC::ADJCALLSTACKUP) {
+ // Add (actually subtract) back the amount the callee popped on return.
+ if (int CalleeAmt = I->getOperand(1).getImm()) {
+ bool is64Bit = Subtarget.isPPC64();
+ CalleeAmt *= -1;
+ unsigned StackReg = is64Bit ? PPC::X1 : PPC::R1;
+ unsigned TmpReg = is64Bit ? PPC::X0 : PPC::R0;
+ unsigned ADDIInstr = is64Bit ? PPC::ADDI8 : PPC::ADDI;
+ unsigned ADDInstr = is64Bit ? PPC::ADD8 : PPC::ADD4;
+ unsigned LISInstr = is64Bit ? PPC::LIS8 : PPC::LIS;
+ unsigned ORIInstr = is64Bit ? PPC::ORI8 : PPC::ORI;
+ MachineInstr *MI = I;
+ DebugLoc dl = MI->getDebugLoc();
+
+ if (isInt16(CalleeAmt)) {
+ BuildMI(MBB, I, dl, TII.get(ADDIInstr), StackReg).addReg(StackReg).
+ addImm(CalleeAmt);
+ } else {
+ MachineBasicBlock::iterator MBBI = I;
+ BuildMI(MBB, MBBI, dl, TII.get(LISInstr), TmpReg)
+ .addImm(CalleeAmt >> 16);
+ BuildMI(MBB, MBBI, dl, TII.get(ORIInstr), TmpReg)
+ .addReg(TmpReg, RegState::Kill)
+ .addImm(CalleeAmt & 0xFFFF);
+ BuildMI(MBB, MBBI, dl, TII.get(ADDInstr))
+ .addReg(StackReg)
+ .addReg(StackReg)
+ .addReg(TmpReg);
+ }
+ }
+ }
// Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions.
MBB.erase(I);
}
/// lowerDynamicAlloc - Generate the code for allocating an object in the
/// current frame. The sequence of code with be in the general form
///
-/// addi R0, SP, #frameSize ; get the address of the previous frame
+/// addi R0, SP, \#frameSize ; get the address of the previous frame
/// stwxu R0, SP, Rnegsize ; add and update the SP with the negated size
-/// addi Rnew, SP, #maxCalFrameSize ; get the top of the allocation
+/// addi Rnew, SP, \#maxCalFrameSize ; get the top of the allocation
///
void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II,
int SPAdj, RegScavenger *RS) const {
MachineFrameInfo *MFI = MF.getFrameInfo();
// Determine whether 64-bit pointers are used.
bool LP64 = Subtarget.isPPC64();
+ DebugLoc dl = MI.getDebugLoc();
// Get the maximum call stack size.
unsigned maxCallFrameSize = MFI->getMaxCallFrameSize();
Reg = PPC::R0;
if (MaxAlign < TargetAlign && isInt16(FrameSize)) {
- BuildMI(MBB, II, TII.get(PPC::ADDI), Reg)
+ BuildMI(MBB, II, dl, TII.get(PPC::ADDI), Reg)
.addReg(PPC::R31)
.addImm(FrameSize);
} else if (LP64) {
if (EnableRegisterScavenging) // FIXME (64-bit): Use "true" part.
- BuildMI(MBB, II, TII.get(PPC::LD), Reg)
- .addImm(0)
- .addReg(PPC::X1);
+ BuildMI(MBB, II, dl, TII.get(PPC::LD), Reg)
+ .addImm(0)
+ .addReg(PPC::X1);
else
- BuildMI(MBB, II, TII.get(PPC::LD), PPC::X0)
- .addImm(0)
- .addReg(PPC::X1);
+ BuildMI(MBB, II, dl, TII.get(PPC::LD), PPC::X0)
+ .addImm(0)
+ .addReg(PPC::X1);
} else {
- BuildMI(MBB, II, TII.get(PPC::LWZ), Reg)
+ BuildMI(MBB, II, dl, TII.get(PPC::LWZ), Reg)
.addImm(0)
.addReg(PPC::R1);
}
// address of new allocated space.
if (LP64) {
if (EnableRegisterScavenging) // FIXME (64-bit): Use "true" part.
