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 BITSTREAM_READER_H
16 #define BITSTREAM_READER_H
18 #include "llvm/Bitcode/BitCodes.h"
27 class BitstreamReader {
29 /// BlockInfo - This contains information emitted to BLOCKINFO_BLOCK blocks.
30 /// These describe abbreviations that all blocks of the specified ID inherit.
33 std::vector<BitCodeAbbrev*> Abbrevs;
36 std::vector<std::pair<unsigned, std::string> > RecordNames;
39 /// FirstChar/LastChar - This remembers the first and last bytes of the
41 const unsigned char *FirstChar, *LastChar;
43 std::vector<BlockInfo> BlockInfoRecords;
45 /// IgnoreBlockInfoNames - This is set to true if we don't care about the
46 /// block/record name information in the BlockInfo block. Only llvm-bcanalyzer
48 bool IgnoreBlockInfoNames;
50 BitstreamReader(const BitstreamReader&); // NOT IMPLEMENTED
51 void operator=(const BitstreamReader&); // NOT IMPLEMENTED
53 BitstreamReader() : FirstChar(0), LastChar(0), IgnoreBlockInfoNames(true) {
56 BitstreamReader(const unsigned char *Start, const unsigned char *End) {
57 IgnoreBlockInfoNames = true;
61 void init(const unsigned char *Start, const unsigned char *End) {
64 assert(((End-Start) & 3) == 0 &&"Bitcode stream not a multiple of 4 bytes");
68 // Free the BlockInfoRecords.
69 while (!BlockInfoRecords.empty()) {
70 BlockInfo &Info = BlockInfoRecords.back();
71 // Free blockinfo abbrev info.
72 for (unsigned i = 0, e = static_cast<unsigned>(Info.Abbrevs.size());
74 Info.Abbrevs[i]->dropRef();
75 BlockInfoRecords.pop_back();
79 const unsigned char *getFirstChar() const { return FirstChar; }
80 const unsigned char *getLastChar() const { return LastChar; }
82 /// CollectBlockInfoNames - This is called by clients that want block/record
84 void CollectBlockInfoNames() { IgnoreBlockInfoNames = false; }
85 bool isIgnoringBlockInfoNames() { return IgnoreBlockInfoNames; }
87 //===--------------------------------------------------------------------===//
89 //===--------------------------------------------------------------------===//
91 /// hasBlockInfoRecords - Return true if we've already read and processed the
92 /// block info block for this Bitstream. We only process it for the first
93 /// cursor that walks over it.
94 bool hasBlockInfoRecords() const { return !BlockInfoRecords.empty(); }
96 /// getBlockInfo - If there is block info for the specified ID, return it,
97 /// otherwise return null.
98 const BlockInfo *getBlockInfo(unsigned BlockID) const {
99 // Common case, the most recent entry matches BlockID.
100 if (!BlockInfoRecords.empty() && BlockInfoRecords.back().BlockID == BlockID)
101 return &BlockInfoRecords.back();
103 for (unsigned i = 0, e = static_cast<unsigned>(BlockInfoRecords.size());
105 if (BlockInfoRecords[i].BlockID == BlockID)
106 return &BlockInfoRecords[i];
110 BlockInfo &getOrCreateBlockInfo(unsigned BlockID) {
111 if (const BlockInfo *BI = getBlockInfo(BlockID))
112 return *const_cast<BlockInfo*>(BI);
114 // Otherwise, add a new record.
115 BlockInfoRecords.push_back(BlockInfo());
116 BlockInfoRecords.back().BlockID = BlockID;
117 return BlockInfoRecords.back();
122 class BitstreamCursor {
123 friend class Deserializer;
124 BitstreamReader *BitStream;
125 const unsigned char *NextChar;
127 /// CurWord - This is the current data we have pulled from the stream but have
128 /// not returned to the client.
131 /// BitsInCurWord - This is the number of bits in CurWord that are valid. This
132 /// is always from [0...31] inclusive.
