llvm.org GIT mirror llvm / 56efe24 lib / Transforms / Scalar / MemCpyOptimizer.cpp
56efe24

Tree @56efe24 (Download .tar.gz)

MemCpyOptimizer.cpp @56efe24raw · 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
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
//===- MemCpyOptimizer.cpp - Optimize use of memcpy and friends -----------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This pass performs various transformations related to eliminating memcpy
// calls, or transforming sets of stores into memset's.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "memcpyopt"
#include "llvm/Transforms/Scalar.h"
#include "llvm/GlobalVariable.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/Instructions.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/IRBuilder.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetLibraryInfo.h"
#include <list>
using namespace llvm;

STATISTIC(NumMemCpyInstr, "Number of memcpy instructions deleted");
STATISTIC(NumMemSetInfer, "Number of memsets inferred");
STATISTIC(NumMoveToCpy,   "Number of memmoves converted to memcpy");
STATISTIC(NumCpyToSet,    "Number of memcpys converted to memset");

static int64_t GetOffsetFromIndex(const GetElementPtrInst *GEP, unsigned Idx,
                                  bool &VariableIdxFound, const TargetData &TD){
  // Skip over the first indices.
  gep_type_iterator GTI = gep_type_begin(GEP);
  for (unsigned i = 1; i != Idx; ++i, ++GTI)
    /*skip along*/;
  
  // Compute the offset implied by the rest of the indices.
  int64_t Offset = 0;
  for (unsigned i = Idx, e = GEP->getNumOperands(); i != e; ++i, ++GTI) {
    ConstantInt *OpC = dyn_cast<ConstantInt>(GEP->getOperand(i));
    if (OpC == 0)
      return VariableIdxFound = true;
    if (OpC->isZero()) continue;  // No offset.

    // Handle struct indices, which add their field offset to the pointer.
    if (StructType *STy = dyn_cast<StructType>(*GTI)) {
      Offset += TD.getStructLayout(STy)->getElementOffset(OpC->getZExtValue());
      continue;
    }
    
    // Otherwise, we have a sequential type like an array or vector.  Multiply
    // the index by the ElementSize.
    uint64_t Size = TD.getTypeAllocSize(GTI.getIndexedType());
    Offset += Size*OpC->getSExtValue();
  }

  return Offset;
}

/// IsPointerOffset - Return true if Ptr1 is provably equal to Ptr2 plus a
/// constant offset, and return that constant offset.  For example, Ptr1 might
/// be &A[42], and Ptr2 might be &A[40].  In this case offset would be -8.
static bool IsPointerOffset(Value *Ptr1, Value *Ptr2, int64_t &Offset,
                            const TargetData &TD) {
  Ptr1 = Ptr1->stripPointerCasts();
  Ptr2 = Ptr2->stripPointerCasts();
  GetElementPtrInst *GEP1 = dyn_cast<GetElementPtrInst>(Ptr1);
  GetElementPtrInst *GEP2 = dyn_cast<GetElementPtrInst>(Ptr2);
  
  bool VariableIdxFound = false;

  // If one pointer is a GEP and the other isn't, then see if the GEP is a
  // constant offset from the base, as in "P" and "gep P, 1".
  if (GEP1 && GEP2 == 0 && GEP1->getOperand(0)->stripPointerCasts() == Ptr2) {
    Offset = -GetOffsetFromIndex(GEP1, 1, VariableIdxFound, TD);
    return !VariableIdxFound;
  }

  if (GEP2 && GEP1 == 0 && GEP2->getOperand(0)->stripPointerCasts() == Ptr1) {
    Offset = GetOffsetFromIndex(GEP2, 1, VariableIdxFound, TD);
    return !VariableIdxFound;
  }
  
  // Right now we handle the case when Ptr1/Ptr2 are both GEPs with an identical
  // base.  After that base, they may have some number of common (and
  // potentially variable) indices.  After that they handle some constant
  // offset, which determines their offset from each other.  At this point, we
  // handle no other case.
  if (!GEP1 || !GEP2 || GEP1->getOperand(0) != GEP2->getOperand(0))
    return false;
  
  // Skip any common indices and track the GEP types.
  unsigned Idx = 1;
  for (; Idx != GEP1->getNumOperands() && Idx != GEP2->getNumOperands(); ++Idx)
    if (GEP1->getOperand(Idx) != GEP2->getOperand(Idx))
      break;

  int64_t Offset1 = GetOffsetFromIndex(GEP1, Idx, VariableIdxFound, TD);
  int64_t Offset2 = GetOffsetFromIndex(GEP2, Idx, VariableIdxFound, TD);
  if (VariableIdxFound) return false;
  
