Home | History | Annotate | Download | only in Reader
      1 //===- BitstreamReader.cpp - BitstreamReader implementation ---------------===//
      2 //
      3 //                     The LLVM Compiler Infrastructure
      4 //
      5 // This file is distributed under the University of Illinois Open Source
      6 // License. See LICENSE.TXT for details.
      7 //
      8 //===----------------------------------------------------------------------===//
      9 
     10 #include "llvm/Bitcode/BitstreamReader.h"
     11 
     12 using namespace llvm;
     13 
     14 //===----------------------------------------------------------------------===//
     15 //  BitstreamCursor implementation
     16 //===----------------------------------------------------------------------===//
     17 
     18 void BitstreamCursor::operator=(const BitstreamCursor &RHS) {
     19   freeState();
     20 
     21   BitStream = RHS.BitStream;
     22   NextChar = RHS.NextChar;
     23   CurWord = RHS.CurWord;
     24   BitsInCurWord = RHS.BitsInCurWord;
     25   CurCodeSize = RHS.CurCodeSize;
     26 
     27   // Copy abbreviations, and bump ref counts.
     28   CurAbbrevs = RHS.CurAbbrevs;
     29   for (size_t i = 0, e = CurAbbrevs.size(); i != e; ++i)
     30     CurAbbrevs[i]->addRef();
     31 
     32   // Copy block scope and bump ref counts.
     33   BlockScope = RHS.BlockScope;
     34   for (size_t S = 0, e = BlockScope.size(); S != e; ++S) {
     35     std::vector<BitCodeAbbrev*> &Abbrevs = BlockScope[S].PrevAbbrevs;
     36     for (size_t i = 0, e = Abbrevs.size(); i != e; ++i)
     37       Abbrevs[i]->addRef();
     38   }
     39 }
     40 
     41 void BitstreamCursor::freeState() {
     42   // Free all the Abbrevs.
     43   for (size_t i = 0, e = CurAbbrevs.size(); i != e; ++i)
     44     CurAbbrevs[i]->dropRef();
     45   CurAbbrevs.clear();
     46 
     47   // Free all the Abbrevs in the block scope.
     48   for (size_t S = 0, e = BlockScope.size(); S != e; ++S) {
     49     std::vector<BitCodeAbbrev*> &Abbrevs = BlockScope[S].PrevAbbrevs;
     50     for (size_t i = 0, e = Abbrevs.size(); i != e; ++i)
     51       Abbrevs[i]->dropRef();
     52   }
     53   BlockScope.clear();
     54 }
     55 
     56 /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter
     57 /// the block, and return true if the block has an error.
     58 bool BitstreamCursor::EnterSubBlock(unsigned BlockID, unsigned *NumWordsP) {
     59   // Save the current block's state on BlockScope.
     60   BlockScope.push_back(Block(CurCodeSize));
     61   BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
     62 
     63   // Add the abbrevs specific to this block to the CurAbbrevs list.
     64   if (const BitstreamReader::BlockInfo *Info =
     65       BitStream->getBlockInfo(BlockID)) {
     66     for (size_t i = 0, e = Info->Abbrevs.size(); i != e; ++i) {
     67       CurAbbrevs.push_back(Info->Abbrevs[i]);
     68       CurAbbrevs.back()->addRef();
     69     }
     70   }
     71 
     72   // Get the codesize of this block.
     73   CurCodeSize = ReadVBR(bitc::CodeLenWidth);
     74   SkipToFourByteBoundary();
     75   unsigned NumWords = Read(bitc::BlockSizeWidth);
     76   if (NumWordsP) *NumWordsP = NumWords;
     77 
     78   // Validate that this block is sane.
     79   if (CurCodeSize == 0 || AtEndOfStream())
     80     return true;
     81 
     82   return false;
     83 }
     84 
     85 void BitstreamCursor::readAbbreviatedLiteral(const BitCodeAbbrevOp &Op,
     86                                              SmallVectorImpl<uint64_t> &Vals) {
     87   assert(Op.isLiteral() && "Not a literal");
     88   // If the abbrev specifies the literal value to use, use it.
