1 //===- BitstreamReader.h - Low-level bitstream reader interface -*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This header defines the BitstreamReader class. This class can be used to
11 // read an arbitrary bitstream, regardless of its contents.
13 //===----------------------------------------------------------------------===//
15 #ifndef LLVM_BITCODE_BITSTREAMREADER_H
16 #define LLVM_BITCODE_BITSTREAMREADER_H
18 #include "llvm/ADT/OwningPtr.h"
19 #include "llvm/Bitcode/BitCodes.h"
20 #include "llvm/Support/Endian.h"
21 #include "llvm/Support/StreamableMemoryObject.h"
30 /// BitstreamReader - This class is used to read from an LLVM bitcode stream,
31 /// maintaining information that is global to decoding the entire file. While
32 /// a file is being read, multiple cursors can be independently advanced or
33 /// skipped around within the file. These are represented by the
34 /// BitstreamCursor class.
35 class BitstreamReader {
37 /// BlockInfo - This contains information emitted to BLOCKINFO_BLOCK blocks.
38 /// These describe abbreviations that all blocks of the specified ID inherit.
41 std::vector<BitCodeAbbrev*> Abbrevs;
44 std::vector<std::pair<unsigned, std::string> > RecordNames;
47 OwningPtr<StreamableMemoryObject> BitcodeBytes;
49 std::vector<BlockInfo> BlockInfoRecords;
51 /// IgnoreBlockInfoNames - This is set to true if we don't care about the
52 /// block/record name information in the BlockInfo block. Only llvm-bcanalyzer
54 bool IgnoreBlockInfoNames;
56 BitstreamReader(const BitstreamReader&) LLVM_DELETED_FUNCTION;
57 void operator=(const BitstreamReader&) LLVM_DELETED_FUNCTION;
59 BitstreamReader() : IgnoreBlockInfoNames(true) {
62 BitstreamReader(const unsigned char *Start, const unsigned char *End) {
63 IgnoreBlockInfoNames = true;
67 BitstreamReader(StreamableMemoryObject *bytes) {
68 BitcodeBytes.reset(bytes);
71 void init(const unsigned char *Start, const unsigned char *End) {
72 assert(((End-Start) & 3) == 0 &&"Bitcode stream not a multiple of 4 bytes");
73 BitcodeBytes.reset(getNonStreamedMemoryObject(Start, End));
76 StreamableMemoryObject &getBitcodeBytes() { return *BitcodeBytes; }
79 // Free the BlockInfoRecords.
80 while (!BlockInfoRecords.empty()) {
81 BlockInfo &Info = BlockInfoRecords.back();
82 // Free blockinfo abbrev info.
83 for (unsigned i = 0, e = static_cast<unsigned>(Info.Abbrevs.size());
85 Info.Abbrevs[i]->dropRef();
86 BlockInfoRecords.pop_back();
90 /// CollectBlockInfoNames - This is called by clients that want block/record
92 void CollectBlockInfoNames() { IgnoreBlockInfoNames = false; }
93 bool isIgnoringBlockInfoNames() { return IgnoreBlockInfoNames; }
95 //===--------------------------------------------------------------------===//
97 //===--------------------------------------------------------------------===//
99 /// hasBlockInfoRecords - Return true if we've already read and processed the
100 /// block info block for this Bitstream. We only process it for the first
101 /// cursor that walks over it.
102 bool hasBlockInfoRecords() const { return !BlockInfoRecords.empty(); }
104 /// getBlockInfo - If there is block info for the specified ID, return it,
105 /// otherwise return null.
106 const BlockInfo *getBlockInfo(unsigned BlockID) const {
107 // Common case, the most recent entry matches BlockID.
108 if (!BlockInfoRecords.empty() && BlockInfoRecords.back().BlockID == BlockID)
109 return &BlockInfoRecords.back();
111 for (unsigned i = 0, e = static_cast<unsigned>(BlockInfoRecords.size());
113 if (BlockInfoRecords[i].BlockID == BlockID)
114 return &BlockInfoRecords[i];
118 BlockInfo &getOrCreateBlockInfo(unsigned BlockID) {
119 if (const BlockInfo *BI = getBlockInfo(BlockID))
120 return *const_cast<BlockInfo*>(BI);
122 // Otherwise, add a new record.
