llvm.org GIT mirror llvm / release_60 lib / Target / ARM / ThumbRegisterInfo.cpp
release_60

Tree @release_60 (Download .tar.gz)

ThumbRegisterInfo.cpp @release_60raw · history · blame

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
//===-- ThumbRegisterInfo.cpp - Thumb-1 Register Information -------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file contains the Thumb-1 implementation of the TargetRegisterInfo
// class.
//
//===----------------------------------------------------------------------===//

#include "ThumbRegisterInfo.h"
#include "ARMBaseInstrInfo.h"
#include "ARMMachineFunctionInfo.h"
#include "ARMSubtarget.h"
#include "MCTargetDesc/ARMAddressingModes.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/RegisterScavenging.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/CodeGen/TargetFrameLowering.h"
#include "llvm/Target/TargetMachine.h"

namespace llvm {
extern cl::opt<bool> ReuseFrameIndexVals;
}

using namespace llvm;

ThumbRegisterInfo::ThumbRegisterInfo() : ARMBaseRegisterInfo() {}

const TargetRegisterClass *
ThumbRegisterInfo::getLargestLegalSuperClass(const TargetRegisterClass *RC,
                                              const MachineFunction &MF) const {
  if (!MF.getSubtarget<ARMSubtarget>().isThumb1Only())
    return ARMBaseRegisterInfo::getLargestLegalSuperClass(RC, MF);

  if (ARM::tGPRRegClass.hasSubClassEq(RC))
    return &ARM::tGPRRegClass;
  return ARMBaseRegisterInfo::getLargestLegalSuperClass(RC, MF);
}

const TargetRegisterClass *
ThumbRegisterInfo::getPointerRegClass(const MachineFunction &MF,
                                      unsigned Kind) const {
  if (!MF.getSubtarget<ARMSubtarget>().isThumb1Only())
    return ARMBaseRegisterInfo::getPointerRegClass(MF, Kind);
  return &ARM::tGPRRegClass;
}

static void emitThumb1LoadConstPool(MachineBasicBlock &MBB,
                                    MachineBasicBlock::iterator &MBBI,
                                    const DebugLoc &dl, unsigned DestReg,
                                    unsigned SubIdx, int Val,
                                    ARMCC::CondCodes Pred, unsigned PredReg,
                                    unsigned MIFlags) {
  MachineFunction &MF = *MBB.getParent();
  const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
  const TargetInstrInfo &TII = *STI.getInstrInfo();
  MachineConstantPool *ConstantPool = MF.getConstantPool();
  const Constant *C = ConstantInt::get(
          Type::getInt32Ty(MBB.getParent()->getFunction().getContext()), Val);
  unsigned Idx = ConstantPool->getConstantPoolIndex(C, 4);

  BuildMI(MBB, MBBI, dl, TII.get(ARM::tLDRpci))
    .addReg(DestReg, getDefRegState(true), SubIdx)
    .addConstantPoolIndex(Idx).addImm(Pred).addReg(PredReg)
    .setMIFlags(MIFlags);
}

static void emitThumb2LoadConstPool(MachineBasicBlock &MBB,
                                    MachineBasicBlock::iterator &MBBI,
                                    const DebugLoc &dl, unsigned DestReg,
                                    unsigned SubIdx, int Val,
                                    ARMCC::CondCodes Pred, unsigned PredReg,
                                    unsigned MIFlags) {
  MachineFunction &MF = *MBB.getParent();
  const TargetInstrInfo &TII = *MF.getSubtarget().getInstrInfo();
  MachineConstantPool *ConstantPool = MF.getConstantPool();
  const Constant *C = ConstantInt::get(
           Type::getInt32Ty(MBB.getParent()->getFunction().getContext()), Val);
  unsigned Idx = ConstantPool->getConstantPoolIndex(C, 4);

  BuildMI(MBB, MBBI, dl, TII.get(ARM::t2LDRpci))
      .addReg(DestReg, getDefRegState(true), SubIdx)
      .addConstantPoolIndex(Idx)
      .add(predOps(ARMCC::AL))
      .setMIFlags(MIFlags);
}