  Offset = Offset2-Offset1;
  return true;
}


/// MemsetRange - Represents a range of memset'd bytes with the ByteVal value.
/// This allows us to analyze stores like:
///   store 0 -> P+1
///   store 0 -> P+0
///   store 0 -> P+3
///   store 0 -> P+2
/// which sometimes happens with stores to arrays of structs etc.  When we see
/// the first store, we make a range [1, 2).  The second store extends the range
/// to [0, 2).  The third makes a new range [2, 3).  The fourth store joins the
/// two ranges into [0, 3) which is memset'able.
namespace {
struct MemsetRange {
  // Start/End - A semi range that describes the span that this range covers.
  // The range is closed at the start and open at the end: [Start, End).  
  int64_t Start, End;

  /// StartPtr - The getelementptr instruction that points to the start of the
  /// range.
  Value *StartPtr;
  
  /// Alignment - The known alignment of the first store.
  unsigned Alignment;
  
  /// TheStores - The actual stores that make up this range.
  SmallVector<Instruction*, 16> TheStores;
  
  bool isProfitableToUseMemset(const TargetData &TD) const;

};
} // end anon namespace

bool MemsetRange::isProfitableToUseMemset(const TargetData &TD) const {
  // If we found more than 8 stores to merge or 64 bytes, use memset.
  if (TheStores.size() >= 8 || End-Start >= 64) return true;

  // If there is nothing to merge, don't do anything.
  if (TheStores.size() < 2) return false;
  
  // If any of the stores are a memset, then it is always good to extend the
  // memset.
  for (unsigned i = 0, e = TheStores.size(); i != e; ++i)
    if (!isa<StoreInst>(TheStores[i]))
      return true;
  
  // Assume that the code generator is capable of merging pairs of stores
  // together if it wants to.
  if (TheStores.size() == 2) return false;
  
  // If we have fewer than 8 stores, it can still be worthwhile to do this.
  // For example, merging 4 i8 stores into an i32 store is useful almost always.
  // However, merging 2 32-bit stores isn't useful on a 32-bit architecture (the
  // memset will be split into 2 32-bit stores anyway) and doing so can
  // pessimize the llvm optimizer.
  //
  // Since we don't have perfect knowledge here, make some assumptions: assume
  // the maximum GPR width is the same size as the pointer size and assume that
  // this width can be stored.  If so, check to see whether we will end up
  // actually reducing the number of stores used.
  unsigned Bytes = unsigned(End-Start);
  unsigned NumPointerStores = Bytes/TD.getPointerSize();
  
  // Assume the remaining bytes if any are done a byte at a time.
  unsigned NumByteStores = Bytes - NumPointerStores*TD.getPointerSize();
  
  // If we will reduce the # stores (according to this heuristic), do the
  // transformation.  This encourages merging 4 x i8 -> i32 and 2 x i16 -> i32
  // etc.
  return TheStores.size() > NumPointerStores+NumByteStores;
}    


namespace {
class MemsetRanges {
  /// Ranges - A sorted list of the memset ranges.  We use std::list here
  /// because each element is relatively large and expensive to copy.
  std::list<MemsetRange> Ranges;
  typedef std::list<MemsetRange>::iterator range_iterator;
  const TargetData &TD;
public:
  MemsetRanges(const TargetData &td) : TD(td) {}
  
  typedef std::list<MemsetRange>::const_iterator const_iterator;
  const_iterator begin() const { return Ranges.begin(); }
  const_iterator end() const { return Ranges.end(); }
  bool empty() const { return Ranges.empty(); }
  
  void addInst(int64_t OffsetFromFirst, Instruction *Inst) {
    if (StoreInst *SI = dyn_cast<StoreInst>(Inst))
      addStore(OffsetFromFirst, SI);
    else
      addMemSet(OffsetFromFirst, cast<MemSetInst>(Inst));
  }

  void addStore(int64_t OffsetFromFirst, StoreInst *SI) {
    int64_t StoreSize = TD.getTypeStoreSize(SI->getOperand(0)->getType());
    
    addRange(OffsetFromFirst, StoreSize,
             SI->getPointerOperand(), SI->getAlignment(), SI);
  }
  
  void addMemSet(int64_t OffsetFromFirst, MemSetInst *MSI) {
    int64_t Size = cast<ConstantInt>(MSI->getLength())->getZExtValue();
    addRange(OffsetFromFirst, Size, MSI->getDest(), MSI->getAlignment(), MSI);
  }
  
  void addRange(int64_t Start, int64_t Size, Value *Ptr,
                unsigned Alignment, Instruction *Inst);

};
  
} // end anon namespace


/// addRange - Add a new store to the MemsetRanges data structure.  This adds a
/// new range for the specified store at the specified offset, merging into
/// existing ranges as appropriate.
///
/// Do a linear search of the ranges to see if this can be joined and/or to
/// find the insertion point in the list.  We keep the ranges sorted for
/// simplicity here.  This is a linear search of a linked list, which is ugly,
/// however the number of ranges is limited, so this won't get crazy slow.
void MemsetRanges::addRange(int64_t Start, int64_t Size, Value *Ptr,
                            unsigned Alignment, Instruction *Inst) {
  int64_t End = Start+Size;
  range_iterator I = Ranges.begin(), E = Ranges.end();
  
  while (I != E && Start > I->End)
    ++I;
  
