llvm.org GIT mirror llvm / release_28 lib / Target / ARM / ARMFastISel.cpp
release_28

Tree @release_28 (Download .tar.gz)

ARMFastISel.cpp @release_28raw · 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
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
//===-- ARMFastISel.cpp - ARM FastISel implementation ---------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the ARM-specific support for the FastISel class. Some
// of the target-specific code is generated by tablegen in the file
// ARMGenFastISel.inc, which is #included here.
//
//===----------------------------------------------------------------------===//

#include "ARM.h"
#include "ARMBaseInstrInfo.h"
#include "ARMRegisterInfo.h"
#include "ARMTargetMachine.h"
#include "ARMSubtarget.h"
#include "llvm/CallingConv.h"
#include "llvm/DerivedTypes.h"
#include "llvm/GlobalVariable.h"
#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/FastISel.h"
#include "llvm/CodeGen/FunctionLoweringInfo.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineConstantPool.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/Support/CallSite.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
using namespace llvm;

static cl::opt<bool>
EnableARMFastISel("arm-fast-isel",
                  cl::desc("Turn on experimental ARM fast-isel support"),
                  cl::init(false), cl::Hidden);

namespace {

class ARMFastISel : public FastISel {

  /// Subtarget - Keep a pointer to the ARMSubtarget around so that we can
  /// make the right decision when generating code for different targets.
  const ARMSubtarget *Subtarget;
  const TargetMachine &TM;
  const TargetInstrInfo &TII;
  const TargetLowering &TLI;
  const ARMFunctionInfo *AFI;

  // Convenience variable to avoid checking all the time.
  bool isThumb;

  public:
    explicit ARMFastISel(FunctionLoweringInfo &funcInfo) 
    : FastISel(funcInfo),
      TM(funcInfo.MF->getTarget()),
      TII(*TM.getInstrInfo()),
      TLI(*TM.getTargetLowering()) {
      Subtarget = &TM.getSubtarget<ARMSubtarget>();
      AFI = funcInfo.MF->getInfo<ARMFunctionInfo>();
      isThumb = AFI->isThumbFunction();
    }

    // Code from FastISel.cpp.
    virtual unsigned FastEmitInst_(unsigned MachineInstOpcode,
                                   const TargetRegisterClass *RC);
    virtual unsigned FastEmitInst_r(unsigned MachineInstOpcode,
                                    const TargetRegisterClass *RC,
                                    unsigned Op0, bool Op0IsKill);
    virtual unsigned FastEmitInst_rr(unsigned MachineInstOpcode,
                                     const TargetRegisterClass *RC,
                                     unsigned Op0, bool Op0IsKill,
                                     unsigned Op1, bool Op1IsKill);
    virtual unsigned FastEmitInst_ri(unsigned MachineInstOpcode,
                                     const TargetRegisterClass *RC,
                                     unsigned Op0, bool Op0IsKill,
                                     uint64_t Imm);
    virtual unsigned FastEmitInst_rf(unsigned MachineInstOpcode,
                                     const TargetRegisterClass *RC,
                                     unsigned Op0, bool Op0IsKill,
                                     const ConstantFP *FPImm);
    virtual unsigned FastEmitInst_i(unsigned MachineInstOpcode,
                                    const TargetRegisterClass *RC,
                                    uint64_t Imm);
    virtual unsigned FastEmitInst_rri(unsigned MachineInstOpcode,
                                      const TargetRegisterClass *RC,
                                      unsigned Op0, bool Op0IsKill,
                                      unsigned Op1, bool Op1IsKill,
                                      uint64_t Imm);
    virtual unsigned FastEmitInst_extractsubreg(MVT RetVT,
                                                unsigned Op0, bool Op0IsKill,
                                                uint32_t Idx);
                                                
    // Backend specific FastISel code.
    virtual bool TargetSelectInstruction(const Instruction *I);
    virtual unsigned TargetMaterializeConstant(const Constant *C);

  #include "ARMGenFastISel.inc"
  
    // Instruction selection routines.
    virtual bool ARMSelectLoad(const Instruction *I);
    virtual bool ARMSelectStore(const Instruction *I);
    virtual bool ARMSelectBranch(const Instruction *I);

