[ARM] Split large truncating MVE stores
[llvm-complete.git] / lib / DebugInfo / CodeView / TypeStreamMerger.cpp
blobaba0e96d606e7129de2c575af6be44985a24a187
1 //===-- TypeStreamMerger.cpp ------------------------------------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
9 #include "llvm/DebugInfo/CodeView/TypeStreamMerger.h"
10 #include "llvm/ADT/SmallString.h"
11 #include "llvm/ADT/StringExtras.h"
12 #include "llvm/DebugInfo/CodeView/GlobalTypeTableBuilder.h"
13 #include "llvm/DebugInfo/CodeView/MergingTypeTableBuilder.h"
14 #include "llvm/DebugInfo/CodeView/TypeDeserializer.h"
15 #include "llvm/DebugInfo/CodeView/TypeIndex.h"
16 #include "llvm/DebugInfo/CodeView/TypeIndexDiscovery.h"
17 #include "llvm/DebugInfo/CodeView/TypeRecord.h"
18 #include "llvm/Support/Error.h"
20 using namespace llvm;
21 using namespace llvm::codeview;
23 static inline size_t slotForIndex(TypeIndex Idx) {
24 assert(!Idx.isSimple() && "simple type indices have no slots");
25 return Idx.getIndex() - TypeIndex::FirstNonSimpleIndex;
28 namespace {
30 /// Implementation of CodeView type stream merging.
31 ///
32 /// A CodeView type stream is a series of records that reference each other
33 /// through type indices. A type index is either "simple", meaning it is less
34 /// than 0x1000 and refers to a builtin type, or it is complex, meaning it
35 /// refers to a prior type record in the current stream. The type index of a
36 /// record is equal to the number of records before it in the stream plus
37 /// 0x1000.
38 ///
39 /// Type records are only allowed to use type indices smaller than their own, so
40 /// a type stream is effectively a topologically sorted DAG. Cycles occuring in
41 /// the type graph of the source program are resolved with forward declarations
42 /// of composite types. This class implements the following type stream merging
43 /// algorithm, which relies on this DAG structure:
44 ///
45 /// - Begin with a new empty stream, and a new empty hash table that maps from
46 /// type record contents to new type index.
47 /// - For each new type stream, maintain a map from source type index to
48 /// destination type index.
49 /// - For each record, copy it and rewrite its type indices to be valid in the
50 /// destination type stream.
51 /// - If the new type record is not already present in the destination stream
52 /// hash table, append it to the destination type stream, assign it the next
53 /// type index, and update the two hash tables.
54 /// - If the type record already exists in the destination stream, discard it
55 /// and update the type index map to forward the source type index to the
56 /// existing destination type index.
57 ///
58 /// As an additional complication, type stream merging actually produces two
59 /// streams: an item (or IPI) stream and a type stream, as this is what is
60 /// actually stored in the final PDB. We choose which records go where by
61 /// looking at the record kind.
62 class TypeStreamMerger {
63 public:
64 explicit TypeStreamMerger(SmallVectorImpl<TypeIndex> &SourceToDest)
65 : IndexMap(SourceToDest) {
66 // When dealing with precompiled headers objects, all data in SourceToDest
67 // belongs to the precompiled headers object, and is assumed to be already
68 // remapped to the target PDB. Any forthcoming type that will be merged in
69 // might potentially back-reference this data. We also don't want to resolve
70 // twice the types in the precompiled object.
