-//===- ARMBaseRegisterInfo.cpp - ARM Register Information -----------*- C++ -*-===//
+//===- ARMBaseRegisterInfo.cpp - ARM Register Information -------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
#include "ARMSubtarget.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
+#include "llvm/Function.h"
+#include "llvm/LLVMContext.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineLocation.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetFrameInfo.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/ADT/BitVector.h"
#include "llvm/ADT/SmallVector.h"
+#include "llvm/Support/CommandLine.h"
using namespace llvm;
-unsigned ARMBaseRegisterInfo::getRegisterNumbering(unsigned RegEnum) {
- using namespace ARM;
- switch (RegEnum) {
- case R0: case S0: case D0: return 0;
- case R1: case S1: case D1: return 1;
- case R2: case S2: case D2: return 2;
- case R3: case S3: case D3: return 3;
- case R4: case S4: case D4: return 4;
- case R5: case S5: case D5: return 5;
- case R6: case S6: case D6: return 6;
- case R7: case S7: case D7: return 7;
- case R8: case S8: case D8: return 8;
- case R9: case S9: case D9: return 9;
- case R10: case S10: case D10: return 10;
- case R11: case S11: case D11: return 11;
- case R12: case S12: case D12: return 12;
- case SP: case S13: case D13: return 13;
- case LR: case S14: case D14: return 14;
- case PC: case S15: case D15: 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;
- default:
- LLVM_UNREACHABLE("Unknown ARM register!");
- }
-}
+static cl::opt<bool>
+ReuseFrameIndexVals("arm-reuse-frame-index-vals", cl::Hidden, cl::init(true),
+ cl::desc("Reuse repeated frame index values"));
unsigned ARMBaseRegisterInfo::getRegisterNumbering(unsigned RegEnum,
- bool &isSPVFP) {
- isSPVFP = false;
+ bool *isSPVFP) {
+ if (isSPVFP)
+ *isSPVFP = false;
using namespace ARM;
switch (RegEnum) {
default:
- LLVM_UNREACHABLE("Unknown ARM register!");
- case R0: case D0: return 0;
- case R1: case D1: return 1;
- case R2: case D2: return 2;
- case R3: case D3: return 3;
- case R4: case D4: return 4;
- case R5: case D5: return 5;
- case R6: case D6: return 6;
- case R7: case D7: return 7;
- case R8: case D8: return 8;
- case R9: case D9: return 9;
- case R10: case D10: return 10;
- case R11: case D11: return 11;
- case R12: case D12: return 12;
- case SP: case D13: return 13;
- case LR: case D14: return 14;
- case PC: case D15: return 15;
+ 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 27;
+ 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 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: {
- isSPVFP = true;
+ 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;
FramePtr((STI.isTargetDarwin() || STI.isThumb()) ? ARM::R7 : ARM::R11) {
}
-unsigned ARMBaseRegisterInfo::
-getOpcode(int Op) const {
- return TII.getOpcode((ARMII::Op)Op);
-}
-
const unsigned*
ARMBaseRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
static const unsigned CalleeSavedRegs[] = {
return Reserved;
}
-bool
-ARMBaseRegisterInfo::isReservedReg(const MachineFunction &MF, unsigned Reg) const {
+bool ARMBaseRegisterInfo::isReservedReg(const MachineFunction &MF,
+ unsigned Reg) const {
switch (Reg) {
default: break;
case ARM::SP:
return false;
}
-const TargetRegisterClass *ARMBaseRegisterInfo::getPointerRegClass() const {
- return &ARM::GPRRegClass;
+const TargetRegisterClass *
+ARMBaseRegisterInfo::getMatchingSuperRegClass(const TargetRegisterClass *A,
+ const TargetRegisterClass *B,
+ unsigned SubIdx) const {
+ switch (SubIdx) {
+ default: return 0;
+ case 1:
+ case 2:
+ case 3:
+ case 4:
+ // S sub-registers.
+ if (A->getSize() == 8) {
+ if (B == &ARM::SPR_8RegClass)
+ return &ARM::DPR_8RegClass;
+ assert(B == &ARM::SPRRegClass && "Expecting SPR register class!");
+ if (A == &ARM::DPR_8RegClass)
+ return A;
+ return &ARM::DPR_VFP2RegClass;
+ }
+
+ assert(A->getSize() == 16 && "Expecting a Q register class!");
+ if (B == &ARM::SPR_8RegClass)
+ return &ARM::QPR_8RegClass;
+ return &ARM::QPR_VFP2RegClass;
+ case 5:
+ case 6:
+ // D sub-registers.
