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
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
|
//===--- CGRecordLayoutBuilder.cpp - CGRecordLayout builder ----*- C++ -*-===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// Builder implementation for CGRecordLayout objects.
//
//===----------------------------------------------------------------------===//
#include "CGRecordLayout.h"
#include "CGCXXABI.h"
#include "CodeGenTypes.h"
#include "clang/AST/ASTContext.h"
#include "clang/AST/Attr.h"
#include "clang/AST/CXXInheritance.h"
#include "clang/AST/DeclCXX.h"
#include "clang/AST/Expr.h"
#include "clang/AST/RecordLayout.h"
#include "clang/Basic/CodeGenOptions.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Type.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
using namespace clang;
using namespace CodeGen;
namespace {
/// The CGRecordLowering is responsible for lowering an ASTRecordLayout to an
/// llvm::Type. Some of the lowering is straightforward, some is not. Here we
/// detail some of the complexities and weirdnesses here.
/// * LLVM does not have unions - Unions can, in theory be represented by any
/// llvm::Type with correct size. We choose a field via a specific heuristic
/// and add padding if necessary.
/// * LLVM does not have bitfields - Bitfields are collected into contiguous
/// runs and allocated as a single storage type for the run. ASTRecordLayout
/// contains enough information to determine where the runs break. Microsoft
/// and Itanium follow different rules and use different codepaths.
/// * It is desired that, when possible, bitfields use the appropriate iN type
/// when lowered to llvm types. For example unsigned x : 24 gets lowered to
/// i24. This isn't always possible because i24 has storage size of 32 bit
/// and if it is possible to use that extra byte of padding we must use
/// [i8 x 3] instead of i24. The function clipTailPadding does this.
/// C++ examples that require clipping:
/// struct { int a : 24; char b; }; // a must be clipped, b goes at offset 3
/// struct A { int a : 24; }; // a must be clipped because a struct like B
// could exist: struct B : A { char b; }; // b goes at offset 3
/// * Clang ignores 0 sized bitfields and 0 sized bases but *not* zero sized
/// fields. The existing asserts suggest that LLVM assumes that *every* field
/// has an underlying storage type. Therefore empty structures containing
/// zero sized subobjects such as empty records or zero sized arrays still get
/// a zero sized (empty struct) storage type.
/// * Clang reads the complete type rather than the base type when generating
/// code to access fields. Bitfields in tail position with tail padding may
/// be clipped in the base class but not the complete class (we may discover
/// that the tail padding is not used in the complete class.) However,
/// because LLVM reads from the complete type it can generate incorrect code
/// if we do not clip the tail padding off of the bitfield in the complete
/// layout. This introduces a somewhat awkward extra unnecessary clip stage.
/// The location of the clip is stored internally as a sentinel of type
/// SCISSOR. If LLVM were updated to read base types (which it probably
/// should because locations of things such as VBases are bogus in the llvm
/// type anyway) then we could eliminate the SCISSOR.
/// * Itanium allows nearly empty primary virtual bases. These bases don't get
/// get their own storage because they're laid out as part of another base
/// or at the beginning of the structure. Determining if a VBase actually
/// gets storage awkwardly involves a walk of all bases.
/// * VFPtrs and VBPtrs do *not* make a record NotZeroInitializable.
struct CGRecordLowering {
// MemberInfo is a helper structure that contains information about a record
// member. In additional to the standard member types, there exists a
// sentinel member type that ensures correct rounding.
struct MemberInfo {
CharUnits Offset;
enum InfoKind { VFPtr, VBPtr, Field, Base, VBase, Scissor } Kind;
llvm::Type *Data;
union {
const FieldDecl *FD;
const CXXRecordDecl *RD;
};
MemberInfo(CharUnits Offset, InfoKind Kind, llvm::Type *Data,
const FieldDecl *FD = nullptr)
: Offset(Offset), Kind(Kind), Data(Data), FD(FD) {}
MemberInfo(CharUnits Offset, InfoKind Kind, llvm::Type *Data,
const CXXRecordDecl *RD)
: Offset(Offset), Kind(Kind), Data(Data), RD(RD) {}
// MemberInfos are sorted so we define a < operator.
bool operator <(const MemberInfo& a) const { return Offset < a.Offset; }
};
// The constructor.
CGRecordLowering(CodeGenTypes &Types, const RecordDecl *D, bool Packed);
// Short helper routines.
/// Constructs a MemberInfo instance from an offset and llvm::Type *.
MemberInfo StorageInfo(CharUnits Offset, llvm::Type *Data) {
return MemberInfo(Offset, MemberInfo::Field, Data);
}
/// The Microsoft bitfield layout rule allocates discrete storage
/// units of the field's formal type and only combines adjacent
/// fields of the same formal type. We want to emit a layout with
/// these discrete storage units instead of combining them into a
/// continuous run.
bool isDiscreteBitFieldABI() {
return Context.getTargetInfo().getCXXABI().isMicrosoft() ||
D->isMsStruct(Context);
}
/// Helper function to check if we are targeting AAPCS.
bool isAAPCS() const {
return Context.getTargetInfo().getABI().startswith("aapcs");
}
/// Helper function to check if the target machine is BigEndian.
bool isBE() const { return Context.getTargetInfo().isBigEndian(); }
/// The Itanium base layout rule allows virtual bases to overlap
/// other bases, which complicates layout in specific ways.
