llvm.org GIT mirror llvm / master lib / Target / AMDGPU / AMDGPUTargetTransformInfo.cpp
master

Tree @master (Download .tar.gz)

AMDGPUTargetTransformInfo.cpp @masterraw · 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
//===- AMDGPUTargetTransformInfo.cpp - AMDGPU specific TTI pass -----------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// \file
// This file implements a TargetTransformInfo analysis pass specific to the
// AMDGPU target machine. It uses the target's detailed information to provide
// more precise answers to certain TTI queries, while letting the target
// independent and default TTI implementations handle the rest.
//
//===----------------------------------------------------------------------===//

#include "AMDGPUTargetTransformInfo.h"
#include "AMDGPUSubtarget.h"
#include "Utils/AMDGPUBaseInfo.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/Value.h"
#include "llvm/MC/SubtargetFeature.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MachineValueType.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
#include <algorithm>
#include <cassert>
#include <limits>
#include <utility>

using namespace llvm;

#define DEBUG_TYPE "AMDGPUtti"

static cl::opt<unsigned> UnrollThresholdPrivate(
  "amdgpu-unroll-threshold-private",
  cl::desc("Unroll threshold for AMDGPU if private memory used in a loop"),
  cl::init(2000), cl::Hidden);

static cl::opt<unsigned> UnrollThresholdLocal(
  "amdgpu-unroll-threshold-local",
  cl::desc("Unroll threshold for AMDGPU if local memory used in a loop"),
  cl::init(1000), cl::Hidden);

static cl::opt<unsigned> UnrollThresholdIf(
  "amdgpu-unroll-threshold-if",
  cl::desc("Unroll threshold increment for AMDGPU for each if statement inside loop"),
  cl::init(150), cl::Hidden);

static bool dependsOnLocalPhi(const Loop *L, const Value *Cond,
                              unsigned Depth = 0) {
  const Instruction *I = dyn_cast<Instruction>(Cond);
  if (!I)
    return false;

  for (const Value *V : I->operand_values()) {
    if (!L->contains(I))
      continue;
    if (const PHINode *PHI = dyn_cast<PHINode>(V)) {
      if (llvm::none_of(L->getSubLoops(), [PHI](const Loop* SubLoop) {
                  return SubLoop->contains(PHI); }))
        return true;
    } else if (Depth < 10 && dependsOnLocalPhi(L, V, Depth+1))
      return true;
  }
  return false;
}

void AMDGPUTTIImpl::getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                            TTI::UnrollingPreferences &UP) {
  UP.Threshold = 300; // Twice the default.
  UP.MaxCount = std::numeric_limits<unsigned>::max();
  UP.Partial = true;

  // TODO: Do we want runtime unrolling?

  // Maximum alloca size than can fit registers. Reserve 16 registers.
  const unsigned MaxAlloca = (256 - 16) * 4;
  unsigned ThresholdPrivate = UnrollThresholdPrivate;
  unsigned ThresholdLocal = UnrollThresholdLocal;
  unsigned MaxBoost = std::max(ThresholdPrivate, ThresholdLocal);
  for (const BasicBlock *BB : L->getBlocks()) {
    const DataLayout &DL = BB->getModule()->getDataLayout();
    unsigned LocalGEPsSeen = 0;

    if (llvm::any_of(L->getSubLoops(), [BB](const Loop* SubLoop) {
               return SubLoop->contains(BB); }))
        continue; // Block belongs to an inner loop.