71 CurIndex += SourceToDest.size();
74 static const TypeIndex Untranslated;
76 // Local hashing entry points
77 Error mergeTypesAndIds(MergingTypeTableBuilder &DestIds,
78 MergingTypeTableBuilder &DestTypes,
79 const CVTypeArray &IdsAndTypes, Optional<uint32_t> &S);
80 Error mergeIdRecords(MergingTypeTableBuilder &Dest,
81 ArrayRef<TypeIndex> TypeSourceToDest,
82 const CVTypeArray &Ids);
83 Error mergeTypeRecords(MergingTypeTableBuilder &Dest,
84 const CVTypeArray &Types);
86 // Global hashing entry points
87 Error mergeTypesAndIds(GlobalTypeTableBuilder &DestIds,
88 GlobalTypeTableBuilder &DestTypes,
89 const CVTypeArray &IdsAndTypes,
90 ArrayRef<GloballyHashedType> Hashes,
91 Optional<uint32_t> &S);
92 Error mergeIdRecords(GlobalTypeTableBuilder &Dest,
93 ArrayRef<TypeIndex> TypeSourceToDest,
94 const CVTypeArray &Ids,
95 ArrayRef<GloballyHashedType> Hashes);
96 Error mergeTypeRecords(GlobalTypeTableBuilder &Dest, const CVTypeArray &Types,
97 ArrayRef<GloballyHashedType> Hashes,
98 Optional<uint32_t> &S);
100 private:
101 Error doit(const CVTypeArray &Types);
103 Error remapAllTypes(const CVTypeArray &Types);
105 Error remapType(const CVType &Type);
107 void addMapping(TypeIndex Idx);
109 inline bool remapTypeIndex(TypeIndex &Idx) {
110 // If we're mapping a pure index stream, then IndexMap only contains
111 // mappings from OldIdStream -> NewIdStream, in which case we will need to
112 // use the special mapping from OldTypeStream -> NewTypeStream which was
113 // computed externally. Regardless, we use this special map if and only if
114 // we are doing an id-only mapping.
115 if (!hasTypeStream())
116 return remapIndex(Idx, TypeLookup);
118 assert(TypeLookup.empty());
119 return remapIndex(Idx, IndexMap);
121 inline bool remapItemIndex(TypeIndex &Idx) {
122 assert(hasIdStream());
123 return remapIndex(Idx, IndexMap);
126 bool hasTypeStream() const {
127 return (UseGlobalHashes) ? (!!DestGlobalTypeStream) : (!!DestTypeStream);
130 bool hasIdStream() const {
131 return (UseGlobalHashes) ? (!!DestGlobalIdStream) : (!!DestIdStream);
134 ArrayRef<uint8_t> remapIndices(const CVType &OriginalType,
135 MutableArrayRef<uint8_t> Storage);
137 inline bool remapIndex(TypeIndex &Idx, ArrayRef<TypeIndex> Map) {
138 if (LLVM_LIKELY(remapIndexSimple(Idx, Map)))
139 return true;
141 return remapIndexFallback(Idx, Map);
144 inline bool remapIndexSimple(TypeIndex &Idx, ArrayRef<TypeIndex> Map) const {
145 // Simple types are unchanged.
146 if (Idx.isSimple())
147 return true;
149 // Check if this type index refers to a record we've already translated
150 // successfully. If it refers to a type later in the stream or a record we
151 // had to defer, defer it until later pass.
152 unsigned MapPos = slotForIndex(Idx);
153 if (LLVM_UNLIKELY(MapPos >= Map.size() || Map[MapPos] == Untranslated))
154 return false;
156 Idx = Map[MapPos];
157 return true;
160 bool remapIndexFallback(TypeIndex &Idx, ArrayRef<TypeIndex> Map);
162 Error errorCorruptRecord() const {
163 return llvm::make_error<CodeViewError>(cv_error_code::corrupt_record);
166 Expected<bool> shouldRemapType(const CVType &Type);
168 Optional<Error> LastError;
170 bool UseGlobalHashes = false;
172 bool IsSecondPass = false;
174 unsigned NumBadIndices = 0;
176 TypeIndex CurIndex{TypeIndex::FirstNonSimpleIndex};
178 MergingTypeTableBuilder *DestIdStream = nullptr;
179 MergingTypeTableBuilder *DestTypeStream = nullptr;
181 GlobalTypeTableBuilder *DestGlobalIdStream = nullptr;
182 GlobalTypeTableBuilder *DestGlobalTypeStream = nullptr;
184 ArrayRef<GloballyHashedType> GlobalHashes;
186 // If we're only mapping id records, this array contains the mapping for
187 // type records.
188 ArrayRef<TypeIndex> TypeLookup;
190 /// Map from source type index to destination type index. Indexed by source
191 /// type index minus 0x1000.
192 SmallVectorImpl<TypeIndex> &IndexMap;
194 /// Temporary storage that we use to copy a record's data while re-writing
195 /// its type indices.
