Merging r261033:
[oota-llvm.git] / lib / Bitcode / Reader / BitstreamReader.cpp
index be70f5213bc844af0acc8db9a84963053aefa4b5..a103fbdf4a93a317710d171058b9788fb1f46148 100644 (file)
@@ -15,47 +15,11 @@ using namespace llvm;
 //  BitstreamCursor implementation
 //===----------------------------------------------------------------------===//
 
-void BitstreamCursor::operator=(const BitstreamCursor &RHS) {
-  freeState();
-  
-  BitStream = RHS.BitStream;
-  NextChar = RHS.NextChar;
-  CurWord = RHS.CurWord;
-  BitsInCurWord = RHS.BitsInCurWord;
-  CurCodeSize = RHS.CurCodeSize;
-  
-  // Copy abbreviations, and bump ref counts.
-  CurAbbrevs = RHS.CurAbbrevs;
-  for (unsigned i = 0, e = static_cast<unsigned>(CurAbbrevs.size());
-       i != e; ++i)
-    CurAbbrevs[i]->addRef();
-  
-  // Copy block scope and bump ref counts.
-  BlockScope = RHS.BlockScope;
-  for (unsigned S = 0, e = static_cast<unsigned>(BlockScope.size());
-       S != e; ++S) {
-    std::vector<BitCodeAbbrev*> &Abbrevs = BlockScope[S].PrevAbbrevs;
-    for (unsigned i = 0, e = static_cast<unsigned>(Abbrevs.size());
-         i != e; ++i)
-      Abbrevs[i]->addRef();
-  }
-}
-
 void BitstreamCursor::freeState() {
   // Free all the Abbrevs.
-  for (unsigned i = 0, e = static_cast<unsigned>(CurAbbrevs.size());
-       i != e; ++i)
-    CurAbbrevs[i]->dropRef();
   CurAbbrevs.clear();
-  
+
   // Free all the Abbrevs in the block scope.
-  for (unsigned S = 0, e = static_cast<unsigned>(BlockScope.size());
-       S != e; ++S) {
-    std::vector<BitCodeAbbrev*> &Abbrevs = BlockScope[S].PrevAbbrevs;
-    for (unsigned i = 0, e = static_cast<unsigned>(Abbrevs.size());
-         i != e; ++i)
-      Abbrevs[i]->dropRef();
-  }
   BlockScope.clear();
 }
 
@@ -65,74 +29,68 @@ bool BitstreamCursor::EnterSubBlock(unsigned BlockID, unsigned *NumWordsP) {
   // Save the current block's state on BlockScope.
   BlockScope.push_back(Block(CurCodeSize));
   BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
-  
+
   // Add the abbrevs specific to this block to the CurAbbrevs list.
   if (const BitstreamReader::BlockInfo *Info =
       BitStream->getBlockInfo(BlockID)) {
-    for (unsigned i = 0, e = static_cast<unsigned>(Info->Abbrevs.size());
-         i != e; ++i) {
-      CurAbbrevs.push_back(Info->Abbrevs[i]);
-      CurAbbrevs.back()->addRef();
-    }
+    CurAbbrevs.insert(CurAbbrevs.end(), Info->Abbrevs.begin(),
+                      Info->Abbrevs.end());
   }
-  
+
   // Get the codesize of this block.
   CurCodeSize = ReadVBR(bitc::CodeLenWidth);
-  SkipToWord();
+  // We can't read more than MaxChunkSize at a time
+  if (CurCodeSize > MaxChunkSize)
+    return true;
+
+  SkipToFourByteBoundary();
   unsigned NumWords = Read(bitc::BlockSizeWidth);
   if (NumWordsP) *NumWordsP = NumWords;
-  
+
   // Validate that this block is sane.
-  if (CurCodeSize == 0 || AtEndOfStream())
-    return true;
-  
-  return false;
+  return CurCodeSize == 0 || AtEndOfStream();
 }
 
-void BitstreamCursor::readAbbreviatedLiteral(const BitCodeAbbrevOp &Op,
-                                             SmallVectorImpl<uint64_t> &Vals) {
-  assert(Op.isLiteral() && "Not a literal");
-  // If the abbrev specifies the literal value to use, use it.
-  Vals.push_back(Op.getLiteralValue());
-}
+static uint64_t readAbbreviatedField(BitstreamCursor &Cursor,
+                                     const BitCodeAbbrevOp &Op) {
+  assert(!Op.isLiteral() && "Not to be used with literals!");
 