+ if (B == &ARM::DPR_VFP2RegClass)
+ return &ARM::QPR_VFP2RegClass;
+ if (B == &ARM::DPR_8RegClass)
+ return &ARM::QPR_8RegClass;
+ return A;
+ }
+ return 0;
+}
+
+const TargetRegisterClass *
+ARMBaseRegisterInfo::getPointerRegClass(unsigned Kind) const {
+ return ARM::GPRRegisterClass;
}
/// getAllocationOrder - Returns the register allocation order for a specified
}
}
+static unsigned calculateMaxStackAlignment(const MachineFrameInfo *FFI) {
+ unsigned MaxAlign = 0;
+
+ for (int i = FFI->getObjectIndexBegin(),
+ e = FFI->getObjectIndexEnd(); i != e; ++i) {
+ if (FFI->isDeadObjectIndex(i))
+ continue;
+
+ unsigned Align = FFI->getObjectAlignment(i);
+ MaxAlign = std::max(MaxAlign, Align);
+ }
+
+ return MaxAlign;
+}
+
/// 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 {
const MachineFrameInfo *MFI = MF.getFrameInfo();
return (NoFramePointerElim ||
+ needsStackRealignment(MF) ||
MFI->hasVarSizedObjects() ||
MFI->isFrameAddressTaken());
}
+bool ARMBaseRegisterInfo::
+needsStackRealignment(const MachineFunction &MF) const {
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
+ const ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ unsigned StackAlign = MF.getTarget().getFrameInfo()->getStackAlignment();
+ return (RealignStack &&
+ !AFI->isThumb1OnlyFunction() &&
+ (MFI->getMaxAlignment() > StackAlign) &&
+ !MFI->hasVarSizedObjects());
+}
+
+bool ARMBaseRegisterInfo::cannotEliminateFrame(const MachineFunction &MF) const {
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
+ if (NoFramePointerElim && MFI->hasCalls())
+ return true;
+ return MFI->hasVarSizedObjects() || MFI->isFrameAddressTaken()
+ || needsStackRealignment(MF);
+}
+
+/// estimateStackSize - Estimate and return the size of the frame.
static unsigned estimateStackSize(MachineFunction &MF, MachineFrameInfo *MFI) {
const MachineFrameInfo *FFI = MF.getFrameInfo();
int Offset = 0;
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 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;
+
+ const TargetInstrDesc &Desc = TII.get(I->getOpcode());
+ unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask);
+ if (AddrMode == ARMII::AddrMode3 ||
+ AddrMode == ARMII::AddrModeT2_i8)
+ return (1 << 8) - 1;
+
+ if (AddrMode == ARMII::AddrMode5 ||
+ AddrMode == ARMII::AddrModeT2_i8s4)
+ Limit = std::min(Limit, ((1U << 8) - 1) * 4);
+
+ if (AddrMode == ARMII::AddrModeT2_i12 && hasFP(MF))
+ // When the stack offset is negative, we will end up using
+ // the i8 instructions instead.
+ return (1 << 8) - 1;
+ break; // At most one FI per instruction
+ }
+ }
+ }
+
+ return Limit;
+}
+
void
ARMBaseRegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
RegScavenger *RS) const {
SmallVector<unsigned, 4> UnspilledCS2GPRs;
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+
+ // Calculate and set max stack object alignment early, so we can decide
+ // whether we will need stack realignment (and thus FP).
+ if (RealignStack) {
+ unsigned MaxAlign = std::max(MFI->getMaxAlignment(),
+ calculateMaxStackAlignment(MFI));
+ MFI->setMaxAlignment(MaxAlign);
+ }
+
// 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();
}
}
- if (CSRegClasses[i] == &ARM::GPRRegClass) {
+ if (CSRegClasses[i] == ARM::GPRRegisterClass ||
+ CSRegClasses[i] == ARM::tGPRRegisterClass) {
if (Spilled) {
NumGPRSpills++;
}
bool ExtraCSSpill = false;
- if (!CanEliminateFrame || hasFP(MF)) {
+ if (!CanEliminateFrame || cannotEliminateFrame(MF)) {
AFI->setHasStackFrame(true);
// If LR is not spilled, but at least one of R4, R5, R6, and R7 is spilled.
// 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.
- if (RS && !ExtraCSSpill && !AFI->isThumb1OnlyFunction()) {
+ // register scavenging. Thumb1 needs a spill slot for stack pointer
+ // adjustments also, even when the frame itself is small.
+ if (RS && !ExtraCSSpill) {
MachineFrameInfo *MFI = MF.getFrameInfo();
- unsigned Size = estimateStackSize(MF, MFI);
- unsigned Limit = (1 << 12) - 1;
- for (MachineFunction::iterator BB = MF.begin(),E = MF.end();BB != E; ++BB)
- for (MachineBasicBlock::iterator I= BB->begin(); I != BB->end(); ++I) {
- for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i)
- if (I->getOperand(i).isFI()) {
- unsigned Opcode = I->getOpcode();
- const TargetInstrDesc &Desc = TII.get(Opcode);
- unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask);
- if (AddrMode == ARMII::AddrMode3) {
- Limit = (1 << 8) - 1;
- goto DoneEstimating;
- } else if (AddrMode == ARMII::AddrMode5) {
- unsigned ThisLimit = ((1 << 8) - 1) * 4;
- if (ThisLimit < Limit)
- Limit = ThisLimit;
- }
- }
- }
- DoneEstimating:
- if (Size >= Limit) {
+ // 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.