/// emitLoadConstPool - Emits a load from constpool to materialize the
/// specified immediate.
void ThumbRegisterInfo::emitLoadConstPool(
    MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
    const DebugLoc &dl, unsigned DestReg, unsigned SubIdx, int Val,
    ARMCC::CondCodes Pred, unsigned PredReg, unsigned MIFlags) const {
  MachineFunction &MF = *MBB.getParent();
  const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
  if (STI.isThumb1Only()) {
    assert((isARMLowRegister(DestReg) || isVirtualRegister(DestReg)) &&
           "Thumb1 does not have ldr to high register");
    return emitThumb1LoadConstPool(MBB, MBBI, dl, DestReg, SubIdx, Val, Pred,
                                   PredReg, MIFlags);
  }
  return emitThumb2LoadConstPool(MBB, MBBI, dl, DestReg, SubIdx, Val, Pred,
                                 PredReg, MIFlags);
}

/// emitThumbRegPlusImmInReg - Emits a series of instructions to materialize
/// a destreg = basereg + immediate in Thumb code. Materialize the immediate
/// in a register using mov / mvn sequences or load the immediate from a
/// constpool entry.
static void emitThumbRegPlusImmInReg(
    MachineBasicBlock &MBB, MachineBasicBlock::iterator &MBBI,
    const DebugLoc &dl, unsigned DestReg, unsigned BaseReg, int NumBytes,
    bool CanChangeCC, const TargetInstrInfo &TII,
    const ARMBaseRegisterInfo &MRI, unsigned MIFlags = MachineInstr::NoFlags) {
  MachineFunction &MF = *MBB.getParent();
  const ARMSubtarget &ST = MF.getSubtarget<ARMSubtarget>();
  bool isHigh = !isARMLowRegister(DestReg) ||
                (BaseReg != 0 && !isARMLowRegister(BaseReg));
  bool isSub = false;
  // Subtract doesn't have high register version. Load the negative value
  // if either base or dest register is a high register. Also, if do not
  // issue sub as part of the sequence if condition register is to be
  // preserved.
  if (NumBytes < 0 && !isHigh && CanChangeCC) {
    isSub = true;
    NumBytes = -NumBytes;
  }
  unsigned LdReg = DestReg;
  if (DestReg == ARM::SP)
    assert(BaseReg == ARM::SP && "Unexpected!");
  if (!isARMLowRegister(DestReg) && !MRI.isVirtualRegister(DestReg))
    LdReg = MF.getRegInfo().createVirtualRegister(&ARM::tGPRRegClass);

  if (NumBytes <= 255 && NumBytes >= 0 && CanChangeCC) {
    BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVi8), LdReg)
        .add(t1CondCodeOp())
        .addImm(NumBytes)
        .setMIFlags(MIFlags);
  } else if (NumBytes < 0 && NumBytes >= -255 && CanChangeCC) {
    BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVi8), LdReg)
        .add(t1CondCodeOp())
        .addImm(NumBytes)
        .setMIFlags(MIFlags);
    BuildMI(MBB, MBBI, dl, TII.get(ARM::tRSB), LdReg)
        .add(t1CondCodeOp())
        .addReg(LdReg, RegState::Kill)
        .setMIFlags(MIFlags);
  } else if (ST.genExecuteOnly()) {
    BuildMI(MBB, MBBI, dl, TII.get(ARM::t2MOVi32imm), LdReg)
      .addImm(NumBytes).setMIFlags(MIFlags);
  } else
    MRI.emitLoadConstPool(MBB, MBBI, dl, LdReg, 0, NumBytes, ARMCC::AL, 0,
                          MIFlags);

  // Emit add / sub.
  int Opc = (isSub) ? ARM::tSUBrr
                    : ((isHigh || !CanChangeCC) ? ARM::tADDhirr : ARM::tADDrr);
  MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(Opc), DestReg);
  if (Opc != ARM::tADDhirr)
    MIB = MIB.add(t1CondCodeOp());
  if (DestReg == ARM::SP || isSub)
    MIB.addReg(BaseReg).addReg(LdReg, RegState::Kill);
  else
    MIB.addReg(LdReg).addReg(BaseReg, RegState::Kill);
  MIB.add(predOps(ARMCC::AL));
}

/// emitThumbRegPlusImmediate - Emits a series of instructions to materialize
/// a destreg = basereg + immediate in Thumb code. Tries a series of ADDs or
/// SUBs first, and uses a constant pool value if the instruction sequence would
/// be too long. This is allowed to modify the condition flags.
void llvm::emitThumbRegPlusImmediate(MachineBasicBlock &MBB,
                                     MachineBasicBlock::iterator &MBBI,
                                     const DebugLoc &dl, unsigned DestReg,
                                     unsigned BaseReg, int NumBytes,
                                     const TargetInstrInfo &TII,
                                     const ARMBaseRegisterInfo &MRI,
                                     unsigned MIFlags) {
  bool isSub = NumBytes < 0;
  unsigned Bytes = (unsigned)NumBytes;
  if (isSub) Bytes = -NumBytes;

  int CopyOpc = 0;
  unsigned CopyBits = 0;
  unsigned CopyScale = 1;
  bool CopyNeedsCC = false;
  int ExtraOpc = 0;
  unsigned ExtraBits = 0;
  unsigned ExtraScale = 1;
  bool ExtraNeedsCC = false;