  // We now know that I == E, in which case we didn't find anything to merge
  // with, or that Start <= I->End.  If End < I->Start or I == E, then we need
  // to insert a new range.  Handle this now.
  if (I == E || End < I->Start) {
    MemsetRange &R = *Ranges.insert(I, MemsetRange());
    R.Start        = Start;
    R.End          = End;
    R.StartPtr     = Ptr;
    R.Alignment    = Alignment;
    R.TheStores.push_back(Inst);
    return;
  }
  
  // This store overlaps with I, add it.
  I->TheStores.push_back(Inst);
  
  // At this point, we may have an interval that completely contains our store.
  // If so, just add it to the interval and return.
  if (I->Start <= Start && I->End >= End)
    return;
  
  // Now we know that Start <= I->End and End >= I->Start so the range overlaps
  // but is not entirely contained within the range.
  
  // See if the range extends the start of the range.  In this case, it couldn't
  // possibly cause it to join the prior range, because otherwise we would have
  // stopped on *it*.
  if (Start < I->Start) {
    I->Start = Start;
    I->StartPtr = Ptr;
    I->Alignment = Alignment;
  }
    
  // Now we know that Start <= I->End and Start >= I->Start (so the startpoint
  // is in or right at the end of I), and that End >= I->Start.  Extend I out to
  // End.
  if (End > I->End) {
    I->End = End;
    range_iterator NextI = I;
    while (++NextI != E && End >= NextI->Start) {
      // Merge the range in.
      I->TheStores.append(NextI->TheStores.begin(), NextI->TheStores.end());
      if (NextI->End > I->End)
        I->End = NextI->End;
      Ranges.erase(NextI);
      NextI = I;
    }
  }
}

//===----------------------------------------------------------------------===//
//                         MemCpyOpt Pass
//===----------------------------------------------------------------------===//

namespace {
  class MemCpyOpt : public FunctionPass {
    MemoryDependenceAnalysis *MD;
    TargetLibraryInfo *TLI;
    const TargetData *TD;
  public:
    static char ID; // Pass identification, replacement for typeid
    MemCpyOpt() : FunctionPass(ID) {
      initializeMemCpyOptPass(*PassRegistry::getPassRegistry());
      MD = 0;
      TLI = 0;
      TD = 0;
    }

    bool runOnFunction(Function &F);

  private:
    // This transformation requires dominator postdominator info
    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
      AU.setPreservesCFG();
      AU.addRequired<DominatorTree>();
      AU.addRequired<MemoryDependenceAnalysis>();
      AU.addRequired<AliasAnalysis>();
      AU.addRequired<TargetLibraryInfo>();
      AU.addPreserved<AliasAnalysis>();
      AU.addPreserved<MemoryDependenceAnalysis>();
    }
  
    // Helper fuctions
    bool processStore(StoreInst *SI, BasicBlock::iterator &BBI);
    bool processMemSet(MemSetInst *SI, BasicBlock::iterator &BBI);
    bool processMemCpy(MemCpyInst *M);
    bool processMemMove(MemMoveInst *M);
    bool performCallSlotOptzn(Instruction *cpy, Value *cpyDst, Value *cpySrc,
                              uint64_t cpyLen, CallInst *C);
    bool processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep,
                                       uint64_t MSize);
    bool processByValArgument(CallSite CS, unsigned ArgNo);
    Instruction *tryMergingIntoMemset(Instruction *I, Value *StartPtr,
                                      Value *ByteVal);

    bool iterateOnFunction(Function &F);
  };
  
  char MemCpyOpt::ID = 0;
}

// createMemCpyOptPass - The public interface to this file...
FunctionPass *llvm::createMemCpyOptPass() { return new MemCpyOpt(); }

INITIALIZE_PASS_BEGIN(MemCpyOpt, "memcpyopt", "MemCpy Optimization",
                      false, false)
INITIALIZE_PASS_DEPENDENCY(DominatorTree)
INITIALIZE_PASS_DEPENDENCY(MemoryDependenceAnalysis)
INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
INITIALIZE_PASS_END(MemCpyOpt, "memcpyopt", "MemCpy Optimization",
                    false, false)

/// tryMergingIntoMemset - When scanning forward over instructions, we look for
/// some other patterns to fold away.  In particular, this looks for stores to
/// neighboring locations of memory.  If it sees enough consecutive ones, it
/// attempts to merge them together into a memcpy/memset.
Instruction *MemCpyOpt::tryMergingIntoMemset(Instruction *StartInst, 
                                             Value *StartPtr, Value *ByteVal) {
  if (TD == 0) return 0;
  
  // Okay, so we now have a single store that can be splatable.  Scan to find
  // all subsequent stores of the same value to offset from the same pointer.
  // Join these together into ranges, so we can decide whether contiguous blocks
  // are stored.
  MemsetRanges Ranges(*TD);
  