133 unsigned BitsInCurWord;
135 // CurCodeSize - This is the declared size of code values used for the current
137 unsigned CurCodeSize;
139 /// CurAbbrevs - Abbrevs installed at in this block.
140 std::vector<BitCodeAbbrev*> CurAbbrevs;
143 unsigned PrevCodeSize;
144 std::vector<BitCodeAbbrev*> PrevAbbrevs;
145 explicit Block(unsigned PCS) : PrevCodeSize(PCS) {}
148 /// BlockScope - This tracks the codesize of parent blocks.
149 SmallVector<Block, 8> BlockScope;
152 BitstreamCursor() : BitStream(0), NextChar(0) {
154 BitstreamCursor(const BitstreamCursor &RHS) : BitStream(0), NextChar(0) {
158 explicit BitstreamCursor(BitstreamReader &R) : BitStream(&R) {
159 NextChar = R.getFirstChar();
160 assert(NextChar && "Bitstream not initialized yet");
166 void init(BitstreamReader &R) {
170 NextChar = R.getFirstChar();
171 assert(NextChar && "Bitstream not initialized yet");
181 void operator=(const BitstreamCursor &RHS) {
184 BitStream = RHS.BitStream;
185 NextChar = RHS.NextChar;
186 CurWord = RHS.CurWord;
187 BitsInCurWord = RHS.BitsInCurWord;
188 CurCodeSize = RHS.CurCodeSize;
190 // Copy abbreviations, and bump ref counts.
191 CurAbbrevs = RHS.CurAbbrevs;
192 for (unsigned i = 0, e = static_cast<unsigned>(CurAbbrevs.size());
194 CurAbbrevs[i]->addRef();
196 // Copy block scope and bump ref counts.
197 for (unsigned S = 0, e = static_cast<unsigned>(BlockScope.size());
199 std::vector<BitCodeAbbrev*> &Abbrevs = BlockScope[S].PrevAbbrevs;
200 for (unsigned i = 0, e = static_cast<unsigned>(Abbrevs.size());
202 Abbrevs[i]->addRef();
207 // Free all the Abbrevs.
208 for (unsigned i = 0, e = static_cast<unsigned>(CurAbbrevs.size());
210 CurAbbrevs[i]->dropRef();
213 // Free all the Abbrevs in the block scope.
214 for (unsigned S = 0, e = static_cast<unsigned>(BlockScope.size());
216 std::vector<BitCodeAbbrev*> &Abbrevs = BlockScope[S].PrevAbbrevs;
217 for (unsigned i = 0, e = static_cast<unsigned>(Abbrevs.size());
219 Abbrevs[i]->dropRef();
224 /// GetAbbrevIDWidth - Return the number of bits used to encode an abbrev #.
225 unsigned GetAbbrevIDWidth() const { return CurCodeSize; }
227 bool AtEndOfStream() const {
228 return NextChar == BitStream->getLastChar() && BitsInCurWord == 0;
231 /// GetCurrentBitNo - Return the bit # of the bit we are reading.
232 uint64_t GetCurrentBitNo() const {
233 return (NextChar-BitStream->getFirstChar())*CHAR_BIT - BitsInCurWord;
236 BitstreamReader *getBitStreamReader() {
239 const BitstreamReader *getBitStreamReader() const {
244 /// JumpToBit - Reset the stream to the specified bit number.
245 void JumpToBit(uint64_t BitNo) {
246 uintptr_t ByteNo = uintptr_t(BitNo/8) & ~3;
247 uintptr_t WordBitNo = uintptr_t(BitNo) & 31;
248 assert(ByteNo <= (uintptr_t)(BitStream->getLastChar()-
249 BitStream->getFirstChar()) &&
252 // Move the cursor to the right word.
253 NextChar = BitStream->getFirstChar()+ByteNo;
257 // Skip over any bits that are already consumed.
259 Read(static_cast<unsigned>(WordBitNo));
263 uint32_t Read(unsigned NumBits) {
264 assert(NumBits <= 32 && "Cannot return more than 32 bits!");
265 // If the field is fully contained by CurWord, return it quickly.
