llvm.org GIT mirror llvm / c313a17 lib / Target / PowerPC / PPCTargetTransformInfo.cpp
c313a17

Tree @c313a17 (Download .tar.gz)

PPCTargetTransformInfo.cpp @c313a17raw · 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
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
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
//===-- PPCTargetTransformInfo.cpp - PPC specific TTI ---------------------===//
//
// 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
//
//===----------------------------------------------------------------------===//

#include "PPCTargetTransformInfo.h"
#include "llvm/Analysis/CodeMetrics.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/CodeGen/BasicTTIImpl.h"
#include "llvm/CodeGen/CostTable.h"
#include "llvm/CodeGen/TargetLowering.h"
#include "llvm/CodeGen/TargetSchedule.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
using namespace llvm;

#define DEBUG_TYPE "ppctti"

static cl::opt<bool> DisablePPCConstHoist("disable-ppc-constant-hoisting",
cl::desc("disable constant hoisting on PPC"), cl::init(false), cl::Hidden);

// This is currently only used for the data prefetch pass which is only enabled
// for BG/Q by default.
static cl::opt<unsigned>
CacheLineSize("ppc-loop-prefetch-cache-line", cl::Hidden, cl::init(64),
              cl::desc("The loop prefetch cache line size"));

static cl::opt<bool>
EnablePPCColdCC("ppc-enable-coldcc", cl::Hidden, cl::init(false),
                cl::desc("Enable using coldcc calling conv for cold "
                         "internal functions"));

// The latency of mtctr is only justified if there are more than 4
// comparisons that will be removed as a result.
static cl::opt<unsigned>
SmallCTRLoopThreshold("min-ctr-loop-threshold", cl::init(4), cl::Hidden,
                      cl::desc("Loops with a constant trip count smaller than "
                               "this value will not use the count register."));

//===----------------------------------------------------------------------===//
//
// PPC cost model.
//
//===----------------------------------------------------------------------===//

TargetTransformInfo::PopcntSupportKind
PPCTTIImpl::getPopcntSupport(unsigned TyWidth) {
  assert(isPowerOf2_32(TyWidth) && "Ty width must be power of 2");
  if (ST->hasPOPCNTD() != PPCSubtarget::POPCNTD_Unavailable && TyWidth <= 64)
    return ST->hasPOPCNTD() == PPCSubtarget::POPCNTD_Slow ?
             TTI::PSK_SlowHardware : TTI::PSK_FastHardware;
  return TTI::PSK_Software;
}

int PPCTTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
  if (DisablePPCConstHoist)
    return BaseT::getIntImmCost(Imm, Ty);

  assert(Ty->isIntegerTy());

  unsigned BitSize = Ty->getPrimitiveSizeInBits();
  if (BitSize == 0)
    return ~0U;

  if (Imm == 0)
    return TTI::TCC_Free;

  if (Imm.getBitWidth() <= 64) {
    if (isInt<16>(Imm.getSExtValue()))
      return TTI::TCC_Basic;

    if (isInt<32>(Imm.getSExtValue())) {
      // A constant that can be materialized using lis.
      if ((Imm.getZExtValue() & 0xFFFF) == 0)
        return TTI::TCC_Basic;

      return 2 * TTI::TCC_Basic;
    }
  }

  return 4 * TTI::TCC_Basic;
}

int PPCTTIImpl::getIntImmCost(Intrinsic::ID IID, unsigned Idx, const APInt &Imm,
                              Type *Ty) {
  if (DisablePPCConstHoist)
    return BaseT::getIntImmCost(IID, Idx, Imm, Ty);

  assert(Ty->isIntegerTy());

  unsigned BitSize = Ty->getPrimitiveSizeInBits();
  if (BitSize == 0)
    return ~0U;

  switch (IID) {
  default:
    return TTI::TCC_Free;
  case Intrinsic::sadd_with_overflow:
  case Intrinsic::uadd_with_overflow:
  case Intrinsic::ssub_with_overflow:
  case Intrinsic::usub_with_overflow:
    if ((Idx == 1) && Imm.getBitWidth() <= 64 && isInt<16>(Imm.getSExtValue()))
      return TTI::TCC_Free;
    break;
  case Intrinsic::experimental_stackmap:
    if ((Idx < 2) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
      return TTI::TCC_Free;
    break;
  case Intrinsic::experimental_patchpoint_void:
  case Intrinsic::experimental_patchpoint_i64:
    if ((Idx < 4) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
      return TTI::TCC_Free;
    break;
  }
  return PPCTTIImpl::getIntImmCost(Imm, Ty);
}