  BasicBlock::iterator BI = StartInst;
  for (++BI; !isa<TerminatorInst>(BI); ++BI) {
    if (!isa<StoreInst>(BI) && !isa<MemSetInst>(BI)) {
      // If the instruction is readnone, ignore it, otherwise bail out.  We
      // don't even allow readonly here because we don't want something like:
      // A[1] = 2; strlen(A); A[2] = 2; -> memcpy(A, ...); strlen(A).
      if (BI->mayWriteToMemory() || BI->mayReadFromMemory())
        break;
      continue;
    }
    
    if (StoreInst *NextStore = dyn_cast<StoreInst>(BI)) {
      // If this is a store, see if we can merge it in.
      if (!NextStore->isSimple()) break;
    
      // Check to see if this stored value is of the same byte-splattable value.
      if (ByteVal != isBytewiseValue(NextStore->getOperand(0)))
        break;
      
      // Check to see if this store is to a constant offset from the start ptr.
      int64_t Offset;
      if (!IsPointerOffset(StartPtr, NextStore->getPointerOperand(),
                           Offset, *TD))
        break;
      
      Ranges.addStore(Offset, NextStore);
    } else {
      MemSetInst *MSI = cast<MemSetInst>(BI);
      
      if (MSI->isVolatile() || ByteVal != MSI->getValue() ||
          !isa<ConstantInt>(MSI->getLength()))
        break;
      
      // Check to see if this store is to a constant offset from the start ptr.
      int64_t Offset;
      if (!IsPointerOffset(StartPtr, MSI->getDest(), Offset, *TD))
        break;
      
      Ranges.addMemSet(Offset, MSI);
    }
  }
  
  // If we have no ranges, then we just had a single store with nothing that
  // could be merged in.  This is a very common case of course.
  if (Ranges.empty())
    return 0;
  
  // If we had at least one store that could be merged in, add the starting
  // store as well.  We try to avoid this unless there is at least something
  // interesting as a small compile-time optimization.
  Ranges.addInst(0, StartInst);

  // If we create any memsets, we put it right before the first instruction that
  // isn't part of the memset block.  This ensure that the memset is dominated
  // by any addressing instruction needed by the start of the block.
  IRBuilder<> Builder(BI);

  // Now that we have full information about ranges, loop over the ranges and
  // emit memset's for anything big enough to be worthwhile.
  Instruction *AMemSet = 0;
  for (MemsetRanges::const_iterator I = Ranges.begin(), E = Ranges.end();
       I != E; ++I) {
    const MemsetRange &Range = *I;
    
    if (Range.TheStores.size() == 1) continue;
    
    // If it is profitable to lower this range to memset, do so now.
    if (!Range.isProfitableToUseMemset(*TD))
      continue;
    
    // Otherwise, we do want to transform this!  Create a new memset.
    // Get the starting pointer of the block.
    StartPtr = Range.StartPtr;
    
    // Determine alignment
    unsigned Alignment = Range.Alignment;
    if (Alignment == 0) {
      Type *EltType = 
        cast<PointerType>(StartPtr->getType())->getElementType();
      Alignment = TD->getABITypeAlignment(EltType);
    }
    
    AMemSet = 
      Builder.CreateMemSet(StartPtr, ByteVal, Range.End-Range.Start, Alignment);
    
    DEBUG(dbgs() << "Replace stores:\n";
          for (unsigned i = 0, e = Range.TheStores.size(); i != e; ++i)
            dbgs() << *Range.TheStores[i] << '\n';
          dbgs() << "With: " << *AMemSet << '\n');

    if (!Range.TheStores.empty())
      AMemSet->setDebugLoc(Range.TheStores[0]->getDebugLoc());

    // Zap all the stores.
    for (SmallVector<Instruction*, 16>::const_iterator
         SI = Range.TheStores.begin(),
         SE = Range.TheStores.end(); SI != SE; ++SI) {
      MD->removeInstruction(*SI);
      (*SI)->eraseFromParent();
    }
    ++NumMemSetInfer;
  }
  
  return AMemSet;
}


bool MemCpyOpt::processStore(StoreInst *SI, BasicBlock::iterator &BBI) {
  if (!SI->isSimple()) return false;
  
  if (TD == 0) return false;

