aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/llvm14/lib/Target/ARM/Thumb2InstrInfo.cpp
blob: ebd139af22190e8278617eee7c315de3ea7270c5 (plain) (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
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
//===- Thumb2InstrInfo.cpp - Thumb-2 Instruction Information --------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file contains the Thumb-2 implementation of the TargetInstrInfo class.
//
//===----------------------------------------------------------------------===//

#include "Thumb2InstrInfo.h"
#include "ARMMachineFunctionInfo.h"
#include "ARMSubtarget.h"
#include "MCTargetDesc/ARMAddressingModes.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/IR/DebugLoc.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/MCInstBuilder.h"
#include "llvm/MC/MCInstrDesc.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetMachine.h"
#include <cassert>

using namespace llvm;

static cl::opt<bool>
OldT2IfCvt("old-thumb2-ifcvt", cl::Hidden,
           cl::desc("Use old-style Thumb2 if-conversion heuristics"),
           cl::init(false));

static cl::opt<bool>
PreferNoCSEL("prefer-no-csel", cl::Hidden,
             cl::desc("Prefer predicated Move to CSEL"),
             cl::init(false));

Thumb2InstrInfo::Thumb2InstrInfo(const ARMSubtarget &STI)
    : ARMBaseInstrInfo(STI) {}

/// Return the noop instruction to use for a noop.
MCInst Thumb2InstrInfo::getNop() const {
  return MCInstBuilder(ARM::tHINT).addImm(0).addImm(ARMCC::AL).addReg(0);
}

unsigned Thumb2InstrInfo::getUnindexedOpcode(unsigned Opc) const {
  // FIXME
  return 0;
}

void
Thumb2InstrInfo::ReplaceTailWithBranchTo(MachineBasicBlock::iterator Tail,
                                         MachineBasicBlock *NewDest) const {
  MachineBasicBlock *MBB = Tail->getParent();
  ARMFunctionInfo *AFI = MBB->getParent()->getInfo<ARMFunctionInfo>();
  if (!AFI->hasITBlocks() || Tail->isBranch()) {
    TargetInstrInfo::ReplaceTailWithBranchTo(Tail, NewDest);
    return;
  }

  // If the first instruction of Tail is predicated, we may have to update
  // the IT instruction.
  Register PredReg;
  ARMCC::CondCodes CC = getInstrPredicate(*Tail, PredReg);
  MachineBasicBlock::iterator MBBI = Tail;
  if (CC != ARMCC::AL)
    // Expecting at least the t2IT instruction before it.
    --MBBI;

  // Actually replace the tail.
  TargetInstrInfo::ReplaceTailWithBranchTo(Tail, NewDest);

  // Fix up IT.
  if (CC != ARMCC::AL) {
    MachineBasicBlock::iterator E = MBB->begin();
    unsigned Count = 4; // At most 4 instructions in an IT block.
    while (Count && MBBI != E) {
      if (MBBI->isDebugInstr()) {
        --MBBI;
        continue;
      }
      if (MBBI->getOpcode() == ARM::t2IT) {
        unsigned Mask = MBBI->getOperand(1).getImm();
        if (Count == 4)
          MBBI->eraseFromParent();
        else {
          unsigned MaskOn = 1 << Count;
          unsigned MaskOff = ~(MaskOn - 1);
          MBBI->getOperand(1).setImm((Mask & MaskOff) | MaskOn);
        }
        return;
      }
      --MBBI;
      --Count;
    }

    // Ctrl flow can reach here if branch folding is run before IT block
    // formation pass.
  }
}

bool
Thumb2InstrInfo::isLegalToSplitMBBAt(MachineBasicBlock &MBB,
                                     MachineBasicBlock::iterator MBBI) const {
  while (MBBI->isDebugInstr()) {
    ++MBBI;
    if (MBBI == MBB.end())
      return false;
  }

  Register PredReg;
  return getITInstrPredicate(*MBBI, PredReg) == ARMCC::AL;
}

MachineInstr *
Thumb2InstrInfo::optimizeSelect(MachineInstr &MI,
                                SmallPtrSetImpl<MachineInstr *> &SeenMIs,
                                bool PreferFalse) const {
  // Try to use the base optimizeSelect, which uses canFoldIntoMOVCC to fold the
  // MOVCC into another instruction. If that fails on 8.1-M fall back to using a
  // CSEL.
  MachineInstr *RV = ARMBaseInstrInfo::optimizeSelect(MI, SeenMIs, PreferFalse);
  if (!RV && getSubtarget().hasV8_1MMainlineOps() && !PreferNoCSEL) {
    Register DestReg = MI.getOperand(0).getReg();

    if (!DestReg.isVirtual())
      return nullptr;

