#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
+#define DEBUG_TYPE "arm-register-info"
+
#define GET_REGINFO_TARGET_DESC
#include "ARMGenRegisterInfo.inc"
using namespace llvm;
-ARMBaseRegisterInfo::ARMBaseRegisterInfo(const ARMSubtarget &sti)
- : ARMGenRegisterInfo(ARM::LR, 0, 0, ARM::PC), STI(sti), BasePtr(ARM::R6) {
+ARMBaseRegisterInfo::ARMBaseRegisterInfo()
+ : ARMGenRegisterInfo(ARM::LR, 0, 0, ARM::PC), BasePtr(ARM::R6) {}
+
+static unsigned getFramePointerReg(const ARMSubtarget &STI) {
if (STI.isTargetMachO()) {
if (STI.isTargetDarwin() || STI.isThumb1Only())
- FramePtr = ARM::R7;
+ return ARM::R7;
else
- FramePtr = ARM::R11;
+ return ARM::R11;
} else if (STI.isTargetWindows())
- FramePtr = ARM::R11;
+ return ARM::R11;
else // ARM EABI
- FramePtr = STI.isThumb() ? ARM::R7 : ARM::R11;
+ return STI.isThumb() ? ARM::R7 : ARM::R11;
}
const MCPhysReg*
ARMBaseRegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
- const MCPhysReg *RegList = (STI.isTargetIOS() && !STI.isAAPCS_ABI())
- ? CSR_iOS_SaveList
- : CSR_AAPCS_SaveList;
-
- if (!MF) return RegList;
+ const ARMSubtarget &STI = MF->getSubtarget<ARMSubtarget>();
+ const MCPhysReg *RegList =
+ STI.isTargetDarwin() ? CSR_iOS_SaveList : CSR_AAPCS_SaveList;
const Function *F = MF->getFunction();
if (F->getCallingConv() == CallingConv::GHC) {
return RegList;
}
-const uint32_t*
-ARMBaseRegisterInfo::getCallPreservedMask(CallingConv::ID CC) const {
+const uint32_t *
+ARMBaseRegisterInfo::getCallPreservedMask(const MachineFunction &MF,
+ CallingConv::ID CC) const {
+ const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
if (CC == CallingConv::GHC)
// This is academic becase all GHC calls are (supposed to be) tail calls
return CSR_NoRegs_RegMask;
- return (STI.isTargetIOS() && !STI.isAAPCS_ABI())
- ? CSR_iOS_RegMask : CSR_AAPCS_RegMask;
+ return STI.isTargetDarwin() ? CSR_iOS_RegMask : CSR_AAPCS_RegMask;
}
const uint32_t*
return CSR_NoRegs_RegMask;
}
-const uint32_t*
-ARMBaseRegisterInfo::getThisReturnPreservedMask(CallingConv::ID CC) const {
+const uint32_t *
+ARMBaseRegisterInfo::getThisReturnPreservedMask(const MachineFunction &MF,
+ CallingConv::ID CC) const {
+ const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
// This should return a register mask that is the same as that returned by
// getCallPreservedMask but that additionally preserves the register used for
// the first i32 argument (which must also be the register used to return a
if (CC == CallingConv::GHC)
// This is academic becase all GHC calls are (supposed to be) tail calls
return nullptr;
- return (STI.isTargetIOS() && !STI.isAAPCS_ABI())
- ? CSR_iOS_ThisReturn_RegMask : CSR_AAPCS_ThisReturn_RegMask;
+ return STI.isTargetDarwin() ? CSR_iOS_ThisReturn_RegMask
+ : CSR_AAPCS_ThisReturn_RegMask;
}
BitVector ARMBaseRegisterInfo::
getReservedRegs(const MachineFunction &MF) const {
- const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+ const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
+ const ARMFrameLowering *TFI = getFrameLowering(MF);
// FIXME: avoid re-calculating this every time.
BitVector Reserved(getNumRegs());
Reserved.set(ARM::FPSCR);
Reserved.set(ARM::APSR_NZCV);
if (TFI->hasFP(MF))
- Reserved.set(FramePtr);
+ Reserved.set(getFramePointerReg(STI));
if (hasBasePointer(MF))
Reserved.set(BasePtr);
// Some targets reserve R9.
return Reserved;
}
-const TargetRegisterClass*
-ARMBaseRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC)
- const {
+const TargetRegisterClass *
+ARMBaseRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC,
+ const MachineFunction &) const {
const TargetRegisterClass *Super = RC;
TargetRegisterClass::sc_iterator I = RC->getSuperClasses();
do {
const TargetRegisterClass *
ARMBaseRegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const {
if (RC == &ARM::CCRRegClass)
- return nullptr; // Can't copy CCR registers.