  // Detect cases where we're performing call slot forwarding, but
  // happen to be using a load-store pair to implement it, rather than
  // a memcpy.
  if (LoadInst *LI = dyn_cast<LoadInst>(SI->getOperand(0))) {
    if (LI->isSimple() && LI->hasOneUse() &&
        LI->getParent() == SI->getParent()) {
      MemDepResult ldep = MD->getDependency(LI);
      CallInst *C = 0;
      if (ldep.isClobber() && !isa<MemCpyInst>(ldep.getInst()))
        C = dyn_cast<CallInst>(ldep.getInst());

      if (C) {
        // Check that nothing touches the dest of the "copy" between
        // the call and the store.
        AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
        AliasAnalysis::Location StoreLoc = AA.getLocation(SI);
        for (BasicBlock::iterator I = --BasicBlock::iterator(SI),
                                  E = C; I != E; --I) {
          if (AA.getModRefInfo(&*I, StoreLoc) != AliasAnalysis::NoModRef) {
            C = 0;
            break;
          }
        }
      }

      if (C) {
        bool changed = performCallSlotOptzn(LI,
                        SI->getPointerOperand()->stripPointerCasts(), 
                        LI->getPointerOperand()->stripPointerCasts(),
                        TD->getTypeStoreSize(SI->getOperand(0)->getType()), C);
        if (changed) {
          MD->removeInstruction(SI);
          SI->eraseFromParent();
          MD->removeInstruction(LI);
          LI->eraseFromParent();
          ++NumMemCpyInstr;
          return true;
        }
      }
    }
  }
  
  // There are two cases that are interesting for this code to handle: memcpy
  // and memset.  Right now we only handle memset.
  
  // Ensure that the value being stored is something that can be memset'able a
  // byte at a time like "0" or "-1" or any width, as well as things like
  // 0xA0A0A0A0 and 0.0.
  if (Value *ByteVal = isBytewiseValue(SI->getOperand(0)))
    if (Instruction *I = tryMergingIntoMemset(SI, SI->getPointerOperand(),
                                              ByteVal)) {
      BBI = I;  // Don't invalidate iterator.
      return true;
    }
  
  return false;
}

bool MemCpyOpt::processMemSet(MemSetInst *MSI, BasicBlock::iterator &BBI) {
  // See if there is another memset or store neighboring this memset which
  // allows us to widen out the memset to do a single larger store.
  if (isa<ConstantInt>(MSI->getLength()) && !MSI->isVolatile())
    if (Instruction *I = tryMergingIntoMemset(MSI, MSI->getDest(),
                                              MSI->getValue())) {
      BBI = I;  // Don't invalidate iterator.
      return true;
    }
  return false;
}


/// performCallSlotOptzn - takes a memcpy and a call that it depends on,
/// and checks for the possibility of a call slot optimization by having
/// the call write its result directly into the destination of the memcpy.
bool MemCpyOpt::performCallSlotOptzn(Instruction *cpy,
                                     Value *cpyDest, Value *cpySrc,
                                     uint64_t cpyLen, CallInst *C) {
  // The general transformation to keep in mind is
  //
  //   call @func(..., src, ...)
  //   memcpy(dest, src, ...)
  //
  // ->
  //
  //   memcpy(dest, src, ...)
  //   call @func(..., dest, ...)
  //
  // Since moving the memcpy is technically awkward, we additionally check that
  // src only holds uninitialized values at the moment of the call, meaning that
  // the memcpy can be discarded rather than moved.

  // Deliberately get the source and destination with bitcasts stripped away,
  // because we'll need to do type comparisons based on the underlying type.
  CallSite CS(C);

  // Require that src be an alloca.  This simplifies the reasoning considerably.
  AllocaInst *srcAlloca = dyn_cast<AllocaInst>(cpySrc);
  if (!srcAlloca)
    return false;

  // Check that all of src is copied to dest.
  if (TD == 0) return false;

  ConstantInt *srcArraySize = dyn_cast<ConstantInt>(srcAlloca->getArraySize());
  if (!srcArraySize)
    return false;

  uint64_t srcSize = TD->getTypeAllocSize(srcAlloca->getAllocatedType()) *
    srcArraySize->getZExtValue();

  if (cpyLen < srcSize)
    return false;

  // Check that accessing the first srcSize bytes of dest will not cause a
  // trap.  Otherwise the transform is invalid since it might cause a trap
  // to occur earlier than it otherwise would.
  if (AllocaInst *A = dyn_cast<AllocaInst>(cpyDest)) {
    // The destination is an alloca.  Check it is larger than srcSize.
    ConstantInt *destArraySize = dyn_cast<ConstantInt>(A->getArraySize());
    if (!destArraySize)
      return false;

    uint64_t destSize = TD->getTypeAllocSize(A->getAllocatedType()) *
      destArraySize->getZExtValue();

    if (destSize < srcSize)
      return false;
  } else if (Argument *A = dyn_cast<Argument>(cpyDest)) {
    // If the destination is an sret parameter then only accesses that are
    // outside of the returned struct type can trap.
    if (!A->hasStructRetAttr())
      return false;