///
/// Note specifically that the ms_struct attribute doesn't change this.
bool isOverlappingVBaseABI() {
return !Context.getTargetInfo().getCXXABI().isMicrosoft();
}
/// Wraps llvm::Type::getIntNTy with some implicit arguments.
llvm::Type *getIntNType(uint64_t NumBits) {
unsigned AlignedBits = llvm::alignTo(NumBits, Context.getCharWidth());
return llvm::Type::getIntNTy(Types.getLLVMContext(), AlignedBits);
}
/// Get the LLVM type sized as one character unit.
llvm::Type *getCharType() {
return llvm::Type::getIntNTy(Types.getLLVMContext(),
Context.getCharWidth());
}
/// Gets an llvm type of size NumChars and alignment 1.
llvm::Type *getByteArrayType(CharUnits NumChars) {
assert(!NumChars.isZero() && "Empty byte arrays aren't allowed.");
llvm::Type *Type = getCharType();
return NumChars == CharUnits::One() ? Type :
(llvm::Type *)llvm::ArrayType::get(Type, NumChars.getQuantity());
}
/// Gets the storage type for a field decl and handles storage
/// for itanium bitfields that are smaller than their declared type.
llvm::Type *getStorageType(const FieldDecl *FD) {
llvm::Type *Type = Types.ConvertTypeForMem(FD->getType());
if (!FD->isBitField()) return Type;
if (isDiscreteBitFieldABI()) return Type;
return getIntNType(std::min(FD->getBitWidthValue(Context),
(unsigned)Context.toBits(getSize(Type))));
}
/// Gets the llvm Basesubobject type from a CXXRecordDecl.
llvm::Type *getStorageType(const CXXRecordDecl *RD) {
return Types.getCGRecordLayout(RD).getBaseSubobjectLLVMType();
}
CharUnits bitsToCharUnits(uint64_t BitOffset) {
return Context.toCharUnitsFromBits(BitOffset);
}
CharUnits getSize(llvm::Type *Type) {
return CharUnits::fromQuantity(DataLayout.getTypeAllocSize(Type));
}
CharUnits getAlignment(llvm::Type *Type) {
return CharUnits::fromQuantity(DataLayout.getABITypeAlign(Type));
}
bool isZeroInitializable(const FieldDecl *FD) {
return Types.isZeroInitializable(FD->getType());
}
bool isZeroInitializable(const RecordDecl *RD) {
return Types.isZeroInitializable(RD);
}
void appendPaddingBytes(CharUnits Size) {
if (!Size.isZero())
FieldTypes.push_back(getByteArrayType(Size));
}
uint64_t getFieldBitOffset(const FieldDecl *FD) {
return Layout.getFieldOffset(FD->getFieldIndex());
}
// Layout routines.
void setBitFieldInfo(const FieldDecl *FD, CharUnits StartOffset,
llvm::Type *StorageType);
/// Lowers an ASTRecordLayout to a llvm type.
void lower(bool NonVirtualBaseType);
void lowerUnion();
void accumulateFields();
void accumulateBitFields(RecordDecl::field_iterator Field,
RecordDecl::field_iterator FieldEnd);
void computeVolatileBitfields();
void accumulateBases();
void accumulateVPtrs();
void accumulateVBases();
/// Recursively searches all of the bases to find out if a vbase is
/// not the primary vbase of some base class.
bool hasOwnStorage(const CXXRecordDecl *Decl, const CXXRecordDecl *Query);
void calculateZeroInit();
/// Lowers bitfield storage types to I8 arrays for bitfields with tail
/// padding that is or can potentially be used.
void clipTailPadding();
/// Determines if we need a packed llvm struct.
void determinePacked(bool NVBaseType);
/// Inserts padding everywhere it's needed.
void insertPadding();
/// Fills out the structures that are ultimately consumed.
void fillOutputFields();
// Input memoization fields.
CodeGenTypes &Types;
const ASTContext &Context;
const RecordDecl *D;
const CXXRecordDecl *RD;
const ASTRecordLayout &Layout;
const llvm::DataLayout &DataLayout;
// Helpful intermediate data-structures.
std::vector<MemberInfo> Members;
// Output fields, consumed by CodeGenTypes::ComputeRecordLayout.
SmallVector<llvm::Type *, 16> FieldTypes;
llvm::DenseMap<const FieldDecl *, unsigned> Fields;
llvm::DenseMap<const FieldDecl *, CGBitFieldInfo> BitFields;
llvm::DenseMap<const CXXRecordDecl *, unsigned> NonVirtualBases;
llvm::DenseMap<const CXXRecordDecl *, unsigned> VirtualBases;
bool IsZeroInitializable : 1;
bool IsZeroInitializableAsBase : 1;
bool Packed : 1;
private:
CGRecordLowering(const CGRecordLowering &) = delete;
void operator =(const CGRecordLowering &) = delete;
};
} // namespace {
CGRecordLowering::CGRecordLowering(CodeGenTypes &Types, const RecordDecl *D,
bool Packed)
: Types(Types), Context(Types.getContext()), D(D),
RD(dyn_cast<CXXRecordDecl>(D)),
Layout(Types.getContext().getASTRecordLayout(D)),
DataLayout(Types.getDataLayout()), IsZeroInitializable(true),
IsZeroInitializableAsBase(true), Packed(Packed) {}
void CGRecordLowering::setBitFieldInfo(
const FieldDecl *FD, CharUnits StartOffset, llvm::Type *StorageType) {
CGBitFieldInfo &Info = BitFields[FD->getCanonicalDecl()];
Info.IsSigned = FD->getType()->isSignedIntegerOrEnumerationType();
Info.Offset = (unsigned)(getFieldBitOffset(FD) - Context.toBits(StartOffset));
Info.Size = FD->getBitWidthValue(Context);
Info.StorageSize = (unsigned)DataLayout.getTypeAllocSizeInBits(StorageType);
Info.StorageOffset = StartOffset;
if (Info.Size > Info.StorageSize)
Info.Size = Info.StorageSize;
// Reverse the bit offsets for big endian machines. Because we represent
// a bitfield as a single large integer load, we can imagine the bits
// counting from the most-significant-bit instead of the
// least-significant-bit.
if (DataLayout.isBigEndian())
Info.Offset = Info.StorageSize - (Info.Offset + Info.Size);
Info.VolatileStorageSize = 0;
Info.VolatileOffset = 0;
Info.VolatileStorageOffset = CharUnits::Zero();
}
void CGRecordLowering::lower(bool NVBaseType) {
// The lowering process implemented in this function takes a variety of
// carefully ordered phases.