    for (const Instruction &I : *BB) {
      // Unroll a loop which contains an "if" statement whose condition
      // defined by a PHI belonging to the loop. This may help to eliminate
      // if region and potentially even PHI itself, saving on both divergence
      // and registers used for the PHI.
      // Add a small bonus for each of such "if" statements.
      if (const BranchInst *Br = dyn_cast<BranchInst>(&I)) {
        if (UP.Threshold < MaxBoost && Br->isConditional()) {
          BasicBlock *Succ0 = Br->getSuccessor(0);
          BasicBlock *Succ1 = Br->getSuccessor(1);
          if ((L->contains(Succ0) && L->isLoopExiting(Succ0)) ||
              (L->contains(Succ1) && L->isLoopExiting(Succ1)))
            continue;
          if (dependsOnLocalPhi(L, Br->getCondition())) {
            UP.Threshold += UnrollThresholdIf;
            LLVM_DEBUG(dbgs() << "Set unroll threshold " << UP.Threshold
                              << " for loop:\n"
                              << *L << " due to " << *Br << '\n');
            if (UP.Threshold >= MaxBoost)
              return;
          }
        }
        continue;
      }

      const GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(&I);
      if (!GEP)
        continue;

      unsigned AS = GEP->getAddressSpace();
      unsigned Threshold = 0;
      if (AS == AMDGPUAS::PRIVATE_ADDRESS)
        Threshold = ThresholdPrivate;
      else if (AS == AMDGPUAS::LOCAL_ADDRESS || AS == AMDGPUAS::REGION_ADDRESS)
        Threshold = ThresholdLocal;
      else
        continue;

      if (UP.Threshold >= Threshold)
        continue;

      if (AS == AMDGPUAS::PRIVATE_ADDRESS) {
        const Value *Ptr = GEP->getPointerOperand();
        const AllocaInst *Alloca =
            dyn_cast<AllocaInst>(GetUnderlyingObject(Ptr, DL));
        if (!Alloca || !Alloca->isStaticAlloca())
          continue;
        Type *Ty = Alloca->getAllocatedType();
        unsigned AllocaSize = Ty->isSized() ? DL.getTypeAllocSize(Ty) : 0;
        if (AllocaSize > MaxAlloca)
          continue;
      } else if (AS == AMDGPUAS::LOCAL_ADDRESS ||
                 AS == AMDGPUAS::REGION_ADDRESS) {
        LocalGEPsSeen++;
        // Inhibit unroll for local memory if we have seen addressing not to
        // a variable, most likely we will be unable to combine it.
        // Do not unroll too deep inner loops for local memory to give a chance
        // to unroll an outer loop for a more important reason.
        if (LocalGEPsSeen > 1 || L->getLoopDepth() > 2 ||
            (!isa<GlobalVariable>(GEP->getPointerOperand()) &&
             !isa<Argument>(GEP->getPointerOperand())))
          continue;
      }

      // Check if GEP depends on a value defined by this loop itself.
      bool HasLoopDef = false;
      for (const Value *Op : GEP->operands()) {
        const Instruction *Inst = dyn_cast<Instruction>(Op);
        if (!Inst || L->isLoopInvariant(Op))
          continue;

        if (llvm::any_of(L->getSubLoops(), [Inst](const Loop* SubLoop) {
             return SubLoop->contains(Inst); }))
          continue;
        HasLoopDef = true;
        break;
      }
      if (!HasLoopDef)
        continue;

      // We want to do whatever we can to limit the number of alloca
      // instructions that make it through to the code generator.  allocas
      // require us to use indirect addressing, which is slow and prone to
      // compiler bugs.  If this loop does an address calculation on an
      // alloca ptr, then we want to use a higher than normal loop unroll
      // threshold. This will give SROA a better chance to eliminate these
      // allocas.
      //
      // We also want to have more unrolling for local memory to let ds
      // instructions with different offsets combine.
      //
      // Don't use the maximum allowed value here as it will make some
      // programs way too big.
      UP.Threshold = Threshold;
      LLVM_DEBUG(dbgs() << "Set unroll threshold " << Threshold
                        << " for loop:\n"
                        << *L << " due to " << *GEP << '\n');
      if (UP.Threshold >= MaxBoost)
        return;
    }
  }
}

unsigned GCNTTIImpl::getHardwareNumberOfRegisters(bool Vec) const {
  // The concept of vector registers doesn't really exist. Some packed vector
  // operations operate on the normal 32-bit registers.
  return 256;
}