123 BlockInfoRecords.push_back(BlockInfo());
124 BlockInfoRecords.back().BlockID = BlockID;
125 return BlockInfoRecords.back();
130 /// BitstreamEntry - When advancing through a bitstream cursor, each advance can
131 /// discover a few different kinds of entries:
132 /// Error - Malformed bitcode was found.
133 /// EndBlock - We've reached the end of the current block, (or the end of the
134 /// file, which is treated like a series of EndBlock records.
135 /// SubBlock - This is the start of a new subblock of a specific ID.
136 /// Record - This is a record with a specific AbbrevID.
138 struct BitstreamEntry {
148 static BitstreamEntry getError() {
149 BitstreamEntry E; E.Kind = Error; return E;
151 static BitstreamEntry getEndBlock() {
152 BitstreamEntry E; E.Kind = EndBlock; return E;
154 static BitstreamEntry getSubBlock(unsigned ID) {
155 BitstreamEntry E; E.Kind = SubBlock; E.ID = ID; return E;
157 static BitstreamEntry getRecord(unsigned AbbrevID) {
158 BitstreamEntry E; E.Kind = Record; E.ID = AbbrevID; return E;
162 /// BitstreamCursor - This represents a position within a bitcode file. There
163 /// may be multiple independent cursors reading within one bitstream, each
164 /// maintaining their own local state.
166 /// Unlike iterators, BitstreamCursors are heavy-weight objects that should not
167 /// be passed by value.
168 class BitstreamCursor {
169 friend class Deserializer;
170 BitstreamReader *BitStream;
174 /// CurWord/word_t - This is the current data we have pulled from the stream
175 /// but have not returned to the client. This is specifically and
176 /// intentionally defined to follow the word size of the host machine for
177 /// efficiency. We use word_t in places that are aware of this to make it
178 /// perfectly explicit what is going on.
179 typedef uint32_t word_t;
182 /// BitsInCurWord - This is the number of bits in CurWord that are valid. This
183 /// is always from [0...31/63] inclusive (depending on word size).
184 unsigned BitsInCurWord;
186 // CurCodeSize - This is the declared size of code values used for the current
188 unsigned CurCodeSize;
190 /// CurAbbrevs - Abbrevs installed at in this block.
191 std::vector<BitCodeAbbrev*> CurAbbrevs;
194 unsigned PrevCodeSize;
195 std::vector<BitCodeAbbrev*> PrevAbbrevs;
196 explicit Block(unsigned PCS) : PrevCodeSize(PCS) {}
199 /// BlockScope - This tracks the codesize of parent blocks.
200 SmallVector<Block, 8> BlockScope;
204 BitstreamCursor() : BitStream(0), NextChar(0) {
206 BitstreamCursor(const BitstreamCursor &RHS) : BitStream(0), NextChar(0) {
210 explicit BitstreamCursor(BitstreamReader &R) : BitStream(&R) {
217 void init(BitstreamReader &R) {
231 void operator=(const BitstreamCursor &RHS);
235 bool isEndPos(size_t pos) {
236 return BitStream->getBitcodeBytes().isObjectEnd(static_cast<uint64_t>(pos));
239 bool canSkipToPos(size_t pos) const {
240 // pos can be skipped to if it is a valid address or one byte past the end.
241 return pos == 0 || BitStream->getBitcodeBytes().isValidAddress(
242 static_cast<uint64_t>(pos - 1));
245 uint32_t getWord(size_t pos) {
246 uint8_t buf[4] = { 0xFF, 0xFF, 0xFF, 0xFF };
247 BitStream->getBitcodeBytes().readBytes(pos, sizeof(buf), buf, NULL);
248 return *reinterpret_cast<support::ulittle32_t *>(buf);
251 bool AtEndOfStream() {
252 return BitsInCurWord == 0 && isEndPos(NextChar);
255 /// getAbbrevIDWidth - Return the number of bits used to encode an abbrev #.
256 unsigned getAbbrevIDWidth() const { return CurCodeSize; }
258 /// GetCurrentBitNo - Return the bit # of the bit we are reading.
259 uint64_t GetCurrentBitNo() const {
260 return NextChar*CHAR_BIT - BitsInCurWord;
263 BitstreamReader *getBitStreamReader() {
266 const BitstreamReader *getBitStreamReader() const {
270 /// Flags that modify the behavior of advance().