     89   Vals.push_back(Op.getLiteralValue());
     90 }
     91 
     92 void BitstreamCursor::readAbbreviatedField(const BitCodeAbbrevOp &Op,
     93                                            SmallVectorImpl<uint64_t> &Vals) {
     94   assert(!Op.isLiteral() && "Use ReadAbbreviatedLiteral for literals!");
     95 
     96   // Decode the value as we are commanded.
     97   switch (Op.getEncoding()) {
     98   case BitCodeAbbrevOp::Array:
     99   case BitCodeAbbrevOp::Blob:
    100     assert(0 && "Should not reach here");
    101   case BitCodeAbbrevOp::Fixed:
    102     Vals.push_back(Read((unsigned)Op.getEncodingData()));
    103     break;
    104   case BitCodeAbbrevOp::VBR:
    105     Vals.push_back(ReadVBR64((unsigned)Op.getEncodingData()));
    106     break;
    107   case BitCodeAbbrevOp::Char6:
    108     Vals.push_back(BitCodeAbbrevOp::DecodeChar6(Read(6)));
    109     break;
    110   }
    111 }
    112 
    113 void BitstreamCursor::skipAbbreviatedField(const BitCodeAbbrevOp &Op) {
    114   assert(!Op.isLiteral() && "Use ReadAbbreviatedLiteral for literals!");
    115 
    116   // Decode the value as we are commanded.
    117   switch (Op.getEncoding()) {
    118   case BitCodeAbbrevOp::Array:
    119   case BitCodeAbbrevOp::Blob:
    120     assert(0 && "Should not reach here");
    121   case BitCodeAbbrevOp::Fixed:
    122     (void)Read((unsigned)Op.getEncodingData());
    123     break;
    124   case BitCodeAbbrevOp::VBR:
    125     (void)ReadVBR64((unsigned)Op.getEncodingData());
    126     break;
    127   case BitCodeAbbrevOp::Char6:
    128     (void)Read(6);
    129     break;
    130   }
    131 }
    132 
    133 
    134 
    135 /// skipRecord - Read the current record and discard it.
    136 void BitstreamCursor::skipRecord(unsigned AbbrevID) {
    137   // Skip unabbreviated records by reading past their entries.
    138   if (AbbrevID == bitc::UNABBREV_RECORD) {
    139     unsigned Code = ReadVBR(6);
    140     (void)Code;
    141     unsigned NumElts = ReadVBR(6);
    142     for (unsigned i = 0; i != NumElts; ++i)
    143       (void)ReadVBR64(6);
    144     return;
    145   }
    146 
    147   const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
    148 
    149   for (unsigned i = 0, e = Abbv->getNumOperandInfos(); i != e; ++i) {
    150     const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
    151     if (Op.isLiteral())
    152       continue;
    153 
    154     if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
    155         Op.getEncoding() != BitCodeAbbrevOp::Blob) {
    156       skipAbbreviatedField(Op);
    157       continue;
    158     }
    159 
    160     if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
    161       // Array case.  Read the number of elements as a vbr6.
    162       unsigned NumElts = ReadVBR(6);
    163 
    164       // Get the element encoding.
    165       assert(i+2 == e && "array op not second to last?");
    166       const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
    167 
    168       // Read all the elements.
    169       for (; NumElts; --NumElts)
    170         skipAbbreviatedField(EltEnc);
    171       continue;
    172     }
    173 
    174     assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
    175     // Blob case.  Read the number of bytes as a vbr6.
    176     unsigned NumElts = ReadVBR(6);
    177     SkipToFourByteBoundary();  // 32-bit alignment
    178 
    179     // Figure out where the end of this blob will be including tail padding.
    180     size_t NewEnd = GetCurrentBitNo()+((NumElts+3)&~3)*8;
    181 
    182     // If this would read off the end of the bitcode file, just set the
    183     // record to empty and return.
    184     if (!canSkipToPos(NewEnd/8)) {
    185       NextChar = BitStream->getBitcodeBytes().getExtent();
    186       break;
    187     }
    188 
    189     // Skip over the blob.
