summaryrefslogtreecommitdiff
path: root/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
blob: 83122281d2b2bb0b17c66ba86d9649d3625b9189 (plain)
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
//===-- AMDGPUSubtarget.cpp - AMDGPU Subtarget Information ----------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
/// \file
/// \brief Implements the AMDGPU specific subclass of TargetSubtarget.
//
//===----------------------------------------------------------------------===//

#include "AMDGPUSubtarget.h"
#include "AMDGPU.h"
#include "AMDGPUTargetMachine.h"
#include "AMDGPUCallLowering.h"
#include "AMDGPUInstructionSelector.h"
#include "AMDGPULegalizerInfo.h"
#include "AMDGPURegisterBankInfo.h"
#include "SIMachineFunctionInfo.h"
#include "llvm/ADT/SmallString.h"
#include "llvm/CodeGen/MachineScheduler.h"
#include "llvm/IR/MDBuilder.h"
#include "llvm/CodeGen/TargetFrameLowering.h"
#include <algorithm>

using namespace llvm;

#define DEBUG_TYPE "amdgpu-subtarget"

#define GET_SUBTARGETINFO_TARGET_DESC
#define GET_SUBTARGETINFO_CTOR
#include "AMDGPUGenSubtargetInfo.inc"

AMDGPUSubtarget::~AMDGPUSubtarget() = default;

AMDGPUSubtarget &
AMDGPUSubtarget::initializeSubtargetDependencies(const Triple &TT,
                                                 StringRef GPU, StringRef FS) {
  // Determine default and user-specified characteristics
  // On SI+, we want FP64 denormals to be on by default. FP32 denormals can be
  // enabled, but some instructions do not respect them and they run at the
  // double precision rate, so don't enable by default.
  //
  // We want to be able to turn these off, but making this a subtarget feature
  // for SI has the unhelpful behavior that it unsets everything else if you
  // disable it.

  SmallString<256> FullFS("+promote-alloca,+fp64-fp16-denormals,+dx10-clamp,+load-store-opt,");
  if (isAmdHsaOS()) // Turn on FlatForGlobal for HSA.
    FullFS += "+flat-address-space,+flat-for-global,+unaligned-buffer-access,+trap-handler,";

  FullFS += FS;

  ParseSubtargetFeatures(GPU, FullFS);

  // Unless +-flat-for-global is specified, turn on FlatForGlobal for all OS-es
  // on VI and newer hardware to avoid assertion failures due to missing ADDR64
  // variants of MUBUF instructions.
  if (!hasAddr64() && !FS.contains("flat-for-global")) {
    FlatForGlobal = true;
  }

  // FIXME: I don't think think Evergreen has any useful support for
  // denormals, but should be checked. Should we issue a warning somewhere
  // if someone tries to enable these?
  if (getGeneration() <= AMDGPUSubtarget::NORTHERN_ISLANDS) {
    FP64FP16Denormals = false;
    FP32Denormals = false;
  }

  // Set defaults if needed.
  if (MaxPrivateElementSize == 0)
    MaxPrivateElementSize = 4;

  if (LDSBankCount == 0)
    LDSBankCount = 32;

  if (TT.getArch() == Triple::amdgcn) {
    if (LocalMemorySize == 0)
      LocalMemorySize = 32768;

    // Do something sensible for unspecified target.
    if (!HasMovrel && !HasVGPRIndexMode)
      HasMovrel = true;
  }

  return *this;
}

AMDGPUSubtarget::AMDGPUSubtarget(const Triple &TT, StringRef GPU, StringRef FS,
                                 const TargetMachine &TM)
  : AMDGPUGenSubtargetInfo(TT, GPU, FS),
    TargetTriple(TT),
    Gen(TT.getArch() == Triple::amdgcn ? SOUTHERN_ISLANDS : R600),
    IsaVersion(ISAVersion0_0_0),
    WavefrontSize(0),
    LocalMemorySize(0),
    LDSBankCount(0),
    MaxPrivateElementSize(0),