    Type *StructTy = cast<PointerType>(A->getType())->getElementType();
    uint64_t destSize = TD->getTypeAllocSize(StructTy);

    if (destSize < srcSize)
      return false;
  } else {
    return false;
  }

  // Check that src is not accessed except via the call and the memcpy.  This
  // guarantees that it holds only undefined values when passed in (so the final
  // memcpy can be dropped), that it is not read or written between the call and
  // the memcpy, and that writing beyond the end of it is undefined.
  SmallVector<User*, 8> srcUseList(srcAlloca->use_begin(),
                                   srcAlloca->use_end());
  while (!srcUseList.empty()) {
    User *UI = srcUseList.pop_back_val();

    if (isa<BitCastInst>(UI)) {
      for (User::use_iterator I = UI->use_begin(), E = UI->use_end();
           I != E; ++I)
        srcUseList.push_back(*I);
    } else if (GetElementPtrInst *G = dyn_cast<GetElementPtrInst>(UI)) {
      if (G->hasAllZeroIndices())
        for (User::use_iterator I = UI->use_begin(), E = UI->use_end();
             I != E; ++I)
          srcUseList.push_back(*I);
      else
        return false;
    } else if (UI != C && UI != cpy) {
      return false;
    }
  }

  // Since we're changing the parameter to the callsite, we need to make sure
  // that what would be the new parameter dominates the callsite.
  DominatorTree &DT = getAnalysis<DominatorTree>();
  if (Instruction *cpyDestInst = dyn_cast<Instruction>(cpyDest))
    if (!DT.dominates(cpyDestInst, C))
      return false;

  // In addition to knowing that the call does not access src in some
  // unexpected manner, for example via a global, which we deduce from
  // the use analysis, we also need to know that it does not sneakily
  // access dest.  We rely on AA to figure this out for us.
  AliasAnalysis &AA = getAnalysis<AliasAnalysis>();
  if (AA.getModRefInfo(C, cpyDest, srcSize) != AliasAnalysis::NoModRef)
    return false;

  // All the checks have passed, so do the transformation.
  bool changedArgument = false;
  for (unsigned i = 0; i < CS.arg_size(); ++i)
    if (CS.getArgument(i)->stripPointerCasts() == cpySrc) {
      if (cpySrc->getType() != cpyDest->getType())
        cpyDest = CastInst::CreatePointerCast(cpyDest, cpySrc->getType(),
                                              cpyDest->getName(), C);
      changedArgument = true;
      if (CS.getArgument(i)->getType() == cpyDest->getType())
        CS.setArgument(i, cpyDest);
      else
        CS.setArgument(i, CastInst::CreatePointerCast(cpyDest, 
                          CS.getArgument(i)->getType(), cpyDest->getName(), C));
    }

  if (!changedArgument)
    return false;

  // Drop any cached information about the call, because we may have changed
  // its dependence information by changing its parameter.
  MD->removeInstruction(C);

  // Remove the memcpy.
  MD->removeInstruction(cpy);
  ++NumMemCpyInstr;

  return true;
}

/// processMemCpyMemCpyDependence - We've found that the (upward scanning)
/// memory dependence of memcpy 'M' is the memcpy 'MDep'.  Try to simplify M to
/// copy from MDep's input if we can.  MSize is the size of M's copy.
/// 
bool MemCpyOpt::processMemCpyMemCpyDependence(MemCpyInst *M, MemCpyInst *MDep,
                                              uint64_t MSize) {
  // We can only transforms memcpy's where the dest of one is the source of the
  // other.
  if (M->getSource() != MDep->getDest() || MDep->isVolatile())
    return false;
  
  // If dep instruction is reading from our current input, then it is a noop
  // transfer and substituting the input won't change this instruction.  Just
  // ignore the input and let someone else zap MDep.  This handles cases like:
  //    memcpy(a <- a)
  //    memcpy(b <- a)
  if (M->getSource() == MDep->getSource())
    return false;
  
  // Second, the length of the memcpy's must be the same, or the preceding one
  // must be larger than the following one.
  ConstantInt *MDepLen = dyn_cast<ConstantInt>(MDep->getLength());
  ConstantInt *MLen = dyn_cast<ConstantInt>(M->getLength());
  if (!MDepLen || !MLen || MDepLen->getZExtValue() < MLen->getZExtValue())
    return false;
  
  AliasAnalysis &AA = getAnalysis<AliasAnalysis>();

  // Verify that the copied-from memory doesn't change in between the two
  // transfers.  For example, in:
  //    memcpy(a <- b)
  //    *b = 42;
  //    memcpy(c <- a)
  // It would be invalid to transform the second memcpy into memcpy(c <- b).
  //
  // TODO: If the code between M and MDep is transparent to the destination "c",
  // then we could still perform the xform by moving M up to the first memcpy.
  //
  // NOTE: This is conservative, it will stop on any read from the source loc,
  // not just the defining memcpy.
  MemDepResult SourceDep =
    MD->getPointerDependencyFrom(AA.getLocationForSource(MDep),
                                 false, M, M->getParent());
  if (!SourceDep.isClobber() || SourceDep.getInst() != MDep)
    return false;
  
  // If the dest of the second might alias the source of the first, then the
  // source and dest might overlap.  We still want to eliminate the intermediate
  // value, but we have to generate a memmove instead of memcpy.
  bool UseMemMove = false;
  if (!AA.isNoAlias(AA.getLocationForDest(M), AA.getLocationForSource(MDep)))
    UseMemMove = true;
  
  // If all checks passed, then we can transform M.
  