    MachineInstrBuilder NewMI = BuildMI(*MI.getParent(), MI, MI.getDebugLoc(),
                                        get(ARM::t2CSEL), DestReg)
                                    .add(MI.getOperand(2))
                                    .add(MI.getOperand(1))
                                    .add(MI.getOperand(3));
    SeenMIs.insert(NewMI);
    return NewMI;
  }
  return RV;
}

void Thumb2InstrInfo::copyPhysReg(MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator I,
                                  const DebugLoc &DL, MCRegister DestReg,
                                  MCRegister SrcReg, bool KillSrc) const {
  // Handle SPR, DPR, and QPR copies.
  if (!ARM::GPRRegClass.contains(DestReg, SrcReg))
    return ARMBaseInstrInfo::copyPhysReg(MBB, I, DL, DestReg, SrcReg, KillSrc);

  BuildMI(MBB, I, DL, get(ARM::tMOVr), DestReg)
      .addReg(SrcReg, getKillRegState(KillSrc))
      .add(predOps(ARMCC::AL));
}

void Thumb2InstrInfo::
storeRegToStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
                    Register SrcReg, bool isKill, int FI,
                    const TargetRegisterClass *RC,
                    const TargetRegisterInfo *TRI) const {
  DebugLoc DL;
  if (I != MBB.end()) DL = I->getDebugLoc();

  MachineFunction &MF = *MBB.getParent();
  MachineFrameInfo &MFI = MF.getFrameInfo();
  MachineMemOperand *MMO = MF.getMachineMemOperand(
      MachinePointerInfo::getFixedStack(MF, FI), MachineMemOperand::MOStore,
      MFI.getObjectSize(FI), MFI.getObjectAlign(FI));

  if (ARM::GPRRegClass.hasSubClassEq(RC)) {
    BuildMI(MBB, I, DL, get(ARM::t2STRi12))
        .addReg(SrcReg, getKillRegState(isKill))
        .addFrameIndex(FI)
        .addImm(0)
        .addMemOperand(MMO)
        .add(predOps(ARMCC::AL));
    return;
  }

  if (ARM::GPRPairRegClass.hasSubClassEq(RC)) {
    // Thumb2 STRD expects its dest-registers to be in rGPR. Not a problem for
    // gsub_0, but needs an extra constraint for gsub_1 (which could be sp
    // otherwise).
    if (Register::isVirtualRegister(SrcReg)) {
      MachineRegisterInfo *MRI = &MF.getRegInfo();
      MRI->constrainRegClass(SrcReg, &ARM::GPRPairnospRegClass);
    }

    MachineInstrBuilder MIB = BuildMI(MBB, I, DL, get(ARM::t2STRDi8));
    AddDReg(MIB, SrcReg, ARM::gsub_0, getKillRegState(isKill), TRI);
    AddDReg(MIB, SrcReg, ARM::gsub_1, 0, TRI);
    MIB.addFrameIndex(FI).addImm(0).addMemOperand(MMO).add(predOps(ARMCC::AL));
    return;
  }

  ARMBaseInstrInfo::storeRegToStackSlot(MBB, I, SrcReg, isKill, FI, RC, TRI);
}

void Thumb2InstrInfo::
loadRegFromStackSlot(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
                     Register DestReg, int FI,
                     const TargetRegisterClass *RC,
                     const TargetRegisterInfo *TRI) const {
  MachineFunction &MF = *MBB.getParent();
  MachineFrameInfo &MFI = MF.getFrameInfo();
  MachineMemOperand *MMO = MF.getMachineMemOperand(
      MachinePointerInfo::getFixedStack(MF, FI), MachineMemOperand::MOLoad,
      MFI.getObjectSize(FI), MFI.getObjectAlign(FI));
  DebugLoc DL;
  if (I != MBB.end()) DL = I->getDebugLoc();

  if (ARM::GPRRegClass.hasSubClassEq(RC)) {
    BuildMI(MBB, I, DL, get(ARM::t2LDRi12), DestReg)
        .addFrameIndex(FI)
        .addImm(0)
        .addMemOperand(MMO)
        .add(predOps(ARMCC::AL));
    return;
  }

  if (ARM::GPRPairRegClass.hasSubClassEq(RC)) {
    // Thumb2 LDRD expects its dest-registers to be in rGPR. Not a problem for
    // gsub_0, but needs an extra constraint for gsub_1 (which could be sp
    // otherwise).
    if (Register::isVirtualRegister(DestReg)) {
      MachineRegisterInfo *MRI = &MF.getRegInfo();
      MRI->constrainRegClass(DestReg, &ARM::GPRPairnospRegClass);
    }