  // Strategy:
  // We need to select two types of instruction, maximizing the available
  // immediate range of each. The instructions we use will depend on whether
  // DestReg and BaseReg are low, high or the stack pointer.
  // * CopyOpc  - DestReg = BaseReg + imm
  //              This will be emitted once if DestReg != BaseReg, and never if
  //              DestReg == BaseReg.
  // * ExtraOpc - DestReg = DestReg + imm
  //              This will be emitted as many times as necessary to add the
  //              full immediate.
  // If the immediate ranges of these instructions are not large enough to cover
  // NumBytes with a reasonable number of instructions, we fall back to using a
  // value loaded from a constant pool.
  if (DestReg == ARM::SP) {
    if (BaseReg == ARM::SP) {
      // sp -> sp
      // Already in right reg, no copy needed
    } else {
      // low -> sp or high -> sp
      CopyOpc = ARM::tMOVr;
      CopyBits = 0;
    }
    ExtraOpc = isSub ? ARM::tSUBspi : ARM::tADDspi;
    ExtraBits = 7;
    ExtraScale = 4;
  } else if (isARMLowRegister(DestReg)) {
    if (BaseReg == ARM::SP) {
      // sp -> low
      assert(!isSub && "Thumb1 does not have tSUBrSPi");
      CopyOpc = ARM::tADDrSPi;
      CopyBits = 8;
      CopyScale = 4;
    } else if (DestReg == BaseReg) {
      // low -> same low
      // Already in right reg, no copy needed
    } else if (isARMLowRegister(BaseReg)) {
      // low -> different low
      CopyOpc = isSub ? ARM::tSUBi3 : ARM::tADDi3;
      CopyBits = 3;
      CopyNeedsCC = true;
    } else {
      // high -> low
      CopyOpc = ARM::tMOVr;
      CopyBits = 0;
    }
    ExtraOpc = isSub ? ARM::tSUBi8 : ARM::tADDi8;
    ExtraBits = 8;
    ExtraNeedsCC = true;
  } else /* DestReg is high */ {
    if (DestReg == BaseReg) {
      // high -> same high
      // Already in right reg, no copy needed
    } else {
      // {low,high,sp} -> high
      CopyOpc = ARM::tMOVr;
      CopyBits = 0;
    }
    ExtraOpc = 0;
  }

  // We could handle an unaligned immediate with an unaligned copy instruction
  // and an aligned extra instruction, but this case is not currently needed.
  assert(((Bytes & 3) == 0 || ExtraScale == 1) &&
         "Unaligned offset, but all instructions require alignment");

  unsigned CopyRange = ((1 << CopyBits) - 1) * CopyScale;
  // If we would emit the copy with an immediate of 0, just use tMOVr.
  if (CopyOpc && Bytes < CopyScale) {
    CopyOpc = ARM::tMOVr;
    CopyScale = 1;
    CopyNeedsCC = false;
    CopyRange = 0;
  }
  unsigned ExtraRange = ((1 << ExtraBits) - 1) * ExtraScale; // per instruction
  unsigned RequiredCopyInstrs = CopyOpc ? 1 : 0;
  unsigned RangeAfterCopy = (CopyRange > Bytes) ? 0 : (Bytes - CopyRange);

  // We could handle this case when the copy instruction does not require an
  // aligned immediate, but we do not currently do this.
  assert(RangeAfterCopy % ExtraScale == 0 &&
         "Extra instruction requires immediate to be aligned");

  unsigned RequiredExtraInstrs;
  if (ExtraRange)
    RequiredExtraInstrs = alignTo(RangeAfterCopy, ExtraRange) / ExtraRange;
  else if (RangeAfterCopy > 0)
    // We need an extra instruction but none is available
    RequiredExtraInstrs = 1000000;
  else
    RequiredExtraInstrs = 0;
  unsigned RequiredInstrs = RequiredCopyInstrs + RequiredExtraInstrs;
  unsigned Threshold = (DestReg == ARM::SP) ? 3 : 2;