196 SmallVector<uint8_t, 256> RemapStorage;
198 Optional<uint32_t> PCHSignature;
201 } // end anonymous namespace
203 const TypeIndex TypeStreamMerger::Untranslated(SimpleTypeKind::NotTranslated);
205 static bool isIdRecord(TypeLeafKind K) {
206 switch (K) {
207 case TypeLeafKind::LF_FUNC_ID:
208 case TypeLeafKind::LF_MFUNC_ID:
209 case TypeLeafKind::LF_STRING_ID:
210 case TypeLeafKind::LF_SUBSTR_LIST:
211 case TypeLeafKind::LF_BUILDINFO:
212 case TypeLeafKind::LF_UDT_SRC_LINE:
213 case TypeLeafKind::LF_UDT_MOD_SRC_LINE:
214 return true;
215 default:
216 return false;
220 void TypeStreamMerger::addMapping(TypeIndex Idx) {
221 if (!IsSecondPass) {
222 assert(IndexMap.size() == slotForIndex(CurIndex) &&
223 "visitKnownRecord should add one index map entry");
224 IndexMap.push_back(Idx);
225 } else {
226 assert(slotForIndex(CurIndex) < IndexMap.size());
227 IndexMap[slotForIndex(CurIndex)] = Idx;
231 bool TypeStreamMerger::remapIndexFallback(TypeIndex &Idx,
232 ArrayRef<TypeIndex> Map) {
233 size_t MapPos = slotForIndex(Idx);
235 // If this is the second pass and this index isn't in the map, then it points
236 // outside the current type stream, and this is a corrupt record.
237 if (IsSecondPass && MapPos >= Map.size()) {
238 // FIXME: Print a more useful error. We can give the current record and the
239 // index that we think its pointing to.
240 if (LastError)
241 LastError = joinErrors(std::move(*LastError), errorCorruptRecord());
242 else
243 LastError = errorCorruptRecord();
246 ++NumBadIndices;
248 // This type index is invalid. Remap this to "not translated by cvpack",
249 // and return failure.
250 Idx = Untranslated;
251 return false;
254 // Local hashing entry points
255 Error TypeStreamMerger::mergeTypeRecords(MergingTypeTableBuilder &Dest,
256 const CVTypeArray &Types) {
257 DestTypeStream = &Dest;
258 UseGlobalHashes = false;
260 return doit(Types);
263 Error TypeStreamMerger::mergeIdRecords(MergingTypeTableBuilder &Dest,
264 ArrayRef<TypeIndex> TypeSourceToDest,
265 const CVTypeArray &Ids) {
266 DestIdStream = &Dest;
267 TypeLookup = TypeSourceToDest;
268 UseGlobalHashes = false;
270 return doit(Ids);
273 Error TypeStreamMerger::mergeTypesAndIds(MergingTypeTableBuilder &DestIds,
274 MergingTypeTableBuilder &DestTypes,
275 const CVTypeArray &IdsAndTypes,
276 Optional<uint32_t> &S) {
277 DestIdStream = &DestIds;
278 DestTypeStream = &DestTypes;
279 UseGlobalHashes = false;
280 auto Err = doit(IdsAndTypes);
281 S = PCHSignature;
282 return Err;
285 // Global hashing entry points
286 Error TypeStreamMerger::mergeTypeRecords(GlobalTypeTableBuilder &Dest,
287 const CVTypeArray &Types,
288 ArrayRef<GloballyHashedType> Hashes,
289 Optional<uint32_t> &S) {
290 DestGlobalTypeStream = &Dest;
291 UseGlobalHashes = true;
292 GlobalHashes = Hashes;
293 auto Err = doit(Types);
294 S = PCHSignature;
295 return Err;
298 Error TypeStreamMerger::mergeIdRecords(GlobalTypeTableBuilder &Dest,
299 ArrayRef<TypeIndex> TypeSourceToDest,
300 const CVTypeArray &Ids,
301 ArrayRef<GloballyHashedType> Hashes) {
302 DestGlobalIdStream = &Dest;
303 TypeLookup = TypeSourceToDest;
304 UseGlobalHashes = true;
305 GlobalHashes = Hashes;
307 return doit(Ids);
310 Error TypeStreamMerger::mergeTypesAndIds(GlobalTypeTableBuilder &DestIds,
311 GlobalTypeTableBuilder &DestTypes,
312 const CVTypeArray &IdsAndTypes,
313 ArrayRef<GloballyHashedType> Hashes,
314 Optional<uint32_t> &S) {
315 DestGlobalIdStream = &DestIds;
316 DestGlobalTypeStream = &DestTypes;
317 UseGlobalHashes = true;
318 GlobalHashes = Hashes;
319 auto Err = doit(IdsAndTypes);
320 S = PCHSignature;
321 return Err;
324 Error TypeStreamMerger::doit(const CVTypeArray &Types) {
325 if (auto EC = remapAllTypes(Types))
326 return EC;
328 // If we found bad indices but no other errors, try doing another pass and see
329 // if we can resolve the indices that weren't in the map on the first pass.