    FastFMAF32(false),
    HalfRate64Ops(false),

    FP32Denormals(false),
    FP64FP16Denormals(false),
    FPExceptions(false),
    DX10Clamp(false),
    FlatForGlobal(false),
    AutoWaitcntBeforeBarrier(false),
    CodeObjectV3(false),
    UnalignedScratchAccess(false),
    UnalignedBufferAccess(false),

    HasApertureRegs(false),
    EnableXNACK(false),
    TrapHandler(false),
    DebuggerInsertNops(false),
    DebuggerReserveRegs(false),
    DebuggerEmitPrologue(false),

    EnableVGPRSpilling(false),
    EnablePromoteAlloca(false),
    EnableLoadStoreOpt(false),
    EnableUnsafeDSOffsetFolding(false),
    EnableSIScheduler(false),
    DumpCode(false),

    FP64(false),
    IsGCN(false),
    GCN3Encoding(false),
    CIInsts(false),
    GFX9Insts(false),
    SGPRInitBug(false),
    HasSMemRealTime(false),
    Has16BitInsts(false),
    HasIntClamp(false),
    HasVOP3PInsts(false),
    HasMadMixInsts(false),
    HasMovrel(false),
    HasVGPRIndexMode(false),
    HasScalarStores(false),
    HasInv2PiInlineImm(false),
    HasSDWA(false),
    HasSDWAOmod(false),
    HasSDWAScalar(false),
    HasSDWASdst(false),
    HasSDWAMac(false),
    HasSDWAOutModsVOPC(false),
    HasDPP(false),
    FlatAddressSpace(false),
    FlatInstOffsets(false),
    FlatGlobalInsts(false),
    FlatScratchInsts(false),
    AddNoCarryInsts(false),

    R600ALUInst(false),
    CaymanISA(false),
    CFALUBug(false),
    HasVertexCache(false),
    TexVTXClauseSize(0),
    ScalarizeGlobal(false),

    FeatureDisable(false),
    InstrItins(getInstrItineraryForCPU(GPU)) {
  AS = AMDGPU::getAMDGPUAS(TT);
  initializeSubtargetDependencies(TT, GPU, FS);
}

unsigned AMDGPUSubtarget::getMaxLocalMemSizeWithWaveCount(unsigned NWaves,
  const Function &F) const {
  if (NWaves == 1)
    return getLocalMemorySize();
  unsigned WorkGroupSize = getFlatWorkGroupSizes(F).second;
  unsigned WorkGroupsPerCu = getMaxWorkGroupsPerCU(WorkGroupSize);
  unsigned MaxWaves = getMaxWavesPerEU();
  return getLocalMemorySize() * MaxWaves / WorkGroupsPerCu / NWaves;
}

unsigned AMDGPUSubtarget::getOccupancyWithLocalMemSize(uint32_t Bytes,
  const Function &F) const {
  unsigned WorkGroupSize = getFlatWorkGroupSizes(F).second;
  unsigned WorkGroupsPerCu = getMaxWorkGroupsPerCU(WorkGroupSize);
  unsigned MaxWaves = getMaxWavesPerEU();
  unsigned Limit = getLocalMemorySize() * MaxWaves / WorkGroupsPerCu;
  unsigned NumWaves = Limit / (Bytes ? Bytes : 1u);
  NumWaves = std::min(NumWaves, MaxWaves);
  NumWaves = std::max(NumWaves, 1u);
  return NumWaves;
}

std::pair<unsigned, unsigned>
AMDGPUSubtarget::getDefaultFlatWorkGroupSize(CallingConv::ID CC) const {
  switch (CC) {
  case CallingConv::AMDGPU_CS:
  case CallingConv::AMDGPU_KERNEL:
  case CallingConv::SPIR_KERNEL:
    return std::make_pair(getWavefrontSize() * 2, getWavefrontSize() * 4);
  case CallingConv::AMDGPU_VS:
  case CallingConv::AMDGPU_LS:
  case CallingConv::AMDGPU_HS:
  case CallingConv::AMDGPU_ES:
  case CallingConv::AMDGPU_GS:
  case CallingConv::AMDGPU_PS:
    return std::make_pair(1, getWavefrontSize());
  default:
    return std::make_pair(1, 16 * getWavefrontSize());
  }
}

std::pair<unsigned, unsigned> AMDGPUSubtarget::getFlatWorkGroupSizes(
  const Function &F) const {
  // FIXME: 1024 if function.
  // Default minimum/maximum flat work group sizes.
  std::pair<unsigned, unsigned> Default =
    getDefaultFlatWorkGroupSize(F.getCallingConv());