+ if (estimateStackSize(MF, MFI) + (hasFP(MF) ? 4 : 0)
+ >= estimateRSStackSizeLimit(MF)) {
// 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;
NumExtras--;
}
}
- while (NumExtras && !UnspilledCS2GPRs.empty()) {
- unsigned Reg = UnspilledCS2GPRs.back();
- UnspilledCS2GPRs.pop_back();
- if (!isReservedReg(MF, Reg)) {
- 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) {
MF.getRegInfo().setPhysRegUsed(Extras[i]);
AFI->setCSRegisterIsSpilled(Extras[i]);
}
- } else {
+ } 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::GPRRegClass;
+ const TargetRegisterClass *RC = ARM::GPRRegisterClass;
RS->setScavengingFrameIndex(MFI->CreateStackObject(RC->getSize(),
- RC->getAlignment()));
+ RC->getAlignment(),
+ false));
}
}
}
return ARM::LR;
}
-unsigned ARMBaseRegisterInfo::getFrameRegister(MachineFunction &MF) const {
+unsigned
+ARMBaseRegisterInfo::getFrameRegister(const MachineFunction &MF) const {
if (STI.isTargetDarwin() || hasFP(MF))
return FramePtr;
return ARM::SP;
}
unsigned ARMBaseRegisterInfo::getEHExceptionRegister() const {
- assert(0 && "What is the exception register");
+ llvm_unreachable("What is the exception register");
return 0;
}
unsigned ARMBaseRegisterInfo::getEHHandlerRegister() const {
- assert(0 && "What is the exception handler register");
+ llvm_unreachable("What is the exception handler register");
return 0;
}
case ARM::R1:
return ARM::R0;
case ARM::R3:
- // FIXME!
- return STI.isThumb1Only() ? 0 : ARM::R2;
+ return ARM::R2;
case ARM::R5:
return ARM::R4;
case ARM::R7:
return ARM::D12;
case ARM::D15:
return ARM::D14;
+ case ARM::D17:
+ return ARM::D16;
+ case ARM::D19:
+ return ARM::D18;
+ case ARM::D21:
+ return ARM::D20;
+ case ARM::D23:
+ return ARM::D22;
+ case ARM::D25:
+ return ARM::D24;
+ case ARM::D27:
+ return ARM::D26;
+ case ARM::D29:
+ return ARM::D28;
+ case ARM::D31:
+ return ARM::D30;
}
return 0;
case ARM::R0:
return ARM::R1;
case ARM::R2:
- // FIXME!
- return STI.isThumb1Only() ? 0 : ARM::R3;
+ return ARM::R3;
case ARM::R4:
return ARM::R5;
case ARM::R6:
return ARM::D13;
case ARM::D14:
return ARM::D15;
+ case ARM::D16:
+ return ARM::D17;
+ case ARM::D18:
+ return ARM::D19;
+ case ARM::D20:
+ return ARM::D21;
+ case ARM::D22:
+ return ARM::D23;
+ case ARM::D24:
+ return ARM::D25;
+ case ARM::D26:
+ return ARM::D27;
+ case ARM::D28:
+ return ARM::D29;
+ case ARM::D30:
+ return ARM::D31;
}
return 0;
}
-
-static inline
-const MachineInstrBuilder &AddDefaultPred(const MachineInstrBuilder &MIB) {
- return MIB.addImm((int64_t)ARMCC::AL).addReg(0);
-}
-
-static inline
-const MachineInstrBuilder &AddDefaultCC(const MachineInstrBuilder &MIB) {
- return MIB.addReg(0);
-}
-
/// emitLoadConstPool - Emits a load from constpool to materialize the
/// specified immediate.
void ARMBaseRegisterInfo::
emitLoadConstPool(MachineBasicBlock &MBB,
MachineBasicBlock::iterator &MBBI,
DebugLoc dl,
- unsigned DestReg, int Val,
+ unsigned DestReg, unsigned SubIdx, int Val,
ARMCC::CondCodes Pred,
unsigned PredReg) const {
MachineFunction &MF = *MBB.getParent();
MachineConstantPool *ConstantPool = MF.getConstantPool();
- Constant *C = ConstantInt::get(Type::Int32Ty, Val);
+ Constant *C =
+ ConstantInt::get(Type::getInt32Ty(MF.getFunction()->getContext()), Val);
unsigned Idx = ConstantPool->getConstantPoolIndex(C, 4);
- BuildMI(MBB, MBBI, dl, TII.get(ARM::LDRcp), DestReg)
+ BuildMI(MBB, MBBI, dl, TII.get(ARM::LDRcp))
+ .addReg(DestReg, getDefRegState(true), SubIdx)
.addConstantPoolIndex(Idx)
.addReg(0).addImm(0).addImm(Pred).addReg(PredReg);
}
return true;
}
+bool ARMBaseRegisterInfo::
+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
return !MF.getFrameInfo()->hasVarSizedObjects();
}
-/// emitARMRegPlusImmediate - Emits a series of instructions to materialize
-/// a destreg = basereg + immediate in ARM code.