+ return &ARM::rGPRRegClass; // Can't copy CCR registers.
return RC;
}
unsigned
ARMBaseRegisterInfo::getRegPressureLimit(const TargetRegisterClass *RC,
MachineFunction &MF) const {
- const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+ const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
+ const ARMFrameLowering *TFI = getFrameLowering(MF);
switch (RC->getID()) {
default:
ArrayRef<MCPhysReg> Order,
SmallVectorImpl<MCPhysReg> &Hints,
const MachineFunction &MF,
- const VirtRegMap *VRM) const {
+ const VirtRegMap *VRM,
+ const LiveRegMatrix *Matrix) const {
const MachineRegisterInfo &MRI = MF.getRegInfo();
std::pair<unsigned, unsigned> Hint = MRI.getRegAllocationHint(VirtReg);
// This register should preferably be even (Odd == 0) or odd (Odd == 1).
// Check if the other part of the pair has already been assigned, and provide
// the paired register as the first hint.
+ unsigned Paired = Hint.second;
+ if (Paired == 0)
+ return;
+
unsigned PairedPhys = 0;
- if (VRM && VRM->hasPhys(Hint.second)) {
- PairedPhys = getPairedGPR(VRM->getPhys(Hint.second), Odd, this);
- if (PairedPhys && MRI.isReserved(PairedPhys))
- PairedPhys = 0;
+ if (TargetRegisterInfo::isPhysicalRegister(Paired)) {
+ PairedPhys = Paired;
+ } else if (VRM && VRM->hasPhys(Paired)) {
+ PairedPhys = getPairedGPR(VRM->getPhys(Paired), Odd, this);
}
// First prefer the paired physreg.
}
void
-ARMBaseRegisterInfo::UpdateRegAllocHint(unsigned Reg, unsigned NewReg,
+ARMBaseRegisterInfo::updateRegAllocHint(unsigned Reg, unsigned NewReg,
MachineFunction &MF) const {
MachineRegisterInfo *MRI = &MF.getRegInfo();
std::pair<unsigned, unsigned> Hint = MRI->getRegAllocationHint(Reg);
// change.
unsigned OtherReg = Hint.second;
Hint = MRI->getRegAllocationHint(OtherReg);
- if (Hint.second == Reg)
- // Make sure the pair has not already divorced.
+ // Make sure the pair has not already divorced.
+ if (Hint.second == Reg) {
MRI->setRegAllocationHint(OtherReg, Hint.first, NewReg);
- }
-}
-
-bool
-ARMBaseRegisterInfo::avoidWriteAfterWrite(const TargetRegisterClass *RC) const {
- // CortexA9 has a Write-after-write hazard for NEON registers.
- if (!STI.isLikeA9())
- return false;
-
- switch (RC->getID()) {
- case ARM::DPRRegClassID:
- case ARM::DPR_8RegClassID:
- case ARM::DPR_VFP2RegClassID:
- case ARM::QPRRegClassID:
- case ARM::QPR_8RegClassID:
- case ARM::QPR_VFP2RegClassID:
- case ARM::SPRRegClassID:
- case ARM::SPR_8RegClassID:
- // Avoid reusing S, D, and Q registers.
- // Don't increase register pressure for QQ and QQQQ.
- return true;
- default:
- return false;
+ if (TargetRegisterInfo::isVirtualRegister(NewReg))
+ MRI->setRegAllocationHint(NewReg,
+ Hint.first == (unsigned)ARMRI::RegPairOdd ? ARMRI::RegPairEven
+ : ARMRI::RegPairOdd, OtherReg);
+ }
}
}
bool ARMBaseRegisterInfo::hasBasePointer(const MachineFunction &MF) const {
const MachineFrameInfo *MFI = MF.getFrameInfo();
const ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
- const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+ const ARMFrameLowering *TFI = getFrameLowering(MF);
// When outgoing call frames are so large that we adjust the stack pointer
// around the call, we can no longer use the stack pointer to reach the
bool ARMBaseRegisterInfo::canRealignStack(const MachineFunction &MF) const {
const MachineRegisterInfo *MRI = &MF.getRegInfo();
const ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ const ARMFrameLowering *TFI = getFrameLowering(MF);
// 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.
- if (MF.getFunction()->hasFnAttribute("no-realign-stack"))
+ if (!TargetRegisterInfo::canRealignStack(MF))
return false;
if (AFI->isThumb1OnlyFunction())
return false;
// Stack realignment requires a frame pointer. If we already started
// register allocation with frame pointer elimination, it is too late now.