- BuildMI(MBB, II, TII.get(PPC::STDUX))
- .addReg(Reg, false, false, true)
- .addReg(PPC::X1)
- .addReg(MI.getOperand(1).getReg());
+ BuildMI(MBB, II, dl, TII.get(PPC::STDUX))
+ .addReg(Reg, RegState::Kill)
+ .addReg(PPC::X1)
+ .addReg(MI.getOperand(1).getReg());
else
- BuildMI(MBB, II, TII.get(PPC::STDUX))
- .addReg(PPC::X0, false, false, true)
- .addReg(PPC::X1)
- .addReg(MI.getOperand(1).getReg());
+ BuildMI(MBB, II, dl, TII.get(PPC::STDUX))
+ .addReg(PPC::X0, RegState::Kill)
+ .addReg(PPC::X1)
+ .addReg(MI.getOperand(1).getReg());
if (!MI.getOperand(1).isKill())
- BuildMI(MBB, II, TII.get(PPC::ADDI8), MI.getOperand(0).getReg())
- .addReg(PPC::X1)
- .addImm(maxCallFrameSize);
+ BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg())
+ .addReg(PPC::X1)
+ .addImm(maxCallFrameSize);
else
// Implicitly kill the register.
- BuildMI(MBB, II, TII.get(PPC::ADDI8), MI.getOperand(0).getReg())
- .addReg(PPC::X1)
- .addImm(maxCallFrameSize)
- .addReg(MI.getOperand(1).getReg(), false, true, true);
+ BuildMI(MBB, II, dl, TII.get(PPC::ADDI8), MI.getOperand(0).getReg())
+ .addReg(PPC::X1)
+ .addImm(maxCallFrameSize)
+ .addReg(MI.getOperand(1).getReg(), RegState::ImplicitKill);
} else {
- BuildMI(MBB, II, TII.get(PPC::STWUX))
- .addReg(Reg, false, false, true)
+ BuildMI(MBB, II, dl, TII.get(PPC::STWUX))
+ .addReg(Reg, RegState::Kill)
.addReg(PPC::R1)
.addReg(MI.getOperand(1).getReg());
if (!MI.getOperand(1).isKill())
- BuildMI(MBB, II, TII.get(PPC::ADDI), MI.getOperand(0).getReg())
- .addReg(PPC::R1)
- .addImm(maxCallFrameSize);
+ BuildMI(MBB, II, dl, TII.get(PPC::ADDI), MI.getOperand(0).getReg())
+ .addReg(PPC::R1)
+ .addImm(maxCallFrameSize);
else
// Implicitly kill the register.
- BuildMI(MBB, II, TII.get(PPC::ADDI), MI.getOperand(0).getReg())
- .addReg(PPC::R1)
- .addImm(maxCallFrameSize)
- .addReg(MI.getOperand(1).getReg(), false, true, true);
+ BuildMI(MBB, II, dl, TII.get(PPC::ADDI), MI.getOperand(0).getReg())
+ .addReg(PPC::R1)
+ .addImm(maxCallFrameSize)
+ .addReg(MI.getOperand(1).getReg(), RegState::ImplicitKill);
}
// Discard the DYNALLOC instruction.
MachineInstr &MI = *II; // ; SPILL_CR <SrcReg>, <offset>, <FI>
// Get the instruction's basic block.
MachineBasicBlock &MBB = *MI.getParent();
+ DebugLoc dl = MI.getDebugLoc();
const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
// We need to store the CR in the low 4-bits of the saved value. First, issue
// an MFCR to save all of the CRBits. Add an implicit kill of the CR.
if (!MI.getOperand(0).isKill())
- BuildMI(MBB, II, TII.get(PPC::MFCR), Reg);
+ BuildMI(MBB, II, dl, TII.get(PPC::MFCR), Reg);
else
// Implicitly kill the CR register.
- BuildMI(MBB, II, TII.get(PPC::MFCR), Reg)
- .addReg(MI.getOperand(0).getReg(), false, true, true);
+ BuildMI(MBB, II, dl, TII.get(PPC::MFCR), Reg)
+ .addReg(MI.getOperand(0).getReg(), RegState::ImplicitKill);
// If the saved register wasn't CR0, shift the bits left so that they are in
// CR0's slot.
unsigned SrcReg = MI.getOperand(0).getReg();
if (SrcReg != PPC::CR0)
// rlwinm rA, rA, ShiftBits, 0, 31.
