llvm.org GIT mirror llvm / 75c4f68 lib / Target / AMDGPU / AMDGPUCodeGenPrepare.cpp
75c4f68

Tree @75c4f68 (Download .tar.gz)

AMDGPUCodeGenPrepare.cpp @75c4f68raw · 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
//===-- AMDGPUCodeGenPrepare.cpp ------------------------------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
/// \file
/// This pass does misc. AMDGPU optimizations on IR before instruction
/// selection.
//
//===----------------------------------------------------------------------===//

#include "AMDGPU.h"
#include "AMDGPUSubtarget.h"
#include "AMDGPUTargetMachine.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/Analysis/DivergenceAnalysis.h"
#include "llvm/Analysis/Loads.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/CodeGen/TargetPassConfig.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/InstVisitor.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/Value.h"
#include "llvm/Pass.h"
#include "llvm/Support/Casting.h"
#include <cassert>
#include <iterator>

#define DEBUG_TYPE "amdgpu-codegenprepare"

using namespace llvm;

namespace {

static cl::opt<bool> WidenLoads(
  "amdgpu-codegenprepare-widen-constant-loads",
  cl::desc("Widen sub-dword constant address space loads in AMDGPUCodeGenPrepare"),
  cl::ReallyHidden,
  cl::init(true));

class AMDGPUCodeGenPrepare : public FunctionPass,
                             public InstVisitor<AMDGPUCodeGenPrepare, bool> {
  const SISubtarget *ST = nullptr;
  DivergenceAnalysis *DA = nullptr;
  Module *Mod = nullptr;
  bool HasUnsafeFPMath = false;
  AMDGPUAS AMDGPUASI;

  /// Copies exact/nsw/nuw flags (if any) from binary operation \p I to
  /// binary operation \p V.
  ///
  /// \returns Binary operation \p V.
  /// \returns \p T's base element bit width.
  unsigned getBaseElementBitWidth(const Type *T) const;

  /// \returns Equivalent 32 bit integer type for given type \p T. For example,
  /// if \p T is i7, then i32 is returned; if \p T is <3 x i12>, then <3 x i32>
  /// is returned.
  Type *getI32Ty(IRBuilder<> &B, const Type *T) const;

  /// \returns True if binary operation \p I is a signed binary operation, false
  /// otherwise.
  bool isSigned(const BinaryOperator &I) const;

  /// \returns True if the condition of 'select' operation \p I comes from a
  /// signed 'icmp' operation, false otherwise.
  bool isSigned(const SelectInst &I) const;

  /// \returns True if type \p T needs to be promoted to 32 bit integer type,
  /// false otherwise.
  bool needsPromotionToI32(const Type *T) const;

  /// Promotes uniform binary operation \p I to equivalent 32 bit binary
  /// operation.
  ///
  /// \details \p I's base element bit width must be greater than 1 and less
  /// than or equal 16. Promotion is done by sign or zero extending operands to
  /// 32 bits, replacing \p I with equivalent 32 bit binary operation, and
  /// truncating the result of 32 bit binary operation back to \p I's original
  /// type. Division operation is not promoted.
  ///
  /// \returns True if \p I is promoted to equivalent 32 bit binary operation,
  /// false otherwise.
  bool promoteUniformOpToI32(BinaryOperator &I) const;

  /// Promotes uniform 'icmp' operation \p I to 32 bit 'icmp' operation.
  ///
  /// \details \p I's base element bit width must be greater than 1 and less
  /// than or equal 16. Promotion is done by sign or zero extending operands to
  /// 32 bits, and replacing \p I with 32 bit 'icmp' operation.
  ///
  /// \returns True.
  bool promoteUniformOpToI32(ICmpInst &I) const;

