-//===- ARMBaseRegisterInfo.cpp - ARM Register Information -----------*- C++ -*-===//
+//===- ARMBaseRegisterInfo.cpp - ARM Register Information -------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
#include "ARM.h"
#include "ARMAddressingModes.h"
+#include "ARMBaseInstrInfo.h"
#include "ARMBaseRegisterInfo.h"
#include "ARMInstrInfo.h"
#include "ARMMachineFunctionInfo.h"
#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/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetFrameInfo.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/ADT/SmallVector.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:
- assert(0 && "Unknown ARM register!");
- abort();
- }
-}
-
unsigned ARMBaseRegisterInfo::getRegisterNumbering(unsigned RegEnum,
- bool &isSPVFP) {
- isSPVFP = false;
+ bool *isSPVFP) {
+ if (isSPVFP)
+ *isSPVFP = false;
using namespace ARM;
switch (RegEnum) {
default:
- assert(0 && "Unknown ARM register!");
- abort();
- 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;
}
}
-ARMBaseRegisterInfo::ARMBaseRegisterInfo(const TargetInstrInfo &tii,
+ARMBaseRegisterInfo::ARMBaseRegisterInfo(const ARMBaseInstrInfo &tii,
const ARMSubtarget &sti)
: ARMGenRegisterInfo(ARM::ADJCALLSTACKDOWN, ARM::ADJCALLSTACKUP),
TII(tii), STI(sti),
0
};
- if (STI.isThumb()) {
+ if (STI.isThumb1Only()) {
return STI.isTargetDarwin()
? DarwinThumbCalleeSavedRegClasses : ThumbCalleeSavedRegClasses;
}
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 {
+const TargetRegisterClass *
+ARMBaseRegisterInfo::getPointerRegClass(unsigned Kind) const {
return &ARM::GPRRegClass;
}
MFI->isFrameAddressTaken());
}
+bool ARMBaseRegisterInfo::hasStackFrame(const MachineFunction &MF) const {
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
+ if (NoFramePointerElim && MFI->hasCalls())
+ return true;
+ return MFI->hasVarSizedObjects() || MFI->isFrameAddressTaken();
+}
+
+/// 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 {
}
bool ForceLRSpill = false;
- if (!LRSpilled && AFI->isThumbFunction()) {
+ 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
}
bool ExtraCSSpill = false;
- if (!CanEliminateFrame || hasFP(MF)) {
+ if (!CanEliminateFrame || hasStackFrame(MF)) {
AFI->setHasStackFrame(true);
// If LR is not spilled, but at least one of R4, R5, R6, and R7 is spilled.
if (CS1Spilled && !UnspilledCS1GPRs.empty()) {
for (unsigned i = 0, e = UnspilledCS1GPRs.size(); i != e; ++i) {
unsigned Reg = UnspilledCS1GPRs[i];
- // Don't spiil high register if the function is thumb
- if (!AFI->isThumbFunction() ||
+ // 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);
}
}
} else if (!UnspilledCS2GPRs.empty() &&
- !AFI->isThumbFunction()) {
+ !AFI->isThumb1OnlyFunction()) {
unsigned Reg = UnspilledCS2GPRs.front();
MF.getRegInfo().setPhysRegUsed(Reg);
AFI->setCSRegisterIsSpilled(Reg);
// 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->isThumbFunction()) {
+ if (RS && !ExtraCSSpill && !AFI->isThumb1OnlyFunction()) {
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 (estimateStackSize(MF, MFI) >= 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;
}
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;
}
return ARM::R0;
case ARM::R3:
// FIXME!
- return STI.isThumb() ? 0 : ARM::R2;
+ return STI.isThumb1Only() ? 0 : 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;
return ARM::R1;
case ARM::R2:
// FIXME!
- return STI.isThumb() ? 0 : ARM::R3;
+ return STI.isThumb1Only() ? 0 : 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;
}
+/// emitLoadConstPool - Emits a load from constpool to materialize the
+/// specified immediate.