  // TODO: Do not process "amdgpu-max-work-group-size" attribute once mesa
  // starts using "amdgpu-flat-work-group-size" attribute.
  Default.second = AMDGPU::getIntegerAttribute(
    F, "amdgpu-max-work-group-size", Default.second);
  Default.first = std::min(Default.first, Default.second);

  // Requested minimum/maximum flat work group sizes.
  std::pair<unsigned, unsigned> Requested = AMDGPU::getIntegerPairAttribute(
    F, "amdgpu-flat-work-group-size", Default);

  // Make sure requested minimum is less than requested maximum.
  if (Requested.first > Requested.second)
    return Default;

  // Make sure requested values do not violate subtarget's specifications.
  if (Requested.first < getMinFlatWorkGroupSize())
    return Default;
  if (Requested.second > getMaxFlatWorkGroupSize())
    return Default;

  return Requested;
}

std::pair<unsigned, unsigned> AMDGPUSubtarget::getWavesPerEU(
  const Function &F) const {
  // Default minimum/maximum number of waves per execution unit.
  std::pair<unsigned, unsigned> Default(1, getMaxWavesPerEU());

  // Default/requested minimum/maximum flat work group sizes.
  std::pair<unsigned, unsigned> FlatWorkGroupSizes = getFlatWorkGroupSizes(F);

  // If minimum/maximum flat work group sizes were explicitly requested using
  // "amdgpu-flat-work-group-size" attribute, then set default minimum/maximum
  // number of waves per execution unit to values implied by requested
  // minimum/maximum flat work group sizes.
  unsigned MinImpliedByFlatWorkGroupSize =
    getMaxWavesPerEU(FlatWorkGroupSizes.second);
  bool RequestedFlatWorkGroupSize = false;

  // TODO: Do not process "amdgpu-max-work-group-size" attribute once mesa
  // starts using "amdgpu-flat-work-group-size" attribute.
  if (F.hasFnAttribute("amdgpu-max-work-group-size") ||
      F.hasFnAttribute("amdgpu-flat-work-group-size")) {
    Default.first = MinImpliedByFlatWorkGroupSize;
    RequestedFlatWorkGroupSize = true;
  }

  // Requested minimum/maximum number of waves per execution unit.
  std::pair<unsigned, unsigned> Requested = AMDGPU::getIntegerPairAttribute(
    F, "amdgpu-waves-per-eu", Default, true);

  // Make sure requested minimum is less than requested maximum.
  if (Requested.second && Requested.first > Requested.second)
    return Default;

  // Make sure requested values do not violate subtarget's specifications.
  if (Requested.first < getMinWavesPerEU() ||
      Requested.first > getMaxWavesPerEU())
    return Default;
  if (Requested.second > getMaxWavesPerEU())
    return Default;

  // Make sure requested values are compatible with values implied by requested
  // minimum/maximum flat work group sizes.
  if (RequestedFlatWorkGroupSize &&
      Requested.first < MinImpliedByFlatWorkGroupSize)
    return Default;

  return Requested;
}

bool AMDGPUSubtarget::makeLIDRangeMetadata(Instruction *I) const {
  Function *Kernel = I->getParent()->getParent();
  unsigned MinSize = 0;
  unsigned MaxSize = getFlatWorkGroupSizes(*Kernel).second;
  bool IdQuery = false;