  // Use a constant pool, if the sequence of ADDs/SUBs is too expensive.
  if (RequiredInstrs > Threshold) {
    emitThumbRegPlusImmInReg(MBB, MBBI, dl,
                             DestReg, BaseReg, NumBytes, true,
                             TII, MRI, MIFlags);
    return;
  }

  // Emit zero or one copy instructions
  if (CopyOpc) {
    unsigned CopyImm = std::min(Bytes, CopyRange) / CopyScale;
    Bytes -= CopyImm * CopyScale;

    MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(CopyOpc), DestReg);
    if (CopyNeedsCC)
      MIB = MIB.add(t1CondCodeOp());
    MIB.addReg(BaseReg, RegState::Kill);
    if (CopyOpc != ARM::tMOVr) {
      MIB.addImm(CopyImm);
    }
    MIB.setMIFlags(MIFlags).add(predOps(ARMCC::AL));

    BaseReg = DestReg;
  }

  // Emit zero or more in-place add/sub instructions
  while (Bytes) {
    unsigned ExtraImm = std::min(Bytes, ExtraRange) / ExtraScale;
    Bytes -= ExtraImm * ExtraScale;

    MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(ExtraOpc), DestReg);
    if (ExtraNeedsCC)
      MIB = MIB.add(t1CondCodeOp());
    MIB.addReg(BaseReg)
       .addImm(ExtraImm)
       .add(predOps(ARMCC::AL))
       .setMIFlags(MIFlags);
  }
}

static void removeOperands(MachineInstr &MI, unsigned i) {
  unsigned Op = i;
  for (unsigned e = MI.getNumOperands(); i != e; ++i)
    MI.RemoveOperand(Op);
}

/// convertToNonSPOpcode - Change the opcode to the non-SP version, because
/// we're replacing the frame index with a non-SP register.
static unsigned convertToNonSPOpcode(unsigned Opcode) {
  switch (Opcode) {
  case ARM::tLDRspi:
    return ARM::tLDRi;

  case ARM::tSTRspi:
    return ARM::tSTRi;
  }

  return Opcode;
}

bool ThumbRegisterInfo::rewriteFrameIndex(MachineBasicBlock::iterator II,
                                          unsigned FrameRegIdx,
                                          unsigned FrameReg, int &Offset,
                                          const ARMBaseInstrInfo &TII) const {
  MachineInstr &MI = *II;
  MachineBasicBlock &MBB = *MI.getParent();
  assert(MBB.getParent()->getSubtarget<ARMSubtarget>().isThumb1Only() &&
         "This isn't needed for thumb2!");
  DebugLoc dl = MI.getDebugLoc();
  MachineInstrBuilder MIB(*MBB.getParent(), &MI);
  unsigned Opcode = MI.getOpcode();
  const MCInstrDesc &Desc = MI.getDesc();
  unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask);

  if (Opcode == ARM::tADDframe) {
    Offset += MI.getOperand(FrameRegIdx+1).getImm();
    unsigned DestReg = MI.getOperand(0).getReg();

    emitThumbRegPlusImmediate(MBB, II, dl, DestReg, FrameReg, Offset, TII,
                              *this);
    MBB.erase(II);
    return true;
  } else {
    if (AddrMode != ARMII::AddrModeT1_s)
      llvm_unreachable("Unsupported addressing mode!");

    unsigned ImmIdx = FrameRegIdx + 1;
    int InstrOffs = MI.getOperand(ImmIdx).getImm();
    unsigned NumBits = (FrameReg == ARM::SP) ? 8 : 5;
    unsigned Scale = 4;

    Offset += InstrOffs * Scale;
    assert((Offset & (Scale - 1)) == 0 && "Can't encode this offset!");

    // 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 the frame register (e.g., sp).
      MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false);
      ImmOp.ChangeToImmediate(ImmedOffset);

      // If we're using a register where sp was stored, convert the instruction
      // to the non-SP version.
      unsigned NewOpc = convertToNonSPOpcode(Opcode);
      if (NewOpc != Opcode && FrameReg != ARM::SP)
        MI.setDesc(TII.get(NewOpc));

      return true;
    }

    NumBits = 5;
    Mask = (1 << NumBits) - 1;