    // Utility routines.
  private:
    bool isTypeLegal(const Type *Ty, EVT &VT);
    bool isLoadTypeLegal(const Type *Ty, EVT &VT);
    bool ARMEmitLoad(EVT VT, unsigned &ResultReg, unsigned Reg, int Offset);
    bool ARMEmitStore(EVT VT, unsigned SrcReg, unsigned Reg, int Offset);
    bool ARMLoadAlloca(const Instruction *I);
    bool ARMStoreAlloca(const Instruction *I, unsigned SrcReg);
    bool ARMComputeRegOffset(const Value *Obj, unsigned &Reg, int &Offset);
    bool ARMMaterializeConstant(const ConstantInt *Val, unsigned &Reg);
    
    bool DefinesOptionalPredicate(MachineInstr *MI, bool *CPSR);
    const MachineInstrBuilder &AddOptionalDefs(const MachineInstrBuilder &MIB);
};

} // end anonymous namespace

// #include "ARMGenCallingConv.inc"

// DefinesOptionalPredicate - This is different from DefinesPredicate in that
// we don't care about implicit defs here, just places we'll need to add a
// default CCReg argument. Sets CPSR if we're setting CPSR instead of CCR.
bool ARMFastISel::DefinesOptionalPredicate(MachineInstr *MI, bool *CPSR) {
  const TargetInstrDesc &TID = MI->getDesc();
  if (!TID.hasOptionalDef())
    return false;

  // Look to see if our OptionalDef is defining CPSR or CCR.
  for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
    const MachineOperand &MO = MI->getOperand(i);
    if (!MO.isReg() || !MO.isDef()) continue;
    if (MO.getReg() == ARM::CPSR)
      *CPSR = true;
  }
  return true;
}

// If the machine is predicable go ahead and add the predicate operands, if
// it needs default CC operands add those.
const MachineInstrBuilder &
ARMFastISel::AddOptionalDefs(const MachineInstrBuilder &MIB) {
  MachineInstr *MI = &*MIB;

  // Do we use a predicate?
  if (TII.isPredicable(MI))
    AddDefaultPred(MIB);
  
  // Do we optionally set a predicate?  Preds is size > 0 iff the predicate
  // defines CPSR. All other OptionalDefines in ARM are the CCR register.
  bool CPSR = false;
  if (DefinesOptionalPredicate(MI, &CPSR)) {
    if (CPSR)
      AddDefaultT1CC(MIB);
    else
      AddDefaultCC(MIB);
  }
  return MIB;
}

unsigned ARMFastISel::FastEmitInst_(unsigned MachineInstOpcode,
                                    const TargetRegisterClass* RC) {
  unsigned ResultReg = createResultReg(RC);
  const TargetInstrDesc &II = TII.get(MachineInstOpcode);

  AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg));
  return ResultReg;
}

unsigned ARMFastISel::FastEmitInst_r(unsigned MachineInstOpcode,
                                     const TargetRegisterClass *RC,
                                     unsigned Op0, bool Op0IsKill) {
  unsigned ResultReg = createResultReg(RC);
  const TargetInstrDesc &II = TII.get(MachineInstOpcode);

  if (II.getNumDefs() >= 1)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
                   .addReg(Op0, Op0IsKill * RegState::Kill));
  else {
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
                   .addReg(Op0, Op0IsKill * RegState::Kill));
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                   TII.get(TargetOpcode::COPY), ResultReg)
                   .addReg(II.ImplicitDefs[0]));
  }
  return ResultReg;
}

unsigned ARMFastISel::FastEmitInst_rr(unsigned MachineInstOpcode,
                                      const TargetRegisterClass *RC,
                                      unsigned Op0, bool Op0IsKill,
                                      unsigned Op1, bool Op1IsKill) {
  unsigned ResultReg = createResultReg(RC);
  const TargetInstrDesc &II = TII.get(MachineInstOpcode);