unsigned GCNTTIImpl::getNumberOfRegisters(bool Vec) const {
  // This is really the number of registers to fill when vectorizing /
  // interleaving loops, so we lie to avoid trying to use all registers.
  return getHardwareNumberOfRegisters(Vec) >> 3;
}

unsigned GCNTTIImpl::getRegisterBitWidth(bool Vector) const {
  return 32;
}

unsigned GCNTTIImpl::getMinVectorRegisterBitWidth() const {
  return 32;
}

unsigned GCNTTIImpl::getLoadVectorFactor(unsigned VF, unsigned LoadSize,
                                            unsigned ChainSizeInBytes,
                                            VectorType *VecTy) const {
  unsigned VecRegBitWidth = VF * LoadSize;
  if (VecRegBitWidth > 128 && VecTy->getScalarSizeInBits() < 32)
    // TODO: Support element-size less than 32bit?
    return 128 / LoadSize;

  return VF;
}

unsigned GCNTTIImpl::getStoreVectorFactor(unsigned VF, unsigned StoreSize,
                                             unsigned ChainSizeInBytes,
                                             VectorType *VecTy) const {
  unsigned VecRegBitWidth = VF * StoreSize;
  if (VecRegBitWidth > 128)
    return 128 / StoreSize;

  return VF;
}

unsigned GCNTTIImpl::getLoadStoreVecRegBitWidth(unsigned AddrSpace) const {
  if (AddrSpace == AMDGPUAS::GLOBAL_ADDRESS ||
      AddrSpace == AMDGPUAS::CONSTANT_ADDRESS ||
      AddrSpace == AMDGPUAS::CONSTANT_ADDRESS_32BIT ||
      AddrSpace == AMDGPUAS::BUFFER_FAT_POINTER) {
    return 512;
  }

  if (AddrSpace == AMDGPUAS::FLAT_ADDRESS ||
      AddrSpace == AMDGPUAS::LOCAL_ADDRESS ||
      AddrSpace == AMDGPUAS::REGION_ADDRESS)
    return 128;

  if (AddrSpace == AMDGPUAS::PRIVATE_ADDRESS)
    return 8 * ST->getMaxPrivateElementSize();

  llvm_unreachable("unhandled address space");
}

bool GCNTTIImpl::isLegalToVectorizeMemChain(unsigned ChainSizeInBytes,
                                               unsigned Alignment,
                                               unsigned AddrSpace) const {
  // We allow vectorization of flat stores, even though we may need to decompose
  // them later if they may access private memory. We don't have enough context
  // here, and legalization can handle it.
  if (AddrSpace == AMDGPUAS::PRIVATE_ADDRESS) {
    return (Alignment >= 4 || ST->hasUnalignedScratchAccess()) &&
      ChainSizeInBytes <= ST->getMaxPrivateElementSize();
  }
  return true;
}

bool GCNTTIImpl::isLegalToVectorizeLoadChain(unsigned ChainSizeInBytes,
                                                unsigned Alignment,
                                                unsigned AddrSpace) const {
  return isLegalToVectorizeMemChain(ChainSizeInBytes, Alignment, AddrSpace);
}

bool GCNTTIImpl::isLegalToVectorizeStoreChain(unsigned ChainSizeInBytes,
                                                 unsigned Alignment,
                                                 unsigned AddrSpace) const {
  return isLegalToVectorizeMemChain(ChainSizeInBytes, Alignment, AddrSpace);
}

unsigned GCNTTIImpl::getMaxInterleaveFactor(unsigned VF) {
  // Disable unrolling if the loop is not vectorized.
  // TODO: Enable this again.
  if (VF == 1)
    return 1;

  return 8;
}

bool GCNTTIImpl::getTgtMemIntrinsic(IntrinsicInst *Inst,
                                       MemIntrinsicInfo &Info) const {
  switch (Inst->getIntrinsicID()) {
  case Intrinsic::amdgcn_atomic_inc:
  case Intrinsic::amdgcn_atomic_dec:
  case Intrinsic::amdgcn_ds_ordered_add:
  case Intrinsic::amdgcn_ds_ordered_swap:
  case Intrinsic::amdgcn_ds_fadd:
  case Intrinsic::amdgcn_ds_fmin:
  case Intrinsic::amdgcn_ds_fmax: {
    auto *Ordering = dyn_cast<ConstantInt>(Inst->getArgOperand(2));
    auto *Volatile = dyn_cast<ConstantInt>(Inst->getArgOperand(4));
    if (!Ordering || !Volatile)
      return false; // Invalid.