  /// Promotes uniform 'select' operation \p I to 32 bit 'select'
  /// operation.
  ///
  /// \details \p I's base element bit width must be greater than 1 and less
  /// than or equal 16. Promotion is done by sign or zero extending operands to
  /// 32 bits, replacing \p I with 32 bit 'select' operation, and truncating the
  /// result of 32 bit 'select' operation back to \p I's original type.
  ///
  /// \returns True.
  bool promoteUniformOpToI32(SelectInst &I) const;

  /// Promotes uniform 'bitreverse' intrinsic \p I to 32 bit 'bitreverse'
  /// intrinsic.
  ///
  /// \details \p I's base element bit width must be greater than 1 and less
  /// than or equal 16. Promotion is done by zero extending the operand to 32
  /// bits, replacing \p I with 32 bit 'bitreverse' intrinsic, shifting the
  /// result of 32 bit 'bitreverse' intrinsic to the right with zero fill (the
  /// shift amount is 32 minus \p I's base element bit width), and truncating
  /// the result of the shift operation back to \p I's original type.
  ///
  /// \returns True.
  bool promoteUniformBitreverseToI32(IntrinsicInst &I) const;
  /// Widen a scalar load.
  ///
  /// \details \p Widen scalar load for uniform, small type loads from constant
  //  memory / to a full 32-bits and then truncate the input to allow a scalar
  //  load instead of a vector load.
  //
  /// \returns True.

  bool canWidenScalarExtLoad(LoadInst &I) const;

public:
  static char ID;

  AMDGPUCodeGenPrepare() : FunctionPass(ID) {}

  bool visitFDiv(BinaryOperator &I);

  bool visitInstruction(Instruction &I) { return false; }
  bool visitBinaryOperator(BinaryOperator &I);
  bool visitLoadInst(LoadInst &I);
  bool visitICmpInst(ICmpInst &I);
  bool visitSelectInst(SelectInst &I);

  bool visitIntrinsicInst(IntrinsicInst &I);
  bool visitBitreverseIntrinsicInst(IntrinsicInst &I);

  bool doInitialization(Module &M) override;
  bool runOnFunction(Function &F) override;

  StringRef getPassName() const override { return "AMDGPU IR optimizations"; }

  void getAnalysisUsage(AnalysisUsage &AU) const override {
    AU.addRequired<DivergenceAnalysis>();
    AU.setPreservesAll();
 }
};

} // end anonymous namespace

unsigned AMDGPUCodeGenPrepare::getBaseElementBitWidth(const Type *T) const {
  assert(needsPromotionToI32(T) && "T does not need promotion to i32");

  if (T->isIntegerTy())
    return T->getIntegerBitWidth();
  return cast<VectorType>(T)->getElementType()->getIntegerBitWidth();
}

Type *AMDGPUCodeGenPrepare::getI32Ty(IRBuilder<> &B, const Type *T) const {
  assert(needsPromotionToI32(T) && "T does not need promotion to i32");

  if (T->isIntegerTy())
    return B.getInt32Ty();
  return VectorType::get(B.getInt32Ty(), cast<VectorType>(T)->getNumElements());
}

bool AMDGPUCodeGenPrepare::isSigned(const BinaryOperator &I) const {
  return I.getOpcode() == Instruction::AShr ||
      I.getOpcode() == Instruction::SDiv || I.getOpcode() == Instruction::SRem;
}

bool AMDGPUCodeGenPrepare::isSigned(const SelectInst &I) const {
  return isa<ICmpInst>(I.getOperand(0)) ?
      cast<ICmpInst>(I.getOperand(0))->isSigned() : false;
}

bool AMDGPUCodeGenPrepare::needsPromotionToI32(const Type *T) const {
  const IntegerType *IntTy = dyn_cast<IntegerType>(T);
  if (IntTy && IntTy->getBitWidth() > 1 && IntTy->getBitWidth() <= 16)
    return true;

  if (const VectorType *VT = dyn_cast<VectorType>(T)) {
    // TODO: The set of packed operations is more limited, so may want to
    // promote some anyway.
    if (ST->hasVOP3PInsts())
      return false;

    return needsPromotionToI32(VT->getElementType());
  }

  return false;
}

// Return true if the op promoted to i32 should have nsw set.
static bool promotedOpIsNSW(const Instruction &I) {
  switch (I.getOpcode()) {
  case Instruction::Shl:
  case Instruction::Add:
  case Instruction::Sub:
    return true;
  case Instruction::Mul:
    return I.hasNoUnsignedWrap();
  default:
    return false;
  }
}

