llvm.org GIT mirror llvm / release_90 lib / Target / ARM / ARMTargetTransformInfo.cpp
release_90

Tree @release_90 (Download .tar.gz)

ARMTargetTransformInfo.cpp @release_90raw · 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
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
//===- ARMTargetTransformInfo.cpp - ARM 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 "ARMTargetTransformInfo.h"
#include "ARMSubtarget.h"
#include "MCTargetDesc/ARMAddressingModes.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/CodeGen/CostTable.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/CallSite.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Type.h"
#include "llvm/MC/SubtargetFeature.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/MachineValueType.h"
#include "llvm/Target/TargetMachine.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <utility>

using namespace llvm;

#define DEBUG_TYPE "armtti"

static cl::opt<bool> DisableLowOverheadLoops(
  "disable-arm-loloops", cl::Hidden, cl::init(true),
  cl::desc("Disable the generation of low-overhead loops"));

bool ARMTTIImpl::areInlineCompatible(const Function *Caller,
                                     const Function *Callee) const {
  const TargetMachine &TM = getTLI()->getTargetMachine();
  const FeatureBitset &CallerBits =
      TM.getSubtargetImpl(*Caller)->getFeatureBits();
  const FeatureBitset &CalleeBits =
      TM.getSubtargetImpl(*Callee)->getFeatureBits();

  // To inline a callee, all features not in the whitelist must match exactly.
  bool MatchExact = (CallerBits & ~InlineFeatureWhitelist) ==
                    (CalleeBits & ~InlineFeatureWhitelist);
  // For features in the whitelist, the callee's features must be a subset of
  // the callers'.
  bool MatchSubset = ((CallerBits & CalleeBits) & InlineFeatureWhitelist) ==
                     (CalleeBits & InlineFeatureWhitelist);
  return MatchExact && MatchSubset;
}

int ARMTTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
  assert(Ty->isIntegerTy());

 unsigned Bits = Ty->getPrimitiveSizeInBits();
 if (Bits == 0 || Imm.getActiveBits() >= 64)
   return 4;

  int64_t SImmVal = Imm.getSExtValue();
  uint64_t ZImmVal = Imm.getZExtValue();
  if (!ST->isThumb()) {
    if ((SImmVal >= 0 && SImmVal < 65536) ||
        (ARM_AM::getSOImmVal(ZImmVal) != -1) ||
        (ARM_AM::getSOImmVal(~ZImmVal) != -1))
      return 1;
    return ST->hasV6T2Ops() ? 2 : 3;
  }
  if (ST->isThumb2()) {
    if ((SImmVal >= 0 && SImmVal < 65536) ||
        (ARM_AM::getT2SOImmVal(ZImmVal) != -1) ||
        (ARM_AM::getT2SOImmVal(~ZImmVal) != -1))
      return 1;
    return ST->hasV6T2Ops() ? 2 : 3;
  }
  // Thumb1, any i8 imm cost 1.
  if (Bits == 8 || (SImmVal >= 0 && SImmVal < 256))
    return 1;
  if ((~SImmVal < 256) || ARM_AM::isThumbImmShiftedVal(ZImmVal))
    return 2;
  // Load from constantpool.
  return 3;
}

// Constants smaller than 256 fit in the immediate field of
// Thumb1 instructions so we return a zero cost and 1 otherwise.
int ARMTTIImpl::getIntImmCodeSizeCost(unsigned Opcode, unsigned Idx,
                                      const APInt &Imm, Type *Ty) {
  if (Imm.isNonNegative() && Imm.getLimitedValue() < 256)
    return 0;

  return 1;
}

int ARMTTIImpl::getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
                              Type *Ty) {
  // Division by a constant can be turned into multiplication, but only if we
  // know it's constant. So it's not so much that the immediate is cheap (it's
  // not), but that the alternative is worse.
  // FIXME: this is probably unneeded with GlobalISel.
  if ((Opcode == Instruction::SDiv || Opcode == Instruction::UDiv ||
       Opcode == Instruction::SRem || Opcode == Instruction::URem) &&
      Idx == 1)
    return 0;

  if (Opcode == Instruction::And) {
    // UXTB/UXTH
    if (Imm == 255 || Imm == 65535)
      return 0;
    // Conversion to BIC is free, and means we can use ~Imm instead.
    return std::min(getIntImmCost(Imm, Ty), getIntImmCost(~Imm, Ty));
  }

  if (Opcode == Instruction::Add)
    // Conversion to SUB is free, and means we can use -Imm instead.
    return std::min(getIntImmCost(Imm, Ty), getIntImmCost(-Imm, Ty));

  if (Opcode == Instruction::ICmp && Imm.isNegative() &&
      Ty->getIntegerBitWidth() == 32) {
    int64_t NegImm = -Imm.getSExtValue();
    if (ST->isThumb2() && NegImm < 1<<12)
      // icmp X, #-C -> cmn X, #C
      return 0;
    if (ST->isThumb() && NegImm < 1<<8)
      // icmp X, #-C -> adds X, #C
      return 0;
  }