    190     JumpToBit(NewEnd);
    191   }
    192 }
    193 
    194 unsigned BitstreamCursor::readRecord(unsigned AbbrevID,
    195                                      SmallVectorImpl<uint64_t> &Vals,
    196                                      StringRef *Blob) {
    197   if (AbbrevID == bitc::UNABBREV_RECORD) {
    198     unsigned Code = ReadVBR(6);
    199     unsigned NumElts = ReadVBR(6);
    200     for (unsigned i = 0; i != NumElts; ++i)
    201       Vals.push_back(ReadVBR64(6));
    202     return Code;
    203   }
    204 
    205   const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
    206 
    207   // Read the record code first.
    208   assert(Abbv->getNumOperandInfos() != 0 && "no record code in abbreviation?");
    209   const BitCodeAbbrevOp &CodeOp = Abbv->getOperandInfo(0);
    210   if (CodeOp.isLiteral())
    211     readAbbreviatedLiteral(CodeOp, Vals);
    212   else
    213     readAbbreviatedField(CodeOp, Vals);
    214   unsigned Code = (unsigned)Vals.pop_back_val();
    215 
    216   for (unsigned i = 1, e = Abbv->getNumOperandInfos(); i != e; ++i) {
    217     const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
    218     if (Op.isLiteral()) {
    219       readAbbreviatedLiteral(Op, Vals);
    220       continue;
    221     }
    222 
    223     if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
    224         Op.getEncoding() != BitCodeAbbrevOp::Blob) {
    225       readAbbreviatedField(Op, Vals);
    226       continue;
    227     }
    228 
    229     if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
    230       // Array case.  Read the number of elements as a vbr6.
    231       unsigned NumElts = ReadVBR(6);
    232 
    233       // Get the element encoding.
    234       assert(i+2 == e && "array op not second to last?");
    235       const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
    236 
    237       // Read all the elements.
    238       for (; NumElts; --NumElts)
    239         readAbbreviatedField(EltEnc, Vals);
    240       continue;
    241     }
    242 
    243     assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
    244     // Blob case.  Read the number of bytes as a vbr6.
    245     unsigned NumElts = ReadVBR(6);
    246     SkipToFourByteBoundary();  // 32-bit alignment
    247 
    248     // Figure out where the end of this blob will be including tail padding.
    249     size_t CurBitPos = GetCurrentBitNo();
    250     size_t NewEnd = CurBitPos+((NumElts+3)&~3)*8;
    251 
    252     // If this would read off the end of the bitcode file, just set the
    253     // record to empty and return.
    254     if (!canSkipToPos(NewEnd/8)) {
    255       Vals.append(NumElts, 0);
    256       NextChar = BitStream->getBitcodeBytes().getExtent();
    257       break;
    258     }
    259 
    260     // Otherwise, inform the streamer that we need these bytes in memory.
    261     const char *Ptr = (const char*)
    262       BitStream->getBitcodeBytes().getPointer(CurBitPos/8, NumElts);
    263 
    264     // If we can return a reference to the data, do so to avoid copying it.
    265     if (Blob) {
    266       *Blob = StringRef(Ptr, NumElts);
    267     } else {
    268       // Otherwise, unpack into Vals with zero extension.
    269       for (; NumElts; --NumElts)
    270         Vals.push_back((unsigned char)*Ptr++);
    271     }
    272     // Skip over tail padding.
    273     JumpToBit(NewEnd);
    274   }
    275 
    276   return Code;
    277 }
    278 
    279 
    280 void BitstreamCursor::ReadAbbrevRecord() {
    281   BitCodeAbbrev *Abbv = new BitCodeAbbrev();
    282   unsigned NumOpInfo = ReadVBR(5);
    283   for (unsigned i = 0; i != NumOpInfo; ++i) {
    284     bool IsLiteral = Read(1) ? true : false;
    285     if (IsLiteral) {
    286       Abbv->Add(BitCodeAbbrevOp(ReadVBR64(8)));
    287       continue;
    288     }
    289 
    290     BitCodeAbbrevOp::Encoding E = (BitCodeAbbrevOp::Encoding)Read(3);
    291     if (BitCodeAbbrevOp::hasEncodingData(E)) {
    292       unsigned Data = ReadVBR64(5);
    293 
    294       // As a special case, handle fixed(0) (i.e., a fixed field with zero bits)
    295       // and vbr(0) as a literal zero.  This is decoded the same way, and avoids
    296       // a slow path in Read() to have to handle reading zero bits.