// Return true if the op promoted to i32 should have nuw set.
static bool promotedOpIsNUW(const Instruction &I) {
  switch (I.getOpcode()) {
  case Instruction::Shl:
  case Instruction::Add:
  case Instruction::Mul:
    return true;
  case Instruction::Sub:
    return I.hasNoUnsignedWrap();
  default:
    return false;
  }
}

bool AMDGPUCodeGenPrepare::canWidenScalarExtLoad(LoadInst &I) const {
  Type *Ty = I.getType();
  const DataLayout &DL = Mod->getDataLayout();
  int TySize = DL.getTypeSizeInBits(Ty);
  unsigned Align = I.getAlignment() ?
                   I.getAlignment() : DL.getABITypeAlignment(Ty);

  return I.isSimple() && TySize < 32 && Align >= 4 && DA->isUniform(&I);
}

bool AMDGPUCodeGenPrepare::promoteUniformOpToI32(BinaryOperator &I) const {
  assert(needsPromotionToI32(I.getType()) &&
         "I does not need promotion to i32");

  if (I.getOpcode() == Instruction::SDiv ||
      I.getOpcode() == Instruction::UDiv)
    return false;

  IRBuilder<> Builder(&I);
  Builder.SetCurrentDebugLocation(I.getDebugLoc());

  Type *I32Ty = getI32Ty(Builder, I.getType());
  Value *ExtOp0 = nullptr;
  Value *ExtOp1 = nullptr;
  Value *ExtRes = nullptr;
  Value *TruncRes = nullptr;

  if (isSigned(I)) {
    ExtOp0 = Builder.CreateSExt(I.getOperand(0), I32Ty);
    ExtOp1 = Builder.CreateSExt(I.getOperand(1), I32Ty);
  } else {
    ExtOp0 = Builder.CreateZExt(I.getOperand(0), I32Ty);
    ExtOp1 = Builder.CreateZExt(I.getOperand(1), I32Ty);
  }

  ExtRes = Builder.CreateBinOp(I.getOpcode(), ExtOp0, ExtOp1);
  if (Instruction *Inst = dyn_cast<Instruction>(ExtRes)) {
    if (promotedOpIsNSW(cast<Instruction>(I)))
      Inst->setHasNoSignedWrap();

    if (promotedOpIsNUW(cast<Instruction>(I)))
      Inst->setHasNoUnsignedWrap();

    if (const auto *ExactOp = dyn_cast<PossiblyExactOperator>(&I))
      Inst->setIsExact(ExactOp->isExact());
  }

  TruncRes = Builder.CreateTrunc(ExtRes, I.getType());

  I.replaceAllUsesWith(TruncRes);
  I.eraseFromParent();

  return true;
}

bool AMDGPUCodeGenPrepare::promoteUniformOpToI32(ICmpInst &I) const {
  assert(needsPromotionToI32(I.getOperand(0)->getType()) &&
         "I does not need promotion to i32");

  IRBuilder<> Builder(&I);
  Builder.SetCurrentDebugLocation(I.getDebugLoc());