int PPCTTIImpl::getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
                              Type *Ty) {
  if (DisablePPCConstHoist)
    return BaseT::getIntImmCost(Opcode, Idx, Imm, Ty);

  assert(Ty->isIntegerTy());

  unsigned BitSize = Ty->getPrimitiveSizeInBits();
  if (BitSize == 0)
    return ~0U;

  unsigned ImmIdx = ~0U;
  bool ShiftedFree = false, RunFree = false, UnsignedFree = false,
       ZeroFree = false;
  switch (Opcode) {
  default:
    return TTI::TCC_Free;
  case Instruction::GetElementPtr:
    // Always hoist the base address of a GetElementPtr. This prevents the
    // creation of new constants for every base constant that gets constant
    // folded with the offset.
    if (Idx == 0)
      return 2 * TTI::TCC_Basic;
    return TTI::TCC_Free;
  case Instruction::And:
    RunFree = true; // (for the rotate-and-mask instructions)
    LLVM_FALLTHROUGH;
  case Instruction::Add:
  case Instruction::Or:
  case Instruction::Xor:
    ShiftedFree = true;
    LLVM_FALLTHROUGH;
  case Instruction::Sub:
  case Instruction::Mul:
  case Instruction::Shl:
  case Instruction::LShr:
  case Instruction::AShr:
    ImmIdx = 1;
    break;
  case Instruction::ICmp:
    UnsignedFree = true;
    ImmIdx = 1;
    // Zero comparisons can use record-form instructions.
    LLVM_FALLTHROUGH;
  case Instruction::Select:
    ZeroFree = true;
    break;
  case Instruction::PHI:
  case Instruction::Call:
  case Instruction::Ret:
  case Instruction::Load:
  case Instruction::Store:
    break;
  }

  if (ZeroFree && Imm == 0)
    return TTI::TCC_Free;

  if (Idx == ImmIdx && Imm.getBitWidth() <= 64) {
    if (isInt<16>(Imm.getSExtValue()))
      return TTI::TCC_Free;

    if (RunFree) {
      if (Imm.getBitWidth() <= 32 &&
          (isShiftedMask_32(Imm.getZExtValue()) ||
           isShiftedMask_32(~Imm.getZExtValue())))
        return TTI::TCC_Free;

      if (ST->isPPC64() &&
          (isShiftedMask_64(Imm.getZExtValue()) ||
           isShiftedMask_64(~Imm.getZExtValue())))
        return TTI::TCC_Free;
    }

    if (UnsignedFree && isUInt<16>(Imm.getZExtValue()))
      return TTI::TCC_Free;

    if (ShiftedFree && (Imm.getZExtValue() & 0xFFFF) == 0)
      return TTI::TCC_Free;
  }

  return PPCTTIImpl::getIntImmCost(Imm, Ty);
}

unsigned PPCTTIImpl::getUserCost(const User *U,
                                 ArrayRef<const Value *> Operands) {
  if (U->getType()->isVectorTy()) {
    // Instructions that need to be split should cost more.
    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, U->getType());
    return LT.first * BaseT::getUserCost(U, Operands);
  }

  return BaseT::getUserCost(U, Operands);
}

bool PPCTTIImpl::mightUseCTR(BasicBlock *BB,
                             TargetLibraryInfo *LibInfo) {
  const PPCTargetMachine &TM = ST->getTargetMachine();

  // Loop through the inline asm constraints and look for something that
  // clobbers ctr.
  auto asmClobbersCTR = [](InlineAsm *IA) {
    InlineAsm::ConstraintInfoVector CIV = IA->ParseConstraints();
    for (unsigned i = 0, ie = CIV.size(); i < ie; ++i) {
      InlineAsm::ConstraintInfo &C = CIV[i];
      if (C.Type != InlineAsm::isInput)
        for (unsigned j = 0, je = C.Codes.size(); j < je; ++j)
          if (StringRef(C.Codes[j]).equals_lower("{ctr}"))
            return true;
    }
    return false;
  };