+void ARMBaseRegisterInfo::
+emitLoadConstPool(MachineBasicBlock &MBB,
+ MachineBasicBlock::iterator &MBBI,
+ DebugLoc dl,
+ 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::getInt32Ty(MF.getFunction()->getContext()), Val);
+ unsigned Idx = ConstantPool->getConstantPoolIndex(C, 4);
+
+ BuildMI(MBB, MBBI, dl, TII.get(ARM::LDRcp))
+ .addReg(DestReg, getDefRegState(true), SubIdx)
+ .addConstantPoolIndex(Idx)
+ .addReg(0).addImm(0).addImm(Pred).addReg(PredReg);
+}
+
+bool ARMBaseRegisterInfo::
+requiresRegisterScavenging(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();
+}
+
+static void
+emitSPUpdate(bool isARM,
+ MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
+ DebugLoc dl, const ARMBaseInstrInfo &TII,
+ int NumBytes,
+ ARMCC::CondCodes Pred = ARMCC::AL, unsigned PredReg = 0) {
+ 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 {
+ if (!hasReservedCallFrame(MF)) {
+ // If we have alloca, convert as follows:
+ // ADJCALLSTACKDOWN -> sub, sp, sp, amount
+ // ADJCALLSTACKUP -> add, sp, sp, amount
+ MachineInstr *Old = I;
+ DebugLoc dl = Old->getDebugLoc();
+ unsigned Amount = Old->getOperand(0).getImm();
+ if (Amount != 0) {
+ // We need to keep the stack aligned properly. To do this, we round the
+ // amount of space needed for the outgoing arguments up to the next
+ // alignment boundary.
+ 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(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(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 i = 0;
+ MachineInstr &MI = *II;
+ MachineBasicBlock &MBB = *MI.getParent();
+ MachineFunction &MF = *MBB.getParent();
+ ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+ assert(!AFI->isThumb1OnlyFunction() &&
+ "This eliminateFrameIndex does not suppor Thumb1!");
+
+ while (!MI.getOperand(i).isFI()) {
+ ++i;
+ assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
+ }
+
+ unsigned FrameReg = ARM::SP;
+ int FrameIndex = MI.getOperand(i).getIndex();
+ int Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex) +
+ MF.getFrameInfo()->getStackSize() + SPAdj;
+
+ 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();
+ }
+
+ // modify MI as necessary to handle as much of 'Offset' as possible
+ if (!AFI->isThumbFunction())
+ Offset = rewriteARMFrameIndex(MI, i, FrameReg, Offset, TII);
+ else {
+ assert(AFI->isThumb2Function());
+ Offset = rewriteT2FrameIndex(MI, i, FrameReg, Offset, TII);
+ }
+ if (Offset == 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
+ // 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);
+ 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();
+ 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);
+}
+
+/// Move iterator pass 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<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();
+ DebugLoc dl = (MBBI != MBB.end() ?
+ MBBI->getDebugLoc() : DebugLoc::getUnknownLoc());
+
+ // 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;
+
+ 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);
+
+ // Darwin ABI requires FP to point to the stack slot that contains the
+ // previous FP.
+ 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;
+ AFI->setFramePtrSpillOffset(MFI->getObjectOffset(FramePtrSpillFI) + NumBytes);
+ AFI->setGPRCalleeSavedArea1Offset(GPRCS1Offset);
+ AFI->setGPRCalleeSavedArea2Offset(GPRCS2Offset);
+ AFI->setDPRCalleeSavedAreaOffset(DPRCSOffset);
+
+ NumBytes = DPRCSOffset;
+ if (NumBytes) {
+ // Insert it after all the callee-save spills.
+ movePastCSLoadStoreOps(MBB, MBBI, ARM::FSTD, 0, 3, STI);
+ 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);
+}
+
+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::FLDD ||
+ 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");
+ 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(isARM, MBB, MBBI, dl, TII, NumBytes);
+ } else {
+ // Unwind MBBI to point to first LDR / FLDD.
+ 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.
+ if ((STI.isTargetDarwin() && NumBytes) || hasFP(MF)) {
+ 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() ||
+ AFI->getDPRCalleeSavedAreaSize() ||
+ AFI->getDPRCalleeSavedAreaOffset()||
+ hasFP(MF)) {
+ 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::FLDD, 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 (VARegSaveSize)
+ emitSPUpdate(isARM, MBB, MBBI, dl, TII, VARegSaveSize);
+}
+
#include "ARMGenRegisterInfo.inc"
projects / oota-llvm.git / blobdiff