- BuildMI(MBB, II, TII.get(PPC::RLWINM), Reg)
- .addReg(Reg, false, false, true)
+ BuildMI(MBB, II, dl, TII.get(PPC::RLWINM), Reg)
+ .addReg(Reg, RegState::Kill)
.addImm(PPCRegisterInfo::getRegisterNumbering(SrcReg) * 4)
.addImm(0)
.addImm(31);
- addFrameReference(BuildMI(MBB, II, TII.get(PPC::STW))
- .addReg(Reg, false, false, MI.getOperand(1).getImm()),
+ addFrameReference(BuildMI(MBB, II, dl, TII.get(PPC::STW))
+ .addReg(Reg, getKillRegState(MI.getOperand(1).getImm())),
FrameIndex);
// Discard the pseudo instruction.
MachineFunction &MF = *MBB.getParent();
// Get the frame info.
MachineFrameInfo *MFI = MF.getFrameInfo();
+ DebugLoc dl = MI.getDebugLoc();
// Find out which operand is the frame index.
unsigned FIOperandNo = 0;
- while (!MI.getOperand(FIOperandNo).isFrameIndex()) {
+ while (!MI.getOperand(FIOperandNo).isFI()) {
++FIOperandNo;
assert(FIOperandNo != MI.getNumOperands() &&
"Instr doesn't have FrameIndex operand!");
SReg = PPC::R0;
// Insert a set of rA with the full offset value before the ld, st, or add
- BuildMI(MBB, II, TII.get(PPC::LIS), SReg)
+ BuildMI(MBB, II, dl, TII.get(PPC::LIS), SReg)
.addImm(Offset >> 16);
- BuildMI(MBB, II, TII.get(PPC::ORI), SReg)
- .addReg(SReg, false, false, true)
+ BuildMI(MBB, II, dl, TII.get(PPC::ORI), SReg)
+ .addReg(SReg, RegState::Kill)
.addImm(Offset);
// Convert into indexed form of the instruction:
// transform this into the appropriate ORI instruction.
static void HandleVRSaveUpdate(MachineInstr *MI, const TargetInstrInfo &TII) {
MachineFunction *MF = MI->getParent()->getParent();
+ DebugLoc dl = MI->getDebugLoc();
unsigned UsedRegMask = 0;
for (unsigned i = 0; i != 32; ++i)
if ((UsedRegMask & 0xFFFF) == UsedRegMask) {
if (DstReg != SrcReg)
- BuildMI(*MI->getParent(), MI, TII.get(PPC::ORI), DstReg)
+ BuildMI(*MI->getParent(), MI, dl, TII.get(PPC::ORI), DstReg)
.addReg(SrcReg)
.addImm(UsedRegMask);
else
- BuildMI(*MI->getParent(), MI, TII.get(PPC::ORI), DstReg)
- .addReg(SrcReg, false, false, true)
+ BuildMI(*MI->getParent(), MI, dl, TII.get(PPC::ORI), DstReg)
+ .addReg(SrcReg, RegState::Kill)
.addImm(UsedRegMask);
} else if ((UsedRegMask & 0xFFFF0000) == UsedRegMask) {
if (DstReg != SrcReg)
- BuildMI(*MI->getParent(), MI, TII.get(PPC::ORIS), DstReg)
+ BuildMI(*MI->getParent(), MI, dl, TII.get(PPC::ORIS), DstReg)
.addReg(SrcReg)
.addImm(UsedRegMask >> 16);
else
- BuildMI(*MI->getParent(), MI, TII.get(PPC::ORIS), DstReg)
- .addReg(SrcReg, false, false, true)
+ BuildMI(*MI->getParent(), MI, dl, TII.get(PPC::ORIS), DstReg)
+ .addReg(SrcReg, RegState::Kill)
.addImm(UsedRegMask >> 16);
} else {
if (DstReg != SrcReg)
- BuildMI(*MI->getParent(), MI, TII.get(PPC::ORIS), DstReg)
+ BuildMI(*MI->getParent(), MI, dl, TII.get(PPC::ORIS), DstReg)
.addReg(SrcReg)
.addImm(UsedRegMask >> 16);
else
- BuildMI(*MI->getParent(), MI, TII.get(PPC::ORIS), DstReg)
- .addReg(SrcReg, false, false, true)
+ BuildMI(*MI->getParent(), MI, dl, TII.get(PPC::ORIS), DstReg)
+ .addReg(SrcReg, RegState::Kill)
.addImm(UsedRegMask >> 16);
- BuildMI(*MI->getParent(), MI, TII.get(PPC::ORI), DstReg)
- .addReg(DstReg, false, false, true)
+ BuildMI(*MI->getParent(), MI, dl, TII.get(PPC::ORI), DstReg)
+ .addReg(DstReg, RegState::Kill)
.addImm(UsedRegMask & 0xFFFF);
}
// If we are a leaf function, and use up to 224 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 (FrameSize <= 224 && // Fits in red zone.