  Type *I32Ty = getI32Ty(Builder, I.getOperand(0)->getType());
  Value *ExtOp0 = nullptr;
  Value *ExtOp1 = nullptr;
  Value *NewICmp  = nullptr;

  if (I.isSigned()) {
    ExtOp0 = Builder.CreateSExt(I.getOperand(0), I32Ty);
    ExtOp1 = Builder.CreateSExt(I.getOperand(1), I32Ty);
  } else {
    ExtOp0 = Builder.CreateZExt(I.getOperand(0), I32Ty);
    ExtOp1 = Builder.CreateZExt(I.getOperand(1), I32Ty);
  }
  NewICmp = Builder.CreateICmp(I.getPredicate(), ExtOp0, ExtOp1);

  I.replaceAllUsesWith(NewICmp);
  I.eraseFromParent();

  return true;
}

bool AMDGPUCodeGenPrepare::promoteUniformOpToI32(SelectInst &I) const {
  assert(needsPromotionToI32(I.getType()) &&
         "I does not need promotion to i32");

  IRBuilder<> Builder(&I);
  Builder.SetCurrentDebugLocation(I.getDebugLoc());

  Type *I32Ty = getI32Ty(Builder, I.getType());
  Value *ExtOp1 = nullptr;
  Value *ExtOp2 = nullptr;
  Value *ExtRes = nullptr;
  Value *TruncRes = nullptr;

  if (isSigned(I)) {
    ExtOp1 = Builder.CreateSExt(I.getOperand(1), I32Ty);
    ExtOp2 = Builder.CreateSExt(I.getOperand(2), I32Ty);
  } else {
    ExtOp1 = Builder.CreateZExt(I.getOperand(1), I32Ty);
    ExtOp2 = Builder.CreateZExt(I.getOperand(2), I32Ty);
  }
  ExtRes = Builder.CreateSelect(I.getOperand(0), ExtOp1, ExtOp2);
  TruncRes = Builder.CreateTrunc(ExtRes, I.getType());

  I.replaceAllUsesWith(TruncRes);
  I.eraseFromParent();

  return true;
}

bool AMDGPUCodeGenPrepare::promoteUniformBitreverseToI32(
    IntrinsicInst &I) const {
  assert(I.getIntrinsicID() == Intrinsic::bitreverse &&
         "I must be bitreverse intrinsic");
  assert(needsPromotionToI32(I.getType()) &&
         "I does not need promotion to i32");

  IRBuilder<> Builder(&I);
  Builder.SetCurrentDebugLocation(I.getDebugLoc());

  Type *I32Ty = getI32Ty(Builder, I.getType());
  Function *I32 =
      Intrinsic::getDeclaration(Mod, Intrinsic::bitreverse, { I32Ty });
  Value *ExtOp = Builder.CreateZExt(I.getOperand(0), I32Ty);
  Value *ExtRes = Builder.CreateCall(I32, { ExtOp });
  Value *LShrOp =
      Builder.CreateLShr(ExtRes, 32 - getBaseElementBitWidth(I.getType()));
  Value *TruncRes =
      Builder.CreateTrunc(LShrOp, I.getType());

  I.replaceAllUsesWith(TruncRes);
  I.eraseFromParent();

  return true;
}

static bool shouldKeepFDivF32(Value *Num, bool UnsafeDiv, bool HasDenormals) {
  const ConstantFP *CNum = dyn_cast<ConstantFP>(Num);
  if (!CNum)
    return HasDenormals;

  if (UnsafeDiv)
    return true;

  bool IsOne = CNum->isExactlyValue(+1.0) || CNum->isExactlyValue(-1.0);

  // Reciprocal f32 is handled separately without denormals.
  return HasDenormals ^ IsOne;
}

// Insert an intrinsic for fast fdiv for safe math situations where we can
// reduce precision. Leave fdiv for situations where the generic node is
// expected to be optimized.
bool AMDGPUCodeGenPrepare::visitFDiv(BinaryOperator &FDiv) {
  Type *Ty = FDiv.getType();

  if (!Ty->getScalarType()->isFloatTy())
    return false;

  MDNode *FPMath = FDiv.getMetadata(LLVMContext::MD_fpmath);
  if (!FPMath)
    return false;

  const FPMathOperator *FPOp = cast<const FPMathOperator>(&FDiv);
  float ULP = FPOp->getFPAccuracy();
  if (ULP < 2.5f)
    return false;