-static
-void emitARMRegPlusImmediate(MachineBasicBlock &MBB,
- MachineBasicBlock::iterator &MBBI,
- unsigned DestReg, unsigned BaseReg, int NumBytes,
- ARMCC::CondCodes Pred, unsigned PredReg,
- const ARMBaseInstrInfo &TII,
- DebugLoc dl) {
- bool isSub = NumBytes < 0;
- if (isSub) NumBytes = -NumBytes;
-
- while (NumBytes) {
- unsigned RotAmt = ARM_AM::getSOImmValRotate(NumBytes);
- unsigned ThisVal = NumBytes & ARM_AM::rotr32(0xFF, RotAmt);
- assert(ThisVal && "Didn't extract field correctly");
-
- // We will handle these bits from offset, clear them.
- NumBytes &= ~ThisVal;
-
- assert(ARM_AM::getSOImmVal(ThisVal) != -1 && "Bit extraction didn't work?");
-
- // Build the new ADD / SUB.
- BuildMI(MBB, MBBI, dl, TII.get(TII.getOpcode(isSub ? ARMII::SUBri : ARMII::ADDri)), DestReg)
- .addReg(BaseReg, RegState::Kill).addImm(ThisVal)
- .addImm((unsigned)Pred).addReg(PredReg).addReg(0);
- BaseReg = DestReg;
- }
-}
-
static void
-emitSPUpdate(MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
- const ARMBaseInstrInfo &TII, DebugLoc dl,
+emitSPUpdate(bool isARM,
+ MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
+ DebugLoc dl, const ARMBaseInstrInfo &TII,
int NumBytes,
ARMCC::CondCodes Pred = ARMCC::AL, unsigned PredReg = 0) {
- emitARMRegPlusImmediate(MBB, MBBI, ARM::SP, ARM::SP, NumBytes,
- Pred, PredReg, TII, dl);
+ if (isARM)
+ emitARMRegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes,
+ Pred, PredReg, TII);
+ else
+ emitT2RegPlusImmediate(MBB, MBBI, dl, ARM::SP, ARM::SP, NumBytes,
+ Pred, PredReg, TII);
}
+
void ARMBaseRegisterInfo::
eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
MachineBasicBlock::iterator I) const {
unsigned Align = MF.getTarget().getFrameInfo()->getStackAlignment();
Amount = (Amount+Align-1)/Align*Align;
+ ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ assert(!AFI->isThumb1OnlyFunction() &&
+ "This eliminateCallFramePseudoInstr does not suppor Thumb1!");
+ bool isARM = !AFI->isThumbFunction();
+
// Replace the pseudo instruction with a new instruction...
unsigned Opc = Old->getOpcode();
ARMCC::CondCodes Pred = (ARMCC::CondCodes)Old->getOperand(1).getImm();
+ // FIXME: Thumb2 version of ADJCALLSTACKUP and ADJCALLSTACKDOWN?
if (Opc == ARM::ADJCALLSTACKDOWN || Opc == ARM::tADJCALLSTACKDOWN) {
// Note: PredReg is operand 2 for ADJCALLSTACKDOWN.
unsigned PredReg = Old->getOperand(2).getReg();
- emitSPUpdate(MBB, I, TII, dl, -Amount, Pred, PredReg);
+ emitSPUpdate(isARM, MBB, I, dl, TII, -Amount, Pred, PredReg);
} else {
// Note: PredReg is operand 3 for ADJCALLSTACKUP.
unsigned PredReg = Old->getOperand(3).getReg();
assert(Opc == ARM::ADJCALLSTACKUP || Opc == ARM::tADJCALLSTACKUP);
- emitSPUpdate(MBB, I, TII, dl, Amount, Pred, PredReg);
+ emitSPUpdate(isARM, MBB, I, dl, TII, Amount, Pred, PredReg);
}
}
}
MBB.erase(I);
}
-/// findScratchRegister - Find a 'free' ARM register. If register scavenger
-/// is not being used, R12 is available. Otherwise, try for a call-clobbered
-/// register first and then a spilled callee-saved register if that fails.
-static
-unsigned findScratchRegister(RegScavenger *RS, const TargetRegisterClass *RC,
- ARMFunctionInfo *AFI) {
- unsigned Reg = RS ? RS->FindUnusedReg(RC, true) : (unsigned) ARM::R12;
- assert (!AFI->isThumb1OnlyFunction());
- if (Reg == 0)
- // Try a already spilled CS register.
