X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FARM%2FARMBaseRegisterInfo.cpp;h=3e7c1c3075b4fe5bd4811f098bef6b8e243d53d9;hb=09aa3f0ef35d9241c92439d74b8d5e9a81d814c2;hp=3f0a2f1569e7dc9d8480d1db393db15cdd105d43;hpb=d4c36cec1db81b4ee48cd4ab462262615d78f22c;p=oota-llvm.git diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp index 3f0a2f1569e..3e7c1c3075b 100644 --- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp +++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp @@ -15,6 +15,7 @@ #include "ARMAddressingModes.h" #include "ARMBaseInstrInfo.h" #include "ARMBaseRegisterInfo.h" +#include "ARMFrameInfo.h" #include "ARMInstrInfo.h" #include "ARMMachineFunctionInfo.h" #include "ARMSubtarget.h" @@ -39,111 +40,24 @@ #include "llvm/ADT/SmallVector.h" #include "llvm/Support/CommandLine.h" -namespace llvm { -cl::opt -ReuseFrameIndexVals("arm-reuse-frame-index-vals", cl::Hidden, cl::init(true), - cl::desc("Reuse repeated frame index values")); -} - using namespace llvm; -unsigned ARMBaseRegisterInfo::getRegisterNumbering(unsigned RegEnum, - bool *isSPVFP) { - if (isSPVFP) - *isSPVFP = false; - - using namespace ARM; - switch (RegEnum) { - default: - llvm_unreachable("Unknown ARM register!"); - case R0: case D0: case Q0: return 0; - case R1: case D1: case Q1: return 1; - case R2: case D2: case Q2: return 2; - case R3: case D3: case Q3: return 3; - case R4: case D4: case Q4: return 4; - case R5: case D5: case Q5: return 5; - case R6: case D6: case Q6: return 6; - case R7: case D7: case Q7: return 7; - case R8: case D8: case Q8: return 8; - case R9: case D9: case Q9: return 9; - case R10: case D10: case Q10: return 10; - case R11: case D11: case Q11: return 11; - case R12: case D12: case Q12: return 12; - case SP: case D13: case Q13: return 13; - case LR: case D14: case Q14: return 14; - case PC: case D15: case Q15: return 15; - - case D16: return 16; - case D17: return 17; - case D18: return 18; - case D19: return 19; - case D20: return 20; - case D21: return 21; - case D22: return 22; - case D23: return 23; - case D24: return 24; - case D25: return 25; - case D26: return 26; - case D27: return 27; - case D28: return 28; - case D29: return 29; - case D30: return 30; - case D31: return 31; - - case S0: case S1: case S2: case S3: - case S4: case S5: case S6: case S7: - case S8: case S9: case S10: case S11: - case S12: case S13: case S14: case S15: - case S16: case S17: case S18: case S19: - case S20: case S21: case S22: case S23: - case S24: case S25: case S26: case S27: - case S28: case S29: case S30: case S31: { - if (isSPVFP) - *isSPVFP = true; - switch (RegEnum) { - default: return 0; // Avoid compile time warning. - case S0: return 0; - case S1: return 1; - case S2: return 2; - case S3: return 3; - case S4: return 4; - case S5: return 5; - case S6: return 6; - case S7: return 7; - case S8: return 8; - case S9: return 9; - case S10: return 10; - case S11: return 11; - case S12: return 12; - case S13: return 13; - case S14: return 14; - case S15: return 15; - case S16: return 16; - case S17: return 17; - case S18: return 18; - case S19: return 19; - case S20: return 20; - case S21: return 21; - case S22: return 22; - case S23: return 23; - case S24: return 24; - case S25: return 25; - case S26: return 26; - case S27: return 27; - case S28: return 28; - case S29: return 29; - case S30: return 30; - case S31: return 31; - } - } - } -} +static cl::opt +ForceAllBaseRegAlloc("arm-force-base-reg-alloc", cl::Hidden, cl::init(false), + cl::desc("Force use of virtual base registers for stack load/store")); +static cl::opt +EnableLocalStackAlloc("enable-local-stack-alloc", cl::init(true), cl::Hidden, + cl::desc("Enable pre-regalloc stack frame index allocation")); +static cl::opt +EnableBasePointer("arm-use-base-pointer", cl::Hidden, cl::init(true), + cl::desc("Enable use of a base pointer for complex stack frames")); ARMBaseRegisterInfo::ARMBaseRegisterInfo(const ARMBaseInstrInfo &tii, const ARMSubtarget &sti) : ARMGenRegisterInfo(ARM::ADJCALLSTACKDOWN, ARM::ADJCALLSTACKUP), TII(tii), STI(sti), - FramePtr((STI.isTargetDarwin() || STI.isThumb()) ? ARM::R7 : ARM::R11) { + FramePtr((STI.isTargetDarwin() || STI.isThumb()) ? ARM::R7 : ARM::R11), + BasePtr(ARM::R6) { } const unsigned* @@ -172,12 +86,17 @@ ARMBaseRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const { BitVector ARMBaseRegisterInfo:: getReservedRegs(const MachineFunction &MF) const { + const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo(); + // FIXME: avoid re-calculating this everytime. BitVector Reserved(getNumRegs()); Reserved.set(ARM::SP); Reserved.set(ARM::PC); - if (STI.isTargetDarwin() || hasFP(MF)) + Reserved.set(ARM::FPSCR); + if (TFI->hasFP(MF)) Reserved.set(FramePtr); + if (hasBasePointer(MF)) + Reserved.set(BasePtr); // Some targets reserve R9. if (STI.isR9Reserved()) Reserved.set(ARM::R9); @@ -186,14 +105,20 @@ getReservedRegs(const MachineFunction &MF) const { bool ARMBaseRegisterInfo::isReservedReg(const MachineFunction &MF, unsigned Reg) const { + const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo(); + switch (Reg) { default: break; case ARM::SP: case ARM::PC: return true; + case ARM::R6: + if (hasBasePointer(MF)) + return true; + break; case ARM::R7: case ARM::R11: - if (FramePtr == Reg && (STI.isTargetDarwin() || hasFP(MF))) + if (FramePtr == Reg && TFI->hasFP(MF)) return true; break; case ARM::R9: @@ -429,6 +354,7 @@ std::pair ARMBaseRegisterInfo::getAllocationOrder(const TargetRegisterClass *RC, unsigned HintType, unsigned HintReg, const MachineFunction &MF) const { + const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo(); // Alternative register allocation orders when favoring even / odd registers // of register pairs. @@ -510,7 +436,7 @@ ARMBaseRegisterInfo::getAllocationOrder(const