  // Determining the address of a TLS variable results in a function call in
  // certain TLS models.
  std::function<bool(const Value*)> memAddrUsesCTR =
    [&memAddrUsesCTR, &TM](const Value *MemAddr) -> bool {
    const auto *GV = dyn_cast<GlobalValue>(MemAddr);
    if (!GV) {
      // Recurse to check for constants that refer to TLS global variables.
      if (const auto *CV = dyn_cast<Constant>(MemAddr))
        for (const auto &CO : CV->operands())
          if (memAddrUsesCTR(CO))
            return true;

      return false;
    }

    if (!GV->isThreadLocal())
      return false;
    TLSModel::Model Model = TM.getTLSModel(GV);
    return Model == TLSModel::GeneralDynamic ||
      Model == TLSModel::LocalDynamic;
  };

  auto isLargeIntegerTy = [](bool Is32Bit, Type *Ty) {
    if (IntegerType *ITy = dyn_cast<IntegerType>(Ty))
      return ITy->getBitWidth() > (Is32Bit ? 32U : 64U);

    return false;
  };

  for (BasicBlock::iterator J = BB->begin(), JE = BB->end();
       J != JE; ++J) {
    if (CallInst *CI = dyn_cast<CallInst>(J)) {
      // Inline ASM is okay, unless it clobbers the ctr register.
      if (InlineAsm *IA = dyn_cast<InlineAsm>(CI->getCalledValue())) {
        if (asmClobbersCTR(IA))
          return true;
        continue;
      }

      if (Function *F = CI->getCalledFunction()) {
        // Most intrinsics don't become function calls, but some might.
        // sin, cos, exp and log are always calls.
        unsigned Opcode = 0;
        if (F->getIntrinsicID() != Intrinsic::not_intrinsic) {
          switch (F->getIntrinsicID()) {
          default: continue;
          // If we have a call to ppc_is_decremented_ctr_nonzero, or ppc_mtctr
          // we're definitely using CTR.
          case Intrinsic::set_loop_iterations:
          case Intrinsic::loop_decrement:
            return true;

// VisualStudio defines setjmp as _setjmp
#if defined(_MSC_VER) && defined(setjmp) && \
                       !defined(setjmp_undefined_for_msvc)
#  pragma push_macro("setjmp")
#  undef setjmp
#  define setjmp_undefined_for_msvc
#endif

          case Intrinsic::setjmp:

#if defined(_MSC_VER) && defined(setjmp_undefined_for_msvc)
 // let's return it to _setjmp state
#  pragma pop_macro("setjmp")
#  undef setjmp_undefined_for_msvc
#endif

          case Intrinsic::longjmp:

          // Exclude eh_sjlj_setjmp; we don't need to exclude eh_sjlj_longjmp
          // because, although it does clobber the counter register, the
          // control can't then return to inside the loop unless there is also
          // an eh_sjlj_setjmp.
          case Intrinsic::eh_sjlj_setjmp:

          case Intrinsic::memcpy:
          case Intrinsic::memmove:
          case Intrinsic::memset:
          case Intrinsic::powi:
          case Intrinsic::log:
          case Intrinsic::log2:
          case Intrinsic::log10:
          case Intrinsic::exp:
          case Intrinsic::exp2:
          case Intrinsic::pow:
          case Intrinsic::sin:
          case Intrinsic::cos:
            return true;
          case Intrinsic::copysign:
            if (CI->getArgOperand(0)->getType()->getScalarType()->
                isPPC_FP128Ty())
              return true;
            else
              continue; // ISD::FCOPYSIGN is never a library call.
          case Intrinsic::sqrt:               Opcode = ISD::FSQRT;      break;
          case Intrinsic::floor:              Opcode = ISD::FFLOOR;     break;
          case Intrinsic::ceil:               Opcode = ISD::FCEIL;      break;
          case Intrinsic::trunc:              Opcode = ISD::FTRUNC;     break;
          case Intrinsic::rint:               Opcode = ISD::FRINT;      break;
          case Intrinsic::nearbyint:          Opcode = ISD::FNEARBYINT; break;
          case Intrinsic::round:              Opcode = ISD::FROUND;     break;
          case Intrinsic::minnum:             Opcode = ISD::FMINNUM;    break;
          case Intrinsic::maxnum:             Opcode = ISD::FMAXNUM;    break;
          case Intrinsic::umul_with_overflow: Opcode = ISD::UMULO;      break;
          case Intrinsic::smul_with_overflow: Opcode = ISD::SMULO;      break;
          }
        }

        // PowerPC does not use [US]DIVREM or other library calls for
        // operations on regular types which are not otherwise library calls
        // (i.e. soft float or atomics). If adapting for targets that do,
        // additional care is required here.