- if (!MRI->canReserveReg(FramePtr))
+ if (!MRI->canReserveReg(getFramePointerReg(MF.getSubtarget<ARMSubtarget>())))
return false;
// We may also need a base pointer if there are dynamic allocas or stack
// pointer adjustments around calls.
- if (MF.getTarget().getFrameLowering()->hasReservedCallFrame(MF))
+ if (TFI->hasReservedCallFrame(MF))
return true;
// A base pointer is required and allowed. Check that it isn't too late to
// reserve it.
return MRI->canReserveReg(BasePtr);
}
-bool ARMBaseRegisterInfo::
-needsStackRealignment(const MachineFunction &MF) const {
- const MachineFrameInfo *MFI = MF.getFrameInfo();
- const Function *F = MF.getFunction();
- unsigned StackAlign = MF.getTarget().getFrameLowering()->getStackAlignment();
- bool requiresRealignment =
- ((MFI->getMaxAlignment() > StackAlign) ||
- F->getAttributes().hasAttribute(AttributeSet::FunctionIndex,
- Attribute::StackAlignment));
-
- return requiresRealignment && canRealignStack(MF);
-}
-
bool ARMBaseRegisterInfo::
cannotEliminateFrame(const MachineFunction &MF) const {
const MachineFrameInfo *MFI = MF.getFrameInfo();
unsigned
ARMBaseRegisterInfo::getFrameRegister(const MachineFunction &MF) const {
- const TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+ const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
+ const ARMFrameLowering *TFI = getFrameLowering(MF);
if (TFI->hasFP(MF))
- return FramePtr;
+ return getFramePointerReg(STI);
return ARM::SP;
}
ARMCC::CondCodes Pred,
unsigned PredReg, unsigned MIFlags) const {
MachineFunction &MF = *MBB.getParent();
- const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+ const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
MachineConstantPool *ConstantPool = MF.getConstantPool();
const Constant *C =
ConstantInt::get(Type::getInt32Ty(MF.getFunction()->getContext()), Val);
.setMIFlags(MIFlags);
}
-bool ARMBaseRegisterInfo::mayOverrideLocalAssignment() const {
- // The native linux build hits a downstream codegen bug when this is enabled.
- return STI.isTargetDarwin();
-}
-
bool ARMBaseRegisterInfo::
requiresRegisterScavenging(const MachineFunction &MF) const {
return true;
// 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 TargetFrameLowering *TFI = MF.getTarget().getFrameLowering();
+ const ARMFrameLowering *TFI = getFrameLowering(MF);
MachineFrameInfo *MFI = MF.getFrameInfo();
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
// 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.
+ // If there's a frame pointer and the addressing mode allows it, 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 = TFI->getStackAlignment();
- if (TFI->hasFP(MF) &&
+ if (TFI->hasFP(MF) &&
!((MFI->getLocalFrameMaxAlign() > StackAlign) && canRealignStack(MF))) {
- if (isFrameOffsetLegal(MI, FPOffset))
+ if (isFrameOffsetLegal(MI, getFrameRegister(MF), FPOffset))
return false;
}
// If we can reference via the stack pointer, try that.
// 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))
+ if (!MFI->hasVarSizedObjects() && isFrameOffsetLegal(MI, ARM::SP, Offset))
return false;
// The offset likely isn't legal, we want to allocate a virtual base register.
int64_t Offset) const {
ARMFunctionInfo *AFI = MBB->getParent()->getInfo<ARMFunctionInfo>();
unsigned ADDriOpc = !AFI->isThumbFunction() ? ARM::ADDri :
- (AFI->isThumb1OnlyFunction() ? ARM::tADDrSPi : ARM::t2ADDri);
+ (AFI->isThumb1OnlyFunction() ? ARM::tADDframe : ARM::t2ADDri);
MachineBasicBlock::iterator Ins = MBB->begin();
DebugLoc DL; // Defaults to "unknown"
const MachineFunction &MF = *MBB->getParent();
MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
- const TargetInstrInfo &TII = *MF.getTarget().getInstrInfo();
+ const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
const MCInstrDesc &MCID = TII.get(ADDriOpc);
MRI.constrainRegClass(BaseReg, TII.getRegClass(MCID, 0, this, MF));
- MachineInstrBuilder MIB = AddDefaultPred(BuildMI(*MBB, Ins, DL, MCID, BaseReg)
- .addFrameIndex(FrameIdx).addImm(Offset));
+ MachineInstrBuilder MIB = BuildMI(*MBB, Ins, DL, MCID, BaseReg)