  FastMathFlags FMF = FPOp->getFastMathFlags();
  bool UnsafeDiv = HasUnsafeFPMath || FMF.isFast() ||
                                      FMF.allowReciprocal();

  // With UnsafeDiv node will be optimized to just rcp and mul.
  if (UnsafeDiv)
    return false;

  IRBuilder<> Builder(FDiv.getParent(), std::next(FDiv.getIterator()), FPMath);
  Builder.setFastMathFlags(FMF);
  Builder.SetCurrentDebugLocation(FDiv.getDebugLoc());

  Function *Decl = Intrinsic::getDeclaration(Mod, Intrinsic::amdgcn_fdiv_fast);

  Value *Num = FDiv.getOperand(0);
  Value *Den = FDiv.getOperand(1);

  Value *NewFDiv = nullptr;

  bool HasDenormals = ST->hasFP32Denormals();
  if (VectorType *VT = dyn_cast<VectorType>(Ty)) {
    NewFDiv = UndefValue::get(VT);

    // FIXME: Doesn't do the right thing for cases where the vector is partially
    // constant. This works when the scalarizer pass is run first.
    for (unsigned I = 0, E = VT->getNumElements(); I != E; ++I) {
      Value *NumEltI = Builder.CreateExtractElement(Num, I);
      Value *DenEltI = Builder.CreateExtractElement(Den, I);
      Value *NewElt;

      if (shouldKeepFDivF32(NumEltI, UnsafeDiv, HasDenormals)) {
        NewElt = Builder.CreateFDiv(NumEltI, DenEltI);
      } else {
        NewElt = Builder.CreateCall(Decl, { NumEltI, DenEltI });
      }

      NewFDiv = Builder.CreateInsertElement(NewFDiv, NewElt, I);
    }
  } else {
    if (!shouldKeepFDivF32(Num, UnsafeDiv, HasDenormals))
      NewFDiv = Builder.CreateCall(Decl, { Num, Den });
  }

  if (NewFDiv) {
    FDiv.replaceAllUsesWith(NewFDiv);
    NewFDiv->takeName(&FDiv);
    FDiv.eraseFromParent();
  }

  return !!NewFDiv;
}

static bool hasUnsafeFPMath(const Function &F) {
  Attribute Attr = F.getFnAttribute("unsafe-fp-math");
  return Attr.getValueAsString() == "true";
}

bool AMDGPUCodeGenPrepare::visitBinaryOperator(BinaryOperator &I) {
  bool Changed = false;

  if (ST->has16BitInsts() && needsPromotionToI32(I.getType()) &&
      DA->isUniform(&I))
    Changed |= promoteUniformOpToI32(I);

  return Changed;
}

bool AMDGPUCodeGenPrepare::visitLoadInst(LoadInst &I) {
  if (!WidenLoads)
    return false;

  if ((I.getPointerAddressSpace() == AMDGPUASI.CONSTANT_ADDRESS ||
       I.getPointerAddressSpace() == AMDGPUASI.CONSTANT_ADDRESS_32BIT) &&
      canWidenScalarExtLoad(I)) {
    IRBuilder<> Builder(&I);
    Builder.SetCurrentDebugLocation(I.getDebugLoc());

    Type *I32Ty = Builder.getInt32Ty();
    Type *PT = PointerType::get(I32Ty, I.getPointerAddressSpace());
    Value *BitCast= Builder.CreateBitCast(I.getPointerOperand(), PT);
    LoadInst *WidenLoad = Builder.CreateLoad(BitCast);
    WidenLoad->copyMetadata(I);