        LibFunc Func;
        if (!F->hasLocalLinkage() && F->hasName() && LibInfo &&
            LibInfo->getLibFunc(F->getName(), Func) &&
            LibInfo->hasOptimizedCodeGen(Func)) {
          // Non-read-only functions are never treated as intrinsics.
          if (!CI->onlyReadsMemory())
            return true;

          // Conversion happens only for FP calls.
          if (!CI->getArgOperand(0)->getType()->isFloatingPointTy())
            return true;

          switch (Func) {
          default: return true;
          case LibFunc_copysign:
          case LibFunc_copysignf:
            continue; // ISD::FCOPYSIGN is never a library call.
          case LibFunc_copysignl:
            return true;
          case LibFunc_fabs:
          case LibFunc_fabsf:
          case LibFunc_fabsl:
            continue; // ISD::FABS is never a library call.
          case LibFunc_sqrt:
          case LibFunc_sqrtf:
          case LibFunc_sqrtl:
            Opcode = ISD::FSQRT; break;
          case LibFunc_floor:
          case LibFunc_floorf:
          case LibFunc_floorl:
            Opcode = ISD::FFLOOR; break;
          case LibFunc_nearbyint:
          case LibFunc_nearbyintf:
          case LibFunc_nearbyintl:
            Opcode = ISD::FNEARBYINT; break;
          case LibFunc_ceil:
          case LibFunc_ceilf:
          case LibFunc_ceill:
            Opcode = ISD::FCEIL; break;
          case LibFunc_rint:
          case LibFunc_rintf:
          case LibFunc_rintl:
            Opcode = ISD::FRINT; break;
          case LibFunc_round:
          case LibFunc_roundf:
          case LibFunc_roundl:
            Opcode = ISD::FROUND; break;
          case LibFunc_trunc:
          case LibFunc_truncf:
          case LibFunc_truncl:
            Opcode = ISD::FTRUNC; break;
          case LibFunc_fmin:
          case LibFunc_fminf:
          case LibFunc_fminl:
            Opcode = ISD::FMINNUM; break;
          case LibFunc_fmax:
          case LibFunc_fmaxf:
          case LibFunc_fmaxl:
            Opcode = ISD::FMAXNUM; break;
          }
        }

        if (Opcode) {
          EVT EVTy =
              TLI->getValueType(DL, CI->getArgOperand(0)->getType(), true);

          if (EVTy == MVT::Other)
            return true;

          if (TLI->isOperationLegalOrCustom(Opcode, EVTy))
            continue;
          else if (EVTy.isVector() &&
                   TLI->isOperationLegalOrCustom(Opcode, EVTy.getScalarType()))
            continue;

          return true;
        }
      }

      return true;
    } else if (isa<BinaryOperator>(J) &&
               J->getType()->getScalarType()->isPPC_FP128Ty()) {
      // Most operations on ppc_f128 values become calls.
      return true;
    } else if (isa<UIToFPInst>(J) || isa<SIToFPInst>(J) ||
               isa<FPToUIInst>(J) || isa<FPToSIInst>(J)) {
      CastInst *CI = cast<CastInst>(J);
      if (CI->getSrcTy()->getScalarType()->isPPC_FP128Ty() ||
          CI->getDestTy()->getScalarType()->isPPC_FP128Ty() ||
          isLargeIntegerTy(!TM.isPPC64(), CI->getSrcTy()->getScalarType()) ||
          isLargeIntegerTy(!TM.isPPC64(), CI->getDestTy()->getScalarType()))
        return true;
    } else if (isLargeIntegerTy(!TM.isPPC64(),
                                J->getType()->getScalarType()) &&
               (J->getOpcode() == Instruction::UDiv ||
                J->getOpcode() == Instruction::SDiv ||
                J->getOpcode() == Instruction::URem ||
                J->getOpcode() == Instruction::SRem)) {
      return true;
    } else if (!TM.isPPC64() &&
               isLargeIntegerTy(false, J->getType()->getScalarType()) &&
               (J->getOpcode() == Instruction::Shl ||
                J->getOpcode() == Instruction::AShr ||
                J->getOpcode() == Instruction::LShr)) {
      // Only on PPC32, for 128-bit integers (specifically not 64-bit
      // integers), these might be runtime calls.
      return true;
    } else if (isa<IndirectBrInst>(J) || isa<InvokeInst>(J)) {
      // On PowerPC, indirect jumps use the counter register.
      return true;
    } else if (SwitchInst *SI = dyn_cast<SwitchInst>(J)) {
      if (SI->getNumCases() + 1 >= (unsigned)TLI->getMinimumJumpTableEntries())
        return true;
    }