- Reg = RS->FindUnusedReg(RC, AFI->getSpilledCSRegisters());
-
- return Reg;
-}
-
-void ARMBaseRegisterInfo::
-eliminateFrameIndex(MachineBasicBlock::iterator II,
- int SPAdj, RegScavenger *RS) const{
+unsigned
+ARMBaseRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
+ int SPAdj, int *Value,
+ RegScavenger *RS) const {
unsigned i = 0;
MachineInstr &MI = *II;
MachineBasicBlock &MBB = *MI.getParent();
MachineFunction &MF = *MBB.getParent();
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
- DebugLoc dl = MI.getDebugLoc();
+ assert(!AFI->isThumb1OnlyFunction() &&
+ "This eliminateFrameIndex does not support Thumb1!");
while (!MI.getOperand(i).isFI()) {
++i;
unsigned FrameReg = ARM::SP;
int FrameIndex = MI.getOperand(i).getIndex();
- int Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex) +
- MF.getFrameInfo()->getStackSize() + SPAdj;
+ int Offset = MFI->getObjectOffset(FrameIndex) + MFI->getStackSize() + SPAdj;
+ bool isFixed = MFI->isFixedObjectIndex(FrameIndex);
+ // When doing dynamic stack realignment, all of these need to change(?)
if (AFI->isGPRCalleeSavedArea1Frame(FrameIndex))
Offset -= AFI->getGPRCalleeSavedArea1Offset();
else if (AFI->isGPRCalleeSavedArea2Frame(FrameIndex))
Offset -= AFI->getGPRCalleeSavedArea2Offset();
else if (AFI->isDPRCalleeSavedAreaFrame(FrameIndex))
Offset -= AFI->getDPRCalleeSavedAreaOffset();
- else if (hasFP(MF)) {
- assert(SPAdj == 0 && "Unexpected");
- // There is alloca()'s in this function, must reference off the frame
- // pointer instead.
- FrameReg = getFrameRegister(MF);
- Offset -= AFI->getFramePtrSpillOffset();
- }
-
- unsigned Opcode = MI.getOpcode();
- const TargetInstrDesc &Desc = MI.getDesc();
- unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask);
- bool isSub = false;
-
- // Memory operands in inline assembly always use AddrMode2.
- if (Opcode == ARM::INLINEASM)
- AddrMode = ARMII::AddrMode2;
-
- if (Opcode == getOpcode(ARMII::ADDri)) {
- Offset += MI.getOperand(i+1).getImm();
- if (Offset == 0) {
- // Turn it into a move.
- MI.setDesc(TII.get(getOpcode(ARMII::MOVr)));
- MI.getOperand(i).ChangeToRegister(FrameReg, false);
- MI.RemoveOperand(i+1);
- return;
- } else if (Offset < 0) {
- Offset = -Offset;
- isSub = true;
- MI.setDesc(TII.get(getOpcode(ARMII::SUBri)));
- }
-
- // Common case: small offset, fits into instruction.
- if (ARM_AM::getSOImmVal(Offset) != -1) {
- // Replace the FrameIndex with sp / fp
- MI.getOperand(i).ChangeToRegister(FrameReg, false);
- MI.getOperand(i+1).ChangeToImmediate(Offset);
- return;
- }
-
- // Otherwise, we fallback to common code below to form the imm offset with
- // a sequence of ADDri instructions. First though, pull as much of the imm
- // into this ADDri as possible.
- unsigned RotAmt = ARM_AM::getSOImmValRotate(Offset);
- unsigned ThisImmVal = Offset & ARM_AM::rotr32(0xFF, RotAmt);
-
- // We will handle these bits from offset, clear them.
- Offset &= ~ThisImmVal;
-
- // Get the properly encoded SOImmVal field.
- assert(ARM_AM::getSOImmVal(ThisImmVal) != -1 &&
- "Bit extraction didn't work?");
- MI.getOperand(i+1).ChangeToImmediate(ThisImmVal);
- } else {
- unsigned ImmIdx = 0;
- int InstrOffs = 0;
- unsigned NumBits = 0;
- unsigned Scale = 1;
- switch (AddrMode) {
- case ARMII::AddrMode2: {
- ImmIdx = i+2;
- InstrOffs = ARM_AM::getAM2Offset(MI.getOperand(ImmIdx).getImm());
- if (ARM_AM::getAM2Op(MI.getOperand(ImmIdx).getImm()) == ARM_AM::sub)
- InstrOffs *= -1;
- NumBits = 12;
- break;
- }
- case ARMII::AddrMode3: {
- ImmIdx = i+2;
- InstrOffs = ARM_AM::getAM3Offset(MI.getOperand(ImmIdx).getImm());
- if (ARM_AM::getAM3Op(MI.getOperand(ImmIdx).getImm()) == ARM_AM::sub)
- InstrOffs *= -1;
- NumBits = 8;
- break;
- }
- case ARMII::AddrMode5: {
- ImmIdx = i+1;
- InstrOffs = ARM_AM::getAM5Offset(MI.getOperand(ImmIdx).getImm());
- if (ARM_AM::getAM5Op(MI.getOperand(ImmIdx).getImm()) == ARM_AM::sub)
- InstrOffs *= -1;
- NumBits = 8;
- Scale = 4;
- break;
- }
- case ARMII::AddrModeT2_i12: {
- ImmIdx = i+2;
- InstrOffs = MI.getOperand(ImmIdx).getImm();
- NumBits = 12;
- break;
+ 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();
+ // When referencing from the frame pointer, stack pointer adjustments
+ // don't matter.