  // If reqd_work_group_size is present it narrows value down.
  if (auto *CI = dyn_cast<CallInst>(I)) {
    const Function *F = CI->getCalledFunction();
    if (F) {
      unsigned Dim = UINT_MAX;
      switch (F->getIntrinsicID()) {
      case Intrinsic::amdgcn_workitem_id_x:
      case Intrinsic::r600_read_tidig_x:
        IdQuery = true;
        LLVM_FALLTHROUGH;
      case Intrinsic::r600_read_local_size_x:
        Dim = 0;
        break;
      case Intrinsic::amdgcn_workitem_id_y:
      case Intrinsic::r600_read_tidig_y:
        IdQuery = true;
        LLVM_FALLTHROUGH;
      case Intrinsic::r600_read_local_size_y:
        Dim = 1;
        break;
      case Intrinsic::amdgcn_workitem_id_z:
      case Intrinsic::r600_read_tidig_z:
        IdQuery = true;
        LLVM_FALLTHROUGH;
      case Intrinsic::r600_read_local_size_z:
        Dim = 2;
        break;
      default:
        break;
      }
      if (Dim <= 3) {
        if (auto Node = Kernel->getMetadata("reqd_work_group_size"))
          if (Node->getNumOperands() == 3)
            MinSize = MaxSize = mdconst::extract<ConstantInt>(
                                  Node->getOperand(Dim))->getZExtValue();
      }
    }
  }

  if (!MaxSize)
    return false;

  // Range metadata is [Lo, Hi). For ID query we need to pass max size
  // as Hi. For size query we need to pass Hi + 1.
  if (IdQuery)
    MinSize = 0;
  else
    ++MaxSize;

  MDBuilder MDB(I->getContext());
  MDNode *MaxWorkGroupSizeRange = MDB.createRange(APInt(32, MinSize),
                                                  APInt(32, MaxSize));
  I->setMetadata(LLVMContext::MD_range, MaxWorkGroupSizeRange);
  return true;
}

R600Subtarget::R600Subtarget(const Triple &TT, StringRef GPU, StringRef FS,
                             const TargetMachine &TM) :
  AMDGPUSubtarget(TT, GPU, FS, TM),
  InstrInfo(*this),
  FrameLowering(TargetFrameLowering::StackGrowsUp, getStackAlignment(), 0),
  TLInfo(TM, *this) {}

SISubtarget::SISubtarget(const Triple &TT, StringRef GPU, StringRef FS,
                         const TargetMachine &TM)
    : AMDGPUSubtarget(TT, GPU, FS, TM), InstrInfo(*this),
      FrameLowering(TargetFrameLowering::StackGrowsUp, getStackAlignment(), 0),
      TLInfo(TM, *this) {
  CallLoweringInfo.reset(new AMDGPUCallLowering(*getTargetLowering()));
  Legalizer.reset(new AMDGPULegalizerInfo());

  RegBankInfo.reset(new AMDGPURegisterBankInfo(*getRegisterInfo()));
  InstSelector.reset(new AMDGPUInstructionSelector(
      *this, *static_cast<AMDGPURegisterBankInfo *>(RegBankInfo.get())));
}

void SISubtarget::overrideSchedPolicy(MachineSchedPolicy &Policy,
                                      unsigned NumRegionInstrs) const {
  // Track register pressure so the scheduler can try to decrease
  // pressure once register usage is above the threshold defined by
  // SIRegisterInfo::getRegPressureSetLimit()
  Policy.ShouldTrackPressure = true;

  // Enabling both top down and bottom up scheduling seems to give us less
  // register spills than just using one of these approaches on its own.
  Policy.OnlyTopDown = false;
  Policy.OnlyBottomUp = false;