    // FREM is always a call.
    if (J->getOpcode() == Instruction::FRem)
      return true;

    if (ST->useSoftFloat()) {
      switch(J->getOpcode()) {
      case Instruction::FAdd:
      case Instruction::FSub:
      case Instruction::FMul:
      case Instruction::FDiv:
      case Instruction::FPTrunc:
      case Instruction::FPExt:
      case Instruction::FPToUI:
      case Instruction::FPToSI:
      case Instruction::UIToFP:
      case Instruction::SIToFP:
      case Instruction::FCmp:
        return true;
      }
    }

    for (Value *Operand : J->operands())
      if (memAddrUsesCTR(Operand))
        return true;
  }

  return false;
}

bool PPCTTIImpl::isHardwareLoopProfitable(Loop *L, ScalarEvolution &SE,
                                          AssumptionCache &AC,
                                          TargetLibraryInfo *LibInfo,
                                          TTI::HardwareLoopInfo &HWLoopInfo) {
  const PPCTargetMachine &TM = ST->getTargetMachine();
  TargetSchedModel SchedModel;
  SchedModel.init(ST);

  // Do not convert small short loops to CTR loop.
  unsigned ConstTripCount = SE.getSmallConstantTripCount(L);
  if (ConstTripCount && ConstTripCount < SmallCTRLoopThreshold) {
    SmallPtrSet<const Value *, 32> EphValues;
    CodeMetrics::collectEphemeralValues(L, &AC, EphValues);
    CodeMetrics Metrics;
    for (BasicBlock *BB : L->blocks())
      Metrics.analyzeBasicBlock(BB, *this, EphValues);
    // 6 is an approximate latency for the mtctr instruction.
    if (Metrics.NumInsts <= (6 * SchedModel.getIssueWidth()))
      return false;
  }

  // We don't want to spill/restore the counter register, and so we don't
  // want to use the counter register if the loop contains calls.
  for (Loop::block_iterator I = L->block_begin(), IE = L->block_end();
       I != IE; ++I)
    if (mightUseCTR(*I, LibInfo))
      return false;

  SmallVector<BasicBlock*, 4> ExitingBlocks;
  L->getExitingBlocks(ExitingBlocks);

  // If there is an exit edge known to be frequently taken,
  // we should not transform this loop.
  for (auto &BB : ExitingBlocks) {
    Instruction *TI = BB->getTerminator();
    if (!TI) continue;

    if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
      uint64_t TrueWeight = 0, FalseWeight = 0;
      if (!BI->isConditional() ||
          !BI->extractProfMetadata(TrueWeight, FalseWeight))
        continue;

      // If the exit path is more frequent than the loop path,
      // we return here without further analysis for this loop.
      bool TrueIsExit = !L->contains(BI->getSuccessor(0));
      if (( TrueIsExit && FalseWeight < TrueWeight) ||
          (!TrueIsExit && FalseWeight > TrueWeight))
        return false;
    }
  }

  LLVMContext &C = L->getHeader()->getContext();
  HWLoopInfo.CountType = TM.isPPC64() ?
    Type::getInt64Ty(C) : Type::getInt32Ty(C);
  HWLoopInfo.LoopDecrement = ConstantInt::get(HWLoopInfo.CountType, 1);
  return true;
}

void PPCTTIImpl::getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                         TTI::UnrollingPreferences &UP) {
  if (ST->getDarwinDirective() == PPC::DIR_A2) {
    // The A2 is in-order with a deep pipeline, and concatenation unrolling
    // helps expose latency-hiding opportunities to the instruction scheduler.
    UP.Partial = UP.Runtime = true;

    // We unroll a lot on the A2 (hundreds of instructions), and the benefits
    // often outweigh the cost of a division to compute the trip count.
    UP.AllowExpensiveTripCount = true;
  }