  // xor a, -1 can always be folded to MVN
  if (Opcode == Instruction::Xor && Imm.isAllOnesValue())
    return 0;

  return getIntImmCost(Imm, Ty);
}

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

  // Single to/from double precision conversions.
  static const CostTblEntry NEONFltDblTbl[] = {
    // Vector fptrunc/fpext conversions.
    { ISD::FP_ROUND,   MVT::v2f64, 2 },
    { ISD::FP_EXTEND,  MVT::v2f32, 2 },
    { ISD::FP_EXTEND,  MVT::v4f32, 4 }
  };

  if (Src->isVectorTy() && ST->hasNEON() && (ISD == ISD::FP_ROUND ||
                                          ISD == ISD::FP_EXTEND)) {
    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Src);
    if (const auto *Entry = CostTableLookup(NEONFltDblTbl, ISD, LT.second))
      return LT.first * Entry->Cost;
  }

  EVT SrcTy = TLI->getValueType(DL, Src);
  EVT DstTy = TLI->getValueType(DL, Dst);

  if (!SrcTy.isSimple() || !DstTy.isSimple())
    return BaseT::getCastInstrCost(Opcode, Dst, Src);

  // Some arithmetic, load and store operations have specific instructions
  // to cast up/down their types automatically at no extra cost.
  // TODO: Get these tables to know at least what the related operations are.
  static const TypeConversionCostTblEntry NEONVectorConversionTbl[] = {
    { ISD::SIGN_EXTEND, MVT::v4i32, MVT::v4i16, 0 },
    { ISD::ZERO_EXTEND, MVT::v4i32, MVT::v4i16, 0 },
    { ISD::SIGN_EXTEND, MVT::v2i64, MVT::v2i32, 1 },
    { ISD::ZERO_EXTEND, MVT::v2i64, MVT::v2i32, 1 },
    { ISD::TRUNCATE,    MVT::v4i32, MVT::v4i64, 0 },
    { ISD::TRUNCATE,    MVT::v4i16, MVT::v4i32, 1 },

    // The number of vmovl instructions for the extension.
    { ISD::SIGN_EXTEND, MVT::v4i64, MVT::v4i16, 3 },
    { ISD::ZERO_EXTEND, MVT::v4i64, MVT::v4i16, 3 },
    { ISD::SIGN_EXTEND, MVT::v8i32, MVT::v8i8, 3 },
    { ISD::ZERO_EXTEND, MVT::v8i32, MVT::v8i8, 3 },
    { ISD::SIGN_EXTEND, MVT::v8i64, MVT::v8i8, 7 },
    { ISD::ZERO_EXTEND, MVT::v8i64, MVT::v8i8, 7 },
    { ISD::SIGN_EXTEND, MVT::v8i64, MVT::v8i16, 6 },
    { ISD::ZERO_EXTEND, MVT::v8i64, MVT::v8i16, 6 },
    { ISD::SIGN_EXTEND, MVT::v16i32, MVT::v16i8, 6 },
    { ISD::ZERO_EXTEND, MVT::v16i32, MVT::v16i8, 6 },

    // Operations that we legalize using splitting.
    { ISD::TRUNCATE,    MVT::v16i8, MVT::v16i32, 6 },
    { ISD::TRUNCATE,    MVT::v8i8, MVT::v8i32, 3 },

    // Vector float <-> i32 conversions.
    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i32, 1 },
    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i32, 1 },

    { ISD::SINT_TO_FP,  MVT::v2f32, MVT::v2i8, 3 },
    { ISD::UINT_TO_FP,  MVT::v2f32, MVT::v2i8, 3 },
    { ISD::SINT_TO_FP,  MVT::v2f32, MVT::v2i16, 2 },
    { ISD::UINT_TO_FP,  MVT::v2f32, MVT::v2i16, 2 },
    { ISD::SINT_TO_FP,  MVT::v2f32, MVT::v2i32, 1 },
    { ISD::UINT_TO_FP,  MVT::v2f32, MVT::v2i32, 1 },
    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i1, 3 },
    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i1, 3 },
    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i8, 3 },
    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i8, 3 },
    { ISD::SINT_TO_FP,  MVT::v4f32, MVT::v4i16, 2 },
    { ISD::UINT_TO_FP,  MVT::v4f32, MVT::v4i16, 2 },
    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i16, 4 },
    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i16, 4 },
    { ISD::SINT_TO_FP,  MVT::v8f32, MVT::v8i32, 2 },
    { ISD::UINT_TO_FP,  MVT::v8f32, MVT::v8i32, 2 },
    { ISD::SINT_TO_FP,  MVT::v16f32, MVT::v16i16, 8 },
    { ISD::UINT_TO_FP,  MVT::v16f32, MVT::v16i16, 8 },
    { ISD::SINT_TO_FP,  MVT::v16f32, MVT::v16i32, 4 },
    { ISD::UINT_TO_FP,  MVT::v16f32, MVT::v16i32, 4 },

    { ISD::FP_TO_SINT,  MVT::v4i32, MVT::v4f32, 1 },
    { ISD::FP_TO_UINT,  MVT::v4i32, MVT::v4f32, 1 },
    { ISD::FP_TO_SINT,  MVT::v4i8, MVT::v4f32, 3 },
    { ISD::FP_TO_UINT,  MVT::v4i8, MVT::v4f32, 3 },
    { ISD::FP_TO_SINT,  MVT::v4i16, MVT::v4f32, 2 },
    { ISD::FP_TO_UINT,  MVT::v4i16, MVT::v4f32, 2 },