266 if (BitsInCurWord >= NumBits) {
267 uint32_t R = CurWord & ((1U << NumBits)-1);
269 BitsInCurWord -= NumBits;
273 // If we run out of data, stop at the end of the stream.
274 if (NextChar == BitStream->getLastChar()) {
280 unsigned R = CurWord;
282 // Read the next word from the stream.
283 CurWord = (NextChar[0] << 0) | (NextChar[1] << 8) |
284 (NextChar[2] << 16) | (NextChar[3] << 24);
287 // Extract NumBits-BitsInCurWord from what we just read.
288 unsigned BitsLeft = NumBits-BitsInCurWord;
290 // Be careful here, BitsLeft is in the range [1..32] inclusive.
291 R |= (CurWord & (~0U >> (32-BitsLeft))) << BitsInCurWord;
293 // BitsLeft bits have just been used up from CurWord.
295 CurWord >>= BitsLeft;
298 BitsInCurWord = 32-BitsLeft;
302 uint64_t Read64(unsigned NumBits) {
303 if (NumBits <= 32) return Read(NumBits);
305 uint64_t V = Read(32);
306 return V | (uint64_t)Read(NumBits-32) << 32;
309 uint32_t ReadVBR(unsigned NumBits) {
310 uint32_t Piece = Read(NumBits);
311 if ((Piece & (1U << (NumBits-1))) == 0)
315 unsigned NextBit = 0;
317 Result |= (Piece & ((1U << (NumBits-1))-1)) << NextBit;
319 if ((Piece & (1U << (NumBits-1))) == 0)
322 NextBit += NumBits-1;
323 Piece = Read(NumBits);
327 // ReadVBR64 - Read a VBR that may have a value up to 64-bits in size. The
328 // chunk size of the VBR must still be <= 32 bits though.
329 uint64_t ReadVBR64(unsigned NumBits) {
330 uint32_t Piece = Read(NumBits);
331 if ((Piece & (1U << (NumBits-1))) == 0)
332 return uint64_t(Piece);
335 unsigned NextBit = 0;
337 Result |= uint64_t(Piece & ((1U << (NumBits-1))-1)) << NextBit;
339 if ((Piece & (1U << (NumBits-1))) == 0)
342 NextBit += NumBits-1;
343 Piece = Read(NumBits);
352 unsigned ReadCode() {
353 return Read(CurCodeSize);
358 // [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen]
360 /// ReadSubBlockID - Having read the ENTER_SUBBLOCK code, read the BlockID for
362 unsigned ReadSubBlockID() {
363 return ReadVBR(bitc::BlockIDWidth);
366 /// SkipBlock - Having read the ENTER_SUBBLOCK abbrevid and a BlockID, skip
367 /// over the body of this block. If the block record is malformed, return
370 // Read and ignore the codelen value. Since we are skipping this block, we
371 // don't care what code widths are used inside of it.
372 ReadVBR(bitc::CodeLenWidth);
374 unsigned NumWords = Read(bitc::BlockSizeWidth);
376 // Check that the block wasn't partially defined, and that the offset isn't
378 if (AtEndOfStream() || NextChar+NumWords*4 > BitStream->getLastChar())
381 NextChar += NumWords*4;
385 /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter
386 /// the block, and return true if the block is valid.
387 bool EnterSubBlock(unsigned BlockID, unsigned *NumWordsP = 0) {
388 // Save the current block's state on BlockScope.
389 BlockScope.push_back(Block(CurCodeSize));
390 BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
392 // Add the abbrevs specific to this block to the CurAbbrevs list.
393 if (const BitstreamReader::BlockInfo *Info =
394 BitStream->getBlockInfo(BlockID)) {
395 for (unsigned i = 0, e = static_cast<unsigned>(Info->Abbrevs.size());
397 CurAbbrevs.push_back(Info->Abbrevs[i]);
398 CurAbbrevs.back()->addRef();
402 // Get the codesize of this block.