  BaseT::getUnrollingPreferences(L, SE, UP);
}

// This function returns true to allow using coldcc calling convention.
// Returning true results in coldcc being used for functions which are cold at
// all call sites when the callers of the functions are not calling any other
// non coldcc functions.
bool PPCTTIImpl::useColdCCForColdCall(Function &F) {
  return EnablePPCColdCC;
}

bool PPCTTIImpl::enableAggressiveInterleaving(bool LoopHasReductions) {
  // On the A2, always unroll aggressively. For QPX unaligned loads, we depend
  // on combining the loads generated for consecutive accesses, and failure to
  // do so is particularly expensive. This makes it much more likely (compared
  // to only using concatenation unrolling).
  if (ST->getDarwinDirective() == PPC::DIR_A2)
    return true;

  return LoopHasReductions;
}

const PPCTTIImpl::TTI::MemCmpExpansionOptions *
PPCTTIImpl::enableMemCmpExpansion(bool IsZeroCmp) const {
  static const auto Options = []() {
    TTI::MemCmpExpansionOptions Options;
    Options.LoadSizes.push_back(8);
    Options.LoadSizes.push_back(4);
    Options.LoadSizes.push_back(2);
    Options.LoadSizes.push_back(1);
    return Options;
  }();
  return &Options;
}

bool PPCTTIImpl::enableInterleavedAccessVectorization() {
  return true;
}

unsigned PPCTTIImpl::getNumberOfRegisters(bool Vector) {
  if (Vector && !ST->hasAltivec() && !ST->hasQPX())
    return 0;
  return ST->hasVSX() ? 64 : 32;
}

unsigned PPCTTIImpl::getRegisterBitWidth(bool Vector) const {
  if (Vector) {
    if (ST->hasQPX()) return 256;
    if (ST->hasAltivec()) return 128;
    return 0;
  }

  if (ST->isPPC64())
    return 64;
  return 32;

}

unsigned PPCTTIImpl::getCacheLineSize() {
  // Check first if the user specified a custom line size.
  if (CacheLineSize.getNumOccurrences() > 0)
    return CacheLineSize;

  // On P7, P8 or P9 we have a cache line size of 128.
  unsigned Directive = ST->getDarwinDirective();
  if (Directive == PPC::DIR_PWR7 || Directive == PPC::DIR_PWR8 ||
      Directive == PPC::DIR_PWR9)
    return 128;

  // On other processors return a default of 64 bytes.
  return 64;
}

unsigned PPCTTIImpl::getPrefetchDistance() {
  // This seems like a reasonable default for the BG/Q (this pass is enabled, by
  // default, only on the BG/Q).
  return 300;
}

unsigned PPCTTIImpl::getMaxInterleaveFactor(unsigned VF) {
  unsigned Directive = ST->getDarwinDirective();
  // The 440 has no SIMD support, but floating-point instructions
  // have a 5-cycle latency, so unroll by 5x for latency hiding.
  if (Directive == PPC::DIR_440)
    return 5;

  // The A2 has no SIMD support, but floating-point instructions
  // have a 6-cycle latency, so unroll by 6x for latency hiding.
  if (Directive == PPC::DIR_A2)
    return 6;

  // FIXME: For lack of any better information, do no harm...
  if (Directive == PPC::DIR_E500mc || Directive == PPC::DIR_E5500)
    return 1;

  // For P7 and P8, floating-point instructions have a 6-cycle latency and
  // there are two execution units, so unroll by 12x for latency hiding.
  // FIXME: the same for P9 as previous gen until POWER9 scheduling is ready
  if (Directive == PPC::DIR_PWR7 || Directive == PPC::DIR_PWR8 ||
      Directive == PPC::DIR_PWR9)
    return 12;

  // For most things, modern systems have two execution units (and
  // out-of-order execution).
  return 2;
}