    // Vector double <-> i32 conversions.
    { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },
    { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },

    { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i8, 4 },
    { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i8, 4 },
    { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i16, 3 },
    { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i16, 3 },
    { ISD::SINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },
    { ISD::UINT_TO_FP,  MVT::v2f64, MVT::v2i32, 2 },

    { ISD::FP_TO_SINT,  MVT::v2i32, MVT::v2f64, 2 },
    { ISD::FP_TO_UINT,  MVT::v2i32, MVT::v2f64, 2 },
    { ISD::FP_TO_SINT,  MVT::v8i16, MVT::v8f32, 4 },
    { ISD::FP_TO_UINT,  MVT::v8i16, MVT::v8f32, 4 },
    { ISD::FP_TO_SINT,  MVT::v16i16, MVT::v16f32, 8 },
    { ISD::FP_TO_UINT,  MVT::v16i16, MVT::v16f32, 8 }
  };

  if (SrcTy.isVector() && ST->hasNEON()) {
    if (const auto *Entry = ConvertCostTableLookup(NEONVectorConversionTbl, ISD,
                                                   DstTy.getSimpleVT(),
                                                   SrcTy.getSimpleVT()))
      return Entry->Cost;
  }

  // Scalar float to integer conversions.
  static const TypeConversionCostTblEntry NEONFloatConversionTbl[] = {
    { ISD::FP_TO_SINT,  MVT::i1, MVT::f32, 2 },
    { ISD::FP_TO_UINT,  MVT::i1, MVT::f32, 2 },
    { ISD::FP_TO_SINT,  MVT::i1, MVT::f64, 2 },
    { ISD::FP_TO_UINT,  MVT::i1, MVT::f64, 2 },
    { ISD::FP_TO_SINT,  MVT::i8, MVT::f32, 2 },
    { ISD::FP_TO_UINT,  MVT::i8, MVT::f32, 2 },
    { ISD::FP_TO_SINT,  MVT::i8, MVT::f64, 2 },
    { ISD::FP_TO_UINT,  MVT::i8, MVT::f64, 2 },
    { ISD::FP_TO_SINT,  MVT::i16, MVT::f32, 2 },
    { ISD::FP_TO_UINT,  MVT::i16, MVT::f32, 2 },
    { ISD::FP_TO_SINT,  MVT::i16, MVT::f64, 2 },
    { ISD::FP_TO_UINT,  MVT::i16, MVT::f64, 2 },
    { ISD::FP_TO_SINT,  MVT::i32, MVT::f32, 2 },
    { ISD::FP_TO_UINT,  MVT::i32, MVT::f32, 2 },
    { ISD::FP_TO_SINT,  MVT::i32, MVT::f64, 2 },
    { ISD::FP_TO_UINT,  MVT::i32, MVT::f64, 2 },
    { ISD::FP_TO_SINT,  MVT::i64, MVT::f32, 10 },
    { ISD::FP_TO_UINT,  MVT::i64, MVT::f32, 10 },
    { ISD::FP_TO_SINT,  MVT::i64, MVT::f64, 10 },
    { ISD::FP_TO_UINT,  MVT::i64, MVT::f64, 10 }
  };
  if (SrcTy.isFloatingPoint() && ST->hasNEON()) {
    if (const auto *Entry = ConvertCostTableLookup(NEONFloatConversionTbl, ISD,
                                                   DstTy.getSimpleVT(),
                                                   SrcTy.getSimpleVT()))
      return Entry->Cost;
  }

  // Scalar integer to float conversions.
  static const TypeConversionCostTblEntry NEONIntegerConversionTbl[] = {
    { ISD::SINT_TO_FP,  MVT::f32, MVT::i1, 2 },
    { ISD::UINT_TO_FP,  MVT::f32, MVT::i1, 2 },
    { ISD::SINT_TO_FP,  MVT::f64, MVT::i1, 2 },
    { ISD::UINT_TO_FP,  MVT::f64, MVT::i1, 2 },
    { ISD::SINT_TO_FP,  MVT::f32, MVT::i8, 2 },
    { ISD::UINT_TO_FP,  MVT::f32, MVT::i8, 2 },
    { ISD::SINT_TO_FP,  MVT::f64, MVT::i8, 2 },
    { ISD::UINT_TO_FP,  MVT::f64, MVT::i8, 2 },
    { ISD::SINT_TO_FP,  MVT::f32, MVT::i16, 2 },
    { ISD::UINT_TO_FP,  MVT::f32, MVT::i16, 2 },
    { ISD::SINT_TO_FP,  MVT::f64, MVT::i16, 2 },
    { ISD::UINT_TO_FP,  MVT::f64, MVT::i16, 2 },
    { ISD::SINT_TO_FP,  MVT::f32, MVT::i32, 2 },
    { ISD::UINT_TO_FP,  MVT::f32, MVT::i32, 2 },
    { ISD::SINT_TO_FP,  MVT::f64, MVT::i32, 2 },
    { ISD::UINT_TO_FP,  MVT::f64, MVT::i32, 2 },
    { ISD::SINT_TO_FP,  MVT::f32, MVT::i64, 10 },
    { ISD::UINT_TO_FP,  MVT::f32, MVT::i64, 10 },
    { ISD::SINT_TO_FP,  MVT::f64, MVT::i64, 10 },
    { ISD::UINT_TO_FP,  MVT::f64, MVT::i64, 10 }
  };

  if (SrcTy.isInteger() && ST->hasNEON()) {
    if (const auto *Entry = ConvertCostTableLookup(NEONIntegerConversionTbl,
                                                   ISD, DstTy.getSimpleVT(),
                                                   SrcTy.getSimpleVT()))
      return Entry->Cost;
  }

