#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/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"
//
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;
+
+ if (isInt16(CalleeAmt)) {
+ BuildMI(MBB, I, TII.get(ADDIInstr), StackReg).addReg(StackReg).
+ addImm(CalleeAmt);
+ } else {
+ MachineBasicBlock::iterator MBBI = I;
+ BuildMI(MBB, MBBI, TII.get(LISInstr), TmpReg)
+ .addImm(CalleeAmt >> 16);
+ BuildMI(MBB, MBBI, TII.get(ORIInstr), TmpReg)
+ .addReg(TmpReg, false, false, true)
+ .addImm(CalleeAmt & 0xFFFF);
+ BuildMI(MBB, MBBI, TII.get(ADDInstr))
+ .addReg(StackReg)
+ .addReg(StackReg)
+ .addReg(TmpReg);
+ }
+ }
+ }
// Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions.
MBB.erase(I);
}
// 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!");
// 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.
MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
bool needsFrameMoves = (MMI && MMI->hasDebugInfo()) ||
!MF.getFunction()->doesNotThrow() ||
- !UnwindTablesOptional;
+ 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)
+ if (MustSaveLR)
BuildMI(MBB, MBBI, TII.get(PPC::MFLR8), PPC::X0);
if (HasFP)
.addImm(FPOffset/4)
.addReg(PPC::X1);
- if (UsesLR)
+ if (MustSaveLR)
BuildMI(MBB, MBBI, TII.get(PPC::STD))
.addReg(PPC::X0)
.addImm(LROffset / 4)
.addReg(PPC::X1);
} else {
- if (UsesLR)
+ if (MustSaveLR)
BuildMI(MBB, MBBI, TII.get(PPC::MFLR), PPC::R0);
if (HasFP)
.addImm(FPOffset)
.addReg(PPC::R1);
- if (UsesLR)
+ if (MustSaveLR)
BuildMI(MBB, MBBI, TII.get(PPC::STW))
.addReg(PPC::R0)
.addImm(LROffset)
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, TII.get(PPC::DBG_LABEL)).addImm(FrameLabelId);
}
// Adjust stack pointer: r1 += 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, TII.get(PPC::DBG_LABEL)).addImm(ReadyLabelId);
MachineLocation FPDst(HasFP ? (IsPPC64 ? PPC::X31 : PPC::R31) :
(IsPPC64 ? PPC::X1 : PPC::R1));
void PPCRegisterInfo::emitEpilogue(MachineFunction &MF,
MachineBasicBlock &MBB) const {
MachineBasicBlock::iterator MBBI = prior(MBB.end());
- assert(MBBI->getOpcode() == PPC::BLR &&
+ unsigned RetOpcode = MBBI->getOpcode();
+
+ 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, TII.get(PPC::ADDI), PPC::R1)
+ .addReg(PPC::R31).addImm(FrameSize);
+ } else if(FI->hasFastCall()) {
+ BuildMI(MBB, MBBI, TII.get(PPC::LIS), PPC::R0)
+ .addImm(FrameSize >> 16);
+ BuildMI(MBB, MBBI, TII.get(PPC::ORI), PPC::R0)
+ .addReg(PPC::R0, false, false, true)
+ .addImm(FrameSize & 0xFFFF);
+ BuildMI(MBB, MBBI, 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, 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);
}
} else {
- if (isInt16(FrameSize) && TargetAlign >= MaxAlign &&
+ if (FI->hasFastCall() && isInt16(FrameSize)) {
+ assert(hasFP(MF) && "Expecting a valid the frame pointer.");
+ BuildMI(MBB, MBBI, TII.get(PPC::ADDI8), PPC::X1)
+ .addReg(PPC::X31).addImm(FrameSize);
+ } else if(FI->hasFastCall()) {
+ BuildMI(MBB, MBBI, TII.get(PPC::LIS8), PPC::X0)
+ .addImm(FrameSize >> 16);
+ BuildMI(MBB, MBBI, TII.get(PPC::ORI8), PPC::X0)
+ .addReg(PPC::X0, false, false, true)
+ .addImm(FrameSize & 0xFFFF);
+ BuildMI(MBB, MBBI, 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)
.addReg(PPC::X1).addImm(FrameSize);
}
if (IsPPC64) {
- if (UsesLR)
+ if (MustSaveLR)
BuildMI(MBB, MBBI, TII.get(PPC::LD), PPC::X0)
.addImm(LROffset/4).addReg(PPC::X1);
BuildMI(MBB, MBBI, TII.get(PPC::LD), PPC::X31)
.addImm(FPOffset/4).addReg(PPC::X1);
- if (UsesLR)
+ if (MustSaveLR)
BuildMI(MBB, MBBI, TII.get(PPC::MTLR8)).addReg(PPC::X0);
} else {
- if (UsesLR)
+ if (MustSaveLR)
BuildMI(MBB, MBBI, TII.get(PPC::LWZ), PPC::R0)
.addImm(LROffset).addReg(PPC::R1);
BuildMI(MBB, MBBI, TII.get(PPC::LWZ), PPC::R31)
.addImm(FPOffset).addReg(PPC::R1);
- if (UsesLR)
+ if (MustSaveLR)
BuildMI(MBB, MBBI, 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, TII.get(ADDIInstr), StackReg)
+ .addReg(StackReg).addImm(CallerAllocatedAmt);
+ } else {
+ BuildMI(MBB, MBBI, TII.get(LISInstr), TmpReg)
+ .addImm(CallerAllocatedAmt >> 16);
+ BuildMI(MBB, MBBI, TII.get(ORIInstr), TmpReg)
+ .addReg(TmpReg, false, false, true)
+ .addImm(CallerAllocatedAmt & 0xFFFF);
+ BuildMI(MBB, MBBI, 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, TII.get(PPC::TAILB)).
+ addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset());
+ } else if (RetOpcode == PPC::TCRETURNri) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ assert(JumpTarget.isReg() && "Expecting register operand.");
+ BuildMI(MBB, MBBI, TII.get(PPC::TAILBCTR));
+ } else if (RetOpcode == PPC::TCRETURNai) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ BuildMI(MBB, MBBI, TII.get(PPC::TAILBA)).addImm(JumpTarget.getImm());
+ } else if (RetOpcode == PPC::TCRETURNdi8) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ BuildMI(MBB, MBBI, TII.get(PPC::TAILB8)).
+ addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset());
+ } else if (RetOpcode == PPC::TCRETURNri8) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ assert(JumpTarget.isReg() && "Expecting register operand.");
+ BuildMI(MBB, MBBI, TII.get(PPC::TAILBCTR8));
+ } else if (RetOpcode == PPC::TCRETURNai8) {
+ MBBI = prior(MBB.end());
+ MachineOperand &JumpTarget = MBBI->getOperand(0);
+ BuildMI(MBB, MBBI, TII.get(PPC::TAILBA8)).addImm(JumpTarget.getImm());
+ }
}
unsigned PPCRegisterInfo::getRARegister() const {
projects / oota-llvm.git / blobdiff