    MachineInstrBuilder MIB = BuildMI(MBB, I, DL, get(ARM::t2LDRDi8));
    AddDReg(MIB, DestReg, ARM::gsub_0, RegState::DefineNoRead, TRI);
    AddDReg(MIB, DestReg, ARM::gsub_1, RegState::DefineNoRead, TRI);
    MIB.addFrameIndex(FI).addImm(0).addMemOperand(MMO).add(predOps(ARMCC::AL));

    if (Register::isPhysicalRegister(DestReg))
      MIB.addReg(DestReg, RegState::ImplicitDefine);
    return;
  }

  ARMBaseInstrInfo::loadRegFromStackSlot(MBB, I, DestReg, FI, RC, TRI);
}

void Thumb2InstrInfo::expandLoadStackGuard(
    MachineBasicBlock::iterator MI) const {
  MachineFunction &MF = *MI->getParent()->getParent();
  Module &M = *MF.getFunction().getParent();

  if (M.getStackProtectorGuard() == "tls") {
    expandLoadStackGuardBase(MI, ARM::t2MRC, ARM::t2LDRi12);
    return;
  }

  const GlobalValue *GV =
      cast<GlobalValue>((*MI->memoperands_begin())->getValue());

  if (MF.getSubtarget<ARMSubtarget>().isGVInGOT(GV))
    expandLoadStackGuardBase(MI, ARM::t2LDRLIT_ga_pcrel, ARM::t2LDRi12);
  else if (MF.getTarget().isPositionIndependent())
    expandLoadStackGuardBase(MI, ARM::t2MOV_ga_pcrel, ARM::t2LDRi12);
  else
    expandLoadStackGuardBase(MI, ARM::t2MOVi32imm, ARM::t2LDRi12);
}

MachineInstr *Thumb2InstrInfo::commuteInstructionImpl(MachineInstr &MI,
                                                      bool NewMI,
                                                      unsigned OpIdx1,
                                                      unsigned OpIdx2) const {
  switch (MI.getOpcode()) {
  case ARM::MVE_VMAXNMAf16:
  case ARM::MVE_VMAXNMAf32:
  case ARM::MVE_VMINNMAf16:
  case ARM::MVE_VMINNMAf32:
    // Don't allow predicated instructions to be commuted.
    if (getVPTInstrPredicate(MI) != ARMVCC::None)
      return nullptr;
  }
  return ARMBaseInstrInfo::commuteInstructionImpl(MI, NewMI, OpIdx1, OpIdx2);
}

void llvm::emitT2RegPlusImmediate(MachineBasicBlock &MBB,
                                  MachineBasicBlock::iterator &MBBI,
                                  const DebugLoc &dl, Register DestReg,
                                  Register BaseReg, int NumBytes,
                                  ARMCC::CondCodes Pred, Register PredReg,
                                  const ARMBaseInstrInfo &TII,
                                  unsigned MIFlags) {
  if (NumBytes == 0 && DestReg != BaseReg) {
    BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), DestReg)
      .addReg(BaseReg, RegState::Kill)
      .addImm((unsigned)Pred).addReg(PredReg).setMIFlags(MIFlags);
    return;
  }

  bool isSub = NumBytes < 0;
  if (isSub) NumBytes = -NumBytes;

  // If profitable, use a movw or movt to materialize the offset.
  // FIXME: Use the scavenger to grab a scratch register.
  if (DestReg != ARM::SP && DestReg != BaseReg &&
      NumBytes >= 4096 &&
      ARM_AM::getT2SOImmVal(NumBytes) == -1) {
    bool Fits = false;
    if (NumBytes < 65536) {
      // Use a movw to materialize the 16-bit constant.
      BuildMI(MBB, MBBI, dl, TII.get(ARM::t2MOVi16), DestReg)
        .addImm(NumBytes)
        .addImm((unsigned)Pred).addReg(PredReg).setMIFlags(MIFlags);
      Fits = true;
    } else if ((NumBytes & 0xffff) == 0) {
      // Use a movt to materialize the 32-bit constant.
      BuildMI(MBB, MBBI, dl, TII.get(ARM::t2MOVTi16), DestReg)
        .addReg(DestReg)
        .addImm(NumBytes >> 16)
        .addImm((unsigned)Pred).addReg(PredReg).setMIFlags(MIFlags);
      Fits = true;
    }