  if (II.getNumDefs() >= 1)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
                   .addReg(Op0, Op0IsKill * RegState::Kill)
                   .addReg(Op1, Op1IsKill * RegState::Kill));
  else {
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
                   .addReg(Op0, Op0IsKill * RegState::Kill)
                   .addReg(Op1, Op1IsKill * RegState::Kill));
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                           TII.get(TargetOpcode::COPY), ResultReg)
                   .addReg(II.ImplicitDefs[0]));
  }
  return ResultReg;
}

unsigned ARMFastISel::FastEmitInst_ri(unsigned MachineInstOpcode,
                                      const TargetRegisterClass *RC,
                                      unsigned Op0, bool Op0IsKill,
                                      uint64_t Imm) {
  unsigned ResultReg = createResultReg(RC);
  const TargetInstrDesc &II = TII.get(MachineInstOpcode);

  if (II.getNumDefs() >= 1)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
                   .addReg(Op0, Op0IsKill * RegState::Kill)
                   .addImm(Imm));
  else {
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
                   .addReg(Op0, Op0IsKill * RegState::Kill)
                   .addImm(Imm));
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                           TII.get(TargetOpcode::COPY), ResultReg)
                   .addReg(II.ImplicitDefs[0]));
  }
  return ResultReg;
}

unsigned ARMFastISel::FastEmitInst_rf(unsigned MachineInstOpcode,
                                      const TargetRegisterClass *RC,
                                      unsigned Op0, bool Op0IsKill,
                                      const ConstantFP *FPImm) {
  unsigned ResultReg = createResultReg(RC);
  const TargetInstrDesc &II = TII.get(MachineInstOpcode);

  if (II.getNumDefs() >= 1)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
                   .addReg(Op0, Op0IsKill * RegState::Kill)
                   .addFPImm(FPImm));
  else {
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
                   .addReg(Op0, Op0IsKill * RegState::Kill)
                   .addFPImm(FPImm));
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                           TII.get(TargetOpcode::COPY), ResultReg)
                   .addReg(II.ImplicitDefs[0]));
  }
  return ResultReg;
}

unsigned ARMFastISel::FastEmitInst_rri(unsigned MachineInstOpcode,
                                       const TargetRegisterClass *RC,
                                       unsigned Op0, bool Op0IsKill,
                                       unsigned Op1, bool Op1IsKill,
                                       uint64_t Imm) {
  unsigned ResultReg = createResultReg(RC);
  const TargetInstrDesc &II = TII.get(MachineInstOpcode);

  if (II.getNumDefs() >= 1)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
                   .addReg(Op0, Op0IsKill * RegState::Kill)
                   .addReg(Op1, Op1IsKill * RegState::Kill)
                   .addImm(Imm));
  else {
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
                   .addReg(Op0, Op0IsKill * RegState::Kill)
                   .addReg(Op1, Op1IsKill * RegState::Kill)
                   .addImm(Imm));
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                           TII.get(TargetOpcode::COPY), ResultReg)
                   .addReg(II.ImplicitDefs[0]));
  }
  return ResultReg;
}

unsigned ARMFastISel::FastEmitInst_i(unsigned MachineInstOpcode,
                                     const TargetRegisterClass *RC,
                                     uint64_t Imm) {
  unsigned ResultReg = createResultReg(RC);
  const TargetInstrDesc &II = TII.get(MachineInstOpcode);
  
  if (II.getNumDefs() >= 1)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II, ResultReg)
                   .addImm(Imm));
  else {
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, II)
                   .addImm(Imm));
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                           TII.get(TargetOpcode::COPY), ResultReg)
                   .addReg(II.ImplicitDefs[0]));
  }
  return ResultReg;
}

unsigned ARMFastISel::FastEmitInst_extractsubreg(MVT RetVT,
                                                 unsigned Op0, bool Op0IsKill,
                                                 uint32_t Idx) {
  unsigned ResultReg = createResultReg(TLI.getRegClassFor(RetVT));
  assert(TargetRegisterInfo::isVirtualRegister(Op0) &&
         "Cannot yet extract from physregs");
  AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
                         DL, TII.get(TargetOpcode::COPY), ResultReg)
                 .addReg(Op0, getKillRegState(Op0IsKill), Idx));
  return ResultReg;
}

unsigned ARMFastISel::TargetMaterializeConstant(const Constant *C) {
  EVT VT = TLI.getValueType(C->getType(), true);

  // Only handle simple types.
  if (!VT.isSimple()) return 0;
  
  // TODO: This should be safe for fp because they're just bits from the
  // Constant.
  // TODO: Theoretically we could materialize fp constants with instructions
  // from VFP3.