272 /// AF_DontPopBlockAtEnd - If this flag is used, the advance() method does
273 /// not automatically pop the block scope when the end of a block is
275 AF_DontPopBlockAtEnd = 1,
277 /// AF_DontAutoprocessAbbrevs - If this flag is used, abbrev entries are
278 /// returned just like normal records.
279 AF_DontAutoprocessAbbrevs = 2
282 /// advance - Advance the current bitstream, returning the next entry in the
284 BitstreamEntry advance(unsigned Flags = 0) {
286 unsigned Code = ReadCode();
287 if (Code == bitc::END_BLOCK) {
288 // Pop the end of the block unless Flags tells us not to.
289 if (!(Flags & AF_DontPopBlockAtEnd) && ReadBlockEnd())
290 return BitstreamEntry::getError();
291 return BitstreamEntry::getEndBlock();
294 if (Code == bitc::ENTER_SUBBLOCK)
295 return BitstreamEntry::getSubBlock(ReadSubBlockID());
297 if (Code == bitc::DEFINE_ABBREV &&
298 !(Flags & AF_DontAutoprocessAbbrevs)) {
299 // We read and accumulate abbrev's, the client can't do anything with
305 return BitstreamEntry::getRecord(Code);
309 /// advanceSkippingSubblocks - This is a convenience function for clients that
310 /// don't expect any subblocks. This just skips over them automatically.
311 BitstreamEntry advanceSkippingSubblocks(unsigned Flags = 0) {
313 // If we found a normal entry, return it.
314 BitstreamEntry Entry = advance(Flags);
315 if (Entry.Kind != BitstreamEntry::SubBlock)
318 // If we found a sub-block, just skip over it and check the next entry.
320 return BitstreamEntry::getError();
324 /// JumpToBit - Reset the stream to the specified bit number.
325 void JumpToBit(uint64_t BitNo) {
326 uintptr_t ByteNo = uintptr_t(BitNo/8) & ~(sizeof(word_t)-1);
327 unsigned WordBitNo = unsigned(BitNo & (sizeof(word_t)*8-1));
328 assert(canSkipToPos(ByteNo) && "Invalid location");
330 // Move the cursor to the right word.
335 // Skip over any bits that are already consumed.
337 if (sizeof(word_t) > 4)
345 uint32_t Read(unsigned NumBits) {
346 assert(NumBits && NumBits <= 32 &&
347 "Cannot return zero or more than 32 bits!");
349 // If the field is fully contained by CurWord, return it quickly.
350 if (BitsInCurWord >= NumBits) {
351 uint32_t R = uint32_t(CurWord) & (~0U >> (32-NumBits));
353 BitsInCurWord -= NumBits;
357 // If we run out of data, stop at the end of the stream.
358 if (isEndPos(NextChar)) {
364 uint32_t R = uint32_t(CurWord);
366 // Read the next word from the stream.
367 uint8_t Array[sizeof(word_t)] = {0};
369 BitStream->getBitcodeBytes().readBytes(NextChar, sizeof(Array),
372 // Handle big-endian byte-swapping if necessary.
373 support::detail::packed_endian_specific_integral
374 <word_t, support::little, support::unaligned> EndianValue;
375 memcpy(&EndianValue, Array, sizeof(Array));
377 CurWord = EndianValue;
379 NextChar += sizeof(word_t);
381 // Extract NumBits-BitsInCurWord from what we just read.
382 unsigned BitsLeft = NumBits-BitsInCurWord;
384 // Be careful here, BitsLeft is in the range [1..32]/[1..64] inclusive.
385 R |= uint32_t((CurWord & (word_t(~0ULL) >> (sizeof(word_t)*8-BitsLeft)))
388 // BitsLeft bits have just been used up from CurWord. BitsLeft is in the
389 // range [1..32]/[1..64] so be careful how we shift.