  // Scalar integer conversion costs.
  static const TypeConversionCostTblEntry ARMIntegerConversionTbl[] = {
    // i16 -> i64 requires two dependent operations.
    { ISD::SIGN_EXTEND, MVT::i64, MVT::i16, 2 },

    // Truncates on i64 are assumed to be free.
    { ISD::TRUNCATE,    MVT::i32, MVT::i64, 0 },
    { ISD::TRUNCATE,    MVT::i16, MVT::i64, 0 },
    { ISD::TRUNCATE,    MVT::i8,  MVT::i64, 0 },
    { ISD::TRUNCATE,    MVT::i1,  MVT::i64, 0 }
  };

  if (SrcTy.isInteger()) {
    if (const auto *Entry = ConvertCostTableLookup(ARMIntegerConversionTbl, ISD,
                                                   DstTy.getSimpleVT(),
                                                   SrcTy.getSimpleVT()))
      return Entry->Cost;
  }

  return BaseT::getCastInstrCost(Opcode, Dst, Src);
}

int ARMTTIImpl::getVectorInstrCost(unsigned Opcode, Type *ValTy,
                                   unsigned Index) {
  // Penalize inserting into an D-subregister. We end up with a three times
  // lower estimated throughput on swift.
  if (ST->hasSlowLoadDSubregister() && Opcode == Instruction::InsertElement &&
      ValTy->isVectorTy() && ValTy->getScalarSizeInBits() <= 32)
    return 3;

  if ((Opcode == Instruction::InsertElement ||
       Opcode == Instruction::ExtractElement)) {
    // Cross-class copies are expensive on many microarchitectures,
    // so assume they are expensive by default.
    if (ValTy->getVectorElementType()->isIntegerTy())
      return 3;

    // Even if it's not a cross class copy, this likely leads to mixing
    // of NEON and VFP code and should be therefore penalized.
    if (ValTy->isVectorTy() &&
        ValTy->getScalarSizeInBits() <= 32)
      return std::max(BaseT::getVectorInstrCost(Opcode, ValTy, Index), 2U);
  }

  return BaseT::getVectorInstrCost(Opcode, ValTy, Index);
}

int ARMTTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy,
                                   const Instruction *I) {
  int ISD = TLI->InstructionOpcodeToISD(Opcode);
  // On NEON a vector select gets lowered to vbsl.
  if (ST->hasNEON() && ValTy->isVectorTy() && ISD == ISD::SELECT) {
    // Lowering of some vector selects is currently far from perfect.
    static const TypeConversionCostTblEntry NEONVectorSelectTbl[] = {
      { ISD::SELECT, MVT::v4i1, MVT::v4i64, 4*4 + 1*2 + 1 },
      { ISD::SELECT, MVT::v8i1, MVT::v8i64, 50 },
      { ISD::SELECT, MVT::v16i1, MVT::v16i64, 100 }
    };

    EVT SelCondTy = TLI->getValueType(DL, CondTy);
    EVT SelValTy = TLI->getValueType(DL, ValTy);
    if (SelCondTy.isSimple() && SelValTy.isSimple()) {
      if (const auto *Entry = ConvertCostTableLookup(NEONVectorSelectTbl, ISD,
                                                     SelCondTy.getSimpleVT(),
                                                     SelValTy.getSimpleVT()))
        return Entry->Cost;
    }

    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, ValTy);
    return LT.first;
  }

  return BaseT::getCmpSelInstrCost(Opcode, ValTy, CondTy, I);
}

int ARMTTIImpl::getAddressComputationCost(Type *Ty, ScalarEvolution *SE,
                                          const SCEV *Ptr) {
  // Address computations in vectorized code with non-consecutive addresses will
  // likely result in more instructions compared to scalar code where the
  // computation can more often be merged into the index mode. The resulting
  // extra micro-ops can significantly decrease throughput.
  unsigned NumVectorInstToHideOverhead = 10;
  int MaxMergeDistance = 64;

  if (Ty->isVectorTy() && SE &&
      !BaseT::isConstantStridedAccessLessThan(SE, Ptr, MaxMergeDistance + 1))
    return NumVectorInstToHideOverhead;

  // In many cases the address computation is not merged into the instruction
  // addressing mode.
  return 1;
}

int ARMTTIImpl::getMemcpyCost(const Instruction *I) {
  const MemCpyInst *MI = dyn_cast<MemCpyInst>(I);
  assert(MI && "MemcpyInst expected");
  ConstantInt *C = dyn_cast<ConstantInt>(MI->getLength());