    if (Fits) {
      if (isSub) {
        BuildMI(MBB, MBBI, dl, TII.get(ARM::t2SUBrr), DestReg)
            .addReg(BaseReg)
            .addReg(DestReg, RegState::Kill)
            .add(predOps(Pred, PredReg))
            .add(condCodeOp())
            .setMIFlags(MIFlags);
      } else {
        // Here we know that DestReg is not SP but we do not
        // know anything about BaseReg. t2ADDrr is an invalid
        // instruction is SP is used as the second argument, but
        // is fine if SP is the first argument. To be sure we
        // do not generate invalid encoding, put BaseReg first.
        BuildMI(MBB, MBBI, dl, TII.get(ARM::t2ADDrr), DestReg)
            .addReg(BaseReg)
            .addReg(DestReg, RegState::Kill)
            .add(predOps(Pred, PredReg))
            .add(condCodeOp())
            .setMIFlags(MIFlags);
      }
      return;
    }
  }

  while (NumBytes) {
    unsigned ThisVal = NumBytes;
    unsigned Opc = 0;
    if (DestReg == ARM::SP && BaseReg != ARM::SP) {
      // mov sp, rn. Note t2MOVr cannot be used.
      BuildMI(MBB, MBBI, dl, TII.get(ARM::tMOVr), DestReg)
          .addReg(BaseReg)
          .setMIFlags(MIFlags)
          .add(predOps(ARMCC::AL));
      BaseReg = ARM::SP;
      continue;
    }

    assert((DestReg != ARM::SP || BaseReg == ARM::SP) &&
           "Writing to SP, from other register.");

    // Try to use T1, as it smaller
    if ((DestReg == ARM::SP) && (ThisVal < ((1 << 7) - 1) * 4)) {
      assert((ThisVal & 3) == 0 && "Stack update is not multiple of 4?");
      Opc = isSub ? ARM::tSUBspi : ARM::tADDspi;
      BuildMI(MBB, MBBI, dl, TII.get(Opc), DestReg)
          .addReg(BaseReg)
          .addImm(ThisVal / 4)
          .setMIFlags(MIFlags)
          .add(predOps(ARMCC::AL));
      break;
    }
    bool HasCCOut = true;
    int ImmIsT2SO = ARM_AM::getT2SOImmVal(ThisVal);
    bool ToSP = DestReg == ARM::SP;
    unsigned t2SUB = ToSP ? ARM::t2SUBspImm : ARM::t2SUBri;
    unsigned t2ADD = ToSP ? ARM::t2ADDspImm : ARM::t2ADDri;
    unsigned t2SUBi12 = ToSP ? ARM::t2SUBspImm12 : ARM::t2SUBri12;
    unsigned t2ADDi12 = ToSP ? ARM::t2ADDspImm12 : ARM::t2ADDri12;
    Opc = isSub ? t2SUB : t2ADD;
    // Prefer T2: sub rd, rn, so_imm | sub sp, sp, so_imm
    if (ImmIsT2SO != -1) {
      NumBytes = 0;
    } else if (ThisVal < 4096) {
      // Prefer T3 if can make it in a single go: subw rd, rn, imm12 | subw sp,
      // sp, imm12
      Opc = isSub ? t2SUBi12 : t2ADDi12;
      HasCCOut = false;
      NumBytes = 0;
    } else {
      // Use one T2 instruction to reduce NumBytes
      // FIXME: Move this to ARMAddressingModes.h?
      unsigned RotAmt = countLeadingZeros(ThisVal);
      ThisVal = ThisVal & ARM_AM::rotr32(0xff000000U, RotAmt);
      NumBytes &= ~ThisVal;
      assert(ARM_AM::getT2SOImmVal(ThisVal) != -1 &&
             "Bit extraction didn't work?");
    }

    // Build the new ADD / SUB.
    MachineInstrBuilder MIB = BuildMI(MBB, MBBI, dl, TII.get(Opc), DestReg)
                                  .addReg(BaseReg, RegState::Kill)
                                  .addImm(ThisVal)
                                  .add(predOps(ARMCC::AL))
                                  .setMIFlags(MIFlags);
    if (HasCCOut)
      MIB.add(condCodeOp());