-void BitstreamCursor::readAbbreviatedField(const BitCodeAbbrevOp &Op,
-                                           SmallVectorImpl<uint64_t> &Vals) {
-  assert(!Op.isLiteral() && "Use ReadAbbreviatedLiteral for literals!");
-  
   // Decode the value as we are commanded.
   switch (Op.getEncoding()) {
   case BitCodeAbbrevOp::Array:
   case BitCodeAbbrevOp::Blob:
-    assert(0 && "Should not reach here");
+    llvm_unreachable("Should not reach here");
   case BitCodeAbbrevOp::Fixed:
-    Vals.push_back(Read((unsigned)Op.getEncodingData()));
-    break;
+    assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
+    return Cursor.Read((unsigned)Op.getEncodingData());
   case BitCodeAbbrevOp::VBR:
-    Vals.push_back(ReadVBR64((unsigned)Op.getEncodingData()));
-    break;
+    assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
+    return Cursor.ReadVBR64((unsigned)Op.getEncodingData());
   case BitCodeAbbrevOp::Char6:
-    Vals.push_back(BitCodeAbbrevOp::DecodeChar6(Read(6)));
-    break;
+    return BitCodeAbbrevOp::DecodeChar6(Cursor.Read(6));
   }
+  llvm_unreachable("invalid abbreviation encoding");
 }
 
-void BitstreamCursor::skipAbbreviatedField(const BitCodeAbbrevOp &Op) {
-  assert(!Op.isLiteral() && "Use ReadAbbreviatedLiteral for literals!");
-  
+static void skipAbbreviatedField(BitstreamCursor &Cursor,
+                                 const BitCodeAbbrevOp &Op) {
+  assert(!Op.isLiteral() && "Not to be used with literals!");
+
   // Decode the value as we are commanded.
   switch (Op.getEncoding()) {
   case BitCodeAbbrevOp::Array:
   case BitCodeAbbrevOp::Blob:
-    assert(0 && "Should not reach here");
+    llvm_unreachable("Should not reach here");
   case BitCodeAbbrevOp::Fixed:
-    (void)Read((unsigned)Op.getEncodingData());
+    assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
+    Cursor.Read((unsigned)Op.getEncodingData());
     break;
   case BitCodeAbbrevOp::VBR:
-    (void)ReadVBR64((unsigned)Op.getEncodingData());
+    assert((unsigned)Op.getEncodingData() <= Cursor.MaxChunkSize);
+    Cursor.ReadVBR64((unsigned)Op.getEncodingData());
     break;
   case BitCodeAbbrevOp::Char6:
-    (void)Read(6);
+    Cursor.Read(6);
     break;
   }
 }
@@ -152,55 +110,55 @@ void BitstreamCursor::skipRecord(unsigned AbbrevID) {
   }
 
   const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
-  
+
   for (unsigned i = 0, e = Abbv->getNumOperandInfos(); i != e; ++i) {
     const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
     if (Op.isLiteral())
       continue;
-    
+
     if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
         Op.getEncoding() != BitCodeAbbrevOp::Blob) {
-      skipAbbreviatedField(Op);
+      skipAbbreviatedField(*this, Op);
       continue;
     }
-    
+
     if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
       // Array case.  Read the number of elements as a vbr6.
       unsigned NumElts = ReadVBR(6);
-      
+
       // Get the element encoding.
       assert(i+2 == e && "array op not second to last?");
       const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
-      
+
       // Read all the elements.
       for (; NumElts; --NumElts)
-        skipAbbreviatedField(EltEnc);
+        skipAbbreviatedField(*this, EltEnc);
       continue;
     }
-    
+
     assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
     // Blob case.  Read the number of bytes as a vbr6.
     unsigned NumElts = ReadVBR(6);
-    SkipToWord();  // 32-bit alignment
-    
+    SkipToFourByteBoundary();  // 32-bit alignment
+
     // Figure out where the end of this blob will be including tail padding.
-    size_t NewEnd = NextChar+((NumElts+3)&~3);
-    
+    size_t NewEnd = GetCurrentBitNo()+((NumElts+3)&~3)*8;
+
     // If this would read off the end of the bitcode file, just set the
     // record to empty and return.
-    if (!canSkipToPos(NewEnd)) {
+    if (!canSkipToPos(NewEnd/8)) {
       NextChar = BitStream->getBitcodeBytes().getExtent();
       break;
     }
-    
+
     // Skip over the blob.
-    NextChar = NewEnd;
+    JumpToBit(NewEnd);
   }
 }
 