// 1) Store all members (fields and bases) in a list and sort them by offset.
// 2) Add a 1-byte capstone member at the Size of the structure.
// 3) Clip bitfield storages members if their tail padding is or might be
// used by another field or base. The clipping process uses the capstone
// by treating it as another object that occurs after the record.
// 4) Determine if the llvm-struct requires packing. It's important that this
// phase occur after clipping, because clipping changes the llvm type.
// This phase reads the offset of the capstone when determining packedness
// and updates the alignment of the capstone to be equal of the alignment
// of the record after doing so.
// 5) Insert padding everywhere it is needed. This phase requires 'Packed' to
// have been computed and needs to know the alignment of the record in
// order to understand if explicit tail padding is needed.
// 6) Remove the capstone, we don't need it anymore.
// 7) Determine if this record can be zero-initialized. This phase could have
// been placed anywhere after phase 1.
// 8) Format the complete list of members in a way that can be consumed by
// CodeGenTypes::ComputeRecordLayout.
CharUnits Size = NVBaseType ? Layout.getNonVirtualSize() : Layout.getSize();
if (D->isUnion()) {
lowerUnion();
computeVolatileBitfields();
return;
}
accumulateFields();
// RD implies C++.
if (RD) {
accumulateVPtrs();
accumulateBases();
if (Members.empty()) {
appendPaddingBytes(Size);
computeVolatileBitfields();
return;
}
if (!NVBaseType)
accumulateVBases();
}
llvm::stable_sort(Members);
Members.push_back(StorageInfo(Size, getIntNType(8)));
clipTailPadding();
determinePacked(NVBaseType);
insertPadding();
Members.pop_back();
calculateZeroInit();
fillOutputFields();
computeVolatileBitfields();
}
void CGRecordLowering::lowerUnion() {
CharUnits LayoutSize = Layout.getSize();
llvm::Type *StorageType = nullptr;
bool SeenNamedMember = false;
// Iterate through the fields setting bitFieldInfo and the Fields array. Also
// locate the "most appropriate" storage type. The heuristic for finding the
// storage type isn't necessary, the first (non-0-length-bitfield) field's
// type would work fine and be simpler but would be different than what we've
// been doing and cause lit tests to change.
for (const auto *Field : D->fields()) {
if (Field->isBitField()) {
if (Field->isZeroLengthBitField(Context))
continue;
llvm::Type *FieldType = getStorageType(Field);
if (LayoutSize < getSize(FieldType))
FieldType = getByteArrayType(LayoutSize);
setBitFieldInfo(Field, CharUnits::Zero(), FieldType);
}
Fields[Field->getCanonicalDecl()] = 0;
llvm::Type *FieldType = getStorageType(Field);
// Compute zero-initializable status.
// This union might not be zero initialized: it may contain a pointer to
// data member which might have some exotic initialization sequence.
// If this is the case, then we aught not to try and come up with a "better"
// type, it might not be very easy to come up with a Constant which
// correctly initializes it.
if (!SeenNamedMember) {
SeenNamedMember = Field->getIdentifier();
if (!SeenNamedMember)
if (const auto *FieldRD = Field->getType()->getAsRecordDecl())
SeenNamedMember = FieldRD->findFirstNamedDataMember();
if (SeenNamedMember && !isZeroInitializable(Field)) {
IsZeroInitializable = IsZeroInitializableAsBase = false;
StorageType = FieldType;
}
}
// Because our union isn't zero initializable, we won't be getting a better
// storage type.
if (!IsZeroInitializable)
continue;
// Conditionally update our storage type if we've got a new "better" one.
if (!StorageType ||
getAlignment(FieldType) > getAlignment(StorageType) ||
(getAlignment(FieldType) == getAlignment(StorageType) &&
getSize(FieldType) > getSize(StorageType)))
StorageType = FieldType;
}
// If we have no storage type just pad to the appropriate size and return.
if (!StorageType)
return appendPaddingBytes(LayoutSize);
// If our storage size was bigger than our required size (can happen in the
// case of packed bitfields on Itanium) then just use an I8 array.
if (LayoutSize < getSize(StorageType))
StorageType = getByteArrayType(LayoutSize);
FieldTypes.push_back(StorageType);
appendPaddingBytes(LayoutSize - getSize(StorageType));
// Set packed if we need it.
if (LayoutSize % getAlignment(StorageType))
Packed = true;
}
void CGRecordLowering::accumulateFields() {
for (RecordDecl::field_iterator Field = D->field_begin(),
FieldEnd = D->field_end();
Field != FieldEnd;) {
if (Field->isBitField()) {
RecordDecl::field_iterator Start = Field;
// Iterate to gather the list of bitfields.
for (++Field; Field != FieldEnd && Field->isBitField(); ++Field);
accumulateBitFields(Start, Field);
} else if (!Field->isZeroSize(Context)) {
Members.push_back(MemberInfo(
bitsToCharUnits(getFieldBitOffset(*Field)), MemberInfo::Field,
getStorageType(*Field), *Field));
++Field;
} else {
++Field;
}
}
}
void
CGRecordLowering::accumulateBitFields(RecordDecl::field_iterator Field,
RecordDecl::field_iterator FieldEnd) {
// Run stores the first element of the current run of bitfields. FieldEnd is
// used as a special value to note that we don't have a current run. A
// bitfield run is a contiguous collection of bitfields that can be stored in
// the same storage block. Zero-sized bitfields and bitfields that would
// cross an alignment boundary break a run and start a new one.