  // To model the cost of a library call, we assume 1 for the call, and
  // 3 for the argument setup.
  const unsigned LibCallCost = 4;

  // If 'size' is not a constant, a library call will be generated.
  if (!C)
    return LibCallCost;

  const unsigned Size = C->getValue().getZExtValue();
  const unsigned DstAlign = MI->getDestAlignment();
  const unsigned SrcAlign = MI->getSourceAlignment();
  const Function *F = I->getParent()->getParent();
  const unsigned Limit = TLI->getMaxStoresPerMemmove(F->hasMinSize());
  std::vector<EVT> MemOps;

  // MemOps will be poplulated with a list of data types that needs to be
  // loaded and stored. That's why we multiply the number of elements by 2 to
  // get the cost for this memcpy.
  if (getTLI()->findOptimalMemOpLowering(
          MemOps, Limit, Size, DstAlign, SrcAlign, false /*IsMemset*/,
          false /*ZeroMemset*/, false /*MemcpyStrSrc*/, false /*AllowOverlap*/,
          MI->getDestAddressSpace(), MI->getSourceAddressSpace(),
          F->getAttributes()))
    return MemOps.size() * 2;

  // If we can't find an optimal memop lowering, return the default cost
  return LibCallCost;
}

int ARMTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index,
                               Type *SubTp) {
  if (Kind == TTI::SK_Broadcast) {
    static const CostTblEntry NEONDupTbl[] = {
        // VDUP handles these cases.
        {ISD::VECTOR_SHUFFLE, MVT::v2i32, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2f32, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2i64, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2f64, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v4i16, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v8i8,  1},

        {ISD::VECTOR_SHUFFLE, MVT::v4i32, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v4f32, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v8i16, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v16i8, 1}};

    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);

    if (const auto *Entry = CostTableLookup(NEONDupTbl, ISD::VECTOR_SHUFFLE,
                                            LT.second))
      return LT.first * Entry->Cost;

    return BaseT::getShuffleCost(Kind, Tp, Index, SubTp);
  }
  if (Kind == TTI::SK_Reverse) {
    static const CostTblEntry NEONShuffleTbl[] = {
        // Reverse shuffle cost one instruction if we are shuffling within a
        // double word (vrev) or two if we shuffle a quad word (vrev, vext).
        {ISD::VECTOR_SHUFFLE, MVT::v2i32, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2f32, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2i64, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2f64, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v4i16, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v8i8,  1},

        {ISD::VECTOR_SHUFFLE, MVT::v4i32, 2},
        {ISD::VECTOR_SHUFFLE, MVT::v4f32, 2},
        {ISD::VECTOR_SHUFFLE, MVT::v8i16, 2},
        {ISD::VECTOR_SHUFFLE, MVT::v16i8, 2}};

    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);

    if (const auto *Entry = CostTableLookup(NEONShuffleTbl, ISD::VECTOR_SHUFFLE,
                                            LT.second))
      return LT.first * Entry->Cost;

    return BaseT::getShuffleCost(Kind, Tp, Index, SubTp);
  }
  if (Kind == TTI::SK_Select) {
    static const CostTblEntry NEONSelShuffleTbl[] = {
        // Select shuffle cost table for ARM. Cost is the number of instructions
        // required to create the shuffled vector.

        {ISD::VECTOR_SHUFFLE, MVT::v2f32, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2i64, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2f64, 1},
        {ISD::VECTOR_SHUFFLE, MVT::v2i32, 1},

        {ISD::VECTOR_SHUFFLE, MVT::v4i32, 2},
        {ISD::VECTOR_SHUFFLE, MVT::v4f32, 2},
        {ISD::VECTOR_SHUFFLE, MVT::v4i16, 2},

        {ISD::VECTOR_SHUFFLE, MVT::v8i16, 16},

        {ISD::VECTOR_SHUFFLE, MVT::v16i8, 32}};

    std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);
    if (const auto *Entry = CostTableLookup(NEONSelShuffleTbl,
                                            ISD::VECTOR_SHUFFLE, LT.second))
      return LT.first * Entry->Cost;
    return BaseT::getShuffleCost(Kind, Tp, Index, SubTp);
  }
  return BaseT::getShuffleCost(Kind, Tp, Index, SubTp);
}

int ARMTTIImpl::getArithmeticInstrCost(
    unsigned Opcode, Type *Ty, TTI::OperandValueKind Op1Info,
    TTI::OperandValueKind Op2Info, TTI::OperandValueProperties Opd1PropInfo,
    TTI::OperandValueProperties Opd2PropInfo,
    ArrayRef<const Value *> Args) {
  int ISDOpcode = TLI->InstructionOpcodeToISD(Opcode);
  std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Ty);