    // If we have range metadata, we need to convert the type, and not make
    // assumptions about the high bits.
    if (auto *Range = WidenLoad->getMetadata(LLVMContext::MD_range)) {
      ConstantInt *Lower =
        mdconst::extract<ConstantInt>(Range->getOperand(0));

      if (Lower->getValue().isNullValue()) {
        WidenLoad->setMetadata(LLVMContext::MD_range, nullptr);
      } else {
        Metadata *LowAndHigh[] = {
          ConstantAsMetadata::get(ConstantInt::get(I32Ty, Lower->getValue().zext(32))),
          // Don't make assumptions about the high bits.
          ConstantAsMetadata::get(ConstantInt::get(I32Ty, 0))
        };

        WidenLoad->setMetadata(LLVMContext::MD_range,
                               MDNode::get(Mod->getContext(), LowAndHigh));
      }
    }

    int TySize = Mod->getDataLayout().getTypeSizeInBits(I.getType());
    Type *IntNTy = Builder.getIntNTy(TySize);
    Value *ValTrunc = Builder.CreateTrunc(WidenLoad, IntNTy);
    Value *ValOrig = Builder.CreateBitCast(ValTrunc, I.getType());
    I.replaceAllUsesWith(ValOrig);
    I.eraseFromParent();
    return true;
  }

  return false;
}

bool AMDGPUCodeGenPrepare::visitICmpInst(ICmpInst &I) {
  bool Changed = false;

  if (ST->has16BitInsts() && needsPromotionToI32(I.getOperand(0)->getType()) &&
      DA->isUniform(&I))
    Changed |= promoteUniformOpToI32(I);

  return Changed;
}

bool AMDGPUCodeGenPrepare::visitSelectInst(SelectInst &I) {
  bool Changed = false;

  if (ST->has16BitInsts() && needsPromotionToI32(I.getType()) &&
      DA->isUniform(&I))
    Changed |= promoteUniformOpToI32(I);

  return Changed;
}

bool AMDGPUCodeGenPrepare::visitIntrinsicInst(IntrinsicInst &I) {
  switch (I.getIntrinsicID()) {
  case Intrinsic::bitreverse:
    return visitBitreverseIntrinsicInst(I);
  default:
    return false;
  }
}

bool AMDGPUCodeGenPrepare::visitBitreverseIntrinsicInst(IntrinsicInst &I) {
  bool Changed = false;

  if (ST->has16BitInsts() && needsPromotionToI32(I.getType()) &&
      DA->isUniform(&I))
    Changed |= promoteUniformBitreverseToI32(I);

  return Changed;
}

bool AMDGPUCodeGenPrepare::doInitialization(Module &M) {
  Mod = &M;
  return false;
}

bool AMDGPUCodeGenPrepare::runOnFunction(Function &F) {
  if (skipFunction(F))
    return false;

  auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
  if (!TPC)
    return false;

  const TargetMachine &TM = TPC->getTM<TargetMachine>();
  ST = &TM.getSubtarget<SISubtarget>(F);
  DA = &getAnalysis<DivergenceAnalysis>();
  HasUnsafeFPMath = hasUnsafeFPMath(F);

  bool MadeChange = false;

  for (BasicBlock &BB : F) {
    BasicBlock::iterator Next;
    for (BasicBlock::iterator I = BB.begin(), E = BB.end(); I != E; I = Next) {
      Next = std::next(I);
      MadeChange |= visit(*I);
    }
  }

  return MadeChange;
}

INITIALIZE_PASS_BEGIN(AMDGPUCodeGenPrepare, DEBUG_TYPE,
                      "AMDGPU IR optimizations", false, false)
INITIALIZE_PASS_DEPENDENCY(DivergenceAnalysis)
INITIALIZE_PASS_END(AMDGPUCodeGenPrepare, DEBUG_TYPE, "AMDGPU IR optimizations",
                    false, false)

char AMDGPUCodeGenPrepare::ID = 0;

FunctionPass *llvm::createAMDGPUCodeGenPreparePass() {
  return new AMDGPUCodeGenPrepare();
}