    // If this is a thumb spill / restore, we will be using a constpool load to
    // materialize the offset.
    if (Opcode == ARM::tLDRspi || Opcode == ARM::tSTRspi) {
      ImmOp.ChangeToImmediate(0);
    } else {
      // Otherwise, it didn't fit. Pull in what we can to simplify the immed.
      ImmedOffset = ImmedOffset & Mask;
      ImmOp.ChangeToImmediate(ImmedOffset);
      Offset &= ~(Mask * Scale);
    }
  }

  return Offset == 0;
}

void ThumbRegisterInfo::resolveFrameIndex(MachineInstr &MI, unsigned BaseReg,
                                           int64_t Offset) const {
  const MachineFunction &MF = *MI.getParent()->getParent();
  const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
  if (!STI.isThumb1Only())
    return ARMBaseRegisterInfo::resolveFrameIndex(MI, BaseReg, Offset);

  const ARMBaseInstrInfo &TII = *STI.getInstrInfo();
  int Off = Offset; // ARM doesn't need the general 64-bit offsets
  unsigned i = 0;

  while (!MI.getOperand(i).isFI()) {
    ++i;
    assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
  }
  bool Done = rewriteFrameIndex(MI, i, BaseReg, Off, TII);
  assert (Done && "Unable to resolve frame index!");
  (void)Done;
}

/// saveScavengerRegister - Spill the register so it can be used by the
/// register scavenger. Return true.
bool ThumbRegisterInfo::saveScavengerRegister(
    MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
    MachineBasicBlock::iterator &UseMI, const TargetRegisterClass *RC,
    unsigned Reg) const {

  const ARMSubtarget &STI = MBB.getParent()->getSubtarget<ARMSubtarget>();
  if (!STI.isThumb1Only())
    return ARMBaseRegisterInfo::saveScavengerRegister(MBB, I, UseMI, RC, Reg);

  // Thumb1 can't use the emergency spill slot on the stack because
  // ldr/str immediate offsets must be positive, and if we're referencing
  // off the frame pointer (if, for example, there are alloca() calls in
  // the function, the offset will be negative. Use R12 instead since that's
  // a call clobbered register that we know won't be used in Thumb1 mode.
  const TargetInstrInfo &TII = *STI.getInstrInfo();
  DebugLoc DL;
  BuildMI(MBB, I, DL, TII.get(ARM::tMOVr))
      .addReg(ARM::R12, RegState::Define)
      .addReg(Reg, RegState::Kill)
      .add(predOps(ARMCC::AL));

  // The UseMI is where we would like to restore the register. If there's
  // interference with R12 before then, however, we'll need to restore it
  // before that instead and adjust the UseMI.
  bool done = false;
  for (MachineBasicBlock::iterator II = I; !done && II != UseMI ; ++II) {
    if (II->isDebugValue())
      continue;
    // If this instruction affects R12, adjust our restore point.
    for (unsigned i = 0, e = II->getNumOperands(); i != e; ++i) {
      const MachineOperand &MO = II->getOperand(i);
      if (MO.isRegMask() && MO.clobbersPhysReg(ARM::R12)) {
        UseMI = II;
        done = true;
        break;
      }
      if (!MO.isReg() || MO.isUndef() || !MO.getReg() ||
          TargetRegisterInfo::isVirtualRegister(MO.getReg()))
        continue;
      if (MO.getReg() == ARM::R12) {
        UseMI = II;
        done = true;
        break;
      }
    }
  }
  // Restore the register from R12
  BuildMI(MBB, UseMI, DL, TII.get(ARM::tMOVr))
      .addReg(Reg, RegState::Define)
      .addReg(ARM::R12, RegState::Kill)
      .add(predOps(ARMCC::AL));

  return true;
}

void ThumbRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
                                            int SPAdj, unsigned FIOperandNum,
                                            RegScavenger *RS) const {
  MachineInstr &MI = *II;
  MachineBasicBlock &MBB = *MI.getParent();
  MachineFunction &MF = *MBB.getParent();
  const ARMSubtarget &STI = MF.getSubtarget<ARMSubtarget>();
  if (!STI.isThumb1Only())
    return ARMBaseRegisterInfo::eliminateFrameIndex(II, SPAdj, FIOperandNum,
                                                    RS);

  unsigned VReg = 0;
  const ARMBaseInstrInfo &TII = *STI.getInstrInfo();
  ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
  DebugLoc dl = MI.getDebugLoc();
  MachineInstrBuilder MIB(*MBB.getParent(), &MI);

  unsigned FrameReg = ARM::SP;
  int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
  int Offset = MF.getFrameInfo().getObjectOffset(FrameIndex) +
               MF.getFrameInfo().getStackSize() + SPAdj;

  if (MF.getFrameInfo().hasVarSizedObjects()) {
    assert(SPAdj == 0 && STI.getFrameLowering()->hasFP(MF) && "Unexpected");
    // There are alloca()'s in this function, must reference off the frame
    // pointer or base pointer instead.
    if (!hasBasePointer(MF)) {
      FrameReg = getFrameRegister(MF);
      Offset -= AFI->getFramePtrSpillOffset();
    } else
      FrameReg = BasePtr;
  }