  // Make sure to use the lesser of the alignment of the source and the dest
  // since we're changing where we're reading from, but don't want to increase
  // the alignment past what can be read from or written to.
  // TODO: Is this worth it if we're creating a less aligned memcpy? For
  // example we could be moving from movaps -> movq on x86.
  unsigned Align = std::min(MDep->getAlignment(), M->getAlignment());
  
  IRBuilder<> Builder(M);
  if (UseMemMove)
    Builder.CreateMemMove(M->getRawDest(), MDep->getRawSource(), M->getLength(),
                          Align, M->isVolatile());
  else
    Builder.CreateMemCpy(M->getRawDest(), MDep->getRawSource(), M->getLength(),
                         Align, M->isVolatile());

  // Remove the instruction we're replacing.
  MD->removeInstruction(M);
  M->eraseFromParent();
  ++NumMemCpyInstr;
  return true;
}


/// processMemCpy - perform simplification of memcpy's.  If we have memcpy A
/// which copies X to Y, and memcpy B which copies Y to Z, then we can rewrite
/// B to be a memcpy from X to Z (or potentially a memmove, depending on
/// circumstances). This allows later passes to remove the first memcpy
/// altogether.
bool MemCpyOpt::processMemCpy(MemCpyInst *M) {
  // We can only optimize statically-sized memcpy's that are non-volatile.
  ConstantInt *CopySize = dyn_cast<ConstantInt>(M->getLength());
  if (CopySize == 0 || M->isVolatile()) return false;

  // If the source and destination of the memcpy are the same, then zap it.
  if (M->getSource() == M->getDest()) {
    MD->removeInstruction(M);
    M->eraseFromParent();
    return false;
  }

  // If copying from a constant, try to turn the memcpy into a memset.
  if (GlobalVariable *GV = dyn_cast<GlobalVariable>(M->getSource()))
    if (GV->isConstant() && GV->hasDefinitiveInitializer())
      if (Value *ByteVal = isBytewiseValue(GV->getInitializer())) {
        IRBuilder<> Builder(M);
        Builder.CreateMemSet(M->getRawDest(), ByteVal, CopySize,
                             M->getAlignment(), false);
        MD->removeInstruction(M);
        M->eraseFromParent();
        ++NumCpyToSet;
        return true;
      }

  // The are two possible optimizations we can do for memcpy:
  //   a) memcpy-memcpy xform which exposes redundance for DSE.
  //   b) call-memcpy xform for return slot optimization.
  MemDepResult DepInfo = MD->getDependency(M);
  if (!DepInfo.isClobber())
    return false;
  
  if (MemCpyInst *MDep = dyn_cast<MemCpyInst>(DepInfo.getInst()))
    return processMemCpyMemCpyDependence(M, MDep, CopySize->getZExtValue());
    
  if (CallInst *C = dyn_cast<CallInst>(DepInfo.getInst())) {
    if (performCallSlotOptzn(M, M->getDest(), M->getSource(),
                             CopySize->getZExtValue(), C)) {
      MD->removeInstruction(M);
      M->eraseFromParent();
      return true;
    }
  }
  
  return false;
}

/// processMemMove - Transforms memmove calls to memcpy calls when the src/dst
/// are guaranteed not to alias.
bool MemCpyOpt::processMemMove(MemMoveInst *M) {
  AliasAnalysis &AA = getAnalysis<AliasAnalysis>();

  if (!TLI->has(LibFunc::memmove))
    return false;
  
  // See if the pointers alias.
  if (!AA.isNoAlias(AA.getLocationForDest(M), AA.getLocationForSource(M)))
    return false;
  
  DEBUG(dbgs() << "MemCpyOpt: Optimizing memmove -> memcpy: " << *M << "\n");
  
  // If not, then we know we can transform this.
  Module *Mod = M->getParent()->getParent()->getParent();
  Type *ArgTys[3] = { M->getRawDest()->getType(),
                      M->getRawSource()->getType(),
                      M->getLength()->getType() };
  M->setCalledFunction(Intrinsic::getDeclaration(Mod, Intrinsic::memcpy,
                                                 ArgTys));

  // MemDep may have over conservative information about this instruction, just
  // conservatively flush it from the cache.
  MD->removeInstruction(M);

  ++NumMoveToCpy;
  return true;
}
  
/// processByValArgument - This is called on every byval argument in call sites.
bool MemCpyOpt::processByValArgument(CallSite CS, unsigned ArgNo) {
  if (TD == 0) return false;

  // Find out what feeds this byval argument.
  Value *ByValArg = CS.getArgument(ArgNo);
  Type *ByValTy =cast<PointerType>(ByValArg->getType())->getElementType();
  uint64_t ByValSize = TD->getTypeAllocSize(ByValTy);
  MemDepResult DepInfo =
    MD->getPointerDependencyFrom(AliasAnalysis::Location(ByValArg, ByValSize),
                                 true, CS.getInstruction(),
                                 CS.getInstruction()->getParent());
  if (!DepInfo.isClobber())
    return false;