+ if (!DisableRedZone &&
+ FrameSize <= 224 && // Fits in red zone.
!MFI->hasVarSizedObjects() && // No dynamic alloca.
!MFI->hasCalls() && // No calls.
(!ALIGN_STACK || MaxAlign <= TargetAlign)) { // No special alignment.
// Save and clear the LR state.
PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
unsigned LR = getRARegister();
- FI->setUsesLR(MF.getRegInfo().isPhysRegUsed(LR));
+ FI->setMustSaveLR(MustSaveLR(MF, LR));
MF.getRegInfo().setPhysRegUnused(LR);
// Save R31 if necessary
FI->setFramePointerSaveIndex(FPSI);
}
+ // Reserve stack space to move the linkage area to in case of a tail call.
+ int TCSPDelta = 0;
+ if (PerformTailCallOpt && (TCSPDelta=FI->getTailCallSPDelta()) < 0) {
+ int AddFPOffsetAmount = IsELF32_ABI ? -4 : 0;
+ MF.getFrameInfo()->CreateFixedObject( -1 * TCSPDelta,
+ AddFPOffsetAmount + TCSPDelta);
+ }
// Reserve a slot closest to SP or frame pointer if we have a dynalloc or
// a large stack, which will require scavenging a register to materialize a
// large offset.
const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
const TargetRegisterClass *RC = IsPPC64 ? G8RC : GPRC;
RS->setScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
- RC->getAlignment()));
+ RC->getAlignment()));
}
}
MachineBasicBlock::iterator MBBI = MBB.begin();
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
+ DebugLoc dl = DebugLoc::getUnknownLoc();
+ bool needsFrameMoves = (MMI && MMI->hasDebugInfo()) ||
+ !MF.getFunction()->doesNotThrow() ||
+ UnwindTablesMandatory;
// Prepare for frame info.
unsigned FrameLabelId = 0;
bool IsPPC64 = Subtarget.isPPC64();
// Get operating system
bool IsMachoABI = Subtarget.isMachoABI();
- // Check if the link register (LR) has been used.
- bool UsesLR = MustSaveLR(MF);
+ // Check if the link register (LR) must be saved.
+ PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
+ bool MustSaveLR = FI->mustSaveLR();
// Do we have a frame pointer for this function?
bool HasFP = hasFP(MF) && FrameSize;
int FPOffset = PPCFrameInfo::getFramePointerSaveOffset(IsPPC64, IsMachoABI);
if (IsPPC64) {
- if (UsesLR)
- BuildMI(MBB, MBBI, TII.get(PPC::MFLR8), PPC::X0);
+ if (MustSaveLR)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::MFLR8), PPC::X0);
if (HasFP)
- BuildMI(MBB, MBBI, TII.get(PPC::STD))
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STD))
.addReg(PPC::X31)
.addImm(FPOffset/4)
.addReg(PPC::X1);
- if (UsesLR)
- BuildMI(MBB, MBBI, TII.get(PPC::STD))
+ if (MustSaveLR)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STD))
.addReg(PPC::X0)
.addImm(LROffset / 4)
.addReg(PPC::X1);
} else {
- if (UsesLR)
- BuildMI(MBB, MBBI, TII.get(PPC::MFLR), PPC::R0);
+ if (MustSaveLR)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::MFLR), PPC::R0);
if (HasFP)
- BuildMI(MBB, MBBI, TII.get(PPC::STW))
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STW))
.addReg(PPC::R31)
.addImm(FPOffset)
.addReg(PPC::R1);
- if (UsesLR)
- BuildMI(MBB, MBBI, TII.get(PPC::STW))
+ if (MustSaveLR)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STW))
.addReg(PPC::R0)
.addImm(LROffset)
.addReg(PPC::R1);
unsigned TargetAlign = MF.getTarget().getFrameInfo()->getStackAlignment();
unsigned MaxAlign = MFI->getMaxAlignment();
- if (MMI && MMI->needsFrameInfo()) {
+ if (needsFrameMoves) {
// Mark effective beginning of when frame pointer becomes valid.