  const unsigned FunctionCallDivCost = 20;
  const unsigned ReciprocalDivCost = 10;
  static const CostTblEntry CostTbl[] = {
    // Division.
    // These costs are somewhat random. Choose a cost of 20 to indicate that
    // vectorizing devision (added function call) is going to be very expensive.
    // Double registers types.
    { ISD::SDIV, MVT::v1i64, 1 * FunctionCallDivCost},
    { ISD::UDIV, MVT::v1i64, 1 * FunctionCallDivCost},
    { ISD::SREM, MVT::v1i64, 1 * FunctionCallDivCost},
    { ISD::UREM, MVT::v1i64, 1 * FunctionCallDivCost},
    { ISD::SDIV, MVT::v2i32, 2 * FunctionCallDivCost},
    { ISD::UDIV, MVT::v2i32, 2 * FunctionCallDivCost},
    { ISD::SREM, MVT::v2i32, 2 * FunctionCallDivCost},
    { ISD::UREM, MVT::v2i32, 2 * FunctionCallDivCost},
    { ISD::SDIV, MVT::v4i16,     ReciprocalDivCost},
    { ISD::UDIV, MVT::v4i16,     ReciprocalDivCost},
    { ISD::SREM, MVT::v4i16, 4 * FunctionCallDivCost},
    { ISD::UREM, MVT::v4i16, 4 * FunctionCallDivCost},
    { ISD::SDIV, MVT::v8i8,      ReciprocalDivCost},
    { ISD::UDIV, MVT::v8i8,      ReciprocalDivCost},
    { ISD::SREM, MVT::v8i8,  8 * FunctionCallDivCost},
    { ISD::UREM, MVT::v8i8,  8 * FunctionCallDivCost},
    // Quad register types.
    { ISD::SDIV, MVT::v2i64, 2 * FunctionCallDivCost},
    { ISD::UDIV, MVT::v2i64, 2 * FunctionCallDivCost},
    { ISD::SREM, MVT::v2i64, 2 * FunctionCallDivCost},
    { ISD::UREM, MVT::v2i64, 2 * FunctionCallDivCost},
    { ISD::SDIV, MVT::v4i32, 4 * FunctionCallDivCost},
    { ISD::UDIV, MVT::v4i32, 4 * FunctionCallDivCost},
    { ISD::SREM, MVT::v4i32, 4 * FunctionCallDivCost},
    { ISD::UREM, MVT::v4i32, 4 * FunctionCallDivCost},
    { ISD::SDIV, MVT::v8i16, 8 * FunctionCallDivCost},
    { ISD::UDIV, MVT::v8i16, 8 * FunctionCallDivCost},
    { ISD::SREM, MVT::v8i16, 8 * FunctionCallDivCost},
    { ISD::UREM, MVT::v8i16, 8 * FunctionCallDivCost},
    { ISD::SDIV, MVT::v16i8, 16 * FunctionCallDivCost},
    { ISD::UDIV, MVT::v16i8, 16 * FunctionCallDivCost},
    { ISD::SREM, MVT::v16i8, 16 * FunctionCallDivCost},
    { ISD::UREM, MVT::v16i8, 16 * FunctionCallDivCost},
    // Multiplication.
  };

  if (ST->hasNEON())
    if (const auto *Entry = CostTableLookup(CostTbl, ISDOpcode, LT.second))
      return LT.first * Entry->Cost;

  int Cost = BaseT::getArithmeticInstrCost(Opcode, Ty, Op1Info, Op2Info,
                                           Opd1PropInfo, Opd2PropInfo);

  // This is somewhat of a hack. The problem that we are facing is that SROA
  // creates a sequence of shift, and, or instructions to construct values.
  // These sequences are recognized by the ISel and have zero-cost. Not so for
  // the vectorized code. Because we have support for v2i64 but not i64 those
  // sequences look particularly beneficial to vectorize.
  // To work around this we increase the cost of v2i64 operations to make them
  // seem less beneficial.
  if (LT.second == MVT::v2i64 &&
      Op2Info == TargetTransformInfo::OK_UniformConstantValue)
    Cost += 4;

  return Cost;
}

int ARMTTIImpl::getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
                                unsigned AddressSpace, const Instruction *I) {
  std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Src);

  if (Src->isVectorTy() && Alignment != 16 &&
      Src->getVectorElementType()->isDoubleTy()) {
    // Unaligned loads/stores are extremely inefficient.
    // We need 4 uops for vst.1/vld.1 vs 1uop for vldr/vstr.
    return LT.first * 4;
  }
  return LT.first;
}

int ARMTTIImpl::getInterleavedMemoryOpCost(unsigned Opcode, Type *VecTy,
                                           unsigned Factor,
                                           ArrayRef<unsigned> Indices,
                                           unsigned Alignment,
                                           unsigned AddressSpace,
                                           bool UseMaskForCond,
                                           bool UseMaskForGaps) {
  assert(Factor >= 2 && "Invalid interleave factor");
  assert(isa<VectorType>(VecTy) && "Expect a vector type");

  // vldN/vstN doesn't support vector types of i64/f64 element.
  bool EltIs64Bits = DL.getTypeSizeInBits(VecTy->getScalarType()) == 64;

  if (Factor <= TLI->getMaxSupportedInterleaveFactor() && !EltIs64Bits &&
      !UseMaskForCond && !UseMaskForGaps) {
    unsigned NumElts = VecTy->getVectorNumElements();
    auto *SubVecTy = VectorType::get(VecTy->getScalarType(), NumElts / Factor);