TargetRegisterClass *RC, return std::make_pair(RC->allocation_order_begin(MF), RC->allocation_order_end(MF)); - if (!STI.isTargetDarwin() && !hasFP(MF)) { + if (!TFI->hasFP(MF)) { if (!STI.isR9Reserved()) return std::make_pair(GPREven1, GPREven1 + (sizeof(GPREven1)/sizeof(unsigned))); @@ -539,7 +465,7 @@ ARMBaseRegisterInfo::getAllocationOrder(const TargetRegisterClass *RC, return std::make_pair(RC->allocation_order_begin(MF), RC->allocation_order_end(MF)); - if (!STI.isTargetDarwin() && !hasFP(MF)) { + if (!TFI->hasFP(MF)) { if (!STI.isR9Reserved()) return std::make_pair(GPROdd1, GPROdd1 + (sizeof(GPROdd1)/sizeof(unsigned))); @@ -604,48 +530,55 @@ ARMBaseRegisterInfo::UpdateRegAllocHint(unsigned Reg, unsigned NewReg, } } -/// hasFP - Return true if the specified function should have a dedicated frame -/// pointer register. This is true if the function has variable sized allocas -/// or if frame pointer elimination is disabled. -/// -bool ARMBaseRegisterInfo::hasFP(const MachineFunction &MF) const { +bool ARMBaseRegisterInfo::hasBasePointer(const MachineFunction &MF) const { const MachineFrameInfo *MFI = MF.getFrameInfo(); - return ((DisableFramePointerElim(MF) && MFI->adjustsStack())|| - needsStackRealignment(MF) || - MFI->hasVarSizedObjects() || - MFI->isFrameAddressTaken()); + const ARMFunctionInfo *AFI = MF.getInfo(); + + if (!EnableBasePointer) + return false; + + if (needsStackRealignment(MF) && MFI->hasVarSizedObjects()) + return true; + + // Thumb has trouble with negative offsets from the FP. Thumb2 has a limited + // negative range for ldr/str (255), and thumb1 is positive offsets only. + // It's going to be better to use the SP or Base Pointer instead. When there + // are variable sized objects, we can't reference off of the SP, so we + // reserve a Base Pointer. + if (AFI->isThumbFunction() && MFI->hasVarSizedObjects()) { + // Conservatively estimate whether the negative offset from the frame + // pointer will be sufficient to reach. If a function has a smallish + // frame, it's less likely to have lots of spills and callee saved + // space, so it's all more likely to be within range of the frame pointer. + // If it's wrong, the scavenger will still enable access to work, it just + // won't be optimal. + if (AFI->isThumb2Function() && MFI->getLocalFrameSize() < 128) + return false; + return true; + } + + return false; } bool ARMBaseRegisterInfo::canRealignStack(const MachineFunction &MF) const { const MachineFrameInfo *MFI = MF.getFrameInfo(); const ARMFunctionInfo *AFI = MF.getInfo(); - return (RealignStack && - !AFI->isThumb1OnlyFunction() && - !MFI->hasVarSizedObjects()); + // We can't realign the stack if: + // 1. Dynamic stack realignment is explicitly disabled, + // 2. This is a Thumb1 function (it's not useful, so we don't bother), or + // 3. There are VLAs in the function and the base pointer is disabled. + return (RealignStack && !AFI->isThumb1OnlyFunction() && + (!MFI->hasVarSizedObjects() || EnableBasePointer)); } bool ARMBaseRegisterInfo:: needsStackRealignment(const MachineFunction &MF) const { const MachineFrameInfo *MFI = MF.getFrameInfo(); const Function *F = MF.getFunction(); - const ARMFunctionInfo *AFI = MF.getInfo(); unsigned StackAlign = MF.getTarget().getFrameInfo()->getStackAlignment(); - bool requiresRealignment = - RealignStack && ((MFI->getMaxAlignment() > StackAlign) || - F->hasFnAttr(Attribute::StackAlignment)); - - // FIXME: Currently we don't support stack realignment for functions with - // variable-sized allocas. - // FIXME: It's more complicated than this... - if (0 && requiresRealignment && MFI->hasVarSizedObjects()) - report_fatal_error( - "Stack realignment in presense of dynamic allocas is not supported"); - - // FIXME: This probably isn't the right place for this. - if (0 && requiresRealignment && AFI->isThumb1OnlyFunction()) - report_fatal_error( - "Stack realignment in thumb1 functions is not supported"); - + bool requiresRealignment = ((MFI->getLocalFrameMaxAlign() > StackAlign) || + F->hasFnAttr(Attribute::StackAlignment)); + return requiresRealignment && canRealignStack(MF); } @@ -658,359 +591,19 @@ cannotEliminateFrame(const MachineFunction &MF) const { || needsStackRealignment(MF); } -/// estimateStackSize - Estimate and return the size of the frame. -static unsigned estimateStackSize(MachineFunction &MF) { - const MachineFrameInfo *FFI = MF.getFrameInfo(); - int Offset = 0; - for (int i = FFI->getObjectIndexBegin(); i != 0; ++i) { - int FixedOff = -FFI->getObjectOffset(i); - if (FixedOff > Offset) Offset = FixedOff; - } - for (unsigned i = 0, e = FFI->getObjectIndexEnd(); i != e; ++i) { - if (FFI->isDeadObjectIndex(i)) - continue; - Offset += FFI->getObjectSize(i); - unsigned Align = FFI->getObjectAlignment(i); - // Adjust to alignment boundary - Offset = (Offset+Align-1)/Align*Align; - } - return (unsigned)Offset; -} - -/// estimateRSStackSizeLimit - Look at each instruction that references stack -/// frames and return the stack size limit beyond which some of these -/// instructions will require a scratch register during their expansion later. -unsigned -ARMBaseRegisterInfo::estimateRSStackSizeLimit(MachineFunction &MF) const { - unsigned Limit = (1 << 12) - 1; - for (MachineFunction::iterator BB = MF.begin(),E = MF.end(); BB != E; ++BB) { - for (MachineBasicBlock::iterator I = BB->begin(), E = BB->end(); - I != E; ++I) { - for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) { - if (!I->getOperand(i).isFI()) continue; - - // When using ADDri to get the address of a stack object, 255 is the - // largest offset guaranteed to fit in the immediate offset. - if (I->getOpcode() == ARM::ADDri) { - Limit = std::min(Limit, (1U << 8) - 1); - break; - } - - // Otherwise check the addressing mode. - switch (I->getDesc().TSFlags & ARMII::AddrModeMask) { - case ARMII::AddrMode3: - case ARMII::AddrModeT2_i8: - Limit = std::min(Limit, (1U << 8) - 1); - break; - case ARMII::AddrMode5: - case ARMII::AddrModeT2_i8s4: - Limit = std::min(Limit, ((1U << 8) - 1) * 4); - break; - case ARMII::AddrModeT2_i12: - if (hasFP(MF)) Limit = std::min(Limit, (1U << 8) - 1); - break; - case ARMII::AddrMode6: - // Addressing mode 6 (load/store) instructions can't encode an - // immediate offset for stack references. - return 0; - default: - break; - } - break; // At most one FI per instruction - } - } - } - - return Limit; -} - -void -ARMBaseRegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF, - RegScavenger *RS) const { - // This tells PEI to spill the FP as if it is any other callee-save register - // to take advantage the eliminateFrameIndex machinery. This also ensures it - // is spilled in the order specified by getCalleeSavedRegs() to make it easier - // to combine multiple loads / stores. - bool CanEliminateFrame = true; - bool CS1Spilled = false; - bool LRSpilled = false; - unsigned NumGPRSpills = 0; - SmallVector UnspilledCS1GPRs; - SmallVector UnspilledCS2GPRs; - ARMFunctionInfo *AFI = MF.getInfo(); - MachineFrameInfo *MFI = MF.getFrameInfo(); - - // Spill R4 if Thumb2 function requires stack realignment - it will be used as - // scratch register. - // FIXME: It will be better just to find spare register here. - if (needsStackRealignment(MF) && - AFI->isThumb2Function()) - MF.getRegInfo().setPhysRegUsed(ARM::R4); - - // Spill LR if Thumb1 function uses variable length argument lists. - if (AFI->isThumb1OnlyFunction() && AFI->getVarArgsRegSaveSize() > 0) - MF.getRegInfo().setPhysRegUsed(ARM::LR); - - // Don't spill FP if the frame can be eliminated. This is determined - // by scanning the callee-save registers to see if any is used. - const unsigned *CSRegs = getCalleeSavedRegs(); - for (unsigned i = 0; CSRegs[i]; ++i) { - unsigned Reg = CSRegs[i]; - bool Spilled = false; - if (MF.getRegInfo().isPhysRegUsed(Reg)) { - AFI->setCSRegisterIsSpilled(Reg); - Spilled = true; - CanEliminateFrame = false; - } else { - // Check alias registers too. - for (const unsigned *Aliases = getAliasSet(Reg); *Aliases; ++Aliases) { - if (MF.getRegInfo().isPhysRegUsed(*Aliases)) { - Spilled = true; - CanEliminateFrame = false; - } - } - } - - if (!ARM::GPRRegisterClass->contains(Reg)) - continue; - - if (Spilled) { - NumGPRSpills++; - - if (!STI.isTargetDarwin()) { - if (Reg == ARM::LR) - LRSpilled = true; - CS1Spilled = true; - continue; - } - - // Keep track if LR and any of R4, R5, R6, and R7 is spilled. - switch (Reg) { - case ARM::LR: - LRSpilled = true; - // Fallthrough - case ARM::R4: - case ARM::R5: - case ARM::R6: - case ARM::R7: - CS1Spilled = true; - break; - default: - break; - } - } else { - if (!STI.isTargetDarwin()) { - UnspilledCS1GPRs.push_back(Reg); - continue; - } - - switch (Reg) { - case ARM::R4: - case ARM::R5: - case ARM::R6: - case ARM::R7: - case ARM::LR: - UnspilledCS1GPRs.push_back(Reg); - break; - default: - UnspilledCS2GPRs.push_back(Reg); - break; - } - } - } - - bool ForceLRSpill = false; - if (!LRSpilled && AFI->isThumb1OnlyFunction()) { - unsigned FnSize = TII.GetFunctionSizeInBytes(MF); - // Force LR to be spilled if the Thumb function size is > 2048. This enables - // use of BL to implement far jump. If it turns out that it's not needed - // then the branch fix up path will undo it. - if (FnSize >= (1 << 11)) { - CanEliminateFrame = false; - ForceLRSpill = true; - } - } - - // If any of the stack slot references may be out of range of an immediate - // offset, make sure a register (or a spill slot) is available for the - // register scavenger. Note that if we're indexing off the frame pointer, the - // effective stack size is 4 bytes larger since the FP points to the stack - // slot of the previous FP. Also, if we have variable sized objects in the - // function, stack slot references will often be negative, and some of - // our instructions are positive-offset only, so conservatively consider - // that case to want a spill slot (or register) as well. - // FIXME: We could add logic to be more precise about negative offsets - // and which instructions will need a scratch register for them. Is it - // worth the effort and added fragility? - bool BigStack = - (RS && (estimateStackSize(MF) + (hasFP(MF) ? 4:0) >= - estimateRSStackSizeLimit(MF))) || MFI->hasVarSizedObjects(); - - bool ExtraCSSpill = false; - if (BigStack || !CanEliminateFrame || cannotEliminateFrame(MF)) { - AFI->setHasStackFrame(true); - - // If LR is not spilled, but at least one of R4, R5, R6, and R7 is spilled. - // Spill LR as well so we can fold BX_RET to the registers restore (LDM). - if (!LRSpilled && CS1Spilled) { - MF.getRegInfo().setPhysRegUsed(ARM::LR); - AFI->setCSRegisterIsSpilled(ARM::LR); - NumGPRSpills++; - UnspilledCS1GPRs.erase(std::find(UnspilledCS1GPRs.begin(), - UnspilledCS1GPRs.end(), (unsigned)ARM::LR)); - ForceLRSpill = false; - ExtraCSSpill = true; - } - - // Darwin ABI requires FP to point to the stack slot that contains the - // previous FP. - if (STI.isTargetDarwin() || hasFP(MF)) { - MF.getRegInfo().setPhysRegUsed(FramePtr); - NumGPRSpills++; - } - - // If stack and double are 8-byte aligned and we are spilling an odd number - // of GPRs. Spill one extra callee save GPR so we won't have to pad between - // the integer and double callee save areas. - unsigned TargetAlign = MF.getTarget().getFrameInfo()->getStackAlignment(); - if (TargetAlign == 8 && (NumGPRSpills & 1)) { - if (CS1Spilled && !UnspilledCS1GPRs.empty()) { - for (unsigned i = 0, e = UnspilledCS1GPRs.size(); i != e; ++i) { - unsigned Reg = UnspilledCS1GPRs[i]; - // Don't spill high register if the function is thumb1 - if (!AFI->isThumb1OnlyFunction() || - isARMLowRegister(Reg) || Reg == ARM::LR) { - MF.getRegInfo().setPhysRegUsed(Reg); - AFI->setCSRegisterIsSpilled(Reg); - if (!isReservedReg(MF, Reg)) - ExtraCSSpill = true; - break; - } - } - } else if (!UnspilledCS2GPRs.empty() && - !AFI->isThumb1OnlyFunction()) { - unsigned Reg = UnspilledCS2GPRs.front(); - MF.getRegInfo().setPhysRegUsed(Reg); - AFI->setCSRegisterIsSpilled(Reg); - if (!isReservedReg(MF, Reg)) - ExtraCSSpill = true; - } - } - - // Estimate if we might need to scavenge a register at some point in order - // to materialize a stack offset. If so, either spill one additional - // callee-saved register or reserve a special spill slot to facilitate - // register scavenging. Thumb1 needs a spill slot for stack pointer - // adjustments also, even when the frame itself is small. - if (BigStack && !ExtraCSSpill) { - // If any non-reserved CS register isn't spilled, just spill one or two - // extra. That should take care of it! - unsigned NumExtras = TargetAlign / 4; - SmallVector Extras; - while (NumExtras && !UnspilledCS1GPRs.empty()) { - unsigned Reg = UnspilledCS1GPRs.back(); - UnspilledCS1GPRs.pop_back(); - if (!isReservedReg(MF, Reg) && - (!AFI->isThumb1OnlyFunction() || isARMLowRegister(Reg) || - Reg == ARM::LR)) { - Extras.push_back(Reg); - NumExtras--; - } - } - // For non-Thumb1 functions, also check for hi-reg CS registers - if (!AFI->isThumb1OnlyFunction()) { - while (NumExtras && !UnspilledCS2GPRs.empty()) { - unsigned Reg = UnspilledCS2GPRs.back(); - UnspilledCS2GPRs.pop_back(); - if (!isReservedReg(MF, Reg)) { - Extras.push_back(Reg); - NumExtras--; - } - } - } - if (Extras.size() && NumExtras == 0) { - for (unsigned i = 0, e = Extras.size(); i != e; ++i) { - MF.getRegInfo().setPhysRegUsed(Extras[i]); - AFI->setCSRegisterIsSpilled(Extras[i]); - } - } else if (!AFI->isThumb1OnlyFunction()) { - // note: Thumb1 functions spill to R12, not the stack. Reserve a slot - // closest to SP or frame pointer. - const TargetRegisterClass *RC = ARM::GPRRegisterClass; - RS->setScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(), - RC->getAlignment(), - false)); - } - } - } - - if (ForceLRSpill) { - MF.getRegInfo().setPhysRegUsed(ARM::LR); - AFI->setCSRegisterIsSpilled(ARM::LR); - AFI->setLRIsSpilledForFarJump(true); - } -} - unsigned ARMBaseRegisterInfo::getRARegister() const { return ARM::LR; } -unsigned +unsigned ARMBaseRegisterInfo::getFrameRegister(const MachineFunction &MF) const { - if (STI.isTargetDarwin() || hasFP(MF)) + const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo(); + + if (TFI->hasFP(MF)) return FramePtr; return ARM::SP; } -int -ARMBaseRegisterInfo::getFrameIndexReference(const MachineFunction &MF, int FI, - unsigned &FrameReg) const { - const MachineFrameInfo *MFI = MF.getFrameInfo(); - const ARMFunctionInfo *AFI = MF.getInfo(); - int Offset = MFI->getObjectOffset(FI) + MFI->getStackSize(); - bool isFixed = MFI->isFixedObjectIndex(FI); - - FrameReg = ARM::SP; - if (AFI->isGPRCalleeSavedArea1Frame(FI)) - Offset -= AFI->getGPRCalleeSavedArea1Offset(); - else if (AFI->isGPRCalleeSavedArea2Frame(FI)) - Offset -= AFI->getGPRCalleeSavedArea2Offset(); - else if (AFI->isDPRCalleeSavedAreaFrame(FI)) - Offset -= AFI->getDPRCalleeSavedAreaOffset(); - else if (needsStackRealignment(MF)) { - // When dynamically realigning the stack, use the frame pointer for - // parameters, and the stack pointer for locals. - assert (hasFP(MF) && "dynamic stack realignment without a FP!"); - if (isFixed) { - FrameReg = getFrameRegister(MF); - Offset -= AFI->getFramePtrSpillOffset(); - } - } else if (hasFP(MF) && AFI->hasStackFrame()) { - if (isFixed || MFI->hasVarSizedObjects()) { - // Use frame pointer to reference fixed objects unless this is a - // frameless function. - FrameReg = getFrameRegister(MF); - Offset -= AFI->getFramePtrSpillOffset(); - } else if (AFI->isThumb2Function()) { - // In Thumb2 mode, the negative offset is very limited. - int FPOffset = Offset - AFI->getFramePtrSpillOffset(); - if (FPOffset >= -255 && FPOffset < 0) { - FrameReg = getFrameRegister(MF); - Offset = FPOffset; - } - } - } - return Offset; -} - - -int -ARMBaseRegisterInfo::getFrameIndexOffset(const MachineFunction &MF, - int FI) const { - unsigned FrameReg; - return getFrameIndexReference(MF, FI, FrameReg); -} - unsigned ARMBaseRegisterInfo::getEHExceptionRegister() const { llvm_unreachable("What is the exception register"); return 0; @@ -1038,7 +631,8 @@ unsigned ARMBaseRegisterInfo::getRegisterPairEven(unsigned Reg, case ARM::R5: return ARM::R4; case ARM::R7: - return isReservedReg(MF, ARM::R7) ? 0 : ARM::R6; + return (isReservedReg(MF, ARM::R7) || isReservedReg(MF, ARM::R6)) + ? 0 : ARM::R6; case ARM::R9: return isReservedReg(MF, ARM::R9) ? 0 :ARM::R8; case ARM::R11: @@ -1127,7 +721,8 @@ unsigned ARMBaseRegisterInfo::getRegisterPairOdd(unsigned Reg, case ARM::R4: return ARM::R5; case ARM::R6: - return isReservedReg(MF, ARM::R7) ? 0 : ARM::R7; + return (isReservedReg(MF, ARM::R7) || isReservedReg(MF, ARM::R6)) + ? 0 : ARM::R7; case ARM::R8: return isReservedReg(MF, ARM::R9) ? 