RecordDecl::field_iterator Run = FieldEnd;
// Tail is the offset of the first bit off the end of the current run. It's
// used to determine if the ASTRecordLayout is treating these two bitfields as
// contiguous. StartBitOffset is offset of the beginning of the Run.
uint64_t StartBitOffset, Tail = 0;
if (isDiscreteBitFieldABI()) {
for (; Field != FieldEnd; ++Field) {
uint64_t BitOffset = getFieldBitOffset(*Field);
// Zero-width bitfields end runs.
if (Field->isZeroLengthBitField(Context)) {
Run = FieldEnd;
continue;
}
llvm::Type *Type =
Types.ConvertTypeForMem(Field->getType(), /*ForBitField=*/true);
// If we don't have a run yet, or don't live within the previous run's
// allocated storage then we allocate some storage and start a new run.
if (Run == FieldEnd || BitOffset >= Tail) {
Run = Field;
StartBitOffset = BitOffset;
Tail = StartBitOffset + DataLayout.getTypeAllocSizeInBits(Type);
// Add the storage member to the record. This must be added to the
// record before the bitfield members so that it gets laid out before
// the bitfields it contains get laid out.
Members.push_back(StorageInfo(bitsToCharUnits(StartBitOffset), Type));
}
// Bitfields get the offset of their storage but come afterward and remain
// there after a stable sort.
Members.push_back(MemberInfo(bitsToCharUnits(StartBitOffset),
MemberInfo::Field, nullptr, *Field));
}
return;
}
// Check if OffsetInRecord (the size in bits of the current run) is better
// as a single field run. When OffsetInRecord has legal integer width, and
// its bitfield offset is naturally aligned, it is better to make the
// bitfield a separate storage component so as it can be accessed directly
// with lower cost.
auto IsBetterAsSingleFieldRun = [&](uint64_t OffsetInRecord,
uint64_t StartBitOffset) {
if (!Types.getCodeGenOpts().FineGrainedBitfieldAccesses)
return false;
if (OffsetInRecord < 8 || !llvm::isPowerOf2_64(OffsetInRecord) ||
!DataLayout.fitsInLegalInteger(OffsetInRecord))
return false;
// Make sure StartBitOffset is naturally aligned if it is treated as an
// IType integer.
if (StartBitOffset %
Context.toBits(getAlignment(getIntNType(OffsetInRecord))) !=
0)
return false;
return true;
};
// The start field is better as a single field run.
bool StartFieldAsSingleRun = false;
for (;;) {
// Check to see if we need to start a new run.
if (Run == FieldEnd) {
// If we're out of fields, return.
if (Field == FieldEnd)
break;
// Any non-zero-length bitfield can start a new run.
if (!Field->isZeroLengthBitField(Context)) {
Run = Field;
StartBitOffset = getFieldBitOffset(*Field);
Tail = StartBitOffset + Field->getBitWidthValue(Context);
StartFieldAsSingleRun = IsBetterAsSingleFieldRun(Tail - StartBitOffset,
StartBitOffset);
}
++Field;
continue;
}
// If the start field of a new run is better as a single run, or
// if current field (or consecutive fields) is better as a single run, or
// if current field has zero width bitfield and either
// UseZeroLengthBitfieldAlignment or UseBitFieldTypeAlignment is set to
// true, or
// if the offset of current field is inconsistent with the offset of
// previous field plus its offset,
// skip the block below and go ahead to emit the storage.
// Otherwise, try to add bitfields to the run.
if (!StartFieldAsSingleRun && Field != FieldEnd &&
!IsBetterAsSingleFieldRun(Tail - StartBitOffset, StartBitOffset) &&
(!Field->isZeroLengthBitField(Context) ||
(!Context.getTargetInfo().useZeroLengthBitfieldAlignment() &&
!Context.getTargetInfo().useBitFieldTypeAlignment())) &&
Tail == getFieldBitOffset(*Field)) {
Tail += Field->getBitWidthValue(Context);
++Field;
continue;
}
// We've hit a break-point in the run and need to emit a storage field.
llvm::Type *Type = getIntNType(Tail - StartBitOffset);
// Add the storage member to the record and set the bitfield info for all of
// the bitfields in the run. Bitfields get the offset of their storage but
// come afterward and remain there after a stable sort.