390 if (BitsLeft != sizeof(word_t)*8)
391 CurWord >>= BitsLeft;
394 BitsInCurWord = sizeof(word_t)*8-BitsLeft;
398 uint64_t Read64(unsigned NumBits) {
399 if (NumBits <= 32) return Read(NumBits);
401 uint64_t V = Read(32);
402 return V | (uint64_t)Read(NumBits-32) << 32;
405 uint32_t ReadVBR(unsigned NumBits) {
406 uint32_t Piece = Read(NumBits);
407 if ((Piece & (1U << (NumBits-1))) == 0)
411 unsigned NextBit = 0;
413 Result |= (Piece & ((1U << (NumBits-1))-1)) << NextBit;
415 if ((Piece & (1U << (NumBits-1))) == 0)
418 NextBit += NumBits-1;
419 Piece = Read(NumBits);
423 // ReadVBR64 - Read a VBR that may have a value up to 64-bits in size. The
424 // chunk size of the VBR must still be <= 32 bits though.
425 uint64_t ReadVBR64(unsigned NumBits) {
426 uint32_t Piece = Read(NumBits);
427 if ((Piece & (1U << (NumBits-1))) == 0)
428 return uint64_t(Piece);
431 unsigned NextBit = 0;
433 Result |= uint64_t(Piece & ((1U << (NumBits-1))-1)) << NextBit;
435 if ((Piece & (1U << (NumBits-1))) == 0)
438 NextBit += NumBits-1;
439 Piece = Read(NumBits);
444 void SkipToFourByteBoundary() {
445 // If word_t is 64-bits and if we've read less than 32 bits, just dump
446 // the bits we have up to the next 32-bit boundary.
447 if (sizeof(word_t) > 4 &&
448 BitsInCurWord >= 32) {
449 CurWord >>= BitsInCurWord-32;
459 unsigned ReadCode() {
460 return Read(CurCodeSize);
465 // [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen]
467 /// ReadSubBlockID - Having read the ENTER_SUBBLOCK code, read the BlockID for
469 unsigned ReadSubBlockID() {
470 return ReadVBR(bitc::BlockIDWidth);
473 /// SkipBlock - Having read the ENTER_SUBBLOCK abbrevid and a BlockID, skip
474 /// over the body of this block. If the block record is malformed, return
477 // Read and ignore the codelen value. Since we are skipping this block, we
478 // don't care what code widths are used inside of it.
479 ReadVBR(bitc::CodeLenWidth);
480 SkipToFourByteBoundary();
481 unsigned NumFourBytes = Read(bitc::BlockSizeWidth);
483 // Check that the block wasn't partially defined, and that the offset isn't
485 size_t SkipTo = GetCurrentBitNo() + NumFourBytes*4*8;
486 if (AtEndOfStream() || !canSkipToPos(SkipTo/8))
493 /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter
494 /// the block, and return true if the block has an error.
495 bool EnterSubBlock(unsigned BlockID, unsigned *NumWordsP = 0);
497 bool ReadBlockEnd() {
498 if (BlockScope.empty()) return true;
501 // [END_BLOCK, <align4bytes>]
502 SkipToFourByteBoundary();
510 void popBlockScope() {
511 CurCodeSize = BlockScope.back().PrevCodeSize;
513 // Delete abbrevs from popped scope.
514 for (unsigned i = 0, e = static_cast<unsigned>(CurAbbrevs.size());
516 CurAbbrevs[i]->dropRef();
518 BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
519 BlockScope.pop_back();
522 //===--------------------------------------------------------------------===//
524 //===--------------------------------------------------------------------===//
527 void readAbbreviatedLiteral(const BitCodeAbbrevOp &Op,
528 SmallVectorImpl<uint64_t> &Vals);
529 void readAbbreviatedField(const BitCodeAbbrevOp &Op,
530 SmallVectorImpl<uint64_t> &Vals);
531 void skipAbbreviatedField(const BitCodeAbbrevOp &Op);
535 /// getAbbrev - Return the abbreviation for the specified AbbrevId.
536 const BitCodeAbbrev *getAbbrev(unsigned AbbrevID) {
537 unsigned AbbrevNo = AbbrevID-bitc::FIRST_APPLICATION_ABBREV;
538 assert(AbbrevNo < CurAbbrevs.size() && "Invalid abbrev #!");
539 return CurAbbrevs[AbbrevNo];
542 /// skipRecord - Read the current record and discard it.
543 void skipRecord(unsigned AbbrevID);
545 unsigned readRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
546 StringRef *Blob = 0);
548 //===--------------------------------------------------------------------===//
550 //===--------------------------------------------------------------------===//
551 void ReadAbbrevRecord();
553 bool ReadBlockInfoBlock();
556 } // End llvm namespace