0 :ARM::R9; case ARM::R10: @@ -1221,7 +816,7 @@ emitLoadConstPool(MachineBasicBlock &MBB, BuildMI(MBB, MBBI, dl, TII.get(ARM::LDRcp)) .addReg(DestReg, getDefRegState(true), SubIdx) .addConstantPoolIndex(Idx) - .addReg(0).addImm(0).addImm(Pred).addReg(PredReg); + .addImm(0).addImm(Pred).addReg(PredReg); } bool ARMBaseRegisterInfo:: @@ -1234,32 +829,9 @@ requiresFrameIndexScavenging(const MachineFunction &MF) const { return true; } -// hasReservedCallFrame - Under normal circumstances, when a frame pointer is -// not required, we reserve argument space for call sites in the function -// immediately on entry to the current function. This eliminates the need for -// add/sub sp brackets around call sites. Returns true if the call frame is -// included as part of the stack frame. -bool ARMBaseRegisterInfo:: -hasReservedCallFrame(MachineFunction &MF) const { - const MachineFrameInfo *FFI = MF.getFrameInfo(); - unsigned CFSize = FFI->getMaxCallFrameSize(); - // It's not always a good idea to include the call frame as part of the - // stack frame. ARM (especially Thumb) has small immediate offset to - // address the stack frame. So a large call frame can cause poor codegen - // and may even makes it impossible to scavenge a register. - if (CFSize >= ((1 << 12) - 1) / 2) // Half of imm12 - return false; - - return !MF.getFrameInfo()->hasVarSizedObjects(); -} - -// canSimplifyCallFramePseudos - If there is a reserved call frame, the -// call frame pseudos can be simplified. Unlike most targets, having a FP -// is not sufficient here since we still may reference some objects via SP -// even when FP is available in Thumb2 mode. bool ARMBaseRegisterInfo:: -canSimplifyCallFramePseudos(MachineFunction &MF) const { - return hasReservedCallFrame(MF) || MF.getFrameInfo()->hasVarSizedObjects(); +requiresVirtualBaseRegisters(const MachineFunction &MF) const { + return EnableLocalStackAlloc; } static void @@ -1280,7 +852,8 @@ emitSPUpdate(bool isARM, void ARMBaseRegisterInfo:: eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB, MachineBasicBlock::iterator I) const { - if (!hasReservedCallFrame(MF)) { + const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo(); + if (!TFI->hasReservedCallFrame(MF)) { // If we have alloca, convert as follows: // ADJCALLSTACKDOWN -> sub, sp, sp, amount // ADJCALLSTACKUP -> add, sp, sp, amount @@ -1319,14 +892,265 @@ eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB, MBB.erase(I); } -unsigned +int64_t ARMBaseRegisterInfo:: +getFrameIndexInstrOffset(const MachineInstr *MI, int Idx) const { + const TargetInstrDesc &Desc = MI->getDesc(); + unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); + int64_t InstrOffs = 0;; + int Scale = 1; + unsigned ImmIdx = 0; + switch (AddrMode) { + case ARMII::AddrModeT2_i8: + case ARMII::AddrModeT2_i12: + case ARMII::AddrMode_i12: + InstrOffs = MI->getOperand(Idx+1).getImm(); + Scale = 1; + break; + case ARMII::AddrMode5: { + // VFP address mode. + const MachineOperand &OffOp = MI->getOperand(Idx+1); + InstrOffs = ARM_AM::getAM5Offset(OffOp.getImm()); + if (ARM_AM::getAM5Op(OffOp.getImm()) == ARM_AM::sub) + InstrOffs = -InstrOffs; + Scale = 4; + break; + } + case ARMII::AddrMode2: { + ImmIdx = Idx+2; + InstrOffs = ARM_AM::getAM2Offset(MI->getOperand(ImmIdx).getImm()); + if (ARM_AM::getAM2Op(MI->getOperand(ImmIdx).getImm()) == ARM_AM::sub) + InstrOffs = -InstrOffs; + break; + } + case ARMII::AddrMode3: { + ImmIdx = Idx+2; + InstrOffs = ARM_AM::getAM3Offset(MI->getOperand(ImmIdx).getImm()); + if (ARM_AM::getAM3Op(MI->getOperand(ImmIdx).getImm()) == ARM_AM::sub) + InstrOffs = -InstrOffs; + break; + } + case ARMII::AddrModeT1_s: { + ImmIdx = Idx+1; + InstrOffs = MI->getOperand(ImmIdx).getImm(); + Scale = 4; + break; + } + default: + llvm_unreachable("Unsupported addressing mode!"); + break; + } + + return InstrOffs * Scale; +} + +/// needsFrameBaseReg - Returns true if the instruction's frame index +/// reference would be better served by a base register other than FP +/// or SP. Used by LocalStackFrameAllocation to determine which frame index +/// references it should create new base registers for. +bool ARMBaseRegisterInfo:: +needsFrameBaseReg(MachineInstr *MI, int64_t Offset) const { + for (unsigned i = 0; !MI->getOperand(i).isFI(); ++i) { + assert(i < MI->getNumOperands() &&"Instr doesn't have FrameIndex operand!"); + } + + // It's the load/store FI references that cause issues, as it can be difficult + // to materialize the offset if it won't fit in the literal field. Estimate + // based on the size of the local frame and some conservative assumptions + // about the rest of the stack frame (note, this is pre-regalloc, so + // we don't know everything for certain yet) whether this offset is likely + // to be out of range of the immediate. Return true if so. + + // We only generate virtual base registers for loads and stores, so + // return false for everything else. + unsigned Opc = MI->getOpcode(); + switch (Opc) { + case ARM::LDRi12: case ARM::LDRH: case ARM::LDRBi12: + case ARM::STRi12: case ARM::STRH: case ARM::STRBi12: + case ARM::t2LDRi12: case ARM::t2LDRi8: + case ARM::t2STRi12: case ARM::t2STRi8: + case ARM::VLDRS: case ARM::VLDRD: + case ARM::VSTRS: case ARM::VSTRD: + case ARM::tSTRspi: case ARM::tLDRspi: + if (ForceAllBaseRegAlloc) + return true; + break; + default: + return false; + } + + // Without a virtual base register, if the function has variable sized + // objects, all fixed-size local references will be via the frame pointer, + // Approximate the offset and see if it's legal for the instruction. + // Note that the incoming offset is based on the SP value at function entry, + // so it'll be negative. + MachineFunction &MF = *MI->getParent()->getParent(); + const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo(); + MachineFrameInfo *MFI = MF.getFrameInfo(); + ARMFunctionInfo *AFI = MF.getInfo(); + + // Estimate an offset from the frame pointer. + // Conservatively assume all callee-saved registers get pushed. R4-R6 + // will be earlier than the FP, so we ignore those. + // R7, LR + int64_t FPOffset = Offset - 8; + // ARM and Thumb2 functions also need to consider R8-R11 and D8-D15 + if (!AFI->isThumbFunction() || !AFI->isThumb1OnlyFunction()) + FPOffset -= 80; + // Estimate an offset from the stack pointer. + // The incoming offset is relating to the SP at the start of the function, + // but when we access the local it'll be relative to the SP after local + // allocation, so adjust our SP-relative offset by that allocation size. + Offset = -Offset; + Offset += MFI->getLocalFrameSize(); + // Assume that we'll have at least some spill slots allocated. + // FIXME: This is a total SWAG number. We should run some statistics + // and pick a real one. + Offset += 128; // 128 bytes of spill slots + + // If there is a frame pointer, try using it. + // The FP is only available if there is no dynamic realignment. We + // don't know for sure yet whether we'll need that, so we guess based + // on whether there are any local variables that would trigger it. + unsigned StackAlign = MF.getTarget().getFrameInfo()->getStackAlignment(); + if (TFI->hasFP(MF) && + !