Members.push_back(StorageInfo(bitsToCharUnits(StartBitOffset), Type));
for (; Run != Field; ++Run)
Members.push_back(MemberInfo(bitsToCharUnits(StartBitOffset),
MemberInfo::Field, nullptr, *Run));
Run = FieldEnd;
StartFieldAsSingleRun = false;
}
}
void CGRecordLowering::accumulateBases() {
// If we've got a primary virtual base, we need to add it with the bases.
if (Layout.isPrimaryBaseVirtual()) {
const CXXRecordDecl *BaseDecl = Layout.getPrimaryBase();
Members.push_back(MemberInfo(CharUnits::Zero(), MemberInfo::Base,
getStorageType(BaseDecl), BaseDecl));
}
// Accumulate the non-virtual bases.
for (const auto &Base : RD->bases()) {
if (Base.isVirtual())
continue;
// Bases can be zero-sized even if not technically empty if they
// contain only a trailing array member.
const CXXRecordDecl *BaseDecl = Base.getType()->getAsCXXRecordDecl();
if (!BaseDecl->isEmpty() &&
!Context.getASTRecordLayout(BaseDecl).getNonVirtualSize().isZero())
Members.push_back(MemberInfo(Layout.getBaseClassOffset(BaseDecl),
MemberInfo::Base, getStorageType(BaseDecl), BaseDecl));
}
}
/// The AAPCS that defines that, when possible, bit-fields should
/// be accessed using containers of the declared type width:
/// When a volatile bit-field is read, and its container does not overlap with
/// any non-bit-field member or any zero length bit-field member, its container
/// must be read exactly once using the access width appropriate to the type of
/// the container. When a volatile bit-field is written, and its container does
/// not overlap with any non-bit-field member or any zero-length bit-field
/// member, its container must be read exactly once and written exactly once
/// using the access width appropriate to the type of the container. The two
/// accesses are not atomic.
///
/// Enforcing the width restriction can be disabled using
/// -fno-aapcs-bitfield-width.
void CGRecordLowering::computeVolatileBitfields() {
if (!isAAPCS() || !Types.getCodeGenOpts().AAPCSBitfieldWidth)
return;
for (auto &I : BitFields) {
const FieldDecl *Field = I.first;
CGBitFieldInfo &Info = I.second;
llvm::Type *ResLTy = Types.ConvertTypeForMem(Field->getType());
// If the record alignment is less than the type width, we can't enforce a
// aligned load, bail out.
if ((uint64_t)(Context.toBits(Layout.getAlignment())) <
ResLTy->getPrimitiveSizeInBits())
continue;
// CGRecordLowering::setBitFieldInfo() pre-adjusts the bit-field offsets
// for big-endian targets, but it assumes a container of width
// Info.StorageSize. Since AAPCS uses a different container size (width
// of the type), we first undo that calculation here and redo it once
// the bit-field offset within the new container is calculated.
const unsigned OldOffset =
isBE() ? Info.StorageSize - (Info.Offset + Info.Size) : Info.Offset;
// Offset to the bit-field from the beginning of the struct.
const unsigned AbsoluteOffset =
Context.toBits(Info.StorageOffset) + OldOffset;
// Container size is the width of the bit-field type.
const unsigned StorageSize = ResLTy->getPrimitiveSizeInBits();
// Nothing to do if the access uses the desired
// container width and is naturally aligned.
if (Info.StorageSize == StorageSize && (OldOffset % StorageSize == 0))
continue;
// Offset within the container.
unsigned Offset = AbsoluteOffset & (StorageSize - 1);
// Bail out if an aligned load of the container cannot cover the entire
// bit-field. This can happen for example, if the bit-field is part of a
// packed struct. AAPCS does not define access rules for such cases, we let
// clang to follow its own rules.
if (Offset + Info.Size > StorageSize)
continue;
// Re-adjust offsets for big-endian targets.
if (isBE())
Offset = StorageSize - (Offset + Info.Size);
const CharUnits StorageOffset =
Context.toCharUnitsFromBits(AbsoluteOffset & ~(StorageSize - 1));
const CharUnits End = StorageOffset +
Context.toCharUnitsFromBits(StorageSize) -
CharUnits::One();
const ASTRecordLayout &Layout =
Context.getASTRecordLayout(Field->getParent());
// If we access outside memory outside the record, than bail out.
const CharUnits RecordSize = Layout.getSize();
if (End >= RecordSize)
continue;
// Bail out if performing this load would access non-bit-fields members.
bool Conflict = false;
for (const auto *F : D->fields()) {
// Allow sized bit-fields overlaps.
if (F->isBitField() && !F->isZeroLengthBitField(Context))
continue;
const CharUnits FOffset = Context.toCharUnitsFromBits(
Layout.getFieldOffset(F->getFieldIndex()));
// As C11 defines, a zero sized bit-field defines a barrier, so
// fields after and before it should be race condition free.
// The AAPCS acknowledges it and imposes no restritions when the
// natural container overlaps a zero-length bit-field.
if (F->isZeroLengthBitField(Context)) {
if (End > FOffset && StorageOffset < FOffset) {
Conflict = true;
break;
}
}
const CharUnits FEnd =
FOffset +
Context.toCharUnitsFromBits(
Types.ConvertTypeForMem(F->getType())->getPrimitiveSizeInBits()) -
CharUnits::One();
// If no overlap, continue.
if (End < FOffset || FEnd < StorageOffset)
continue;
// The desired load overlaps a non-bit-field member, bail out.
Conflict = true;
break;
}
if (Conflict)
continue;
// Write the new bit-field access parameters.
// As the storage offset now is defined as the number of elements from the
// start of the structure, we should divide the Offset by the element size.