  // Enabling ShouldTrackLaneMasks crashes the SI Machine Scheduler.
  if (!enableSIScheduler())
    Policy.ShouldTrackLaneMasks = true;
}

bool SISubtarget::isVGPRSpillingEnabled(const Function& F) const {
  return EnableVGPRSpilling || !AMDGPU::isShader(F.getCallingConv());
}

unsigned SISubtarget::getKernArgSegmentSize(const MachineFunction &MF,
                                            unsigned ExplicitArgBytes) const {
  unsigned ImplicitBytes = getImplicitArgNumBytes(MF);
  if (ImplicitBytes == 0)
    return ExplicitArgBytes;

  unsigned Alignment = getAlignmentForImplicitArgPtr();
  return alignTo(ExplicitArgBytes, Alignment) + ImplicitBytes;
}

unsigned SISubtarget::getOccupancyWithNumSGPRs(unsigned SGPRs) const {
  if (getGeneration() >= SISubtarget::VOLCANIC_ISLANDS) {
    if (SGPRs <= 80)
      return 10;
    if (SGPRs <= 88)
      return 9;
    if (SGPRs <= 100)
      return 8;
    return 7;
  }
  if (SGPRs <= 48)
    return 10;
  if (SGPRs <= 56)
    return 9;
  if (SGPRs <= 64)
    return 8;
  if (SGPRs <= 72)
    return 7;
  if (SGPRs <= 80)
    return 6;
  return 5;
}

unsigned SISubtarget::getOccupancyWithNumVGPRs(unsigned VGPRs) const {
  if (VGPRs <= 24)
    return 10;
  if (VGPRs <= 28)
    return 9;
  if (VGPRs <= 32)
    return 8;
  if (VGPRs <= 36)
    return 7;
  if (VGPRs <= 40)
    return 6;
  if (VGPRs <= 48)
    return 5;
  if (VGPRs <= 64)
    return 4;
  if (VGPRs <= 84)
    return 3;
  if (VGPRs <= 128)
    return 2;
  return 1;
}

unsigned SISubtarget::getReservedNumSGPRs(const MachineFunction &MF) const {
  const SIMachineFunctionInfo &MFI = *MF.getInfo<SIMachineFunctionInfo>();
  if (MFI.hasFlatScratchInit()) {
    if (getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS)
      return 6; // FLAT_SCRATCH, XNACK, VCC (in that order).
    if (getGeneration() == AMDGPUSubtarget::SEA_ISLANDS)
      return 4; // FLAT_SCRATCH, VCC (in that order).
  }

  if (isXNACKEnabled())
    return 4; // XNACK, VCC (in that order).
  return 2; // VCC.
}

unsigned SISubtarget::getMaxNumSGPRs(const MachineFunction &MF) const {
  const Function &F = *MF.getFunction();
  const SIMachineFunctionInfo &MFI = *MF.getInfo<SIMachineFunctionInfo>();

  // Compute maximum number of SGPRs function can use using default/requested
  // minimum number of waves per execution unit.
  std::pair<unsigned, unsigned> WavesPerEU = MFI.getWavesPerEU();
  unsigned MaxNumSGPRs = getMaxNumSGPRs(WavesPerEU.first, false);
  unsigned MaxAddressableNumSGPRs = getMaxNumSGPRs(WavesPerEU.first, true);

  // Check if maximum number of SGPRs was explicitly requested using
  // "amdgpu-num-sgpr" attribute.
  if (F.hasFnAttribute("amdgpu-num-sgpr")) {
    unsigned Requested = AMDGPU::getIntegerAttribute(
      F, "amdgpu-num-sgpr", MaxNumSGPRs);

    // Make sure requested value does not violate subtarget's specifications.
    if (Requested && (Requested <= getReservedNumSGPRs(MF)))
      Requested = 0;

    // If more SGPRs are required to support the input user/system SGPRs,
    // increase to accommodate them.
    //
    // FIXME: This really ends up using the requested number of SGPRs + number
    // of reserved special registers in total. Theoretically you could re-use
    // the last input registers for these special registers, but this would
    // require a lot of complexity to deal with the weird aliasing.
    unsigned InputNumSGPRs = MFI.getNumPreloadedSGPRs();
    if (Requested && Requested < InputNumSGPRs)
      Requested = InputNumSGPRs;