    unsigned OrderingVal = Ordering->getZExtValue();
    if (OrderingVal > static_cast<unsigned>(AtomicOrdering::SequentiallyConsistent))
      return false;

    Info.PtrVal = Inst->getArgOperand(0);
    Info.Ordering = static_cast<AtomicOrdering>(OrderingVal);
    Info.ReadMem = true;
    Info.WriteMem = true;
    Info.IsVolatile = !Volatile->isNullValue();
    return true;
  }
  default:
    return false;
  }
}

int GCNTTIImpl::getArithmeticInstrCost(
    unsigned Opcode, Type *Ty, TTI::OperandValueKind Opd1Info,
    TTI::OperandValueKind Opd2Info, TTI::OperandValueProperties Opd1PropInfo,
    TTI::OperandValueProperties Opd2PropInfo, ArrayRef<const Value *> Args ) {
  EVT OrigTy = TLI->getValueType(DL, Ty);
  if (!OrigTy.isSimple()) {
    return BaseT::getArithmeticInstrCost(Opcode, Ty, Opd1Info, Opd2Info,
                                         Opd1PropInfo, Opd2PropInfo);
  }

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

  // Because we don't have any legal vector operations, but the legal types, we
  // need to account for split vectors.
  unsigned NElts = LT.second.isVector() ?
    LT.second.getVectorNumElements() : 1;

  MVT::SimpleValueType SLT = LT.second.getScalarType().SimpleTy;

  switch (ISD) {
  case ISD::SHL:
  case ISD::SRL:
  case ISD::SRA:
    if (SLT == MVT::i64)
      return get64BitInstrCost() * LT.first * NElts;

    // i32
    return getFullRateInstrCost() * LT.first * NElts;
  case ISD::ADD:
  case ISD::SUB:
  case ISD::AND:
  case ISD::OR:
  case ISD::XOR:
    if (SLT == MVT::i64){
      // and, or and xor are typically split into 2 VALU instructions.
      return 2 * getFullRateInstrCost() * LT.first * NElts;
    }

    return LT.first * NElts * getFullRateInstrCost();
  case ISD::MUL: {
    const int QuarterRateCost = getQuarterRateInstrCost();
    if (SLT == MVT::i64) {
      const int FullRateCost = getFullRateInstrCost();
      return (4 * QuarterRateCost + (2 * 2) * FullRateCost) * LT.first * NElts;
    }

    // i32
    return QuarterRateCost * NElts * LT.first;
  }
  case ISD::FADD:
  case ISD::FSUB:
  case ISD::FMUL:
    if (SLT == MVT::f64)
      return LT.first * NElts * get64BitInstrCost();

    if (SLT == MVT::f32 || SLT == MVT::f16)
      return LT.first * NElts * getFullRateInstrCost();
    break;
  case ISD::FDIV:
  case ISD::FREM:
    // FIXME: frem should be handled separately. The fdiv in it is most of it,
    // but the current lowering is also not entirely correct.
    if (SLT == MVT::f64) {
      int Cost = 4 * get64BitInstrCost() + 7 * getQuarterRateInstrCost();
      // Add cost of workaround.
      if (!ST->hasUsableDivScaleConditionOutput())
        Cost += 3 * getFullRateInstrCost();

      return LT.first * Cost * NElts;
    }

    if (!Args.empty() && match(Args[0], PatternMatch::m_FPOne())) {
      // TODO: This is more complicated, unsafe flags etc.
      if ((SLT == MVT::f32 && !ST->hasFP32Denormals()) ||
          (SLT == MVT::f16 && ST->has16BitInsts())) {
        return LT.first * getQuarterRateInstrCost() * NElts;
      }
    }