403 CurCodeSize = ReadVBR(bitc::CodeLenWidth);
405 unsigned NumWords = Read(bitc::BlockSizeWidth);
406 if (NumWordsP) *NumWordsP = NumWords;
408 // Validate that this block is sane.
409 if (CurCodeSize == 0 || AtEndOfStream() ||
410 NextChar+NumWords*4 > BitStream->getLastChar())
416 bool ReadBlockEnd() {
417 if (BlockScope.empty()) return true;
420 // [END_BLOCK, <align4bytes>]
428 void PopBlockScope() {
429 CurCodeSize = BlockScope.back().PrevCodeSize;
431 // Delete abbrevs from popped scope.
432 for (unsigned i = 0, e = static_cast<unsigned>(CurAbbrevs.size());
434 CurAbbrevs[i]->dropRef();
436 BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
437 BlockScope.pop_back();
440 //===--------------------------------------------------------------------===//
442 //===--------------------------------------------------------------------===//
445 void ReadAbbreviatedLiteral(const BitCodeAbbrevOp &Op,
446 SmallVectorImpl<uint64_t> &Vals) {
447 assert(Op.isLiteral() && "Not a literal");
448 // If the abbrev specifies the literal value to use, use it.
449 Vals.push_back(Op.getLiteralValue());
452 void ReadAbbreviatedField(const BitCodeAbbrevOp &Op,
453 SmallVectorImpl<uint64_t> &Vals) {
454 assert(!Op.isLiteral() && "Use ReadAbbreviatedLiteral for literals!");
456 // Decode the value as we are commanded.
457 switch (Op.getEncoding()) {
458 default: assert(0 && "Unknown encoding!");
459 case BitCodeAbbrevOp::Fixed:
460 Vals.push_back(Read((unsigned)Op.getEncodingData()));
462 case BitCodeAbbrevOp::VBR:
463 Vals.push_back(ReadVBR64((unsigned)Op.getEncodingData()));
465 case BitCodeAbbrevOp::Char6:
466 Vals.push_back(BitCodeAbbrevOp::DecodeChar6(Read(6)));
472 /// getAbbrev - Return the abbreviation for the specified AbbrevId.
473 const BitCodeAbbrev *getAbbrev(unsigned AbbrevID) {
474 unsigned AbbrevNo = AbbrevID-bitc::FIRST_APPLICATION_ABBREV;
475 assert(AbbrevNo < CurAbbrevs.size() && "Invalid abbrev #!");
476 return CurAbbrevs[AbbrevNo];
479 unsigned ReadRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
480 const char **BlobStart = 0, unsigned *BlobLen = 0) {
481 if (AbbrevID == bitc::UNABBREV_RECORD) {
482 unsigned Code = ReadVBR(6);
483 unsigned NumElts = ReadVBR(6);
484 for (unsigned i = 0; i != NumElts; ++i)
485 Vals.push_back(ReadVBR64(6));
489 const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
491 for (unsigned i = 0, e = Abbv->getNumOperandInfos(); i != e; ++i) {
492 const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
493 if (Op.isLiteral()) {
494 ReadAbbreviatedLiteral(Op, Vals);
495 } else if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
496 // Array case. Read the number of elements as a vbr6.
497 unsigned NumElts = ReadVBR(6);
499 // Get the element encoding.
500 assert(i+2 == e && "array op not second to last?");
501 const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
503 // Read all the elements.
504 for (; NumElts; --NumElts)
505 ReadAbbreviatedField(EltEnc, Vals);
506 } else if (Op.getEncoding() == BitCodeAbbrevOp::Blob) {
507 // Blob case. Read the number of bytes as a vbr6.
508 unsigned NumElts = ReadVBR(6);
509 SkipToWord(); // 32-bit alignment
511 // Figure out where the end of this blob will be including tail padding.
512 const unsigned char *NewEnd = NextChar+((NumElts+3)&~3);
514 // If this would read off the end of the bitcode file, just set the
515 // record to empty and return.