    // Make sure requested value is compatible with values implied by
    // default/requested minimum/maximum number of waves per execution unit.
    if (Requested && Requested > getMaxNumSGPRs(WavesPerEU.first, false))
      Requested = 0;
    if (WavesPerEU.second &&
        Requested && Requested < getMinNumSGPRs(WavesPerEU.second))
      Requested = 0;

    if (Requested)
      MaxNumSGPRs = Requested;
  }

  if (hasSGPRInitBug())
    MaxNumSGPRs = AMDGPU::IsaInfo::FIXED_NUM_SGPRS_FOR_INIT_BUG;

  return std::min(MaxNumSGPRs - getReservedNumSGPRs(MF),
                  MaxAddressableNumSGPRs);
}

unsigned SISubtarget::getMaxNumVGPRs(const MachineFunction &MF) const {
  const Function &F = *MF.getFunction();
  const SIMachineFunctionInfo &MFI = *MF.getInfo<SIMachineFunctionInfo>();

  // Compute maximum number of VGPRs function can use using default/requested
  // minimum number of waves per execution unit.
  std::pair<unsigned, unsigned> WavesPerEU = MFI.getWavesPerEU();
  unsigned MaxNumVGPRs = getMaxNumVGPRs(WavesPerEU.first);

  // Check if maximum number of VGPRs was explicitly requested using
  // "amdgpu-num-vgpr" attribute.
  if (F.hasFnAttribute("amdgpu-num-vgpr")) {
    unsigned Requested = AMDGPU::getIntegerAttribute(
      F, "amdgpu-num-vgpr", MaxNumVGPRs);

    // Make sure requested value does not violate subtarget's specifications.
    if (Requested && Requested <= getReservedNumVGPRs(MF))
      Requested = 0;

    // Make sure requested value is compatible with values implied by
    // default/requested minimum/maximum number of waves per execution unit.
    if (Requested && Requested > getMaxNumVGPRs(WavesPerEU.first))
      Requested = 0;
    if (WavesPerEU.second &&
        Requested && Requested < getMinNumVGPRs(WavesPerEU.second))
      Requested = 0;

    if (Requested)
      MaxNumVGPRs = Requested;
  }

  return MaxNumVGPRs - getReservedNumVGPRs(MF);
}

namespace {
struct MemOpClusterMutation : ScheduleDAGMutation {
  const SIInstrInfo *TII;

  MemOpClusterMutation(const SIInstrInfo *tii) : TII(tii) {}

  void apply(ScheduleDAGInstrs *DAGInstrs) override {
    ScheduleDAGMI *DAG = static_cast<ScheduleDAGMI*>(DAGInstrs);

    SUnit *SUa = nullptr;
    // Search for two consequent memory operations and link them
    // to prevent scheduler from moving them apart.
    // In DAG pre-process SUnits are in the original order of
    // the instructions before scheduling.
    for (SUnit &SU : DAG->SUnits) {
      MachineInstr &MI2 = *SU.getInstr();
      if (!MI2.mayLoad() && !MI2.mayStore()) {
        SUa = nullptr;
        continue;
      }
      if (!SUa) {
        SUa = &SU;
        continue;
      }

      MachineInstr &MI1 = *SUa->getInstr();
      if ((TII->isVMEM(MI1) && TII->isVMEM(MI2)) ||
          (TII->isFLAT(MI1) && TII->isFLAT(MI2)) ||
          (TII->isSMRD(MI1) && TII->isSMRD(MI2)) ||
          (TII->isDS(MI1)   && TII->isDS(MI2))) {
        SU.addPredBarrier(SUa);

        for (const SDep &SI : SU.Preds) {
          if (SI.getSUnit() != SUa)
            SUa->addPred(SDep(SI.getSUnit(), SDep::Artificial));
        }

        if (&SU != &DAG->ExitSU) {
          for (const SDep &SI : SUa->Succs) {
            if (SI.getSUnit() != &SU)
              SI.getSUnit()->addPred(SDep(&SU, SDep::Artificial));
          }
        }
      }

      SUa = &SU;
    }
  }
};
} // namespace

void SISubtarget::getPostRAMutations(
    std::vector<std::unique_ptr<ScheduleDAGMutation>> &Mutations) const {
  Mutations.push_back(llvm::make_unique<MemOpClusterMutation>(&InstrInfo));
}