Info.VolatileStorageOffset =
StorageOffset / Context.toCharUnitsFromBits(StorageSize).getQuantity();
Info.VolatileStorageSize = StorageSize;
Info.VolatileOffset = Offset;
}
}
void CGRecordLowering::accumulateVPtrs() {
if (Layout.hasOwnVFPtr())
Members.push_back(MemberInfo(CharUnits::Zero(), MemberInfo::VFPtr,
llvm::FunctionType::get(getIntNType(32), /*isVarArg=*/true)->
getPointerTo()->getPointerTo()));
if (Layout.hasOwnVBPtr())
Members.push_back(MemberInfo(Layout.getVBPtrOffset(), MemberInfo::VBPtr,
llvm::Type::getInt32PtrTy(Types.getLLVMContext())));
}
void CGRecordLowering::accumulateVBases() {
CharUnits ScissorOffset = Layout.getNonVirtualSize();
// In the itanium ABI, it's possible to place a vbase at a dsize that is
// smaller than the nvsize. Here we check to see if such a base is placed
// before the nvsize and set the scissor offset to that, instead of the
// nvsize.
if (isOverlappingVBaseABI())
for (const auto &Base : RD->vbases()) {
const CXXRecordDecl *BaseDecl = Base.getType()->getAsCXXRecordDecl();
if (BaseDecl->isEmpty())
continue;
// If the vbase is a primary virtual base of some base, then it doesn't
// get its own storage location but instead lives inside of that base.
if (Context.isNearlyEmpty(BaseDecl) && !hasOwnStorage(RD, BaseDecl))
continue;
ScissorOffset = std::min(ScissorOffset,
Layout.getVBaseClassOffset(BaseDecl));
}
Members.push_back(MemberInfo(ScissorOffset, MemberInfo::Scissor, nullptr,
RD));
for (const auto &Base : RD->vbases()) {
const CXXRecordDecl *BaseDecl = Base.getType()->getAsCXXRecordDecl();
if (BaseDecl->isEmpty())
continue;
CharUnits Offset = Layout.getVBaseClassOffset(BaseDecl);
// If the vbase is a primary virtual base of some base, then it doesn't
// get its own storage location but instead lives inside of that base.
if (isOverlappingVBaseABI() &&
Context.isNearlyEmpty(BaseDecl) &&
!hasOwnStorage(RD, BaseDecl)) {
Members.push_back(MemberInfo(Offset, MemberInfo::VBase, nullptr,
BaseDecl));
continue;
}
// If we've got a vtordisp, add it as a storage type.
if (Layout.getVBaseOffsetsMap().find(BaseDecl)->second.hasVtorDisp())
Members.push_back(StorageInfo(Offset - CharUnits::fromQuantity(4),
getIntNType(32)));
Members.push_back(MemberInfo(Offset, MemberInfo::VBase,
getStorageType(BaseDecl), BaseDecl));
}
}
bool CGRecordLowering::hasOwnStorage(const CXXRecordDecl *Decl,
const CXXRecordDecl *Query) {
const ASTRecordLayout &DeclLayout = Context.getASTRecordLayout(Decl);
if (DeclLayout.isPrimaryBaseVirtual() && DeclLayout.getPrimaryBase() == Query)
return false;
for (const auto &Base : Decl->bases())
if (!hasOwnStorage(Base.getType()->getAsCXXRecordDecl(), Query))
return false;
return true;
}
void CGRecordLowering::calculateZeroInit() {
for (std::vector<MemberInfo>::const_iterator Member = Members.begin(),
MemberEnd = Members.end();
IsZeroInitializableAsBase && Member != MemberEnd; ++Member) {
if (Member->Kind == MemberInfo::Field) {
if (!Member->FD || isZeroInitializable(Member->FD))
continue;
IsZeroInitializable = IsZeroInitializableAsBase = false;
} else if (Member->Kind == MemberInfo::Base ||
Member->Kind == MemberInfo::VBase) {
if (isZeroInitializable(Member->RD))
continue;
IsZeroInitializable = false;
if (Member->Kind == MemberInfo::Base)
IsZeroInitializableAsBase = false;
}
}
}
void CGRecordLowering::clipTailPadding() {
std::vector<MemberInfo>::iterator Prior = Members.begin();
CharUnits Tail = getSize(Prior->Data);
for (std::vector<MemberInfo>::iterator Member = Prior + 1,
MemberEnd = Members.end();
Member != MemberEnd; ++Member) {
// Only members with data and the scissor can cut into tail padding.
if (!Member->Data && Member->Kind != MemberInfo::Scissor)
continue;
if (Member->Offset < Tail) {
assert(Prior->Kind == MemberInfo::Field &&
"Only storage fields have tail padding!");
if (!Prior->FD || Prior->FD->isBitField())
Prior->Data = getByteArrayType(bitsToCharUnits(llvm::alignTo(
cast<llvm::IntegerType>(Prior->Data)->getIntegerBitWidth(), 8)));
else {
assert(Prior->FD->hasAttr<NoUniqueAddressAttr>() &&
"should not have reused this field's tail padding");
Prior->Data = getByteArrayType(
Context.getTypeInfoDataSizeInChars(Prior->FD->getType()).Width);
}
}
if (Member->Data)
Prior = Member;
Tail = Prior->Offset + getSize(Prior->Data);
}
}
void CGRecordLowering::determinePacked(bool NVBaseType) {
if (Packed)
return;
CharUnits Alignment = CharUnits::One();
CharUnits NVAlignment = CharUnits::One();
CharUnits NVSize =
!NVBaseType && RD ? Layout.getNonVirtualSize() : CharUnits::Zero();
for (std::vector<MemberInfo>::const_iterator Member = Members.begin(),
MemberEnd = Members.end();
Member != MemberEnd; ++Member) {
if (!Member->Data)
continue;
// If any member falls at an offset that it not a multiple of its alignment,
// then the entire record must be packed.
if (Member->Offset % getAlignment(Member->Data))
Packed = true;
if (Member->Offset < NVSize)
NVAlignment = std::max(NVAlignment, getAlignment(Member->Data));
Alignment = std::max(Alignment, getAlignment(Member->Data));
}
// If the size of the record (the capstone's offset) is not a multiple of the
// record's alignment, it must be packed.