330 // This may require multiple passes, but we should always make progress. MASM
331 // is the only known CodeView producer that makes type streams that aren't
332 // topologically sorted. The standard library contains MASM-produced objects,
333 // so this is important to handle correctly, but we don't have to be too
334 // efficient. MASM type streams are usually very small.
335 while (!LastError && NumBadIndices > 0) {
336 unsigned BadIndicesRemaining = NumBadIndices;
337 IsSecondPass = true;
338 NumBadIndices = 0;
339 CurIndex = TypeIndex(TypeIndex::FirstNonSimpleIndex);
341 if (auto EC = remapAllTypes(Types))
342 return EC;
344 assert(NumBadIndices <= BadIndicesRemaining &&
345 "second pass found more bad indices");
346 if (!LastError && NumBadIndices == BadIndicesRemaining) {
347 return llvm::make_error<CodeViewError>(
348 cv_error_code::corrupt_record, "Input type graph contains cycles");
352 if (LastError)
353 return std::move(*LastError);
354 return Error::success();
357 Error TypeStreamMerger::remapAllTypes(const CVTypeArray &Types) {
358 BinaryStreamRef Stream = Types.getUnderlyingStream();
359 ArrayRef<uint8_t> Buffer;
360 cantFail(Stream.readBytes(0, Stream.getLength(), Buffer));
362 return forEachCodeViewRecord<CVType>(
363 Buffer, [this](const CVType &T) { return remapType(T); });
366 Error TypeStreamMerger::remapType(const CVType &Type) {
367 auto R = shouldRemapType(Type);
368 if (!R)
369 return R.takeError();
371 TypeIndex DestIdx = Untranslated;
372 if (*R) {
373 auto DoSerialize =
374 [this, Type](MutableArrayRef<uint8_t> Storage) -> ArrayRef<uint8_t> {
375 return remapIndices(Type, Storage);
377 if (LLVM_LIKELY(UseGlobalHashes)) {
378 GlobalTypeTableBuilder &Dest =
379 isIdRecord(Type.kind()) ? *DestGlobalIdStream : *DestGlobalTypeStream;
380 GloballyHashedType H = GlobalHashes[CurIndex.toArrayIndex()];
381 DestIdx = Dest.insertRecordAs(H, Type.RecordData.size(), DoSerialize);
382 } else {
383 MergingTypeTableBuilder &Dest =
384 isIdRecord(Type.kind()) ? *DestIdStream : *DestTypeStream;
386 RemapStorage.resize(Type.RecordData.size());
387 ArrayRef<uint8_t> Result = DoSerialize(RemapStorage);
388 if (!Result.empty())
389 DestIdx = Dest.insertRecordBytes(Result);
392 addMapping(DestIdx);
394 ++CurIndex;
395 assert((IsSecondPass || IndexMap.size() == slotForIndex(CurIndex)) &&
396 "visitKnownRecord should add one index map entry");
397 return Error::success();
400 ArrayRef<uint8_t>
401 TypeStreamMerger::remapIndices(const CVType &OriginalType,
402 MutableArrayRef<uint8_t> Storage) {
403 SmallVector<TiReference, 4> Refs;
404 discoverTypeIndices(OriginalType.RecordData, Refs);
405 if (Refs.empty())
406 return OriginalType.RecordData;
408 ::memcpy(Storage.data(), OriginalType.RecordData.data(),
409 OriginalType.RecordData.size());
411 uint8_t *DestContent = Storage.data() + sizeof(RecordPrefix);
413 for (auto &Ref : Refs) {