    BaseReg = DestReg;
  }
}

static unsigned
negativeOffsetOpcode(unsigned opcode)
{
  switch (opcode) {
  case ARM::t2LDRi12:   return ARM::t2LDRi8;
  case ARM::t2LDRHi12:  return ARM::t2LDRHi8;
  case ARM::t2LDRBi12:  return ARM::t2LDRBi8;
  case ARM::t2LDRSHi12: return ARM::t2LDRSHi8;
  case ARM::t2LDRSBi12: return ARM::t2LDRSBi8;
  case ARM::t2STRi12:   return ARM::t2STRi8;
  case ARM::t2STRBi12:  return ARM::t2STRBi8;
  case ARM::t2STRHi12:  return ARM::t2STRHi8;
  case ARM::t2PLDi12:   return ARM::t2PLDi8;
  case ARM::t2PLDWi12:  return ARM::t2PLDWi8;
  case ARM::t2PLIi12:   return ARM::t2PLIi8;

  case ARM::t2LDRi8:
  case ARM::t2LDRHi8:
  case ARM::t2LDRBi8:
  case ARM::t2LDRSHi8:
  case ARM::t2LDRSBi8:
  case ARM::t2STRi8:
  case ARM::t2STRBi8:
  case ARM::t2STRHi8:
  case ARM::t2PLDi8:
  case ARM::t2PLDWi8:
  case ARM::t2PLIi8:
    return opcode;

  default:
    llvm_unreachable("unknown thumb2 opcode.");
  }
}

static unsigned
positiveOffsetOpcode(unsigned opcode)
{
  switch (opcode) {
  case ARM::t2LDRi8:   return ARM::t2LDRi12;
  case ARM::t2LDRHi8:  return ARM::t2LDRHi12;
  case ARM::t2LDRBi8:  return ARM::t2LDRBi12;
  case ARM::t2LDRSHi8: return ARM::t2LDRSHi12;
  case ARM::t2LDRSBi8: return ARM::t2LDRSBi12;
  case ARM::t2STRi8:   return ARM::t2STRi12;
  case ARM::t2STRBi8:  return ARM::t2STRBi12;
  case ARM::t2STRHi8:  return ARM::t2STRHi12;
  case ARM::t2PLDi8:   return ARM::t2PLDi12;
  case ARM::t2PLDWi8:  return ARM::t2PLDWi12;
  case ARM::t2PLIi8:   return ARM::t2PLIi12;

  case ARM::t2LDRi12:
  case ARM::t2LDRHi12:
  case ARM::t2LDRBi12:
  case ARM::t2LDRSHi12:
  case ARM::t2LDRSBi12:
  case ARM::t2STRi12:
  case ARM::t2STRBi12:
  case ARM::t2STRHi12:
  case ARM::t2PLDi12:
  case ARM::t2PLDWi12:
  case ARM::t2PLIi12:
    return opcode;

  default:
    llvm_unreachable("unknown thumb2 opcode.");
  }
}

static unsigned
immediateOffsetOpcode(unsigned opcode)
{
  switch (opcode) {
  case ARM::t2LDRs:   return ARM::t2LDRi12;
  case ARM::t2LDRHs:  return ARM::t2LDRHi12;
  case ARM::t2LDRBs:  return ARM::t2LDRBi12;
  case ARM::t2LDRSHs: return ARM::t2LDRSHi12;
  case ARM::t2LDRSBs: return ARM::t2LDRSBi12;
  case ARM::t2STRs:   return ARM::t2STRi12;
  case ARM::t2STRBs:  return ARM::t2STRBi12;
  case ARM::t2STRHs:  return ARM::t2STRHi12;
  case ARM::t2PLDs:   return ARM::t2PLDi12;
  case ARM::t2PLDWs:  return ARM::t2PLDWi12;
  case ARM::t2PLIs:   return ARM::t2PLIi12;

  case ARM::t2LDRi12:
  case ARM::t2LDRHi12:
  case ARM::t2LDRBi12:
  case ARM::t2LDRSHi12:
  case ARM::t2LDRSBi12:
  case ARM::t2STRi12:
  case ARM::t2STRBi12:
  case ARM::t2STRHi12:
  case ARM::t2PLDi12:
  case ARM::t2PLDWi12:
  case ARM::t2PLIi12:
  case ARM::t2LDRi8:
  case ARM::t2LDRHi8:
  case ARM::t2LDRBi8:
  case ARM::t2LDRSHi8:
  case ARM::t2LDRSBi8:
  case ARM::t2STRi8:
  case ARM::t2STRBi8:
  case ARM::t2STRHi8:
  case ARM::t2PLDi8:
  case ARM::t2PLDWi8:
  case ARM::t2PLIi8:
    return opcode;

  default:
    llvm_unreachable("unknown thumb2 opcode.");
  }
}

bool llvm::rewriteT2FrameIndex(MachineInstr &MI, unsigned FrameRegIdx,
                               Register FrameReg, int &Offset,
                               const ARMBaseInstrInfo &TII,
                               const TargetRegisterInfo *TRI) {
  unsigned Opcode = MI.getOpcode();
  const MCInstrDesc &Desc = MI.getDesc();
  unsigned AddrMode = (Desc.TSFlags & ARMII::AddrModeMask);
  bool isSub = false;

