Adding a collector name attribute to Function in the IR. These
[oota-llvm.git] / include / llvm / Bitcode / BitstreamReader.h
index c6394b6c99a539109f55dfe7564838005f9f48a1..d06b46604baf4b7eb04a9805aadfa097a99335fe 100644 (file)
 
 namespace llvm {
   
+  class Deserializer;
+  
 class BitstreamReader {
   const unsigned char *NextChar;
   const unsigned char *LastChar;
+  friend class Deserializer;
   
   /// CurWord - This is the current data we have pulled from the stream but have
   /// not returned to the client.
@@ -48,11 +51,31 @@ class BitstreamReader {
   /// BlockScope - This tracks the codesize of parent blocks.
   SmallVector<Block, 8> BlockScope;
 
+  /// BlockInfo - This contains information emitted to BLOCKINFO_BLOCK blocks.
+  /// These describe abbreviations that all blocks of the specified ID inherit.
+  struct BlockInfo {
+    unsigned BlockID;
+    std::vector<BitCodeAbbrev*> Abbrevs;
+  };
+  std::vector<BlockInfo> BlockInfoRecords;
+  
   /// FirstChar - This remembers the first byte of the stream.
   const unsigned char *FirstChar;
 public:
-  BitstreamReader(const unsigned char *Start, const unsigned char *End)
-    : NextChar(Start), LastChar(End), FirstChar(Start) {
+  BitstreamReader() {
+    NextChar = FirstChar = LastChar = 0;
+    CurWord = 0;
+    BitsInCurWord = 0;
+    CurCodeSize = 0;
+  }
+
+  BitstreamReader(const unsigned char *Start, const unsigned char *End) {
+    init(Start, End);
+  }
+  
+  void init(const unsigned char *Start, const unsigned char *End) {
+    NextChar = FirstChar = Start;
+    LastChar = End;
     assert(((End-Start) & 3) == 0 &&"Bitcode stream not a multiple of 4 bytes");
     CurWord = 0;
     BitsInCurWord = 0;
@@ -63,22 +86,53 @@ public:
     // Abbrevs could still exist if the stream was broken.  If so, don't leak
     // them.
     for (unsigned i = 0, e = CurAbbrevs.size(); i != e; ++i)
-      delete CurAbbrevs[i];
+      CurAbbrevs[i]->dropRef();
 
     for (unsigned S = 0, e = BlockScope.size(); S != e; ++S) {
       std::vector<BitCodeAbbrev*> &Abbrevs = BlockScope[S].PrevAbbrevs;
       for (unsigned i = 0, e = Abbrevs.size(); i != e; ++i)
-        delete Abbrevs[i];
+        Abbrevs[i]->dropRef();
+    }
+    
+    // Free the BlockInfoRecords.
+    while (!BlockInfoRecords.empty()) {
+      BlockInfo &Info = BlockInfoRecords.back();
+      // Free blockinfo abbrev info.
+      for (unsigned i = 0, e = Info.Abbrevs.size(); i != e; ++i)
+        Info.Abbrevs[i]->dropRef();
+      BlockInfoRecords.pop_back();
     }
   }
   
-  bool AtEndOfStream() const { return NextChar == LastChar; }
+  bool AtEndOfStream() const {
+    return NextChar == LastChar && BitsInCurWord == 0;
+  }
   
   /// GetCurrentBitNo - Return the bit # of the bit we are reading.
   uint64_t GetCurrentBitNo() const {
-    return (NextChar-FirstChar)*8 + (32-BitsInCurWord);
+    return (NextChar-FirstChar)*8 + ((32-BitsInCurWord) & 31);
+  }
+  
+  /// JumpToBit - Reset the stream to the specified bit number.
+  void JumpToBit(uint64_t BitNo) {
+    uintptr_t ByteNo = uintptr_t(BitNo/8) & ~3;
+    uintptr_t WordBitNo = uintptr_t(BitNo) & 31;
+    assert(ByteNo < (uintptr_t)(LastChar-FirstChar) && "Invalid location");
+    
+    // Move the cursor to the right word.
+    NextChar = FirstChar+ByteNo;
+    BitsInCurWord = 0;
+    CurWord = 0;
+    
+    // Skip over any bits that are already consumed.
+    if (WordBitNo) {
+      NextChar -= 4;
+      Read(WordBitNo);
+    }
   }
   