+ SPAdj = 0;
}
- case ARMII::AddrModeT2_so:
- case ARMII::AddrModeT2_i8: {
- ImmIdx = i+2;
- InstrOffs = MI.getOperand(ImmIdx).getImm();
- NumBits = 8;
- break;
- }
- default:
- LLVM_UNREACHABLE("Unsupported addressing mode!");
- break;
- }
-
- Offset += InstrOffs * Scale;
- assert((Offset & (Scale-1)) == 0 && "Can't encode this offset!");
- if (Offset < 0) {
- Offset = -Offset;
- isSub = true;
- }
-
- // Common case: small offset, fits into instruction.
- MachineOperand &ImmOp = MI.getOperand(ImmIdx);
- int ImmedOffset = Offset / Scale;
- unsigned Mask = (1 << NumBits) - 1;
- if ((unsigned)Offset <= Mask * Scale) {
- // Replace the FrameIndex with sp
- MI.getOperand(i).ChangeToRegister(FrameReg, false);
- if (isSub)
- ImmedOffset |= 1 << NumBits;
- ImmOp.ChangeToImmediate(ImmedOffset);
- return;
+ } else if (hasFP(MF) && AFI->hasStackFrame()) {
+ assert(SPAdj == 0 && "Unexpected stack offset!");
+ 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;
+ }
}
+ }
- // Otherwise, it didn't fit. Pull in what we can to simplify the immed.
- ImmedOffset = ImmedOffset & Mask;
- if (isSub)
- ImmedOffset |= 1 << NumBits;
- ImmOp.ChangeToImmediate(ImmedOffset);
- Offset &= ~(Mask*Scale);
+ // Modify MI as necessary to handle as much of 'Offset' as possible
+ bool Done = false;
+ if (!AFI->isThumbFunction())
+ Done = rewriteARMFrameIndex(MI, i, FrameReg, Offset, TII);
+ else {
+ assert(AFI->isThumb2Function());
+ Done = rewriteT2FrameIndex(MI, i, FrameReg, Offset, TII);
}
+ if (Done)
+ return 0;
// 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
// SP+LargeImm.
- assert(Offset && "This code isn't needed if offset already handled!");
-
- // Insert a set of r12 with the full address: r12 = sp + offset
- // If the offset we have is too large to fit into the instruction, we need
- // to form it with a series of ADDri's. Do this by taking 8-bit chunks
- // out of 'Offset'.
- unsigned ScratchReg = findScratchRegister(RS, &ARM::GPRRegClass, AFI);
- if (ScratchReg == 0)
- // No register is "free". Scavenge a register.
- ScratchReg = RS->scavengeRegister(&ARM::GPRRegClass, II, SPAdj);
+ assert((Offset ||
+ (MI.getDesc().TSFlags & ARMII::AddrModeMask) == ARMII::AddrMode4 ||
+ (MI.getDesc().TSFlags & ARMII::AddrModeMask) == ARMII::AddrMode6) &&
+ "This code isn't needed if offset already handled!");
+
+ unsigned ScratchReg = 0;
int PIdx = MI.findFirstPredOperandIdx();
ARMCC::CondCodes Pred = (PIdx == -1)
? ARMCC::AL : (ARMCC::CondCodes)MI.getOperand(PIdx).getImm();
unsigned PredReg = (PIdx == -1) ? 0 : MI.getOperand(PIdx+1).getReg();
- emitARMRegPlusImmediate(MBB, II, ScratchReg, FrameReg,
- isSub ? -Offset : Offset, Pred, PredReg, TII, dl);
- MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true);
+ if (Offset == 0)
+ // Must be addrmode4/6.