FrameLabelId = MMI->NextLabelID();
- BuildMI(MBB, MBBI, TII.get(PPC::LABEL)).addImm(FrameLabelId).addImm(0);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::DBG_LABEL)).addImm(FrameLabelId);
}
// Adjust stack pointer: r1 += NegFrameSize.
assert(isPowerOf2_32(MaxAlign)&&isInt16(MaxAlign)&&"Invalid alignment!");
assert(isInt16(NegFrameSize) && "Unhandled stack size and alignment!");
- BuildMI(MBB, MBBI, TII.get(PPC::RLWINM), PPC::R0)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::RLWINM), PPC::R0)
.addReg(PPC::R1)
.addImm(0)
.addImm(32 - Log2_32(MaxAlign))
.addImm(31);
- BuildMI(MBB, MBBI, TII.get(PPC::SUBFIC) ,PPC::R0)
- .addReg(PPC::R0, false, false, true)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::SUBFIC) ,PPC::R0)
+ .addReg(PPC::R0, RegState::Kill)
.addImm(NegFrameSize);
- BuildMI(MBB, MBBI, TII.get(PPC::STWUX))
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STWUX))
.addReg(PPC::R1)
.addReg(PPC::R1)
.addReg(PPC::R0);
} else if (isInt16(NegFrameSize)) {
- BuildMI(MBB, MBBI, TII.get(PPC::STWU), PPC::R1)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STWU), PPC::R1)
.addReg(PPC::R1)
.addImm(NegFrameSize)
.addReg(PPC::R1);
} else {
- BuildMI(MBB, MBBI, TII.get(PPC::LIS), PPC::R0)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LIS), PPC::R0)
.addImm(NegFrameSize >> 16);
- BuildMI(MBB, MBBI, TII.get(PPC::ORI), PPC::R0)
- .addReg(PPC::R0, false, false, true)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ORI), PPC::R0)
+ .addReg(PPC::R0, RegState::Kill)
.addImm(NegFrameSize & 0xFFFF);
- BuildMI(MBB, MBBI, TII.get(PPC::STWUX))
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STWUX))
.addReg(PPC::R1)
.addReg(PPC::R1)
.addReg(PPC::R0);
assert(isPowerOf2_32(MaxAlign)&&isInt16(MaxAlign)&&"Invalid alignment!");
assert(isInt16(NegFrameSize) && "Unhandled stack size and alignment!");
- BuildMI(MBB, MBBI, TII.get(PPC::RLDICL), PPC::X0)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::RLDICL), PPC::X0)
.addReg(PPC::X1)
.addImm(0)
.addImm(64 - Log2_32(MaxAlign));
- BuildMI(MBB, MBBI, TII.get(PPC::SUBFIC8), PPC::X0)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::SUBFIC8), PPC::X0)
.addReg(PPC::X0)
.addImm(NegFrameSize);
- BuildMI(MBB, MBBI, TII.get(PPC::STDUX))
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STDUX))
.addReg(PPC::X1)
.addReg(PPC::X1)
.addReg(PPC::X0);
} else if (isInt16(NegFrameSize)) {
- BuildMI(MBB, MBBI, TII.get(PPC::STDU), PPC::X1)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STDU), PPC::X1)
.addReg(PPC::X1)
.addImm(NegFrameSize / 4)
.addReg(PPC::X1);
} else {
- BuildMI(MBB, MBBI, TII.get(PPC::LIS8), PPC::X0)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LIS8), PPC::X0)
.addImm(NegFrameSize >> 16);
- BuildMI(MBB, MBBI, TII.get(PPC::ORI8), PPC::X0)
- .addReg(PPC::X0, false, false, true)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ORI8), PPC::X0)
+ .addReg(PPC::X0, RegState::Kill)
.addImm(NegFrameSize & 0xFFFF);
- BuildMI(MBB, MBBI, TII.get(PPC::STDUX))
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::STDUX))
.addReg(PPC::X1)
.addReg(PPC::X1)
.addReg(PPC::X0);
}
}