  // MachineConstantPool wants an explicit alignment.
  unsigned Align = TD.getPrefTypeAlignment(C->getType());
  if (Align == 0) {
    // TODO: Figure out if this is correct.
    Align = TD.getTypeAllocSize(C->getType());
  }
  unsigned Idx = MCP.getConstantPoolIndex(C, Align);

  unsigned DestReg = createResultReg(TLI.getRegClassFor(VT));
  // Different addressing modes between ARM/Thumb2 for constant pool loads.
  if (isThumb)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                            TII.get(ARM::t2LDRpci))
                    .addReg(DestReg).addConstantPoolIndex(Idx));
  else
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                            TII.get(ARM::LDRcp))
                    .addReg(DestReg).addConstantPoolIndex(Idx)
                    .addReg(0).addImm(0));
    
  return DestReg;
}

bool ARMFastISel::isTypeLegal(const Type *Ty, EVT &VT) {
  VT = TLI.getValueType(Ty, true);
  
  // Only handle simple types.
  if (VT == MVT::Other || !VT.isSimple()) return false;
    
  // Handle all legal types, i.e. a register that will directly hold this
  // value.
  return TLI.isTypeLegal(VT);
}

bool ARMFastISel::isLoadTypeLegal(const Type *Ty, EVT &VT) {
  if (isTypeLegal(Ty, VT)) return true;
  
  // If this is a type than can be sign or zero-extended to a basic operation
  // go ahead and accept it now.
  if (VT == MVT::i8 || VT == MVT::i16)
    return true;
  
  return false;
}

// Computes the Reg+Offset to get to an object.
bool ARMFastISel::ARMComputeRegOffset(const Value *Obj, unsigned &Reg,
                                      int &Offset) {
  // Some boilerplate from the X86 FastISel.
  const User *U = NULL;
  unsigned Opcode = Instruction::UserOp1;
  if (const Instruction *I = dyn_cast<Instruction>(Obj)) {
    // Don't walk into other basic blocks; it's possible we haven't
    // visited them yet, so the instructions may not yet be assigned
    // virtual registers.
    if (FuncInfo.MBBMap[I->getParent()] != FuncInfo.MBB)
      return false;

    Opcode = I->getOpcode();
    U = I;
  } else if (const ConstantExpr *C = dyn_cast<ConstantExpr>(Obj)) {
    Opcode = C->getOpcode();
    U = C;
  }

  if (const PointerType *Ty = dyn_cast<PointerType>(Obj->getType()))
    if (Ty->getAddressSpace() > 255)
      // Fast instruction selection doesn't support the special
      // address spaces.
      return false;
  
  switch (Opcode) {
    default: 
    //errs() << "Failing Opcode is: " << *Op1 << "\n";
    break;
    case Instruction::Alloca: {
      assert(false && "Alloca should have been handled earlier!");
      return false;
    }
  }
  
  if (const GlobalValue *GV = dyn_cast<GlobalValue>(Obj)) {
    //errs() << "Failing GV is: " << GV << "\n";
    (void)GV;
    return false;
  }
  
  // Try to get this in a register if nothing else has worked.
  Reg = getRegForValue(Obj);
  if (Reg == 0) return false;