+ .addFrameIndex(FrameIdx).addImm(Offset);
if (!AFI->isThumb1OnlyFunction())
- AddDefaultCC(MIB);
+ AddDefaultCC(AddDefaultPred(MIB));
}
void ARMBaseRegisterInfo::resolveFrameIndex(MachineInstr &MI, unsigned BaseReg,
MachineBasicBlock &MBB = *MI.getParent();
MachineFunction &MF = *MBB.getParent();
const ARMBaseInstrInfo &TII =
- *static_cast<const ARMBaseInstrInfo*>(MF.getTarget().getInstrInfo());
+ *static_cast<const ARMBaseInstrInfo *>(MF.getSubtarget().getInstrInfo());
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
int Off = Offset; // ARM doesn't need the general 64-bit offsets
unsigned i = 0;
(void)Done;
}
-bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI,
+bool ARMBaseRegisterInfo::isFrameOffsetLegal(const MachineInstr *MI, unsigned BaseReg,
int64_t Offset) const {
const MCInstrDesc &Desc = MI->getDesc();
unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask);
NumBits = 8;
break;
case ARMII::AddrModeT1_s:
- NumBits = 5;
+ NumBits = (BaseReg == ARM::SP ? 8 : 5);
Scale = 4;
isSigned = false;
break;
MachineBasicBlock &MBB = *MI.getParent();
MachineFunction &MF = *MBB.getParent();
const ARMBaseInstrInfo &TII =
- *static_cast<const ARMBaseInstrInfo*>(MF.getTarget().getInstrInfo());
- const ARMFrameLowering *TFI =
- static_cast<const ARMFrameLowering*>(MF.getTarget().getFrameLowering());
+ *static_cast<const ARMBaseInstrInfo *>(MF.getSubtarget().getInstrInfo());
+ const ARMFrameLowering *TFI = getFrameLowering(MF);
ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
assert(!AFI->isThumb1OnlyFunction() &&
"This eliminateFrameIndex does not support Thumb1!");
MI.getOperand(FIOperandNum).ChangeToRegister(ScratchReg, false, false,true);
}
}
+
+bool ARMBaseRegisterInfo::shouldCoalesce(MachineInstr *MI,
+ const TargetRegisterClass *SrcRC,
+ unsigned SubReg,
+ const TargetRegisterClass *DstRC,
+ unsigned DstSubReg,
+ const TargetRegisterClass *NewRC) const {
+ auto MBB = MI->getParent();
+ auto MF = MBB->getParent();
+ const MachineRegisterInfo &MRI = MF->getRegInfo();
+ // If not copying into a sub-register this should be ok because we shouldn't
+ // need to split the reg.
+ if (!DstSubReg)
+ return true;
+ // Small registers don't frequently cause a problem, so we can coalesce them.
+ if (NewRC->getSize() < 32 && DstRC->getSize() < 32 && SrcRC->getSize() < 32)
+ return true;
+
+ auto NewRCWeight =
+ MRI.getTargetRegisterInfo()->getRegClassWeight(NewRC);
+ auto SrcRCWeight =
+ MRI.getTargetRegisterInfo()->getRegClassWeight(SrcRC);
+ auto DstRCWeight =
+ MRI.getTargetRegisterInfo()->getRegClassWeight(DstRC);
+ // If the source register class is more expensive than the destination, the
+ // coalescing is probably profitable.
+ if (SrcRCWeight.RegWeight > NewRCWeight.RegWeight)
+ return true;
+ if (DstRCWeight.RegWeight > NewRCWeight.RegWeight)
+ return true;
+
+ // If the register allocator isn't constrained, we can always allow coalescing
+ // unfortunately we don't know yet if we will be constrained.
+ // The goal of this heuristic is to restrict how many expensive registers
+ // we allow to coalesce in a given basic block.
+ auto AFI = MF->getInfo<ARMFunctionInfo>();
+ auto It = AFI->getCoalescedWeight(MBB);
+
+ DEBUG(dbgs() << "\tARM::shouldCoalesce - Coalesced Weight: "
+ << It->second << "\n");
+ DEBUG(dbgs() << "\tARM::shouldCoalesce - Reg Weight: "
+ << NewRCWeight.RegWeight << "\n");
+
+ // This number is the largest round number that which meets the criteria:
+ // (1) addresses PR18825
+ // (2) generates better code in some test cases (like vldm-shed-a9.ll)
+ // (3) Doesn't regress any test cases (in-tree, test-suite, and SPEC)
+ // In practice the SizeMultiplier will only factor in for straight line code
+ // that uses a lot of NEON vectors, which isn't terribly common.
+ unsigned SizeMultiplier = MBB->size()/100;
+ SizeMultiplier = SizeMultiplier ? SizeMultiplier : 1;
+ if (It->second < NewRCWeight.WeightLimit * SizeMultiplier) {
+ It->second += NewRCWeight.RegWeight;
+ return true;
+ }
+ return false;
+}