+ MI.getOperand(i).ChangeToRegister(FrameReg, false, false, false);
+ else {
+ ScratchReg = MF.getRegInfo().createVirtualRegister(ARM::GPRRegisterClass);
+ if (Value) *Value = Offset;
+ if (!AFI->isThumbFunction())
+ emitARMRegPlusImmediate(MBB, II, MI.getDebugLoc(), ScratchReg, FrameReg,
+ Offset, Pred, PredReg, TII);
+ else {
+ assert(AFI->isThumb2Function());
+ emitT2RegPlusImmediate(MBB, II, MI.getDebugLoc(), ScratchReg, FrameReg,
+ Offset, Pred, PredReg, TII);
+ }
+ MI.getOperand(i).ChangeToRegister(ScratchReg, false, false, true);
+ if (!ReuseFrameIndexVals)
+ ScratchReg = 0;
+ }
+ return ScratchReg;
}
-/// Move iterator pass the next bunch of callee save load / store ops for
+/// 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 Opc, unsigned Area,
+ int Opc1, int Opc2, unsigned Area,
const ARMSubtarget &STI) {
while (MBBI != MBB.end() &&
- MBBI->getOpcode() == Opc && MBBI->getOperand(1).isFI()) {
+ ((MBBI->getOpcode() == Opc1) || (MBBI->getOpcode() == Opc2)) &&
+ MBBI->getOperand(1).isFI()) {
if (Area != 0) {
bool Done = false;
unsigned Category = 0;
MachineBasicBlock::iterator MBBI = MBB.begin();
MachineFrameInfo *MFI = MF.getFrameInfo();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ assert(!AFI->isThumb1OnlyFunction() &&
+ "This emitPrologue does not suppor Thumb1!");
+ bool isARM = !AFI->isThumbFunction();
unsigned VARegSaveSize = AFI->getVarArgsRegSaveSize();
unsigned NumBytes = MFI->getStackSize();
const std::vector<CalleeSavedInfo> &CSI = MFI->getCalleeSavedInfo();
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(MBB, MBBI, TII, dl, -VARegSaveSize);
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, -VARegSaveSize);
if (!AFI->hasStackFrame()) {
if (NumBytes != 0)
- emitSPUpdate(MBB, MBBI, TII, dl, -NumBytes);
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, -NumBytes);
return;
}
}
// Build the new SUBri to adjust SP for integer callee-save spill area 1.
- emitSPUpdate(MBB, MBBI, TII, dl, -GPRCS1Size);
- movePastCSLoadStoreOps(MBB, MBBI, getOpcode(ARMII::STR), 1, STI);
-
- // Darwin ABI requires FP to point to the stack slot that contains the
- // previous FP.
+ 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(getOpcode(ARMII::ADDri)), FramePtr)
+ 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(MBB, MBBI, TII, dl, -GPRCS2Size);
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, -GPRCS2Size);
// Build the new SUBri to adjust SP for FP callee-save spill area.
- movePastCSLoadStoreOps(MBB, MBBI, getOpcode(ARMII::STR), 2, STI);
- emitSPUpdate(MBB, MBBI, TII, dl, -DPRCSSize);
+ 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);
AFI->setGPRCalleeSavedArea2Offset(GPRCS2Offset);
AFI->setDPRCalleeSavedAreaOffset(DPRCSOffset);
+ movePastCSLoadStoreOps(MBB, MBBI, ARM::VSTRD, 0, 3, STI);
NumBytes = DPRCSOffset;
if (NumBytes) {
- // Insert it after all the callee-save spills.
- movePastCSLoadStoreOps(MBB, MBBI, getOpcode(ARMII::FSTD), 3, STI);
- emitSPUpdate(MBB, MBBI, TII, dl, -NumBytes);
+ // Adjust SP after all the callee-save spills.
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, -NumBytes);
}
if (STI.isTargetELF() && hasFP(MF)) {
AFI->setGPRCalleeSavedArea1Size(GPRCS1Size);
AFI->setGPRCalleeSavedArea2Size(GPRCS2Size);
AFI->setDPRCalleeSavedAreaSize(DPRCSSize);
+
+ // If we need dynamic stack realignment, do it here.
+ if (needsStackRealignment(MF)) {
+ unsigned Opc;
+ unsigned MaxAlign = MFI->getMaxAlignment();
+ assert (!AFI->isThumb1OnlyFunction());
+ Opc = AFI->isThumbFunction() ? ARM::t2BICri : ARM::BICri;
+
+ AddDefaultCC(AddDefaultPred(BuildMI(MBB, MBBI, dl, TII.get(Opc), ARM::SP)
+ .addReg(ARM::SP, RegState::Kill)
+ .addImm(MaxAlign-1)));
+ }
}
static bool isCalleeSavedRegister(unsigned Reg, const unsigned *CSRegs) {
}
static bool isCSRestore(MachineInstr *MI,
- const ARMBaseInstrInfo &TII,
+ const ARMBaseInstrInfo &TII,
const unsigned *CSRegs) {
- return ((MI->getOpcode() == (int)TII.getOpcode(ARMII::FLDD) ||
- MI->getOpcode() == (int)TII.getOpcode(ARMII::LDR)) &&
+ 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 {
+emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const {
MachineBasicBlock::iterator MBBI = prior(MBB.end());
- assert(MBBI->getOpcode() == (int)getOpcode(ARMII::BX_RET) &&
+ assert(MBBI->getDesc().isReturn() &&
"Can only insert epilog into returning blocks");
DebugLoc dl = MBBI->getDebugLoc();
MachineFrameInfo *MFI = MF.getFrameInfo();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ assert(!AFI->isThumb1OnlyFunction() &&
+ "This emitEpilogue does not suppor Thumb1!");
+ bool isARM = !AFI->isThumbFunction();
+
unsigned VARegSaveSize = AFI->getVarArgsRegSaveSize();
int NumBytes = (int)MFI->getStackSize();
if (!AFI->hasStackFrame()) {
if (NumBytes != 0)
- emitSPUpdate(MBB, MBBI, TII, dl, NumBytes);
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, NumBytes);
} else {
- // Unwind MBBI to point to first LDR / FLDD.