- if (MMI && MMI->needsFrameInfo()) {
+ if (needsFrameMoves) {
std::vector<MachineMove> &Moves = MMI->getFrameMoves();
if (NegFrameSize) {
for (unsigned I = 0, E = CSI.size(); I != E; ++I) {
int Offset = MFI->getObjectOffset(CSI[I].getFrameIdx());
unsigned Reg = CSI[I].getReg();
- if (Reg == PPC::LR || Reg == PPC::LR8) continue;
+ if (Reg == PPC::LR || Reg == PPC::LR8 || Reg == PPC::RM) continue;
MachineLocation CSDst(MachineLocation::VirtualFP, Offset);
MachineLocation CSSrc(Reg);
Moves.push_back(MachineMove(FrameLabelId, CSDst, CSSrc));
// Mark effective beginning of when frame pointer is ready.
unsigned ReadyLabelId = MMI->NextLabelID();
- BuildMI(MBB, MBBI, TII.get(PPC::LABEL)).addImm(ReadyLabelId).addImm(0);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::DBG_LABEL)).addImm(ReadyLabelId);
MachineLocation FPDst(HasFP ? (IsPPC64 ? PPC::X31 : PPC::R31) :
(IsPPC64 ? PPC::X1 : PPC::R1));
// If there is a frame pointer, copy R1 into R31
if (HasFP) {
if (!IsPPC64) {
- BuildMI(MBB, MBBI, TII.get(PPC::OR), PPC::R31)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::OR), PPC::R31)
.addReg(PPC::R1)
.addReg(PPC::R1);
} else {
- BuildMI(MBB, MBBI, TII.get(PPC::OR8), PPC::X31)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::OR8), PPC::X31)
.addReg(PPC::X1)
.addReg(PPC::X1);
}
void PPCRegisterInfo::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
MachineBasicBlock::iterator MBBI = prior(MBB.end());
- assert(MBBI->getOpcode() == PPC::BLR &&
+ unsigned RetOpcode = MBBI->getOpcode();
+ DebugLoc dl = DebugLoc::getUnknownLoc();
+
+ assert( (RetOpcode == PPC::BLR ||
+ RetOpcode == PPC::TCRETURNri ||
+ RetOpcode == PPC::TCRETURNdi ||
+ RetOpcode == PPC::TCRETURNai ||
+ RetOpcode == PPC::TCRETURNri8 ||
+ RetOpcode == PPC::TCRETURNdi8 ||
+ RetOpcode == PPC::TCRETURNai8) &&
"Can only insert epilog into returning blocks");
// Get alignment info so we know how to restore r1
unsigned MaxAlign = MFI->getMaxAlignment();
// Get the number of bytes allocated from the FrameInfo.
- unsigned FrameSize = MFI->getStackSize();
+ int FrameSize = MFI->getStackSize();
// Get processor type.
bool IsPPC64 = Subtarget.isPPC64();
// Get operating system
bool IsMachoABI = Subtarget.isMachoABI();
- // Check if the link register (LR) has been used.
- bool UsesLR = MustSaveLR(MF);
+ // Check if the link register (LR) has been saved.
+ PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
+ bool MustSaveLR = FI->mustSaveLR();
// Do we have a frame pointer for this function?
bool HasFP = hasFP(MF) && FrameSize;
int LROffset = PPCFrameInfo::getReturnSaveOffset(IsPPC64, IsMachoABI);
int FPOffset = PPCFrameInfo::getFramePointerSaveOffset(IsPPC64, IsMachoABI);
+ bool UsesTCRet = RetOpcode == PPC::TCRETURNri ||
+ RetOpcode == PPC::TCRETURNdi ||
+ RetOpcode == PPC::TCRETURNai ||
+ RetOpcode == PPC::TCRETURNri8 ||
+ RetOpcode == PPC::TCRETURNdi8 ||
+ RetOpcode == PPC::TCRETURNai8;
+
+ if (UsesTCRet) {
+ int MaxTCRetDelta = FI->getTailCallSPDelta();
+ MachineOperand &StackAdjust = MBBI->getOperand(1);
+ assert(StackAdjust.isImm() && "Expecting immediate value.");
+ // Adjust stack pointer.