414 TypeIndex *DestTIs =
415 reinterpret_cast<TypeIndex *>(DestContent + Ref.Offset);
417 for (size_t I = 0; I < Ref.Count; ++I) {
418 TypeIndex &TI = DestTIs[I];
419 bool Success = (Ref.Kind == TiRefKind::IndexRef) ? remapItemIndex(TI)
420 : remapTypeIndex(TI);
421 if (LLVM_UNLIKELY(!Success))
422 return {};
425 return Storage;
428 Error llvm::codeview::mergeTypeRecords(MergingTypeTableBuilder &Dest,
429 SmallVectorImpl<TypeIndex> &SourceToDest,
430 const CVTypeArray &Types) {
431 TypeStreamMerger M(SourceToDest);
432 return M.mergeTypeRecords(Dest, Types);
435 Error llvm::codeview::mergeIdRecords(MergingTypeTableBuilder &Dest,
436 ArrayRef<TypeIndex> TypeSourceToDest,
437 SmallVectorImpl<TypeIndex> &SourceToDest,
438 const CVTypeArray &Ids) {
439 TypeStreamMerger M(SourceToDest);
440 return M.mergeIdRecords(Dest, TypeSourceToDest, Ids);
443 Error llvm::codeview::mergeTypeAndIdRecords(
444 MergingTypeTableBuilder &DestIds, MergingTypeTableBuilder &DestTypes,
445 SmallVectorImpl<TypeIndex> &SourceToDest, const CVTypeArray &IdsAndTypes,
446 Optional<uint32_t> &PCHSignature) {
447 TypeStreamMerger M(SourceToDest);
448 return M.mergeTypesAndIds(DestIds, DestTypes, IdsAndTypes, PCHSignature);
451 Error llvm::codeview::mergeTypeAndIdRecords(
452 GlobalTypeTableBuilder &DestIds, GlobalTypeTableBuilder &DestTypes,
453 SmallVectorImpl<TypeIndex> &SourceToDest, const CVTypeArray &IdsAndTypes,
454 ArrayRef<GloballyHashedType> Hashes, Optional<uint32_t> &PCHSignature) {
455 TypeStreamMerger M(SourceToDest);
456 return M.mergeTypesAndIds(DestIds, DestTypes, IdsAndTypes, Hashes,
457 PCHSignature);
460 Error llvm::codeview::mergeTypeRecords(GlobalTypeTableBuilder &Dest,
461 SmallVectorImpl<TypeIndex> &SourceToDest,
462 const CVTypeArray &Types,
463 ArrayRef<GloballyHashedType> Hashes,
464 Optional<uint32_t> &PCHSignature) {
465 TypeStreamMerger M(SourceToDest);
466 return M.mergeTypeRecords(Dest, Types, Hashes, PCHSignature);
469 Error llvm::codeview::mergeIdRecords(GlobalTypeTableBuilder &Dest,
470 ArrayRef<TypeIndex> Types,
471 SmallVectorImpl<TypeIndex> &SourceToDest,
472 const CVTypeArray &Ids,
473 ArrayRef<GloballyHashedType> Hashes) {
474 TypeStreamMerger M(SourceToDest);
475 return M.mergeIdRecords(Dest, Types, Ids, Hashes);
478 Expected<bool> TypeStreamMerger::shouldRemapType(const CVType &Type) {
479 // For object files containing precompiled types, we need to extract the
480 // signature, through EndPrecompRecord. This is done here for performance
481 // reasons, to avoid re-parsing the Types stream.
482 if (Type.kind() == LF_ENDPRECOMP) {
483 EndPrecompRecord EP;
484 if (auto EC = TypeDeserializer::deserializeAs(const_cast<CVType &>(Type),
485 EP))
486 return joinErrors(std::move(EC), errorCorruptRecord());
487 if (PCHSignature.hasValue())
488 return errorCorruptRecord();
489 PCHSignature.emplace(EP.getSignature());
490 return false;
492 return true;