    if (SLT == MVT::f16 && ST->has16BitInsts()) {
      // 2 x v_cvt_f32_f16
      // f32 rcp
      // f32 fmul
      // v_cvt_f16_f32
      // f16 div_fixup
      int Cost = 4 * getFullRateInstrCost() + 2 * getQuarterRateInstrCost();
      return LT.first * Cost * NElts;
    }

    if (SLT == MVT::f32 || SLT == MVT::f16) {
      int Cost = 7 * getFullRateInstrCost() + 1 * getQuarterRateInstrCost();

      if (!ST->hasFP32Denormals()) {
        // FP mode switches.
        Cost += 2 * getFullRateInstrCost();
      }

      return LT.first * NElts * Cost;
    }
    break;
  default:
    break;
  }

  return BaseT::getArithmeticInstrCost(Opcode, Ty, Opd1Info, Opd2Info,
                                       Opd1PropInfo, Opd2PropInfo);
}

unsigned GCNTTIImpl::getCFInstrCost(unsigned Opcode) {
  // XXX - For some reason this isn't called for switch.
  switch (Opcode) {
  case Instruction::Br:
  case Instruction::Ret:
    return 10;
  default:
    return BaseT::getCFInstrCost(Opcode);
  }
}

int GCNTTIImpl::getArithmeticReductionCost(unsigned Opcode, Type *Ty,
                                              bool IsPairwise) {
  EVT OrigTy = TLI->getValueType(DL, Ty);

  // Computes cost on targets that have packed math instructions(which support
  // 16-bit types only).
  if (IsPairwise ||
      !ST->hasVOP3PInsts() ||
      OrigTy.getScalarSizeInBits() != 16)
    return BaseT::getArithmeticReductionCost(Opcode, Ty, IsPairwise);

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

int GCNTTIImpl::getMinMaxReductionCost(Type *Ty, Type *CondTy,
                                          bool IsPairwise,
                                          bool IsUnsigned) {
  EVT OrigTy = TLI->getValueType(DL, Ty);

  // Computes cost on targets that have packed math instructions(which support
  // 16-bit types only).
  if (IsPairwise ||
      !ST->hasVOP3PInsts() ||
      OrigTy.getScalarSizeInBits() != 16)
    return BaseT::getMinMaxReductionCost(Ty, CondTy, IsPairwise, IsUnsigned);

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

int GCNTTIImpl::getVectorInstrCost(unsigned Opcode, Type *ValTy,
                                      unsigned Index) {
  switch (Opcode) {
  case Instruction::ExtractElement:
  case Instruction::InsertElement: {
    unsigned EltSize
      = DL.getTypeSizeInBits(cast<VectorType>(ValTy)->getElementType());
    if (EltSize < 32) {
      if (EltSize == 16 && Index == 0 && ST->has16BitInsts())
        return 0;
      return BaseT::getVectorInstrCost(Opcode, ValTy, Index);
    }

    // Extracts are just reads of a subregister, so are free. Inserts are
    // considered free because we don't want to have any cost for scalarizing
    // operations, and we don't have to copy into a different register class.

    // Dynamic indexing isn't free and is best avoided.
    return Index == ~0u ? 2 : 0;
  }
  default:
    return BaseT::getVectorInstrCost(Opcode, ValTy, Index);
  }
}



static bool isArgPassedInSGPR(const Argument *A) {
  const Function *F = A->getParent();

  // Arguments to compute shaders are never a source of divergence.
  CallingConv::ID CC = F->getCallingConv();
  switch (CC) {
  case CallingConv::AMDGPU_KERNEL:
  case CallingConv::SPIR_KERNEL:
    return true;
  case CallingConv::AMDGPU_VS:
  case CallingConv::AMDGPU_LS:
  case CallingConv::AMDGPU_HS:
  case CallingConv::AMDGPU_ES:
  case CallingConv::AMDGPU_GS:
  case CallingConv::AMDGPU_PS:
  case CallingConv::AMDGPU_CS:
    // For non-compute shaders, SGPR inputs are marked with either inreg or byval.
    // Everything else is in VGPRs.
    return F->getAttributes().hasParamAttribute(A->getArgNo(), Attribute::InReg) ||
           F->getAttributes().hasParamAttribute(A->getArgNo(), Attribute::ByVal);
  default:
    // TODO: Should calls support inreg for SGPR inputs?
    return false;
  }
}