+ // Unwind MBBI to point to first LDR / VLDRD.
const unsigned *CSRegs = getCalleeSavedRegs();
if (MBBI != MBB.begin()) {
do
// Darwin ABI requires FP to point to the stack slot that contains the
// previous FP.
- if ((STI.isTargetDarwin() && NumBytes) || hasFP(MF)) {
+ 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 (AFI->getGPRCalleeSavedArea2Size() ||
+ if (HasFP ||
+ AFI->getGPRCalleeSavedArea2Size() ||
AFI->getDPRCalleeSavedAreaSize() ||
- AFI->getDPRCalleeSavedAreaOffset()||
- hasFP(MF)) {
- if (NumBytes)
- BuildMI(MBB, MBBI, dl, TII.get(getOpcode(ARMII::SUBri)), ARM::SP).addReg(FramePtr)
- .addImm(NumBytes)
- .addImm((unsigned)ARMCC::AL).addReg(0).addReg(0);
- else
- BuildMI(MBB, MBBI, dl, TII.get(getOpcode(ARMII::MOVr)), ARM::SP).addReg(FramePtr)
- .addImm((unsigned)ARMCC::AL).addReg(0).addReg(0);
+ 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(MBB, MBBI, TII, dl, NumBytes);
- }
+ } else if (NumBytes)
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, NumBytes);
// Move SP to start of integer callee save spill area 2.
- movePastCSLoadStoreOps(MBB, MBBI, getOpcode(ARMII::FLDD), 3, STI);
- emitSPUpdate(MBB, MBBI, TII, dl, AFI->getDPRCalleeSavedAreaSize());
+ 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, getOpcode(ARMII::LDR), 2, STI);
- emitSPUpdate(MBB, MBBI, TII, dl, AFI->getGPRCalleeSavedArea2Size());
+ 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, getOpcode(ARMII::LDR), 1, STI);
- emitSPUpdate(MBB, MBBI, TII, dl, AFI->getGPRCalleeSavedArea1Size());
+ movePastCSLoadStoreOps(MBB, MBBI, ARM::LDR, ARM::t2LDRi12, 1, STI);
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, AFI->getGPRCalleeSavedArea1Size());
}
if (VARegSaveSize)
- emitSPUpdate(MBB, MBBI, TII, dl, VARegSaveSize);
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, VARegSaveSize);
+}
+
+namespace {
+ struct MaximalStackAlignmentCalculator : public MachineFunctionPass {
+ static char ID;
+ MaximalStackAlignmentCalculator() : MachineFunctionPass(&ID) {}
+
+ virtual bool runOnMachineFunction(MachineFunction &MF) {
+ MachineFrameInfo *FFI = MF.getFrameInfo();
+ MachineRegisterInfo &RI = MF.getRegInfo();
+
+ // Calculate max stack alignment of all already allocated stack objects.
+ unsigned MaxAlign = calculateMaxStackAlignment(FFI);
+
+ // Be over-conservative: scan over all vreg defs and find, whether vector
+ // registers are used. If yes - there is probability, that vector register
+ // will be spilled and thus stack needs to be aligned properly.
+ for (unsigned RegNum = TargetRegisterInfo::FirstVirtualRegister;
+ RegNum < RI.getLastVirtReg(); ++RegNum)
+ MaxAlign = std::max(MaxAlign, RI.getRegClass(RegNum)->getAlignment());
+
+ if (FFI->getMaxAlignment() == MaxAlign)
+ return false;
+
+ FFI->setMaxAlignment(MaxAlign);
+ return true;
+ }
+
+ virtual const char *getPassName() const {
+ return "ARM Stack Required Alignment Auto-Detector";
+ }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.setPreservesCFG();
+ MachineFunctionPass::getAnalysisUsage(AU);
+ }
+ };
+
+ char MaximalStackAlignmentCalculator::ID = 0;
+}
+FunctionPass*
+llvm::createARMMaxStackAlignmentCalculatorPass() {
+ return new MaximalStackAlignmentCalculator();
}
#include "ARMGenRegisterInfo.inc"
projects / oota-llvm.git / blobdiff