((MFI->getLocalFrameMaxAlign() > StackAlign) && canRealignStack(MF))) { + if (isFrameOffsetLegal(MI, FPOffset)) + return false; + } + // If we can reference via the stack pointer, try that. + // FIXME: This (and the code that resolves the references) can be improved + // to only disallow SP relative references in the live range of + // the VLA(s). In practice, it's unclear how much difference that + // would make, but it may be worth doing. + if (!MFI->hasVarSizedObjects() && isFrameOffsetLegal(MI, Offset)) + return false; + + // The offset likely isn't legal, we want to allocate a virtual base register. + return true; +} + +/// materializeFrameBaseRegister - Insert defining instruction(s) for +/// BaseReg to be a pointer to FrameIdx before insertion point I. +void ARMBaseRegisterInfo:: +materializeFrameBaseRegister(MachineBasicBlock::iterator I, unsigned BaseReg, + int FrameIdx, int64_t Offset) const { + ARMFunctionInfo *AFI = + I->getParent()->getParent()->getInfo(); + unsigned ADDriOpc = !AFI->isThumbFunction() ? ARM::ADDri : + (AFI->isThumb1OnlyFunction() ? ARM::tADDrSPi : ARM::t2ADDri); + + MachineInstrBuilder MIB = + BuildMI(*I->getParent(), I, I->getDebugLoc(), TII.get(ADDriOpc), BaseReg) + .addFrameIndex(FrameIdx).addImm(Offset); + if (!AFI->isThumb1OnlyFunction()) + AddDefaultCC(AddDefaultPred(MIB)); +} + +void +ARMBaseRegisterInfo::resolveFrameIndex(MachineBasicBlock::iterator I, + unsigned BaseReg, int64_t Offset) const { + MachineInstr &MI = *I; + MachineBasicBlock &MBB = *MI.getParent(); + MachineFunction &MF = *MBB.getParent(); + ARMFunctionInfo *AFI = MF.getInfo(); + int Off = Offset; // ARM doesn't need the general 64-bit offsets + unsigned i = 0; + + assert(!AFI->isThumb1OnlyFunction() && + "This resolveFrameIndex does not support Thumb1!"); + + while (!MI.getOperand(i).isFI()) { + ++i; + assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!"); + } + bool Done = false; + if (!AFI->isThumbFunction()) + Done = rewriteARMFrameIndex(MI, i, BaseReg, Off, TII); + else { + assert(AFI->isThumb2Function()); + Done = rewriteT2FrameIndex(MI, i, BaseReg, Off, TII); + } + assert (Done && "Unable to resolve frame index!"); +} + +bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, + int64_t Offset) const { + const TargetInstrDesc &Desc = MI->getDesc(); + unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask); + unsigned i = 0; + + while (!MI->getOperand(i).isFI()) { + ++i; + assert(i < MI->getNumOperands() &&"Instr doesn't have FrameIndex operand!"); + } + + // AddrMode4 and AddrMode6 cannot handle any offset. + if (AddrMode == ARMII::AddrMode4 || AddrMode == ARMII::AddrMode6) + return Offset == 0; + + unsigned NumBits = 0; + unsigned Scale = 1; + bool isSigned = true; + switch (AddrMode) { + case ARMII::AddrModeT2_i8: + case ARMII::AddrModeT2_i12: + // i8 supports only negative, and i12 supports only positive, so + // based on Offset sign, consider the appropriate instruction + Scale = 1; + if (Offset < 0) { + NumBits = 8; + Offset = -Offset; + } else { + NumBits = 12; + } + break; + case ARMII::AddrMode5: + // VFP address mode. + NumBits = 8; + Scale = 4; + break; + case ARMII::AddrMode_i12: + case ARMII::AddrMode2: + NumBits = 12; + break; + case ARMII::AddrMode3: + NumBits = 8; + break; + case ARMII::AddrModeT1_s: + NumBits = 5; + Scale = 4; + isSigned = false; + break; + default: + llvm_unreachable("Unsupported addressing mode!"); + break; + } + + Offset += getFrameIndexInstrOffset(MI, i); + // Make sure the offset is encodable for instructions that scale the + // immediate. + if ((Offset & (Scale-1)) != 0) + return false; + + if (isSigned && Offset < 0) + Offset = -Offset; + + unsigned Mask = (1 << NumBits) - 1; + if ((unsigned)Offset <= Mask * Scale) + return true; + + return false; +} + +void ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, - int SPAdj, FrameIndexValue *Value, - RegScavenger *RS) const { + int SPAdj, RegScavenger *RS) const { unsigned i = 0; MachineInstr &MI = *II; MachineBasicBlock &MBB = *MI.getParent(); MachineFunction &MF = *MBB.getParent(); + const ARMFrameInfo *TFI = + static_cast(MF.getTarget().getFrameInfo()); ARMFunctionInfo *AFI = MF.getInfo(); assert(!AFI->isThumb1OnlyFunction() && "This eliminateFrameIndex does not support Thumb1!"); @@ -1339,16 +1163,13 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, int FrameIndex = MI.getOperand(i).getIndex(); unsigned FrameReg; - int Offset = getFrameIndexReference(MF, FrameIndex, FrameReg); - if (FrameReg != ARM::SP) - SPAdj = 0; - Offset += SPAdj; + int Offset = TFI->ResolveFrameIndexReference(MF, FrameIndex, FrameReg, SPAdj); // Special handling of dbg_value instructions. if (MI.isDebugValue()) { MI.getOperand(i). ChangeToRegister(FrameReg, false /*isDef*/); MI.getOperand(i+1).ChangeToImmediate(Offset); - return 0; + return; } // Modify MI as necessary to handle as much of 'Offset' as possible @@ -1360,7 +1181,7 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, Done = rewriteT2FrameIndex(MI, i, FrameReg, Offset, TII); } if (Done) - return 0; + return; // If we get here, the immediate doesn't fit into the