  // Since the offset may be too large for the load instruction
  // get the reg+offset into a register.
  // TODO: Verify the additions work, otherwise we'll need to add the
  // offset instead of 0 to the instructions and do all sorts of operand
  // munging.
  // TODO: Optimize this somewhat.
  if (Offset != 0) {
    ARMCC::CondCodes Pred = ARMCC::AL;
    unsigned PredReg = 0;

    if (!isThumb)
      emitARMRegPlusImmediate(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                              Reg, Reg, Offset, Pred, PredReg,
                              static_cast<const ARMBaseInstrInfo&>(TII));
    else {
      assert(AFI->isThumb2Function());
      emitT2RegPlusImmediate(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                             Reg, Reg, Offset, Pred, PredReg,
                             static_cast<const ARMBaseInstrInfo&>(TII));
    }
  }
  
  return true;
}

bool ARMFastISel::ARMLoadAlloca(const Instruction *I) {
  Value *Op0 = I->getOperand(0);

  // Verify it's an alloca.
  if (const AllocaInst *AI = dyn_cast<AllocaInst>(Op0)) {
    DenseMap<const AllocaInst*, int>::iterator SI =
      FuncInfo.StaticAllocaMap.find(AI);

    if (SI != FuncInfo.StaticAllocaMap.end()) {
      TargetRegisterClass* RC = TLI.getRegClassFor(TLI.getPointerTy());
      unsigned ResultReg = createResultReg(RC);
      TII.loadRegFromStackSlot(*FuncInfo.MBB, *FuncInfo.InsertPt,
                               ResultReg, SI->second, RC,
                               TM.getRegisterInfo());
      UpdateValueMap(I, ResultReg);
      return true;
    }
  }
  return false;
}

bool ARMFastISel::ARMEmitLoad(EVT VT, unsigned &ResultReg,
                              unsigned Reg, int Offset) {
  
  assert(VT.isSimple() && "Non-simple types are invalid here!");
  unsigned Opc;
  
  switch (VT.getSimpleVT().SimpleTy) {
    default: 
      assert(false && "Trying to emit for an unhandled type!");
      return false;
    case MVT::i16:
      Opc = isThumb ? ARM::tLDRH : ARM::LDRH;
      VT = MVT::i32;
      break;
    case MVT::i8:
      Opc = isThumb ? ARM::tLDRB : ARM::LDRB;
      VT = MVT::i32;
      break;
    case MVT::i32:
      Opc = isThumb ? ARM::tLDR : ARM::LDR;
      break;
  }
  
  ResultReg = createResultReg(TLI.getRegClassFor(VT));
  
  // TODO: Fix the Addressing modes so that these can share some code.
  // Since this is a Thumb1 load this will work in Thumb1 or 2 mode.
  if (isThumb)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                            TII.get(Opc), ResultReg)
                    .addReg(Reg).addImm(Offset).addReg(0));
  else
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                            TII.get(Opc), ResultReg)
                    .addReg(Reg).addReg(0).addImm(Offset));
                    
  return true;
}

bool ARMFastISel::ARMStoreAlloca(const Instruction *I, unsigned SrcReg) {
  Value *Op1 = I->getOperand(1);

  // Verify it's an alloca.
  if (const AllocaInst *AI = dyn_cast<AllocaInst>(Op1)) {
    DenseMap<const AllocaInst*, int>::iterator SI =
      FuncInfo.StaticAllocaMap.find(AI);

    if (SI != FuncInfo.StaticAllocaMap.end()) {
      TargetRegisterClass* RC = TLI.getRegClassFor(TLI.getPointerTy());
      assert(SrcReg != 0 && "Nothing to store!");
      TII.storeRegToStackSlot(*FuncInfo.MBB, *FuncInfo.InsertPt,
                              SrcReg, true /*isKill*/, SI->second, RC,
                              TM.getRegisterInfo());
      return true;
    }
  }
  return false;
}

bool ARMFastISel::ARMEmitStore(EVT VT, unsigned SrcReg,
                               unsigned DstReg, int Offset) {
  unsigned StrOpc;
  switch (VT.getSimpleVT().SimpleTy) {
    default: return false;
    case MVT::i1:
    case MVT::i8: StrOpc = isThumb ? ARM::tSTRB : ARM::STRB; break;
    case MVT::i16: StrOpc = isThumb ? ARM::tSTRH : ARM::STRH; break;
    case MVT::i32: StrOpc = isThumb ? ARM::tSTR : ARM::STR; break;
    case MVT::f32:
      if (!Subtarget->hasVFP2()) return false;
      StrOpc = ARM::VSTRS;
      break;
    case MVT::f64:
      if (!Subtarget->hasVFP2()) return false;
      StrOpc = ARM::VSTRD;
      break;
  }
  