-unsigned BitstreamCursor::ReadRecord(unsigned AbbrevID,
+unsigned BitstreamCursor::readRecord(unsigned AbbrevID,
                                      SmallVectorImpl<uint64_t> &Vals,
-                                     const char **BlobStart, unsigned *BlobLen){
+                                     StringRef *Blob) {
   if (AbbrevID == bitc::UNABBREV_RECORD) {
     unsigned Code = ReadVBR(6);
     unsigned NumElts = ReadVBR(6);
@@ -208,68 +166,89 @@ unsigned BitstreamCursor::ReadRecord(unsigned AbbrevID,
       Vals.push_back(ReadVBR64(6));
     return Code;
   }
-  
+
   const BitCodeAbbrev *Abbv = getAbbrev(AbbrevID);
-  
-  for (unsigned i = 0, e = Abbv->getNumOperandInfos(); i != e; ++i) {
+
+  // Read the record code first.
+  assert(Abbv->getNumOperandInfos() != 0 && "no record code in abbreviation?");
+  const BitCodeAbbrevOp &CodeOp = Abbv->getOperandInfo(0);
+  unsigned Code;
+  if (CodeOp.isLiteral())
+    Code = CodeOp.getLiteralValue();
+  else {
+    if (CodeOp.getEncoding() == BitCodeAbbrevOp::Array ||
+        CodeOp.getEncoding() == BitCodeAbbrevOp::Blob)
+      report_fatal_error("Abbreviation starts with an Array or a Blob");
+    Code = readAbbreviatedField(*this, CodeOp);
+  }
+
+  for (unsigned i = 1, e = Abbv->getNumOperandInfos(); i != e; ++i) {
     const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
     if (Op.isLiteral()) {
-      readAbbreviatedLiteral(Op, Vals);
+      Vals.push_back(Op.getLiteralValue());
       continue;
     }
-    
+
     if (Op.getEncoding() != BitCodeAbbrevOp::Array &&
         Op.getEncoding() != BitCodeAbbrevOp::Blob) {
-      readAbbreviatedField(Op, Vals);
+      Vals.push_back(readAbbreviatedField(*this, Op));
       continue;
     }
-    
+
     if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
       // Array case.  Read the number of elements as a vbr6.
       unsigned NumElts = ReadVBR(6);
-      
+
       // Get the element encoding.
-      assert(i+2 == e && "array op not second to last?");
+      if (i + 2 != e)
+        report_fatal_error("Array op not second to last");
       const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
-      
+      if (!EltEnc.isEncoding())
+        report_fatal_error(
+            "Array element type has to be an encoding of a type");
+      if (EltEnc.getEncoding() == BitCodeAbbrevOp::Array ||
+          EltEnc.getEncoding() == BitCodeAbbrevOp::Blob)
+        report_fatal_error("Array element type can't be an Array or a Blob");
+
       // Read all the elements.
       for (; NumElts; --NumElts)
-        readAbbreviatedField(EltEnc, Vals);
+        Vals.push_back(readAbbreviatedField(*this, EltEnc));
       continue;
     }
-    
+
     assert(Op.getEncoding() == BitCodeAbbrevOp::Blob);
     // Blob case.  Read the number of bytes as a vbr6.
     unsigned NumElts = ReadVBR(6);
-    SkipToWord();  // 32-bit alignment
-    
+    SkipToFourByteBoundary();  // 32-bit alignment
+
     // Figure out where the end of this blob will be including tail padding.
-    size_t NewEnd = NextChar+((NumElts+3)&~3);
-    
+    size_t CurBitPos = GetCurrentBitNo();
+    size_t NewEnd = CurBitPos+((NumElts+3)&~3)*8;
+
     // If this would read off the end of the bitcode file, just set the
     // record to empty and return.
-    if (!canSkipToPos(NewEnd)) {
+    if (!canSkipToPos(NewEnd/8)) {
       Vals.append(NumElts, 0);
       NextChar = BitStream->getBitcodeBytes().getExtent();
       break;
     }
-    
-    // Otherwise, read the number of bytes.  If we can return a reference to
-    // the data, do so to avoid copying it.
-    if (BlobStart) {
-      *BlobStart = (const char*)BitStream->getBitcodeBytes().getPointer(
-                                                            NextChar, NumElts);
-      *BlobLen = NumElts;
+
+    // Otherwise, inform the streamer that we need these bytes in memory.
+    const char *Ptr = (const char*)
+      BitStream->getBitcodeBytes().getPointer(CurBitPos/8, NumElts);
+
+    // If we can return a reference to the data, do so to avoid copying it.
+    if (Blob) {
+      *Blob = StringRef(Ptr, NumElts);
     } else {
-      for (; NumElts; ++NextChar, --NumElts)
-        Vals.push_back(getByte(NextChar));
+      // Otherwise, unpack into Vals with zero extension.
+      for (; NumElts; --NumElts)
+        Vals.push_back((unsigned char)*Ptr++);
     }
     // Skip over tail padding.
-    NextChar = NewEnd;
+    JumpToBit(NewEnd);
   }
-  
-  unsigned Code = (unsigned)Vals[0];
-  Vals.erase(Vals.begin());
+
   return Code;
 }
 