  MachineFunction &MF = *MI.getParent()->getParent();
  const TargetRegisterClass *RegClass =
      TII.getRegClass(Desc, FrameRegIdx, TRI, MF);

  // Memory operands in inline assembly always use AddrModeT2_i12.
  if (Opcode == ARM::INLINEASM || Opcode == ARM::INLINEASM_BR)
    AddrMode = ARMII::AddrModeT2_i12; // FIXME. mode for thumb2?

  const bool IsSP = Opcode == ARM::t2ADDspImm12 || Opcode == ARM::t2ADDspImm;
  if (IsSP || Opcode == ARM::t2ADDri || Opcode == ARM::t2ADDri12) {
    Offset += MI.getOperand(FrameRegIdx+1).getImm();

    Register PredReg;
    if (Offset == 0 && getInstrPredicate(MI, PredReg) == ARMCC::AL &&
        !MI.definesRegister(ARM::CPSR)) {
      // Turn it into a move.
      MI.setDesc(TII.get(ARM::tMOVr));
      MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false);
      // Remove offset and remaining explicit predicate operands.
      do MI.RemoveOperand(FrameRegIdx+1);
      while (MI.getNumOperands() > FrameRegIdx+1);
      MachineInstrBuilder MIB(*MI.getParent()->getParent(), &MI);
      MIB.add(predOps(ARMCC::AL));
      return true;
    }

    bool HasCCOut = (Opcode != ARM::t2ADDspImm12 && Opcode != ARM::t2ADDri12);

    if (Offset < 0) {
      Offset = -Offset;
      isSub = true;
      MI.setDesc(IsSP ? TII.get(ARM::t2SUBspImm) : TII.get(ARM::t2SUBri));
    } else {
      MI.setDesc(IsSP ? TII.get(ARM::t2ADDspImm) : TII.get(ARM::t2ADDri));
    }

    // Common case: small offset, fits into instruction.
    if (ARM_AM::getT2SOImmVal(Offset) != -1) {
      MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false);
      MI.getOperand(FrameRegIdx+1).ChangeToImmediate(Offset);
      // Add cc_out operand if the original instruction did not have one.
      if (!HasCCOut)
        MI.addOperand(MachineOperand::CreateReg(0, false));
      Offset = 0;
      return true;
    }
    // Another common case: imm12.
    if (Offset < 4096 &&
        (!HasCCOut || MI.getOperand(MI.getNumOperands()-1).getReg() == 0)) {
      unsigned NewOpc = isSub ? IsSP ? ARM::t2SUBspImm12 : ARM::t2SUBri12
                              : IsSP ? ARM::t2ADDspImm12 : ARM::t2ADDri12;
      MI.setDesc(TII.get(NewOpc));
      MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false);
      MI.getOperand(FrameRegIdx+1).ChangeToImmediate(Offset);
      // Remove the cc_out operand.
      if (HasCCOut)
        MI.RemoveOperand(MI.getNumOperands()-1);
      Offset = 0;
      return true;
    }

    // Otherwise, extract 8 adjacent bits from the immediate into this
    // t2ADDri/t2SUBri.
    unsigned RotAmt = countLeadingZeros<unsigned>(Offset);
    unsigned ThisImmVal = Offset & ARM_AM::rotr32(0xff000000U, RotAmt);

    // We will handle these bits from offset, clear them.
    Offset &= ~ThisImmVal;

    assert(ARM_AM::getT2SOImmVal(ThisImmVal) != -1 &&
           "Bit extraction didn't work?");
    MI.getOperand(FrameRegIdx+1).ChangeToImmediate(ThisImmVal);
    // Add cc_out operand if the original instruction did not have one.
    if (!HasCCOut)
      MI.addOperand(MachineOperand::CreateReg(0, false));
  } else {
    // AddrMode4 and AddrMode6 cannot handle any offset.
    if (AddrMode == ARMII::AddrMode4 || AddrMode == ARMII::AddrMode6)
      return false;

    // AddrModeT2_so cannot handle any offset. If there is no offset
    // register then we change to an immediate version.
    unsigned NewOpc = Opcode;
    if (AddrMode == ARMII::AddrModeT2_so) {
      Register OffsetReg = MI.getOperand(FrameRegIdx + 1).getReg();
      if (OffsetReg != 0) {
        MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false);
        return Offset == 0;
      }