// Adjust the cost of vector instructions on targets which there is overlap
// between the vector and scalar units, thereby reducing the overall throughput
// of vector code wrt. scalar code.
int PPCTTIImpl::vectorCostAdjustment(int Cost, unsigned Opcode, Type *Ty1,
                                     Type *Ty2) {
  if (!ST->vectorsUseTwoUnits() || !Ty1->isVectorTy())
    return Cost;

  std::pair<int, MVT> LT1 = TLI->getTypeLegalizationCost(DL, Ty1);
  // If type legalization involves splitting the vector, we don't want to
  // double the cost at every step - only the last step.
  if (LT1.first != 1 || !LT1.second.isVector())
    return Cost;

  int ISD = TLI->InstructionOpcodeToISD(Opcode);
  if (TLI->isOperationExpand(ISD, LT1.second))
    return Cost;

  if (Ty2) {
    std::pair<int, MVT> LT2 = TLI->getTypeLegalizationCost(DL, Ty2);
    if (LT2.first != 1 || !LT2.second.isVector())
      return Cost;
  }

  return Cost * 2;
}

int PPCTTIImpl::getArithmeticInstrCost(
    unsigned Opcode, Type *Ty, TTI::OperandValueKind Op1Info,
    TTI::OperandValueKind Op2Info, TTI::OperandValueProperties Opd1PropInfo,
    TTI::OperandValueProperties Opd2PropInfo, ArrayRef<const Value *> Args) {
  assert(TLI->InstructionOpcodeToISD(Opcode) && "Invalid opcode");

  // Fallback to the default implementation.
  int Cost = BaseT::getArithmeticInstrCost(Opcode, Ty, Op1Info, Op2Info,
                                           Opd1PropInfo, Opd2PropInfo);
  return vectorCostAdjustment(Cost, Opcode, Ty, nullptr);
}

int PPCTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index,
                               Type *SubTp) {
  // Legalize the type.
  std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);

  // PPC, for both Altivec/VSX and QPX, support cheap arbitrary permutations
  // (at least in the sense that there need only be one non-loop-invariant
  // instruction). We need one such shuffle instruction for each actual
  // register (this is not true for arbitrary shuffles, but is true for the
  // structured types of shuffles covered by TTI::ShuffleKind).
  return vectorCostAdjustment(LT.first, Instruction::ShuffleVector, Tp,
                              nullptr);
}

int PPCTTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src,
                                 const Instruction *I) {
  assert(TLI->InstructionOpcodeToISD(Opcode) && "Invalid opcode");

  int Cost = BaseT::getCastInstrCost(Opcode, Dst, Src);
  return vectorCostAdjustment(Cost, Opcode, Dst, Src);
}

int PPCTTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy,
                                   const Instruction *I) {
  int Cost = BaseT::getCmpSelInstrCost(Opcode, ValTy, CondTy, I);
  return vectorCostAdjustment(Cost, Opcode, ValTy, nullptr);
}

int PPCTTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val, unsigned Index) {
  assert(Val->isVectorTy() && "This must be a vector type");

  int ISD = TLI->InstructionOpcodeToISD(Opcode);
  assert(ISD && "Invalid opcode");

  int Cost = BaseT::getVectorInstrCost(Opcode, Val, Index);
  Cost = vectorCostAdjustment(Cost, Opcode, Val, nullptr);

  if (ST->hasVSX() && Val->getScalarType()->isDoubleTy()) {
    // Double-precision scalars are already located in index #0 (or #1 if LE).
    if (ISD == ISD::EXTRACT_VECTOR_ELT &&
        Index == (ST->isLittleEndian() ? 1 : 0))
      return 0;

    return Cost;

  } else if (ST->hasQPX() && Val->getScalarType()->isFloatingPointTy()) {
    // Floating point scalars are already located in index #0.
    if (Index == 0)
      return 0;

    return Cost;
  }

  // Estimated cost of a load-hit-store delay.  This was obtained
  // experimentally as a minimum needed to prevent unprofitable
  // vectorization for the paq8p benchmark.  It may need to be
  // raised further if other unprofitable cases remain.
  unsigned LHSPenalty = 2;
  if (ISD == ISD::INSERT_VECTOR_ELT)
    LHSPenalty += 7;

  // Vector element insert/extract with Altivec is very expensive,
  // because they require store and reload with the attendant
  // processor stall for load-hit-store.  Until VSX is available,
  // these need to be estimated as very costly.
  if (ISD == ISD::EXTRACT_VECTOR_ELT ||
      ISD == ISD::INSERT_VECTOR_ELT)
    return LHSPenalty + Cost;

  return Cost;
}

int PPCTTIImpl::getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
                                unsigned AddressSpace, const Instruction *I) {
  // Legalize the type.
  std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Src);
  assert((Opcode == Instruction::Load || Opcode == Instruction::Store) &&
         "Invalid Opcode");

  int Cost = BaseT::getMemoryOpCost(Opcode, Src, Alignment, AddressSpace);
  Cost = vectorCostAdjustment(Cost, Opcode, Src, nullptr);

  bool IsAltivecType = ST->hasAltivec() &&
                       (LT.second == MVT::v16i8 || LT.second == MVT::v8i16 ||
                        LT.second == MVT::v4i32 || LT.second == MVT::v4f32);
  bool IsVSXType = ST->hasVSX() &&
                   (LT.second == MVT::v2f64 || LT.second == MVT::v2i64);
  bool IsQPXType = ST->hasQPX() &&
                   (LT.second == MVT::v4f64 || LT.second == MVT::v4f32);