    // vldN/vstN only support legal vector types of size 64 or 128 in bits.
    // Accesses having vector types that are a multiple of 128 bits can be
    // matched to more than one vldN/vstN instruction.
    if (NumElts % Factor == 0 &&
        TLI->isLegalInterleavedAccessType(SubVecTy, DL))
      return Factor * TLI->getNumInterleavedAccesses(SubVecTy, DL);
  }

  return BaseT::getInterleavedMemoryOpCost(Opcode, VecTy, Factor, Indices,
                                           Alignment, AddressSpace,
                                           UseMaskForCond, UseMaskForGaps);
}

bool ARMTTIImpl::isLoweredToCall(const Function *F) {
  if (!F->isIntrinsic())
    BaseT::isLoweredToCall(F);

  // Assume all Arm-specific intrinsics map to an instruction.
  if (F->getName().startswith("llvm.arm"))
    return false;

  switch (F->getIntrinsicID()) {
  default: break;
  case Intrinsic::powi:
  case Intrinsic::sin:
  case Intrinsic::cos:
  case Intrinsic::pow:
  case Intrinsic::log:
  case Intrinsic::log10:
  case Intrinsic::log2:
  case Intrinsic::exp:
  case Intrinsic::exp2:
    return true;
  case Intrinsic::sqrt:
  case Intrinsic::fabs:
  case Intrinsic::copysign:
  case Intrinsic::floor:
  case Intrinsic::ceil:
  case Intrinsic::trunc:
  case Intrinsic::rint:
  case Intrinsic::nearbyint:
  case Intrinsic::round:
  case Intrinsic::canonicalize:
  case Intrinsic::lround:
  case Intrinsic::llround:
  case Intrinsic::lrint:
  case Intrinsic::llrint:
    if (F->getReturnType()->isDoubleTy() && !ST->hasFP64())
      return true;
    if (F->getReturnType()->isHalfTy() && !ST->hasFullFP16())
      return true;
    // Some operations can be handled by vector instructions and assume
    // unsupported vectors will be expanded into supported scalar ones.
    // TODO Handle scalar operations properly.
    return !ST->hasFPARMv8Base() && !ST->hasVFP2Base();
  case Intrinsic::masked_store:
  case Intrinsic::masked_load:
  case Intrinsic::masked_gather:
  case Intrinsic::masked_scatter:
    return !ST->hasMVEIntegerOps();
  case Intrinsic::sadd_with_overflow:
  case Intrinsic::uadd_with_overflow:
  case Intrinsic::ssub_with_overflow:
  case Intrinsic::usub_with_overflow:
  case Intrinsic::sadd_sat:
  case Intrinsic::uadd_sat:
  case Intrinsic::ssub_sat:
  case Intrinsic::usub_sat:
    return false;
  }

  return BaseT::isLoweredToCall(F);
}

bool ARMTTIImpl::isHardwareLoopProfitable(Loop *L, ScalarEvolution &SE,
                                          AssumptionCache &AC,
                                          TargetLibraryInfo *LibInfo,
                                          HardwareLoopInfo &HWLoopInfo) {
  // Low-overhead branches are only supported in the 'low-overhead branch'
  // extension of v8.1-m.
  if (!ST->hasLOB() || DisableLowOverheadLoops)
    return false;

  if (!SE.hasLoopInvariantBackedgeTakenCount(L))
    return false;

  const SCEV *BackedgeTakenCount = SE.getBackedgeTakenCount(L);
  if (isa<SCEVCouldNotCompute>(BackedgeTakenCount))
    return false;

  const SCEV *TripCountSCEV =
    SE.getAddExpr(BackedgeTakenCount,
                  SE.getOne(BackedgeTakenCount->getType()));

  // We need to store the trip count in LR, a 32-bit register.
  if (SE.getUnsignedRangeMax(TripCountSCEV).getBitWidth() > 32)
    return false;

  // Making a call will trash LR and clear LO_BRANCH_INFO, so there's little
  // point in generating a hardware loop if that's going to happen.
  auto MaybeCall = [this](Instruction &I) {
    const ARMTargetLowering *TLI = getTLI();
    unsigned ISD = TLI->InstructionOpcodeToISD(I.getOpcode());
    EVT VT = TLI->getValueType(DL, I.getType(), true);
    if (TLI->getOperationAction(ISD, VT) == TargetLowering::LibCall)
      return true;

    // Check if an intrinsic will be lowered to a call and assume that any
    // other CallInst will generate a bl.
    if (auto *Call = dyn_cast<CallInst>(&I)) {
      if (isa<IntrinsicInst>(Call)) {
        if (const Function *F = Call->getCalledFunction())
          return isLoweredToCall(F);
      }
      return true;
    }

    // FPv5 provides conversions between integer, double-precision,
    // single-precision, and half-precision formats.
    switch (I.getOpcode()) {
    default:
      break;
    case Instruction::FPToSI:
    case Instruction::FPToUI:
    case Instruction::SIToFP:
    case Instruction::UIToFP:
    case Instruction::FPTrunc:
    case Instruction::FPExt:
      return !ST->hasFPARMv8Base();
    }