      MI.RemoveOperand(FrameRegIdx+1);
      MI.getOperand(FrameRegIdx+1).ChangeToImmediate(0);
      NewOpc = immediateOffsetOpcode(Opcode);
      AddrMode = ARMII::AddrModeT2_i12;
    }

    unsigned NumBits = 0;
    unsigned Scale = 1;
    if (AddrMode == ARMII::AddrModeT2_i8neg ||
        AddrMode == ARMII::AddrModeT2_i12) {
      // i8 supports only negative, and i12 supports only positive, so
      // based on Offset sign convert Opcode to the appropriate
      // instruction
      Offset += MI.getOperand(FrameRegIdx+1).getImm();
      if (Offset < 0) {
        NewOpc = negativeOffsetOpcode(Opcode);
        NumBits = 8;
        isSub = true;
        Offset = -Offset;
      } else {
        NewOpc = positiveOffsetOpcode(Opcode);
        NumBits = 12;
      }
    } else if (AddrMode == ARMII::AddrMode5) {
      // VFP address mode.
      const MachineOperand &OffOp = MI.getOperand(FrameRegIdx+1);
      int InstrOffs = ARM_AM::getAM5Offset(OffOp.getImm());
      if (ARM_AM::getAM5Op(OffOp.getImm()) == ARM_AM::sub)
        InstrOffs *= -1;
      NumBits = 8;
      Scale = 4;
      Offset += InstrOffs * 4;
      assert((Offset & (Scale-1)) == 0 && "Can't encode this offset!");
      if (Offset < 0) {
        Offset = -Offset;
        isSub = true;
      }
    } else if (AddrMode == ARMII::AddrMode5FP16) {
      // VFP address mode.
      const MachineOperand &OffOp = MI.getOperand(FrameRegIdx+1);
      int InstrOffs = ARM_AM::getAM5FP16Offset(OffOp.getImm());
      if (ARM_AM::getAM5FP16Op(OffOp.getImm()) == ARM_AM::sub)
        InstrOffs *= -1;
      NumBits = 8;
      Scale = 2;
      Offset += InstrOffs * 2;
      assert((Offset & (Scale-1)) == 0 && "Can't encode this offset!");
      if (Offset < 0) {
        Offset = -Offset;
        isSub = true;
      }
    } else if (AddrMode == ARMII::AddrModeT2_i7s4 ||
               AddrMode == ARMII::AddrModeT2_i7s2 ||
               AddrMode == ARMII::AddrModeT2_i7) {
      Offset += MI.getOperand(FrameRegIdx + 1).getImm();
      unsigned OffsetMask;
      switch (AddrMode) {
      case ARMII::AddrModeT2_i7s4: NumBits = 9; OffsetMask = 0x3; break;
      case ARMII::AddrModeT2_i7s2: NumBits = 8; OffsetMask = 0x1; break;
      default:                     NumBits = 7; OffsetMask = 0x0; break;
      }
      // MCInst operand expects already scaled value.
      Scale = 1;
      assert((Offset & OffsetMask) == 0 && "Can't encode this offset!");
      (void)OffsetMask; // squash unused-variable warning at -NDEBUG
    } else if (AddrMode == ARMII::AddrModeT2_i8s4) {
      Offset += MI.getOperand(FrameRegIdx + 1).getImm();
      NumBits = 8 + 2;
      // MCInst operand expects already scaled value.
      Scale = 1;
      assert((Offset & 3) == 0 && "Can't encode this offset!");
    } else if (AddrMode == ARMII::AddrModeT2_ldrex) {
      Offset += MI.getOperand(FrameRegIdx + 1).getImm() * 4;
      NumBits = 8; // 8 bits scaled by 4
      Scale = 4;
      assert((Offset & 3) == 0 && "Can't encode this offset!");
    } else {
      llvm_unreachable("Unsupported addressing mode!");
    }

    if (NewOpc != Opcode)
      MI.setDesc(TII.get(NewOpc));

    MachineOperand &ImmOp = MI.getOperand(FrameRegIdx+1);