  // PEI::scavengeFrameVirtualRegs() cannot accurately track SPAdj because the
  // call frame setup/destroy instructions have already been eliminated.  That
  // means the stack pointer cannot be used to access the emergency spill slot
  // when !hasReservedCallFrame().
#ifndef NDEBUG
  if (RS && FrameReg == ARM::SP && RS->isScavengingFrameIndex(FrameIndex)){
    assert(STI.getFrameLowering()->hasReservedCallFrame(MF) &&
           "Cannot use SP to access the emergency spill slot in "
           "functions without a reserved call frame");
    assert(!MF.getFrameInfo().hasVarSizedObjects() &&
           "Cannot use SP to access the emergency spill slot in "
           "functions with variable sized frame objects");
  }
#endif // NDEBUG

  // Special handling of dbg_value instructions.
  if (MI.isDebugValue()) {
    MI.getOperand(FIOperandNum).  ChangeToRegister(FrameReg, false /*isDef*/);
    MI.getOperand(FIOperandNum+1).ChangeToImmediate(Offset);
    return;
  }

  // Modify MI as necessary to handle as much of 'Offset' as possible
  assert(AFI->isThumbFunction() &&
         "This eliminateFrameIndex only supports Thumb1!");
  if (rewriteFrameIndex(MI, FIOperandNum, FrameReg, Offset, TII))
    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!");

  unsigned Opcode = MI.getOpcode();

  // Remove predicate first.
  int PIdx = MI.findFirstPredOperandIdx();
  if (PIdx != -1)
    removeOperands(MI, PIdx);

  if (MI.mayLoad()) {
    // Use the destination register to materialize sp + offset.
    unsigned TmpReg = MI.getOperand(0).getReg();
    bool UseRR = false;
    if (Opcode == ARM::tLDRspi) {
      if (FrameReg == ARM::SP || STI.genExecuteOnly())
        emitThumbRegPlusImmInReg(MBB, II, dl, TmpReg, FrameReg,
                                 Offset, false, TII, *this);
      else {
        emitLoadConstPool(MBB, II, dl, TmpReg, 0, Offset);
        UseRR = true;
      }
    } else {
      emitThumbRegPlusImmediate(MBB, II, dl, TmpReg, FrameReg, Offset, TII,
                                *this);
    }

    MI.setDesc(TII.get(UseRR ? ARM::tLDRr : ARM::tLDRi));
    MI.getOperand(FIOperandNum).ChangeToRegister(TmpReg, false, false, true);
    if (UseRR)
      // Use [reg, reg] addrmode. Replace the immediate operand w/ the frame
      // register. The offset is already handled in the vreg value.
      MI.getOperand(FIOperandNum+1).ChangeToRegister(FrameReg, false, false,
                                                     false);
  } else if (MI.mayStore()) {
      VReg = MF.getRegInfo().createVirtualRegister(&ARM::tGPRRegClass);
      bool UseRR = false;

      if (Opcode == ARM::tSTRspi) {
        if (FrameReg == ARM::SP || STI.genExecuteOnly())
          emitThumbRegPlusImmInReg(MBB, II, dl, VReg, FrameReg,
                                   Offset, false, TII, *this);
        else {
          emitLoadConstPool(MBB, II, dl, VReg, 0, Offset);
          UseRR = true;
        }
      } else
        emitThumbRegPlusImmediate(MBB, II, dl, VReg, FrameReg, Offset, TII,
                                  *this);
      MI.setDesc(TII.get(UseRR ? ARM::tSTRr : ARM::tSTRi));
      MI.getOperand(FIOperandNum).ChangeToRegister(VReg, false, false, true);
      if (UseRR)
        // Use [reg, reg] addrmode. Replace the immediate operand w/ the frame
        // register. The offset is already handled in the vreg value.
        MI.getOperand(FIOperandNum+1).ChangeToRegister(FrameReg, false, false,
                                                       false);
  } else {
    llvm_unreachable("Unexpected opcode!");
  }

  // Add predicate back if it's needed.
  if (MI.isPredicable())
    MIB.add(predOps(ARMCC::AL));
}