@@ -278,18 +257,37 @@ void BitstreamCursor::ReadAbbrevRecord() {
   BitCodeAbbrev *Abbv = new BitCodeAbbrev();
   unsigned NumOpInfo = ReadVBR(5);
   for (unsigned i = 0; i != NumOpInfo; ++i) {
-    bool IsLiteral = Read(1) ? true : false;
+    bool IsLiteral = Read(1);
     if (IsLiteral) {
       Abbv->Add(BitCodeAbbrevOp(ReadVBR64(8)));
       continue;
     }
-    
+
     BitCodeAbbrevOp::Encoding E = (BitCodeAbbrevOp::Encoding)Read(3);
-    if (BitCodeAbbrevOp::hasEncodingData(E))
-      Abbv->Add(BitCodeAbbrevOp(E, ReadVBR64(5)));
-    else
+    if (BitCodeAbbrevOp::hasEncodingData(E)) {
+      uint64_t Data = ReadVBR64(5);
+
+      // As a special case, handle fixed(0) (i.e., a fixed field with zero bits)
+      // and vbr(0) as a literal zero.  This is decoded the same way, and avoids
+      // a slow path in Read() to have to handle reading zero bits.
+      if ((E == BitCodeAbbrevOp::Fixed || E == BitCodeAbbrevOp::VBR) &&
+          Data == 0) {
+        Abbv->Add(BitCodeAbbrevOp(0));
+        continue;
+      }
+
+      if ((E == BitCodeAbbrevOp::Fixed || E == BitCodeAbbrevOp::VBR) &&
+          Data > MaxChunkSize)
+        report_fatal_error(
+            "Fixed or VBR abbrev record with size > MaxChunkData");
+
+      Abbv->Add(BitCodeAbbrevOp(E, Data));
+    } else
       Abbv->Add(BitCodeAbbrevOp(E));
   }
+
+  if (Abbv->getNumOperandInfos() == 0)
+    report_fatal_error("Abbrev record with no operands");
   CurAbbrevs.push_back(Abbv);
 }
 
@@ -297,39 +295,42 @@ bool BitstreamCursor::ReadBlockInfoBlock() {
   // If this is the second stream to get to the block info block, skip it.
   if (BitStream->hasBlockInfoRecords())
     return SkipBlock();
-  
+
   if (EnterSubBlock(bitc::BLOCKINFO_BLOCK_ID)) return true;
-  
+
   SmallVector<uint64_t, 64> Record;
-  BitstreamReader::BlockInfo *CurBlockInfo = 0;
-  
+  BitstreamReader::BlockInfo *CurBlockInfo = nullptr;
+
   // Read all the records for this module.
   while (1) {
-    unsigned Code = ReadCode();
-    if (Code == bitc::END_BLOCK)
-      return ReadBlockEnd();
-    if (Code == bitc::ENTER_SUBBLOCK) {
-      ReadSubBlockID();
-      if (SkipBlock()) return true;
-      continue;
+    BitstreamEntry Entry = advanceSkippingSubblocks(AF_DontAutoprocessAbbrevs);
+
+    switch (Entry.Kind) {
+    case llvm::BitstreamEntry::SubBlock: // Handled for us already.
+    case llvm::BitstreamEntry::Error:
+      return true;
+    case llvm::BitstreamEntry::EndBlock:
+      return false;
+    case llvm::BitstreamEntry::Record:
+      // The interesting case.
+      break;
     }
-    
+
     // Read abbrev records, associate them with CurBID.
-    if (Code == bitc::DEFINE_ABBREV) {
+    if (Entry.ID == bitc::DEFINE_ABBREV) {
       if (!CurBlockInfo) return true;
       ReadAbbrevRecord();
-      
+
       // ReadAbbrevRecord installs the abbrev in CurAbbrevs.  Move it to the
       // appropriate BlockInfo.
-      BitCodeAbbrev *Abbv = CurAbbrevs.back();
+      CurBlockInfo->Abbrevs.push_back(std::move(CurAbbrevs.back()));
       CurAbbrevs.pop_back();
-      CurBlockInfo->Abbrevs.push_back(Abbv);
       continue;
     }
-    
+
     // Read a record.
     Record.clear();
-    switch (ReadRecord(Code, Record)) {
+    switch (readRecord(Entry.ID, Record)) {
       default: break;  // Default behavior, ignore unknown content.
       case bitc::BLOCKINFO_CODE_SETBID:
         if (Record.size() < 1) return true;
@@ -358,4 +359,3 @@ bool BitstreamCursor::ReadBlockInfoBlock() {
   }
 }
 
-