    297       if ((E == BitCodeAbbrevOp::Fixed || E == BitCodeAbbrevOp::VBR) &&
    298           Data == 0) {
    299         Abbv->Add(BitCodeAbbrevOp(0));
    300         continue;
    301       }
    302 
    303       Abbv->Add(BitCodeAbbrevOp(E, Data));
    304     } else
    305       Abbv->Add(BitCodeAbbrevOp(E));
    306   }
    307   CurAbbrevs.push_back(Abbv);
    308 }
    309 
    310 bool BitstreamCursor::ReadBlockInfoBlock() {
    311   // If this is the second stream to get to the block info block, skip it.
    312   if (BitStream->hasBlockInfoRecords())
    313     return SkipBlock();
    314 
    315   if (EnterSubBlock(bitc::BLOCKINFO_BLOCK_ID)) return true;
    316 
    317   SmallVector<uint64_t, 64> Record;
    318   BitstreamReader::BlockInfo *CurBlockInfo = 0;
    319 
    320   // Read all the records for this module.
    321   while (1) {
    322     BitstreamEntry Entry = advanceSkippingSubblocks(AF_DontAutoprocessAbbrevs);
    323 
    324     switch (Entry.Kind) {
    325     case llvm::BitstreamEntry::SubBlock: // Handled for us already.
    326     case llvm::BitstreamEntry::Error:
    327       return true;
    328     case llvm::BitstreamEntry::EndBlock:
    329       return false;
    330     case llvm::BitstreamEntry::Record:
    331       // The interesting case.
    332       break;
    333     }
    334 
    335     // Read abbrev records, associate them with CurBID.
    336     if (Entry.ID == bitc::DEFINE_ABBREV) {
    337       if (!CurBlockInfo) return true;
    338       ReadAbbrevRecord();
    339 
    340       // ReadAbbrevRecord installs the abbrev in CurAbbrevs.  Move it to the
    341       // appropriate BlockInfo.
    342       BitCodeAbbrev *Abbv = CurAbbrevs.back();
    343       CurAbbrevs.pop_back();
    344       CurBlockInfo->Abbrevs.push_back(Abbv);
    345       continue;
    346     }
    347 
    348     // Read a record.
    349     Record.clear();
    350     switch (readRecord(Entry.ID, Record)) {
    351       default: break;  // Default behavior, ignore unknown content.
    352       case bitc::BLOCKINFO_CODE_SETBID:
    353         if (Record.size() < 1) return true;
    354         CurBlockInfo = &BitStream->getOrCreateBlockInfo((unsigned)Record[0]);
    355         break;
    356       case bitc::BLOCKINFO_CODE_BLOCKNAME: {
    357         if (!CurBlockInfo) return true;
    358         if (BitStream->isIgnoringBlockInfoNames()) break;  // Ignore name.
    359         std::string Name;
    360         for (unsigned i = 0, e = Record.size(); i != e; ++i)
    361           Name += (char)Record[i];
    362         CurBlockInfo->Name = Name;
    363         break;
    364       }
    365       case bitc::BLOCKINFO_CODE_SETRECORDNAME: {
    366         if (!CurBlockInfo) return true;
    367         if (BitStream->isIgnoringBlockInfoNames()) break;  // Ignore name.
    368         std::string Name;
    369         for (unsigned i = 1, e = Record.size(); i != e; ++i)
    370           Name += (char)Record[i];
    371         CurBlockInfo->RecordNames.push_back(std::make_pair((unsigned)Record[0],
    372                                                            Name));
    373         break;
    374       }
    375     }
    376   }
    377 }
    378 
    379