/// \returns true if the result of the value could potentially be
/// different across workitems in a wavefront.
bool GCNTTIImpl::isSourceOfDivergence(const Value *V) const {
  if (const Argument *A = dyn_cast<Argument>(V))
    return !isArgPassedInSGPR(A);

  // Loads from the private and flat address spaces are divergent, because
  // threads can execute the load instruction with the same inputs and get
  // different results.
  //
  // All other loads are not divergent, because if threads issue loads with the
  // same arguments, they will always get the same result.
  if (const LoadInst *Load = dyn_cast<LoadInst>(V))
    return Load->getPointerAddressSpace() == AMDGPUAS::PRIVATE_ADDRESS ||
           Load->getPointerAddressSpace() == AMDGPUAS::FLAT_ADDRESS;

  // Atomics are divergent because they are executed sequentially: when an
  // atomic operation refers to the same address in each thread, then each
  // thread after the first sees the value written by the previous thread as
  // original value.
  if (isa<AtomicRMWInst>(V) || isa<AtomicCmpXchgInst>(V))
    return true;

  if (const IntrinsicInst *Intrinsic = dyn_cast<IntrinsicInst>(V))
    return AMDGPU::isIntrinsicSourceOfDivergence(Intrinsic->getIntrinsicID());

  // Assume all function calls are a source of divergence.
  if (isa<CallInst>(V) || isa<InvokeInst>(V))
    return true;

  return false;
}

bool GCNTTIImpl::isAlwaysUniform(const Value *V) const {
  if (const IntrinsicInst *Intrinsic = dyn_cast<IntrinsicInst>(V)) {
    switch (Intrinsic->getIntrinsicID()) {
    default:
      return false;
    case Intrinsic::amdgcn_readfirstlane:
    case Intrinsic::amdgcn_readlane:
    case Intrinsic::amdgcn_icmp:
    case Intrinsic::amdgcn_fcmp:
      return true;
    }
  }
  return false;
}

bool GCNTTIImpl::collectFlatAddressOperands(SmallVectorImpl<int> &OpIndexes,
                                            Intrinsic::ID IID) const {
  switch (IID) {
  case Intrinsic::amdgcn_atomic_inc:
  case Intrinsic::amdgcn_atomic_dec:
  case Intrinsic::amdgcn_ds_fadd:
  case Intrinsic::amdgcn_ds_fmin:
  case Intrinsic::amdgcn_ds_fmax:
  case Intrinsic::amdgcn_is_shared:
  case Intrinsic::amdgcn_is_private:
    OpIndexes.push_back(0);
    return true;
  default:
    return false;
  }
}

bool GCNTTIImpl::rewriteIntrinsicWithAddressSpace(
  IntrinsicInst *II, Value *OldV, Value *NewV) const {
  auto IntrID = II->getIntrinsicID();
  switch (IntrID) {
  case Intrinsic::amdgcn_atomic_inc:
  case Intrinsic::amdgcn_atomic_dec:
  case Intrinsic::amdgcn_ds_fadd:
  case Intrinsic::amdgcn_ds_fmin:
  case Intrinsic::amdgcn_ds_fmax: {
    const ConstantInt *IsVolatile = cast<ConstantInt>(II->getArgOperand(4));
    if (!IsVolatile->isZero())
      return false;
    Module *M = II->getParent()->getParent()->getParent();
    Type *DestTy = II->getType();
    Type *SrcTy = NewV->getType();
    Function *NewDecl =
        Intrinsic::getDeclaration(M, II->getIntrinsicID(), {DestTy, SrcTy});
    II->setArgOperand(0, NewV);
    II->setCalledFunction(NewDecl);
    return true;
  }
  case Intrinsic::amdgcn_is_shared:
  case Intrinsic::amdgcn_is_private: {
    unsigned TrueAS = IntrID == Intrinsic::amdgcn_is_shared ?
      AMDGPUAS::LOCAL_ADDRESS : AMDGPUAS::PRIVATE_ADDRESS;
    unsigned NewAS = NewV->getType()->getPointerAddressSpace();
    LLVMContext &Ctx = NewV->getType()->getContext();
    ConstantInt *NewVal = (TrueAS == NewAS) ?
      ConstantInt::getTrue(Ctx) : ConstantInt::getFalse(Ctx);
    II->replaceAllUsesWith(NewVal);
    II->eraseFromParent();
    return true;
  }
  default:
    return false;
  }
}