if (Members.back().Offset % Alignment)
Packed = true;
// If the non-virtual sub-object is not a multiple of the non-virtual
// sub-object's alignment, it must be packed. We cannot have a packed
// non-virtual sub-object and an unpacked complete object or vise versa.
if (NVSize % NVAlignment)
Packed = true;
// Update the alignment of the sentinel.
if (!Packed)
Members.back().Data = getIntNType(Context.toBits(Alignment));
}
void CGRecordLowering::insertPadding() {
std::vector<std::pair<CharUnits, CharUnits> > Padding;
CharUnits Size = CharUnits::Zero();
for (std::vector<MemberInfo>::const_iterator Member = Members.begin(),
MemberEnd = Members.end();
Member != MemberEnd; ++Member) {
if (!Member->Data)
continue;
CharUnits Offset = Member->Offset;
assert(Offset >= Size);
// Insert padding if we need to.
if (Offset !=
Size.alignTo(Packed ? CharUnits::One() : getAlignment(Member->Data)))
Padding.push_back(std::make_pair(Size, Offset - Size));
Size = Offset + getSize(Member->Data);
}
if (Padding.empty())
return;
// Add the padding to the Members list and sort it.
for (std::vector<std::pair<CharUnits, CharUnits> >::const_iterator
Pad = Padding.begin(), PadEnd = Padding.end();
Pad != PadEnd; ++Pad)
Members.push_back(StorageInfo(Pad->first, getByteArrayType(Pad->second)));
llvm::stable_sort(Members);
}
void CGRecordLowering::fillOutputFields() {
for (std::vector<MemberInfo>::const_iterator Member = Members.begin(),
MemberEnd = Members.end();
Member != MemberEnd; ++Member) {
if (Member->Data)
FieldTypes.push_back(Member->Data);
if (Member->Kind == MemberInfo::Field) {
if (Member->FD)
Fields[Member->FD->getCanonicalDecl()] = FieldTypes.size() - 1;
// A field without storage must be a bitfield.
if (!Member->Data)
setBitFieldInfo(Member->FD, Member->Offset, FieldTypes.back());
} else if (Member->Kind == MemberInfo::Base)
NonVirtualBases[Member->RD] = FieldTypes.size() - 1;
else if (Member->Kind == MemberInfo::VBase)
VirtualBases[Member->RD] = FieldTypes.size() - 1;
}
}
CGBitFieldInfo CGBitFieldInfo::MakeInfo(CodeGenTypes &Types,
const FieldDecl *FD,
uint64_t Offset, uint64_t Size,
uint64_t StorageSize,
CharUnits StorageOffset) {
// This function is vestigial from CGRecordLayoutBuilder days but is still
// used in GCObjCRuntime.cpp. That usage has a "fixme" attached to it that
// when addressed will allow for the removal of this function.
llvm::Type *Ty = Types.ConvertTypeForMem(FD->getType());
CharUnits TypeSizeInBytes =
CharUnits::fromQuantity(Types.getDataLayout().getTypeAllocSize(Ty));
uint64_t TypeSizeInBits = Types.getContext().toBits(TypeSizeInBytes);
bool IsSigned = FD->getType()->isSignedIntegerOrEnumerationType();
if (Size > TypeSizeInBits) {
// We have a wide bit-field. The extra bits are only used for padding, so
// if we have a bitfield of type T, with size N:
//
// T t : N;
//
// We can just assume that it's:
//
// T t : sizeof(T);
//
Size = TypeSizeInBits;
}
// Reverse the bit offsets for big endian machines. Because we represent
// a bitfield as a single large integer load, we can imagine the bits
// counting from the most-significant-bit instead of the
// least-significant-bit.
if (Types.getDataLayout().isBigEndian()) {
Offset = StorageSize - (Offset + Size);
}
return CGBitFieldInfo(Offset, Size, IsSigned, StorageSize, StorageOffset);
}
std::unique_ptr<CGRecordLayout>
CodeGenTypes::ComputeRecordLayout(const RecordDecl *D, llvm::StructType *Ty) {
CGRecordLowering Builder(*this, D, /*Packed=*/false);
Builder.lower(/*NonVirtualBaseType=*/false);
// If we're in C++, compute the base subobject type.
llvm::StructType *BaseTy = nullptr;
if (isa<CXXRecordDecl>(D) && !D->isUnion() && !D->hasAttr<FinalAttr>()) {
BaseTy = Ty;
if (Builder.Layout.getNonVirtualSize() != Builder.Layout.getSize()) {
CGRecordLowering BaseBuilder(*this, D, /*Packed=*/Builder.Packed);
BaseBuilder.lower(/*NonVirtualBaseType=*/true);
BaseTy = llvm::StructType::create(
getLLVMContext(), BaseBuilder.FieldTypes, "", BaseBuilder.Packed);
addRecordTypeName(D, BaseTy, ".base");
// BaseTy and Ty must agree on their packedness for getLLVMFieldNo to work
// on both of them with the same index.
assert(Builder.Packed == BaseBuilder.Packed &&
"Non-virtual and complete types must agree on packedness");
}
}
// Fill in the struct *after* computing the base type. Filling in the body
// signifies that the type is no longer opaque and record layout is complete,
// but we may need to recursively layout D while laying D out as a base type.
Ty->setBody(Builder.FieldTypes, Builder.Packed);
auto RL = std::make_unique<CGRecordLayout>(
Ty, BaseTy, (bool)Builder.IsZeroInitializable,
(bool)Builder.IsZeroInitializableAsBase);
RL->NonVirtualBases.swap(Builder.NonVirtualBases);
RL->CompleteObjectVirtualBases.swap(Builder.VirtualBases);
// Add all the field numbers.