  if (isThumb)
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                            TII.get(StrOpc), SrcReg)
                    .addReg(DstReg).addImm(Offset).addReg(0));
  else
    AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
                            TII.get(StrOpc), SrcReg)
                    .addReg(DstReg).addReg(0).addImm(Offset));
  
  return true;
}

bool ARMFastISel::ARMSelectStore(const Instruction *I) {
  Value *Op0 = I->getOperand(0);
  unsigned SrcReg = 0;

  // Yay type legalization
  EVT VT;
  if (!isLoadTypeLegal(I->getOperand(0)->getType(), VT))
    return false;

  // Get the value to be stored into a register.
  SrcReg = getRegForValue(Op0);
  if (SrcReg == 0)
    return false;
    
  // If we're an alloca we know we have a frame index and can emit the store
  // quickly.
  if (ARMStoreAlloca(I, SrcReg))
    return true;
    
  // Our register and offset with innocuous defaults.
  unsigned Reg = 0;
  int Offset = 0;
  
  // See if we can handle this as Reg + Offset
  if (!ARMComputeRegOffset(I->getOperand(1), Reg, Offset))
    return false;
    
  if (!ARMEmitStore(VT, SrcReg, Reg, Offset /* 0 */)) return false;
    
  return false;
  
}

bool ARMFastISel::ARMSelectLoad(const Instruction *I) {
  // If we're an alloca we know we have a frame index and can emit the load
  // directly in short order.
  if (ARMLoadAlloca(I))
    return true;
    
  // Verify we have a legal type before going any further.
  EVT VT;
  if (!isLoadTypeLegal(I->getType(), VT))
    return false;
  
  // Our register and offset with innocuous defaults.
  unsigned Reg = 0;
  int Offset = 0;
  
  // See if we can handle this as Reg + Offset
  if (!ARMComputeRegOffset(I->getOperand(0), Reg, Offset))
    return false;
  
  unsigned ResultReg;
  if (!ARMEmitLoad(VT, ResultReg, Reg, Offset /* 0 */)) return false;
  
  UpdateValueMap(I, ResultReg);
  return true;
}

bool ARMFastISel::ARMSelectBranch(const Instruction *I) {
  const BranchInst *BI = cast<BranchInst>(I);
  MachineBasicBlock *TBB = FuncInfo.MBBMap[BI->getSuccessor(0)];
  MachineBasicBlock *FBB = FuncInfo.MBBMap[BI->getSuccessor(1)];
  
  // Simple branch support.
  unsigned CondReg = getRegForValue(BI->getCondition());
  if (CondReg == 0) return false;
  
  unsigned CmpOpc = isThumb ? ARM::t2CMPrr : ARM::CMPrr;
  unsigned BrOpc = isThumb ? ARM::t2Bcc : ARM::Bcc;
  AddOptionalDefs(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(CmpOpc))
                  .addReg(CondReg).addReg(CondReg));
  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL, TII.get(BrOpc))
                  .addMBB(TBB).addImm(ARMCC::NE).addReg(ARM::CPSR);
  FastEmitBranch(FBB, DL);
  FuncInfo.MBB->addSuccessor(TBB);
  return true;
}

// TODO: SoftFP support.
bool ARMFastISel::TargetSelectInstruction(const Instruction *I) {
  // No Thumb-1 for now.
  if (isThumb && !AFI->isThumb2Function()) return false;
  
  switch (I->getOpcode()) {
    case Instruction::Load:
      return ARMSelectLoad(I);
    case Instruction::Store:
      return ARMSelectStore(I);
    case Instruction::Br:
      return ARMSelectBranch(I);
    default: break;
  }
  return false;
}

namespace llvm {
  llvm::FastISel *ARM::createFastISel(FunctionLoweringInfo &funcInfo) {
    if (EnableARMFastISel) return new ARMFastISel(funcInfo);
    return 0;
  }
}