unsigned GCNTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index,
                                       Type *SubTp) {
  if (ST->hasVOP3PInsts()) {
    VectorType *VT = cast<VectorType>(Tp);
    if (VT->getNumElements() == 2 &&
        DL.getTypeSizeInBits(VT->getElementType()) == 16) {
      // With op_sel VOP3P instructions freely can access the low half or high
      // half of a register, so any swizzle is free.

      switch (Kind) {
      case TTI::SK_Broadcast:
      case TTI::SK_Reverse:
      case TTI::SK_PermuteSingleSrc:
        return 0;
      default:
        break;
      }
    }
  }

  return BaseT::getShuffleCost(Kind, Tp, Index, SubTp);
}

bool GCNTTIImpl::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();

  FeatureBitset RealCallerBits = CallerBits & ~InlineFeatureIgnoreList;
  FeatureBitset RealCalleeBits = CalleeBits & ~InlineFeatureIgnoreList;
  if ((RealCallerBits & RealCalleeBits) != RealCalleeBits)
    return false;

  // FIXME: dx10_clamp can just take the caller setting, but there seems to be
  // no way to support merge for backend defined attributes.
  AMDGPU::SIModeRegisterDefaults CallerMode(*Caller);
  AMDGPU::SIModeRegisterDefaults CalleeMode(*Callee);
  return CallerMode.isInlineCompatible(CalleeMode);
}

void GCNTTIImpl::getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                         TTI::UnrollingPreferences &UP) {
  CommonTTI.getUnrollingPreferences(L, SE, UP);
}

unsigned GCNTTIImpl::getUserCost(const User *U,
                                 ArrayRef<const Value *> Operands) {
  // Estimate extractelement elimination
  if (const ExtractElementInst *EE = dyn_cast<ExtractElementInst>(U)) {
    ConstantInt *CI = dyn_cast<ConstantInt>(EE->getOperand(1));
    unsigned Idx = -1;
    if (CI)
      Idx = CI->getZExtValue();
    return getVectorInstrCost(EE->getOpcode(), EE->getOperand(0)->getType(),
                              Idx);
  }

  // Estimate insertelement elimination
  if (const InsertElementInst *IE = dyn_cast<InsertElementInst>(U)) {
    ConstantInt *CI = dyn_cast<ConstantInt>(IE->getOperand(2));
    unsigned Idx = -1;
    if (CI)
      Idx = CI->getZExtValue();
    return getVectorInstrCost(IE->getOpcode(), IE->getType(), Idx);
  }

  // Estimate different intrinsics, e.g. llvm.fabs
  if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
    SmallVector<Value *, 4> Args(II->arg_operands());
    FastMathFlags FMF;
    if (auto *FPMO = dyn_cast<FPMathOperator>(II))
      FMF = FPMO->getFastMathFlags();
    return getIntrinsicInstrCost(II->getIntrinsicID(), II->getType(), Args,
                                 FMF);
  }
  return BaseT::getUserCost(U, Operands);
}

unsigned R600TTIImpl::getHardwareNumberOfRegisters(bool Vec) const {
  return 4 * 128; // XXX - 4 channels. Should these count as vector instead?
}