516 if (NewEnd > BitStream->getLastChar()) {
517 Vals.append(NumElts, 0);
518 NextChar = BitStream->getLastChar();
522 // Otherwise, read the number of bytes. If we can return a reference to
523 // the data, do so to avoid copying it.
525 *BlobStart = (const char*)NextChar;
528 for (; NumElts; ++NextChar, --NumElts)
529 Vals.push_back(*NextChar);
531 // Skip over tail padding.
534 ReadAbbreviatedField(Op, Vals);
538 unsigned Code = (unsigned)Vals[0];
539 Vals.erase(Vals.begin());
543 unsigned ReadRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
544 const char *&BlobStart, unsigned &BlobLen) {
545 return ReadRecord(AbbrevID, Vals, &BlobStart, &BlobLen);
549 //===--------------------------------------------------------------------===//
551 //===--------------------------------------------------------------------===//
553 void ReadAbbrevRecord() {
554 BitCodeAbbrev *Abbv = new BitCodeAbbrev();
555 unsigned NumOpInfo = ReadVBR(5);
556 for (unsigned i = 0; i != NumOpInfo; ++i) {
557 bool IsLiteral = Read(1) ? true : false;
559 Abbv->Add(BitCodeAbbrevOp(ReadVBR64(8)));
563 BitCodeAbbrevOp::Encoding E = (BitCodeAbbrevOp::Encoding)Read(3);
564 if (BitCodeAbbrevOp::hasEncodingData(E))
565 Abbv->Add(BitCodeAbbrevOp(E, ReadVBR64(5)));
567 Abbv->Add(BitCodeAbbrevOp(E));
569 CurAbbrevs.push_back(Abbv);
574 bool ReadBlockInfoBlock() {
575 // If this is the second stream to get to the block info block, skip it.
576 if (BitStream->hasBlockInfoRecords())
579 if (EnterSubBlock(bitc::BLOCKINFO_BLOCK_ID)) return true;
581 SmallVector<uint64_t, 64> Record;
582 BitstreamReader::BlockInfo *CurBlockInfo = 0;
584 // Read all the records for this module.
586 unsigned Code = ReadCode();
587 if (Code == bitc::END_BLOCK)
588 return ReadBlockEnd();
589 if (Code == bitc::ENTER_SUBBLOCK) {
591 if (SkipBlock()) return true;
595 // Read abbrev records, associate them with CurBID.
596 if (Code == bitc::DEFINE_ABBREV) {
597 if (!CurBlockInfo) return true;
600 // ReadAbbrevRecord installs the abbrev in CurAbbrevs. Move it to the
601 // appropriate BlockInfo.
602 BitCodeAbbrev *Abbv = CurAbbrevs.back();
603 CurAbbrevs.pop_back();
604 CurBlockInfo->Abbrevs.push_back(Abbv);
610 switch (ReadRecord(Code, Record)) {
611 default: break; // Default behavior, ignore unknown content.
612 case bitc::BLOCKINFO_CODE_SETBID:
613 if (Record.size() < 1) return true;
614 CurBlockInfo = &BitStream->getOrCreateBlockInfo((unsigned)Record[0]);
616 case bitc::BLOCKINFO_CODE_BLOCKNAME: {
617 if (!CurBlockInfo) return true;
618 if (BitStream->isIgnoringBlockInfoNames()) break; // Ignore name.
620 for (unsigned i = 0, e = Record.size(); i != e; ++i)
621 Name += (char)Record[i];
622 CurBlockInfo->Name = Name;
625 case bitc::BLOCKINFO_CODE_SETRECORDNAME: {
626 if (!CurBlockInfo) return true;
627 if (BitStream->isIgnoringBlockInfoNames()) break; // Ignore name.
629 for (unsigned i = 1, e = Record.size(); i != e; ++i)
630 Name += (char)Record[i];
631 CurBlockInfo->RecordNames.push_back(std::make_pair((unsigned)Record[0],
640 } // End llvm namespace