instruction. We folded // as much as possible above, handle the rest, providing a register that is @@ -1380,10 +1201,6 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, MI.getOperand(i).ChangeToRegister(FrameReg, false, false, false); else { ScratchReg = MF.getRegInfo().createVirtualRegister(ARM::GPRRegisterClass); - if (Value) { - Value->first = FrameReg; // use the frame register as a kind indicator - Value->second = Offset; - } if (!AFI->isThumbFunction()) emitARMRegPlusImmediate(MBB, II, MI.getDebugLoc(), ScratchReg, FrameReg, Offset, Pred, PredReg, TII); @@ -1393,324 +1210,7 @@ ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, Offset, Pred, PredReg, TII); } MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true); - if (!ReuseFrameIndexVals) - ScratchReg = 0; - } - return ScratchReg; -} - -/// Move iterator past the next bunch of callee save load / store ops for -/// the particular spill area (1: integer area 1, 2: integer area 2, -/// 3: fp area, 0: don't care). -static void movePastCSLoadStoreOps(MachineBasicBlock &MBB, - MachineBasicBlock::iterator &MBBI, - int Opc1, int Opc2, unsigned Area, - const ARMSubtarget &STI) { - while (MBBI != MBB.end() && - ((MBBI->getOpcode() == Opc1) || (MBBI->getOpcode() == Opc2)) && - MBBI->getOperand(1).isFI()) { - if (Area != 0) { - bool Done = false; - unsigned Category = 0; - switch (MBBI->getOperand(0).getReg()) { - case ARM::R4: case ARM::R5: case ARM::R6: case ARM::R7: - case ARM::LR: - Category = 1; - break; - case ARM::R8: case ARM::R9: case ARM::R10: case ARM::R11: - Category = STI.isTargetDarwin() ? 2 : 1; - break; - case ARM::D8: case ARM::D9: case ARM::D10: case ARM::D11: - case ARM::D12: case ARM::D13: case ARM::D14: case ARM::D15: - Category = 3; - break; - default: - Done = true; - break; - } - if (Done || Category != Area) - break; - } - - ++MBBI; - } -} - -void ARMBaseRegisterInfo:: -emitPrologue(MachineFunction &MF) const { - MachineBasicBlock &MBB = MF.front(); - MachineBasicBlock::iterator MBBI = MBB.begin(); - MachineFrameInfo *MFI = MF.getFrameInfo(); - ARMFunctionInfo *AFI = MF.getInfo(); - assert(!AFI->isThumb1OnlyFunction() && - "This emitPrologue does not support Thumb1!"); - bool isARM = !AFI->isThumbFunction(); - unsigned VARegSaveSize = AFI->getVarArgsRegSaveSize(); - unsigned NumBytes = MFI->getStackSize(); - const std::vector &CSI = MFI->getCalleeSavedInfo(); - DebugLoc dl = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc(); - - // Determine the sizes of each callee-save spill areas and record which frame - // belongs to which callee-save spill areas. - unsigned GPRCS1Size = 0, GPRCS2Size = 0, DPRCSSize = 0; - int FramePtrSpillFI = 0; - - // Allocate the vararg register save area. This is not counted in NumBytes. - if (VARegSaveSize) - emitSPUpdate(isARM, MBB, MBBI, dl, TII, -VARegSaveSize); - - if (!AFI->hasStackFrame()) { - if (NumBytes != 0) - emitSPUpdate(isARM, MBB, MBBI, dl, TII, -NumBytes); - return; - } - - for (unsigned i = 0, e = CSI.size(); i != e; ++i) { - unsigned Reg = CSI[i].getReg(); - int FI = CSI[i].getFrameIdx(); - switch (Reg) { - case ARM::R4: - case ARM::R5: - case ARM::R6: - case ARM::R7: - case ARM::LR: - if (Reg == FramePtr) - FramePtrSpillFI = FI; - AFI->addGPRCalleeSavedArea1Frame(FI); - GPRCS1Size += 4; - break; - case ARM::R8: - case ARM::R9: - case ARM::R10: - case ARM::R11: - if (Reg == FramePtr) - FramePtrSpillFI = FI; - if (STI.isTargetDarwin()) { - AFI->addGPRCalleeSavedArea2Frame(FI); - GPRCS2Size += 4; - } else { - AFI->addGPRCalleeSavedArea1Frame(FI); - GPRCS1Size += 4; - } - break; - default: - AFI->addDPRCalleeSavedAreaFrame(FI); - DPRCSSize += 8; - } - } - - // Build the new SUBri to adjust SP for integer callee-save spill area 1. - emitSPUpdate(isARM, MBB, MBBI, dl, TII, -GPRCS1Size); - movePastCSLoadStoreOps(MBB, MBBI, ARM::STR, ARM::t2STRi12, 1, STI); - - // Set FP to point to the stack slot that contains the previous FP. - // For Darwin, FP is R7, which has now been stored in spill area 1. - // Otherwise, if this is not Darwin, all the callee-saved registers go - // into spill area 1, including the FP in R11. In either case, it is - // now safe to emit this assignment. - if (STI.isTargetDarwin() || hasFP(MF)) { - unsigned ADDriOpc = !AFI->isThumbFunction() ? ARM::ADDri : ARM::t2ADDri; - MachineInstrBuilder MIB = - BuildMI(MBB, MBBI, dl, TII.get(ADDriOpc), FramePtr) - .addFrameIndex(FramePtrSpillFI).addImm(0); - AddDefaultCC(AddDefaultPred(MIB)); - } - - // Build the new SUBri to adjust SP for integer callee-save spill area 2. - emitSPUpdate(isARM, MBB, MBBI, dl, TII, -GPRCS2Size); - - // Build the new SUBri to adjust SP for FP callee-save spill area. - movePastCSLoadStoreOps(MBB, MBBI, ARM::STR, ARM::t2STRi12, 2, STI); - emitSPUpdate(isARM, MBB, MBBI, dl, TII, -DPRCSSize); - - // Determine starting offsets of spill areas. - unsigned DPRCSOffset = NumBytes - (GPRCS1Size + GPRCS2Size + DPRCSSize); - unsigned GPRCS2Offset = DPRCSOffset + DPRCSSize; - unsigned GPRCS1Offset = GPRCS2Offset + GPRCS2Size; - if (STI.isTargetDarwin() || hasFP(MF)) - AFI->setFramePtrSpillOffset(MFI->getObjectOffset(FramePtrSpillFI) + - NumBytes); - AFI->setGPRCalleeSavedArea1Offset(GPRCS1Offset); - AFI->setGPRCalleeSavedArea2Offset(GPRCS2Offset); - AFI->setDPRCalleeSavedAreaOffset(DPRCSOffset); - - movePastCSLoadStoreOps(MBB, MBBI, ARM::VSTRD, 0, 3, STI); - NumBytes = DPRCSOffset; - if (NumBytes) { - // Adjust SP after all the callee-save spills. - emitSPUpdate(isARM, MBB, MBBI, dl, TII, -NumBytes); - } - - if (STI.isTargetELF() && hasFP(MF)) { - MFI->setOffsetAdjustment(MFI->getOffsetAdjustment() - - AFI->getFramePtrSpillOffset()); - } - - AFI->setGPRCalleeSavedArea1Size(GPRCS1Size); - AFI->setGPRCalleeSavedArea2Size(GPRCS2Size); - AFI->setDPRCalleeSavedAreaSize(DPRCSSize); - - // If