unsigned R600TTIImpl::getNumberOfRegisters(bool Vec) const {
  return getHardwareNumberOfRegisters(Vec);
}

unsigned R600TTIImpl::getRegisterBitWidth(bool Vector) const {
  return 32;
}

unsigned R600TTIImpl::getMinVectorRegisterBitWidth() const {
  return 32;
}

unsigned R600TTIImpl::getLoadStoreVecRegBitWidth(unsigned AddrSpace) const {
  if (AddrSpace == AMDGPUAS::GLOBAL_ADDRESS ||
      AddrSpace == AMDGPUAS::CONSTANT_ADDRESS)
    return 128;
  if (AddrSpace == AMDGPUAS::LOCAL_ADDRESS ||
      AddrSpace == AMDGPUAS::REGION_ADDRESS)
    return 64;
  if (AddrSpace == AMDGPUAS::PRIVATE_ADDRESS)
    return 32;

  if ((AddrSpace == AMDGPUAS::PARAM_D_ADDRESS ||
      AddrSpace == AMDGPUAS::PARAM_I_ADDRESS ||
      (AddrSpace >= AMDGPUAS::CONSTANT_BUFFER_0 &&
      AddrSpace <= AMDGPUAS::CONSTANT_BUFFER_15)))
    return 128;
  llvm_unreachable("unhandled address space");
}

bool R600TTIImpl::isLegalToVectorizeMemChain(unsigned ChainSizeInBytes,
                                             unsigned Alignment,
                                             unsigned AddrSpace) const {
  // We allow vectorization of flat stores, even though we may need to decompose
  // them later if they may access private memory. We don't have enough context
  // here, and legalization can handle it.
  return (AddrSpace != AMDGPUAS::PRIVATE_ADDRESS);
}

bool R600TTIImpl::isLegalToVectorizeLoadChain(unsigned ChainSizeInBytes,
                                              unsigned Alignment,
                                              unsigned AddrSpace) const {
  return isLegalToVectorizeMemChain(ChainSizeInBytes, Alignment, AddrSpace);
}

bool R600TTIImpl::isLegalToVectorizeStoreChain(unsigned ChainSizeInBytes,
                                               unsigned Alignment,
                                               unsigned AddrSpace) const {
  return isLegalToVectorizeMemChain(ChainSizeInBytes, Alignment, AddrSpace);
}

unsigned R600TTIImpl::getMaxInterleaveFactor(unsigned VF) {
  // Disable unrolling if the loop is not vectorized.
  // TODO: Enable this again.
  if (VF == 1)
    return 1;

  return 8;
}

unsigned R600TTIImpl::getCFInstrCost(unsigned Opcode) {
  // XXX - For some reason this isn't called for switch.
  switch (Opcode) {
  case Instruction::Br:
  case Instruction::Ret:
    return 10;
  default:
    return BaseT::getCFInstrCost(Opcode);
  }
}

int R600TTIImpl::getVectorInstrCost(unsigned Opcode, Type *ValTy,
                                    unsigned Index) {
  switch (Opcode) {
  case Instruction::ExtractElement:
  case Instruction::InsertElement: {
    unsigned EltSize
      = DL.getTypeSizeInBits(cast<VectorType>(ValTy)->getElementType());
    if (EltSize < 32) {
      return BaseT::getVectorInstrCost(Opcode, ValTy, Index);
    }

    // Extracts are just reads of a subregister, so are free. Inserts are
    // considered free because we don't want to have any cost for scalarizing
    // operations, and we don't have to copy into a different register class.

    // Dynamic indexing isn't free and is best avoided.
    return Index == ~0u ? 2 : 0;
  }
  default:
    return BaseT::getVectorInstrCost(Opcode, ValTy, Index);
  }
}

void R600TTIImpl::getUnrollingPreferences(Loop *L, ScalarEvolution &SE,
                                          TTI::UnrollingPreferences &UP) {
  CommonTTI.getUnrollingPreferences(L, SE, UP);
}