RL->FieldInfo.swap(Builder.Fields);
// Add bitfield info.
RL->BitFields.swap(Builder.BitFields);
// Dump the layout, if requested.
if (getContext().getLangOpts().DumpRecordLayouts) {
llvm::outs() << "\n*** Dumping IRgen Record Layout\n";
llvm::outs() << "Record: ";
D->dump(llvm::outs());
llvm::outs() << "\nLayout: ";
RL->print(llvm::outs());
}
#ifndef NDEBUG
// Verify that the computed LLVM struct size matches the AST layout size.
const ASTRecordLayout &Layout = getContext().getASTRecordLayout(D);
uint64_t TypeSizeInBits = getContext().toBits(Layout.getSize());
assert(TypeSizeInBits == getDataLayout().getTypeAllocSizeInBits(Ty) &&
"Type size mismatch!");
if (BaseTy) {
CharUnits NonVirtualSize = Layout.getNonVirtualSize();
uint64_t AlignedNonVirtualTypeSizeInBits =
getContext().toBits(NonVirtualSize);
assert(AlignedNonVirtualTypeSizeInBits ==
getDataLayout().getTypeAllocSizeInBits(BaseTy) &&
"Type size mismatch!");
}
// Verify that the LLVM and AST field offsets agree.
llvm::StructType *ST = RL->getLLVMType();
const llvm::StructLayout *SL = getDataLayout().getStructLayout(ST);
const ASTRecordLayout &AST_RL = getContext().getASTRecordLayout(D);
RecordDecl::field_iterator it = D->field_begin();
for (unsigned i = 0, e = AST_RL.getFieldCount(); i != e; ++i, ++it) {
const FieldDecl *FD = *it;
// Ignore zero-sized fields.
if (FD->isZeroSize(getContext()))
continue;
// For non-bit-fields, just check that the LLVM struct offset matches the
// AST offset.
if (!FD->isBitField()) {
unsigned FieldNo = RL->getLLVMFieldNo(FD);
assert(AST_RL.getFieldOffset(i) == SL->getElementOffsetInBits(FieldNo) &&
"Invalid field offset!");
continue;
}
// Ignore unnamed bit-fields.
if (!FD->getDeclName())
continue;
const CGBitFieldInfo &Info = RL->getBitFieldInfo(FD);
llvm::Type *ElementTy = ST->getTypeAtIndex(RL->getLLVMFieldNo(FD));
// Unions have overlapping elements dictating their layout, but for
// non-unions we can verify that this section of the layout is the exact
// expected size.
if (D->isUnion()) {
// For unions we verify that the start is zero and the size
// is in-bounds. However, on BE systems, the offset may be non-zero, but
// the size + offset should match the storage size in that case as it
// "starts" at the back.
if (getDataLayout().isBigEndian())
assert(static_cast<unsigned>(Info.Offset + Info.Size) ==
Info.StorageSize &&
"Big endian union bitfield does not end at the back");
else
assert(Info.Offset == 0 &&
"Little endian union bitfield with a non-zero offset");
assert(Info.StorageSize <= SL->getSizeInBits() &&
"Union not large enough for bitfield storage");
} else {
assert((Info.StorageSize ==
getDataLayout().getTypeAllocSizeInBits(ElementTy) ||
Info.VolatileStorageSize ==
getDataLayout().getTypeAllocSizeInBits(ElementTy)) &&
"Storage size does not match the element type size");
}
assert(Info.Size > 0 && "Empty bitfield!");
assert(static_cast<unsigned>(Info.Offset) + Info.Size <= Info.StorageSize &&
"Bitfield outside of its allocated storage");
}
#endif
return RL;
}
void CGRecordLayout::print(raw_ostream &OS) const {
OS << "<CGRecordLayout\n";
OS << " LLVMType:" << *CompleteObjectType << "\n";
if (BaseSubobjectType)
OS << " NonVirtualBaseLLVMType:" << *BaseSubobjectType << "\n";
OS << " IsZeroInitializable:" << IsZeroInitializable << "\n";
OS << " BitFields:[\n";
// Print bit-field infos in declaration order.
std::vector<std::pair<unsigned, const CGBitFieldInfo*> > BFIs;
for (llvm::DenseMap<const FieldDecl*, CGBitFieldInfo>::const_iterator
it = BitFields.begin(), ie = BitFields.end();
it != ie; ++it) {
const RecordDecl *RD = it->first->getParent();
unsigned Index = 0;
for (RecordDecl::field_iterator
it2 = RD->field_begin(); *it2 != it->first; ++it2)
++Index;
BFIs.push_back(std::make_pair(Index, &it->second));
}
llvm::array_pod_sort(BFIs.begin(), BFIs.end());
for (unsigned i = 0, e = BFIs.size(); i != e; ++i) {
OS.indent(4);
BFIs[i].second->print(OS);
OS << "\n";
}
OS << "]>\n";
}
LLVM_DUMP_METHOD void CGRecordLayout::dump() const {
print(llvm::errs());
}
void CGBitFieldInfo::print(raw_ostream &OS) const {
OS << "<CGBitFieldInfo"
<< " Offset:" << Offset << " Size:" << Size << " IsSigned:" << IsSigned
<< " StorageSize:" << StorageSize
<< " StorageOffset:" << StorageOffset.getQuantity()
<< " VolatileOffset:" << VolatileOffset
<< " VolatileStorageSize:" << VolatileStorageSize
<< " VolatileStorageOffset:" << VolatileStorageOffset.getQuantity() << ">";
}
LLVM_DUMP_METHOD void CGBitFieldInfo::dump() const {
print(llvm::errs());
}
|