  // VSX has 32b/64b load instructions. Legalization can handle loading of
  // 32b/64b to VSR correctly and cheaply. But BaseT::getMemoryOpCost and
  // PPCTargetLowering can't compute the cost appropriately. So here we
  // explicitly check this case.
  unsigned MemBytes = Src->getPrimitiveSizeInBits();
  if (Opcode == Instruction::Load && ST->hasVSX() && IsAltivecType &&
      (MemBytes == 64 || (ST->hasP8Vector() && MemBytes == 32)))
    return 1;

  // Aligned loads and stores are easy.
  unsigned SrcBytes = LT.second.getStoreSize();
  if (!SrcBytes || !Alignment || Alignment >= SrcBytes)
    return Cost;

  // If we can use the permutation-based load sequence, then this is also
  // relatively cheap (not counting loop-invariant instructions): one load plus
  // one permute (the last load in a series has extra cost, but we're
  // neglecting that here). Note that on the P7, we could do unaligned loads
  // for Altivec types using the VSX instructions, but that's more expensive
  // than using the permutation-based load sequence. On the P8, that's no
  // longer true.
  if (Opcode == Instruction::Load &&
      ((!ST->hasP8Vector() && IsAltivecType) || IsQPXType) &&
      Alignment >= LT.second.getScalarType().getStoreSize())
    return Cost + LT.first; // Add the cost of the permutations.

  // For VSX, we can do unaligned loads and stores on Altivec/VSX types. On the
  // P7, unaligned vector loads are more expensive than the permutation-based
  // load sequence, so that might be used instead, but regardless, the net cost
  // is about the same (not counting loop-invariant instructions).
  if (IsVSXType || (ST->hasVSX() && IsAltivecType))
    return Cost;

  // Newer PPC supports unaligned memory access.
  if (TLI->allowsMisalignedMemoryAccesses(LT.second, 0))
    return Cost;

  // PPC in general does not support unaligned loads and stores. They'll need
  // to be decomposed based on the alignment factor.

  // Add the cost of each scalar load or store.
  Cost += LT.first*(SrcBytes/Alignment-1);

  // For a vector type, there is also scalarization overhead (only for
  // stores, loads are expanded using the vector-load + permutation sequence,
  // which is much less expensive).
  if (Src->isVectorTy() && Opcode == Instruction::Store)
    for (int i = 0, e = Src->getVectorNumElements(); i < e; ++i)
      Cost += getVectorInstrCost(Instruction::ExtractElement, Src, i);

  return Cost;
}

int PPCTTIImpl::getInterleavedMemoryOpCost(unsigned Opcode, Type *VecTy,
                                           unsigned Factor,
                                           ArrayRef<unsigned> Indices,
                                           unsigned Alignment,
                                           unsigned AddressSpace,
                                           bool UseMaskForCond,
                                           bool UseMaskForGaps) {
  if (UseMaskForCond || UseMaskForGaps)
    return BaseT::getInterleavedMemoryOpCost(Opcode, VecTy, Factor, Indices,
                                             Alignment, AddressSpace,
                                             UseMaskForCond, UseMaskForGaps);

  assert(isa<VectorType>(VecTy) &&
         "Expect a vector type for interleaved memory op");

  // Legalize the type.
  std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, VecTy);

  // Firstly, the cost of load/store operation.
  int Cost = getMemoryOpCost(Opcode, VecTy, Alignment, AddressSpace);

  // PPC, for both Altivec/VSX and QPX, support cheap arbitrary permutations
  // (at least in the sense that there need only be one non-loop-invariant
  // instruction). For each result vector, we need one shuffle per incoming
  // vector (except that the first shuffle can take two incoming vectors
  // because it does not need to take itself).
  Cost += Factor*(LT.first-1);

  return Cost;
}