we need dynamic stack realignment, do it here. - if (needsStackRealignment(MF)) { - unsigned MaxAlign = MFI->getMaxAlignment(); - assert (!AFI->isThumb1OnlyFunction()); - if (!AFI->isThumbFunction()) { - // Emit bic sp, sp, MaxAlign - AddDefaultCC(AddDefaultPred(BuildMI(MBB, MBBI, dl, - TII.get(ARM::BICri), ARM::SP) - .addReg(ARM::SP, RegState::Kill) - .addImm(MaxAlign-1))); - } else { - // We cannot use sp as source/dest register here, thus we're emitting the - // following sequence: - // mov r4, sp - // bic r4, r4, MaxAlign - // mov sp, r4 - // FIXME: It will be better just to find spare register here. - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2tgpr), ARM::R4) - .addReg(ARM::SP, RegState::Kill); - AddDefaultCC(AddDefaultPred(BuildMI(MBB, MBBI, dl, - TII.get(ARM::t2BICri), ARM::R4) - .addReg(ARM::R4, RegState::Kill) - .addImm(MaxAlign-1))); - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVtgpr2gpr), ARM::SP) - .addReg(ARM::R4, RegState::Kill); - } } } -static bool isCalleeSavedRegister(unsigned Reg, const unsigned *CSRegs) { - for (unsigned i = 0; CSRegs[i]; ++i) - if (Reg == CSRegs[i]) - return true; - return false; -} - -static bool isCSRestore(MachineInstr *MI, - const ARMBaseInstrInfo &TII, - const unsigned *CSRegs) { - return ((MI->getOpcode() == (int)ARM::VLDRD || - MI->getOpcode() == (int)ARM::LDR || - MI->getOpcode() == (int)ARM::t2LDRi12) && - MI->getOperand(1).isFI() && - isCalleeSavedRegister(MI->getOperand(0).getReg(), CSRegs)); -} - -void ARMBaseRegisterInfo:: -emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const { - MachineBasicBlock::iterator MBBI = prior(MBB.end()); - assert(MBBI->getDesc().isReturn() && - "Can only insert epilog into returning blocks"); - unsigned RetOpcode = MBBI->getOpcode(); - DebugLoc dl = MBBI->getDebugLoc(); - MachineFrameInfo *MFI = MF.getFrameInfo(); - ARMFunctionInfo *AFI = MF.getInfo(); - assert(!AFI->isThumb1OnlyFunction() && - "This emitEpilogue does not support Thumb1!"); - bool isARM = !AFI->isThumbFunction(); - - unsigned VARegSaveSize = AFI->getVarArgsRegSaveSize(); - int NumBytes = (int)MFI->getStackSize(); - - if (!AFI->hasStackFrame()) { - if (NumBytes != 0) - emitSPUpdate(isARM, MBB, MBBI, dl, TII, NumBytes); - } else { - // Unwind MBBI to point to first LDR / VLDRD. - const unsigned *CSRegs = getCalleeSavedRegs(); - if (MBBI != MBB.begin()) { - do - --MBBI; - while (MBBI != MBB.begin() && isCSRestore(MBBI, TII, CSRegs)); - if (!isCSRestore(MBBI, TII, CSRegs)) - ++MBBI; - } - - // Move SP to start of FP callee save spill area. - NumBytes -= (AFI->getGPRCalleeSavedArea1Size() + - AFI->getGPRCalleeSavedArea2Size() + - AFI->getDPRCalleeSavedAreaSize()); - - // Darwin ABI requires FP to point to the stack slot that contains the - // previous FP. - bool HasFP = hasFP(MF); - if ((STI.isTargetDarwin() && NumBytes) || HasFP) { - NumBytes = AFI->getFramePtrSpillOffset() - NumBytes; - // Reset SP based on frame pointer only if the stack frame extends beyond - // frame pointer stack slot or target is ELF and the function has FP. - if (HasFP || - AFI->getGPRCalleeSavedArea2Size() || - AFI->getDPRCalleeSavedAreaSize() || - AFI->getDPRCalleeSavedAreaOffset()) { - if (NumBytes) { - if (isARM) - emitARMRegPlusImmediate(MBB, MBBI, dl, ARM::SP, FramePtr, -NumBytes, - ARMCC::AL, 0, TII); - else - emitT2RegPlusImmediate(MBB, MBBI, dl, ARM::SP, FramePtr, -NumBytes, - ARMCC::AL, 0, TII); - } else { - // Thumb2 or ARM. - if (isARM) - BuildMI(MBB, MBBI, dl, TII.get(ARM::MOVr), ARM::SP) - .addReg(FramePtr) - .addImm((unsigned)ARMCC::AL).addReg(0).addReg(0); - else - BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVgpr2gpr), ARM::SP) - .addReg(FramePtr); - } - } - } else if (NumBytes) - emitSPUpdate(isARM, MBB, MBBI, dl, TII, NumBytes); - - // Move SP to start of integer callee save spill area 2. - movePastCSLoadStoreOps(MBB, MBBI, ARM::VLDRD, 0, 3, STI); - emitSPUpdate(isARM, MBB, MBBI, dl, TII, AFI->getDPRCalleeSavedAreaSize()); - - // Move SP to start of integer callee save spill area 1. - movePastCSLoadStoreOps(MBB, MBBI, ARM::LDR, ARM::t2LDRi12, 2, STI); - emitSPUpdate(isARM, MBB, MBBI, dl, TII, AFI->getGPRCalleeSavedArea2Size()); - - // Move SP to SP upon entry to the function. - movePastCSLoadStoreOps(MBB, MBBI, ARM::LDR, ARM::t2LDRi12, 1, STI); - emitSPUpdate(isARM, MBB, MBBI, dl, TII, AFI->getGPRCalleeSavedArea1Size()); - } - - if (RetOpcode == ARM::TCRETURNdi || RetOpcode == ARM::TCRETURNdiND || - RetOpcode == ARM::TCRETURNri || RetOpcode == ARM::TCRETURNriND) { - // Tail call return: adjust the stack pointer and jump to callee. - MBBI = prior(MBB.end()); - MachineOperand &JumpTarget = MBBI->getOperand(0); - - // Jump to label or value in register. - if (RetOpcode == ARM::TCRETURNdi) { - BuildMI(MBB, MBBI, dl, - TII.get(STI.isThumb() ? ARM::TAILJMPdt : ARM::TAILJMPd)). - addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), - JumpTarget.getTargetFlags()); - } else if (RetOpcode == ARM::TCRETURNdiND) { - BuildMI(MBB, MBBI, dl, - TII.get(STI.isThumb() ? ARM::TAILJMPdNDt : ARM::TAILJMPdND)). - addGlobalAddress(JumpTarget.getGlobal(), JumpTarget.getOffset(), - JumpTarget.getTargetFlags()); - } else if (RetOpcode == ARM::TCRETURNri) { - BuildMI(MBB, MBBI, dl, TII.get(ARM::TAILJMPr)). - addReg(JumpTarget.getReg(), RegState::Kill); - } else if (RetOpcode == ARM::TCRETURNriND) { - BuildMI(MBB, MBBI, dl, TII.get(ARM::TAILJMPrND)). - addReg(JumpTarget.getReg(), RegState::Kill); - } - - MachineInstr *NewMI = prior(MBBI); - for (unsigned i = 1, e = MBBI->getNumOperands(); i != e; ++i) - NewMI->addOperand(MBBI->getOperand(i)); - - // Delete the pseudo instruction TCRETURN. - MBB.erase(MBBI); - } - - if (VARegSaveSize) - emitSPUpdate(isARM, MBB, MBBI, dl, TII, VARegSaveSize); -} - #include "ARMGenRegisterInfo.inc"