+ int StackAdj = StackAdjust.getImm();
+ int Delta = StackAdj - MaxTCRetDelta;
+ assert((Delta >= 0) && "Delta must be positive");
+ if (MaxTCRetDelta>0)
+ FrameSize += (StackAdj +Delta);
+ else
+ FrameSize += StackAdj;
+ }
+
if (FrameSize) {
// The loaded (or persistent) stack pointer value is offset by the 'stwu'
// on entry to the function. Add this offset back now.
- if (!Subtarget.isPPC64()) {
- if (isInt16(FrameSize) && (!ALIGN_STACK || TargetAlign >= MaxAlign) &&
- !MFI->hasVarSizedObjects()) {
- BuildMI(MBB, MBBI, TII.get(PPC::ADDI), PPC::R1)
- .addReg(PPC::R1).addImm(FrameSize);
+ if (!IsPPC64) {
+ // If this function contained a fastcc call and PerformTailCallOpt is
+ // enabled (=> hasFastCall()==true) the fastcc call might contain a tail
+ // call which invalidates the stack pointer value in SP(0). So we use the
+ // value of R31 in this case.
+ if (FI->hasFastCall() && isInt16(FrameSize)) {
+ assert(hasFP(MF) && "Expecting a valid the frame pointer.");
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ADDI), PPC::R1)
+ .addReg(PPC::R31).addImm(FrameSize);
+ } else if(FI->hasFastCall()) {
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LIS), PPC::R0)
+ .addImm(FrameSize >> 16);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ORI), PPC::R0)
+ .addReg(PPC::R0, RegState::Kill)
+ .addImm(FrameSize & 0xFFFF);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ADD4))
+ .addReg(PPC::R1)
+ .addReg(PPC::R31)
+ .addReg(PPC::R0);
+ } else if (isInt16(FrameSize) &&
+ (!ALIGN_STACK || TargetAlign >= MaxAlign) &&
+ !MFI->hasVarSizedObjects()) {
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ADDI), PPC::R1)
+ .addReg(PPC::R1).addImm(FrameSize);
} else {
- BuildMI(MBB, MBBI, TII.get(PPC::LWZ),PPC::R1).addImm(0).addReg(PPC::R1);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LWZ),PPC::R1)
+ .addImm(0).addReg(PPC::R1);
}
} else {
- if (isInt16(FrameSize) && TargetAlign >= MaxAlign &&
+ if (FI->hasFastCall() && isInt16(FrameSize)) {
+ assert(hasFP(MF) && "Expecting a valid the frame pointer.");
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ADDI8), PPC::X1)
+ .addReg(PPC::X31).addImm(FrameSize);
+ } else if(FI->hasFastCall()) {
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LIS8), PPC::X0)
+ .addImm(FrameSize >> 16);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ORI8), PPC::X0)
+ .addReg(PPC::X0, RegState::Kill)
+ .addImm(FrameSize & 0xFFFF);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ADD8))
+ .addReg(PPC::X1)
+ .addReg(PPC::X31)
+ .addReg(PPC::X0);
+ } else if (isInt16(FrameSize) && TargetAlign >= MaxAlign &&
!MFI->hasVarSizedObjects()) {
- BuildMI(MBB, MBBI, TII.get(PPC::ADDI8), PPC::X1)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::ADDI8), PPC::X1)
.addReg(PPC::X1).addImm(FrameSize);
} else {
- BuildMI(MBB, MBBI, TII.get(PPC::LD), PPC::X1).addImm(0).addReg(PPC::X1);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LD), PPC::X1)
+ .addImm(0).addReg(PPC::X1);
}
}
}
if (IsPPC64) {
- if (UsesLR)
- BuildMI(MBB, MBBI, TII.get(PPC::LD), PPC::X0)
+ if (MustSaveLR)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LD), PPC::X0)
.addImm(LROffset/4).addReg(PPC::X1);