  // If the byval argument isn't fed by a memcpy, ignore it.  If it is fed by
  // a memcpy, see if we can byval from the source of the memcpy instead of the
  // result.
  MemCpyInst *MDep = dyn_cast<MemCpyInst>(DepInfo.getInst());
  if (MDep == 0 || MDep->isVolatile() ||
      ByValArg->stripPointerCasts() != MDep->getDest())
    return false;
  
  // The length of the memcpy must be larger or equal to the size of the byval.
  ConstantInt *C1 = dyn_cast<ConstantInt>(MDep->getLength());
  if (C1 == 0 || C1->getValue().getZExtValue() < ByValSize)
    return false;

  // Get the alignment of the byval.  If the call doesn't specify the alignment,
  // then it is some target specific value that we can't know.
  unsigned ByValAlign = CS.getParamAlignment(ArgNo+1);
  if (ByValAlign == 0) return false;
  
  // If it is greater than the memcpy, then we check to see if we can force the
  // source of the memcpy to the alignment we need.  If we fail, we bail out.
  if (MDep->getAlignment() < ByValAlign &&
      getOrEnforceKnownAlignment(MDep->getSource(),ByValAlign, TD) < ByValAlign)
    return false;
  
  // Verify that the copied-from memory doesn't change in between the memcpy and
  // the byval call.
  //    memcpy(a <- b)
  //    *b = 42;
  //    foo(*a)
  // It would be invalid to transform the second memcpy into foo(*b).
  //
  // NOTE: This is conservative, it will stop on any read from the source loc,
  // not just the defining memcpy.
  MemDepResult SourceDep =
    MD->getPointerDependencyFrom(AliasAnalysis::getLocationForSource(MDep),
                                 false, CS.getInstruction(), MDep->getParent());
  if (!SourceDep.isClobber() || SourceDep.getInst() != MDep)
    return false;
  
  Value *TmpCast = MDep->getSource();
  if (MDep->getSource()->getType() != ByValArg->getType())
    TmpCast = new BitCastInst(MDep->getSource(), ByValArg->getType(),
                              "tmpcast", CS.getInstruction());
  
  DEBUG(dbgs() << "MemCpyOpt: Forwarding memcpy to byval:\n"
               << "  " << *MDep << "\n"
               << "  " << *CS.getInstruction() << "\n");
  
  // Otherwise we're good!  Update the byval argument.
  CS.setArgument(ArgNo, TmpCast);
  ++NumMemCpyInstr;
  return true;
}

/// iterateOnFunction - Executes one iteration of MemCpyOpt.
bool MemCpyOpt::iterateOnFunction(Function &F) {
  bool MadeChange = false;

  // Walk all instruction in the function.
  for (Function::iterator BB = F.begin(), BBE = F.end(); BB != BBE; ++BB) {
    for (BasicBlock::iterator BI = BB->begin(), BE = BB->end(); BI != BE;) {
      // Avoid invalidating the iterator.
      Instruction *I = BI++;
      
      bool RepeatInstruction = false;
      
      if (StoreInst *SI = dyn_cast<StoreInst>(I))
        MadeChange |= processStore(SI, BI);
      else if (MemSetInst *M = dyn_cast<MemSetInst>(I))
        RepeatInstruction = processMemSet(M, BI);
      else if (MemCpyInst *M = dyn_cast<MemCpyInst>(I))
        RepeatInstruction = processMemCpy(M);
      else if (MemMoveInst *M = dyn_cast<MemMoveInst>(I))
        RepeatInstruction = processMemMove(M);
      else if (CallSite CS = (Value*)I) {
        for (unsigned i = 0, e = CS.arg_size(); i != e; ++i)
          if (CS.paramHasAttr(i+1, Attribute::ByVal))
            MadeChange |= processByValArgument(CS, i);
      }

      // Reprocess the instruction if desired.
      if (RepeatInstruction) {
        if (BI != BB->begin()) --BI;
        MadeChange = true;
      }
    }
  }
  
  return MadeChange;
}

// MemCpyOpt::runOnFunction - This is the main transformation entry point for a
// function.
//
bool MemCpyOpt::runOnFunction(Function &F) {
  bool MadeChange = false;
  MD = &getAnalysis<MemoryDependenceAnalysis>();
  TD = getAnalysisIfAvailable<TargetData>();
  TLI = &getAnalysis<TargetLibraryInfo>();
  
  // If we don't have at least memset and memcpy, there is little point of doing
  // anything here.  These are required by a freestanding implementation, so if
  // even they are disabled, there is no point in trying hard.
  if (!TLI->has(LibFunc::memset) || !TLI->has(LibFunc::memcpy))
    return false;
  
  while (1) {
    if (!iterateOnFunction(F))
      break;
    MadeChange = true;
  }
  
  MD = 0;
  return MadeChange;
}