    // FIXME: Unfortunately the approach of checking the Operation Action does
    // not catch all cases of Legalization that use library calls. Our
    // Legalization step categorizes some transformations into library calls as
    // Custom, Expand or even Legal when doing type legalization. So for now
    // we have to special case for instance the SDIV of 64bit integers and the
    // use of floating point emulation.
    if (VT.isInteger() && VT.getSizeInBits() >= 64) {
      switch (ISD) {
      default:
        break;
      case ISD::SDIV:
      case ISD::UDIV:
      case ISD::SREM:
      case ISD::UREM:
      case ISD::SDIVREM:
      case ISD::UDIVREM:
        return true;
      }
    }

    // Assume all other non-float operations are supported.
    if (!VT.isFloatingPoint())
      return false;

    // We'll need a library call to handle most floats when using soft.
    if (TLI->useSoftFloat()) {
      switch (I.getOpcode()) {
      default:
        return true;
      case Instruction::Alloca:
      case Instruction::Load:
      case Instruction::Store:
      case Instruction::Select:
      case Instruction::PHI:
        return false;
      }
    }

    // We'll need a libcall to perform double precision operations on a single
    // precision only FPU.
    if (I.getType()->isDoubleTy() && !ST->hasFP64())
      return true;

    // Likewise for half precision arithmetic.
    if (I.getType()->isHalfTy() && !ST->hasFullFP16())
      return true;

    return false;
  };

  auto IsHardwareLoopIntrinsic = [](Instruction &I) {
    if (auto *Call = dyn_cast<IntrinsicInst>(&I)) {
      switch (Call->getIntrinsicID()) {
      default:
        break;
      case Intrinsic::set_loop_iterations:
      case Intrinsic::test_set_loop_iterations:
      case Intrinsic::loop_decrement:
      case Intrinsic::loop_decrement_reg:
        return true;
      }
    }
    return false;
  };

  // Scan the instructions to see if there's any that we know will turn into a
  // call or if this loop is already a low-overhead loop.
  auto ScanLoop = [&](Loop *L) {
    for (auto *BB : L->getBlocks()) {
      for (auto &I : *BB) {
        if (MaybeCall(I) || IsHardwareLoopIntrinsic(I))
          return false;
      }
    }
    return true;
  };

  // Visit inner loops.
  for (auto Inner : *L)
    if (!ScanLoop(Inner))
      return false;

  if (!ScanLoop(L))
    return false;

  // TODO: Check whether the trip count calculation is expensive. If L is the
  // inner loop but we know it has a low trip count, calculating that trip
  // count (in the parent loop) may be detrimental.

  LLVMContext &C = L->getHeader()->getContext();
  HWLoopInfo.CounterInReg = true;
  HWLoopInfo.IsNestingLegal = false;
  HWLoopInfo.PerformEntryTest = true;
  HWLoopInfo.CountType = Type::getInt32Ty(C);
  HWLoopInfo.LoopDecrement = ConstantInt::get(HWLoopInfo.CountType, 1);
  return true;
}

void ARMTTIImpl::getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                         TTI::UnrollingPreferences &UP) {
  // Only currently enable these preferences for M-Class cores.
  if (!ST->isMClass())
    return BasicTTIImplBase::getUnrollingPreferences(L, SE, UP);

  // Disable loop unrolling for Oz and Os.
  UP.OptSizeThreshold = 0;
  UP.PartialOptSizeThreshold = 0;
  if (L->getHeader()->getParent()->hasOptSize())
    return;

  // Only enable on Thumb-2 targets.
  if (!ST->isThumb2())
    return;

  SmallVector<BasicBlock*, 4> ExitingBlocks;
  L->getExitingBlocks(ExitingBlocks);
  LLVM_DEBUG(dbgs() << "Loop has:\n"
                    << "Blocks: " << L->getNumBlocks() << "\n"
                    << "Exit blocks: " << ExitingBlocks.size() << "\n");

  // Only allow another exit other than the latch. This acts as an early exit
  // as it mirrors the profitability calculation of the runtime unroller.
  if (ExitingBlocks.size() > 2)
    return;

  // Limit the CFG of the loop body for targets with a branch predictor.
  // Allowing 4 blocks permits if-then-else diamonds in the body.
  if (ST->hasBranchPredictor() && L->getNumBlocks() > 4)
    return;

  // Scan the loop: don't unroll loops with calls as this could prevent
  // inlining.
  unsigned Cost = 0;
  for (auto *BB : L->getBlocks()) {
    for (auto &I : *BB) {
      if (isa<CallInst>(I) || isa<InvokeInst>(I)) {
        ImmutableCallSite CS(&I);
        if (const Function *F = CS.getCalledFunction()) {
          if (!isLoweredToCall(F))
            continue;
        }
        return;
      }
      SmallVector<const Value*, 4> Operands(I.value_op_begin(),
                                            I.value_op_end());
      Cost += getUserCost(&I, Operands);
    }
  }

  LLVM_DEBUG(dbgs() << "Cost of loop: " << Cost << "\n");

  UP.Partial = true;
  UP.Runtime = true;
  UP.UpperBound = true;
  UP.UnrollRemainder = true;
  UP.DefaultUnrollRuntimeCount = 4;
  UP.UnrollAndJam = true;
  UP.UnrollAndJamInnerLoopThreshold = 60;

  // Force unrolling small loops can be very useful because of the branch
  // taken cost of the backedge.
  if (Cost < 12)
    UP.Force = true;
}