if (HasFP)
- BuildMI(MBB, MBBI, TII.get(PPC::LD), PPC::X31)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LD), PPC::X31)
.addImm(FPOffset/4).addReg(PPC::X1);
- if (UsesLR)
- BuildMI(MBB, MBBI, TII.get(PPC::MTLR8)).addReg(PPC::X0);
+ if (MustSaveLR)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::MTLR8)).addReg(PPC::X0);
} else {
- if (UsesLR)
- BuildMI(MBB, MBBI, TII.get(PPC::LWZ), PPC::R0)
+ if (MustSaveLR)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LWZ), PPC::R0)
.addImm(LROffset).addReg(PPC::R1);
if (HasFP)
- BuildMI(MBB, MBBI, TII.get(PPC::LWZ), PPC::R31)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::LWZ), PPC::R31)
.addImm(FPOffset).addReg(PPC::R1);
- if (UsesLR)
- BuildMI(MBB, MBBI, TII.get(PPC::MTLR)).addReg(PPC::R0);
+ if (MustSaveLR)
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::MTLR)).addReg(PPC::R0);
+ }
+
+ // Callee pop calling convention. Pop parameter/linkage area. Used for tail
+ // call optimization
+ if (PerformTailCallOpt && RetOpcode == PPC::BLR &&
+ MF.getFunction()->getCallingConv() == CallingConv::Fast) {
+ PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
+ unsigned CallerAllocatedAmt = FI->getMinReservedArea();
+ unsigned StackReg = IsPPC64 ? PPC::X1 : PPC::R1;
+ unsigned FPReg = IsPPC64 ? PPC::X31 : PPC::R31;
+ unsigned TmpReg = IsPPC64 ? PPC::X0 : PPC::R0;
+ unsigned ADDIInstr = IsPPC64 ? PPC::ADDI8 : PPC::ADDI;
+ unsigned ADDInstr = IsPPC64 ? PPC::ADD8 : PPC::ADD4;
+ unsigned LISInstr = IsPPC64 ? PPC::LIS8 : PPC::LIS;
+ unsigned ORIInstr = IsPPC64 ? PPC::ORI8 : PPC::ORI;
+
+ if (CallerAllocatedAmt && isInt16(CallerAllocatedAmt)) {
+ BuildMI(MBB, MBBI, dl, TII.get(ADDIInstr), StackReg)
+ .addReg(StackReg).addImm(CallerAllocatedAmt);
+ } else {
+ BuildMI(MBB, MBBI, dl, TII.get(LISInstr), TmpReg)
+ .addImm(CallerAllocatedAmt >> 16);
+ BuildMI(MBB, MBBI, dl, TII.get(ORIInstr), TmpReg)
+ .addReg(TmpReg, RegState::Kill)
+ .addImm(CallerAllocatedAmt & 0xFFFF);
+ BuildMI(MBB, MBBI, dl, TII.get(ADDInstr))
+ .addReg(StackReg)
+ .addReg(FPReg)
+ .addReg(TmpReg);
+ }
+ } else if (RetOpcode == PPC::TCRETURNdi) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::TAILB)).
+ addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset());
+ } else if (RetOpcode == PPC::TCRETURNri) {
+ MBBI = prior(MBB.end());
+ assert(MBBI->getOperand(0).isReg() && "Expecting register operand.");
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::TAILBCTR));
+ } else if (RetOpcode == PPC::TCRETURNai) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::TAILBA)).addImm(JumpTarget.getImm());
+ } else if (RetOpcode == PPC::TCRETURNdi8) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::TAILB8)).
+ addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset());
+ } else if (RetOpcode == PPC::TCRETURNri8) {
+ MBBI = prior(MBB.end());
+ assert(MBBI->getOperand(0).isReg() && "Expecting register operand.");
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::TAILBCTR8));
+ } else if (RetOpcode == PPC::TCRETURNai8) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ BuildMI(MBB, MBBI, dl, TII.get(PPC::TAILBA8)).addImm(JumpTarget.getImm());
}
}
projects / oota-llvm.git / blobdiff