    // Attempt to fold address computation
    // Common case: small offset, fits into instruction. We need to make sure
    // the register class is correct too, for instructions like the MVE
    // VLDRH.32, which only accepts low tGPR registers.
    int ImmedOffset = Offset / Scale;
    unsigned Mask = (1 << NumBits) - 1;
    if ((unsigned)Offset <= Mask * Scale &&
        (Register::isVirtualRegister(FrameReg) ||
         RegClass->contains(FrameReg))) {
      if (Register::isVirtualRegister(FrameReg)) {
        // Make sure the register class for the virtual register is correct
        MachineRegisterInfo *MRI = &MF.getRegInfo();
        if (!MRI->constrainRegClass(FrameReg, RegClass))
          llvm_unreachable("Unable to constrain virtual register class.");
      }

      // Replace the FrameIndex with fp/sp
      MI.getOperand(FrameRegIdx).ChangeToRegister(FrameReg, false);
      if (isSub) {
        if (AddrMode == ARMII::AddrMode5 || AddrMode == ARMII::AddrMode5FP16)
          // FIXME: Not consistent.
          ImmedOffset |= 1 << NumBits;
        else
          ImmedOffset = -ImmedOffset;
      }
      ImmOp.ChangeToImmediate(ImmedOffset);
      Offset = 0;
      return true;
    }

    // Otherwise, offset doesn't fit. Pull in what we can to simplify
    ImmedOffset = ImmedOffset & Mask;
    if (isSub) {
      if (AddrMode == ARMII::AddrMode5 || AddrMode == ARMII::AddrMode5FP16)
        // FIXME: Not consistent.
        ImmedOffset |= 1 << NumBits;
      else {
        ImmedOffset = -ImmedOffset;
        if (ImmedOffset == 0)
          // Change the opcode back if the encoded offset is zero.
          MI.setDesc(TII.get(positiveOffsetOpcode(NewOpc)));
      }
    }
    ImmOp.ChangeToImmediate(ImmedOffset);
    Offset &= ~(Mask*Scale);
  }

  Offset = (isSub) ? -Offset : Offset;
  return Offset == 0 && (Register::isVirtualRegister(FrameReg) ||
                         RegClass->contains(FrameReg));
}

ARMCC::CondCodes llvm::getITInstrPredicate(const MachineInstr &MI,
                                           Register &PredReg) {
  unsigned Opc = MI.getOpcode();
  if (Opc == ARM::tBcc || Opc == ARM::t2Bcc)
    return ARMCC::AL;
  return getInstrPredicate(MI, PredReg);
}

int llvm::findFirstVPTPredOperandIdx(const MachineInstr &MI) {
  const MCInstrDesc &MCID = MI.getDesc();

  if (!MCID.OpInfo)
    return -1;

  for (unsigned i = 0, e = MCID.getNumOperands(); i != e; ++i)
    if (ARM::isVpred(MCID.OpInfo[i].OperandType))
      return i;

  return -1;
}

ARMVCC::VPTCodes llvm::getVPTInstrPredicate(const MachineInstr &MI,
                                            Register &PredReg) {
  int PIdx = findFirstVPTPredOperandIdx(MI);
  if (PIdx == -1) {
    PredReg = 0;
    return ARMVCC::None;
  }

  PredReg = MI.getOperand(PIdx+1).getReg();
  return (ARMVCC::VPTCodes)MI.getOperand(PIdx).getImm();
}

void llvm::recomputeVPTBlockMask(MachineInstr &Instr) {
  assert(isVPTOpcode(Instr.getOpcode()) && "Not a VPST or VPT Instruction!");

  MachineOperand &MaskOp = Instr.getOperand(0);
  assert(MaskOp.isImm() && "Operand 0 is not the block mask of the VPT/VPST?!");

  MachineBasicBlock::iterator Iter = ++Instr.getIterator(),
                              End = Instr.getParent()->end();

  while (Iter != End && Iter->isDebugInstr())
    ++Iter;

  // Verify that the instruction after the VPT/VPST is predicated (it should
  // be), and skip it.
  assert(Iter != End && "Expected some instructions in any VPT block");
  assert(
      getVPTInstrPredicate(*Iter) == ARMVCC::Then &&
      "VPT/VPST should be followed by an instruction with a 'then' predicate!");
  ++Iter;

  // Iterate over the predicated instructions, updating the BlockMask as we go.
  ARM::PredBlockMask BlockMask = ARM::PredBlockMask::T;
  while (Iter != End) {
    if (Iter->isDebugInstr()) {
      ++Iter;
      continue;
    }
    ARMVCC::VPTCodes Pred = getVPTInstrPredicate(*Iter);
    if (Pred == ARMVCC::None)
      break;
    BlockMask = expandPredBlockMask(BlockMask, Pred);
    ++Iter;
  }

  // Rewrite the BlockMask.
  MaskOp.setImm((int64_t)(BlockMask));
}