+  /// GetAbbrevIDWidth - Return the number of bits used to encode an abbrev #.
+  unsigned GetAbbrevIDWidth() const { return CurCodeSize; }
   
   uint32_t Read(unsigned NumBits) {
     // If the field is fully contained by CurWord, return it quickly.
@@ -175,6 +229,22 @@ public:
   // Block Manipulation
   //===--------------------------------------------------------------------===//
   
+private:
+  /// getBlockInfo - If there is block info for the specified ID, return it,
+  /// otherwise return null.
+  BlockInfo *getBlockInfo(unsigned BlockID) {
+    // Common case, the most recent entry matches BlockID.
+    if (!BlockInfoRecords.empty() && BlockInfoRecords.back().BlockID == BlockID)
+      return &BlockInfoRecords.back();
+    
+    for (unsigned i = 0, e = BlockInfoRecords.size(); i != e; ++i)
+      if (BlockInfoRecords[i].BlockID == BlockID)
+        return &BlockInfoRecords[i];
+    return 0;
+  }
+public:
+  
+  
   // Block header:
   //    [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen]
 
@@ -203,16 +273,26 @@ public:
     return false;
   }
   
-  /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, read and enter
-  /// the block, returning the BlockID of the block we just entered.
-  bool EnterSubBlock() {
+  /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter
+  /// the block, and return true if the block is valid.
+  bool EnterSubBlock(unsigned BlockID, unsigned *NumWordsP = 0) {
+    // 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 (BlockInfo *Info = getBlockInfo(BlockID)) {
+      for (unsigned i = 0, e = Info->Abbrevs.size(); i != e; ++i) {
+        CurAbbrevs.push_back(Info->Abbrevs[i]);
+        CurAbbrevs.back()->addRef();
+      }
+    }
+    
     // Get the codesize of this block.
     CurCodeSize = ReadVBR(bitc::CodeLenWidth);
     SkipToWord();
     unsigned NumWords = Read(bitc::BlockSizeWidth);
+    if (NumWordsP) *NumWordsP = NumWords;
     
     // Validate that this block is sane.
     if (CurCodeSize == 0 || AtEndOfStream() || NextChar+NumWords*4 > LastChar)
@@ -227,21 +307,50 @@ public:
     // Block tail:
     //    [END_BLOCK, <align4bytes>]
     SkipToWord();
+    
+    PopBlockScope();
+    return false;
+  }
+  
+private:
+  void PopBlockScope() {
     CurCodeSize = BlockScope.back().PrevCodeSize;
     
     // Delete abbrevs from popped scope.
     for (unsigned i = 0, e = CurAbbrevs.size(); i != e; ++i)
-      delete CurAbbrevs[i];
+      CurAbbrevs[i]->dropRef();
     
     BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
     BlockScope.pop_back();
-    return false;
-  }
-  
+  }  
+    
   //===--------------------------------------------------------------------===//
   // Record Processing
   //===--------------------------------------------------------------------===//
   
+private:
+  void ReadAbbreviatedField(const BitCodeAbbrevOp &Op, 
+                            SmallVectorImpl<uint64_t> &Vals) {
+    if (Op.isLiteral()) {
+      // If the abbrev specifies the literal value to use, use it.
+      Vals.push_back(Op.getLiteralValue());
+    } else {
+      // Decode the value as we are commanded.
+      switch (Op.getEncoding()) {
+      default: assert(0 && "Unknown encoding!");
+      case BitCodeAbbrevOp::Fixed:
+        Vals.push_back(Read((unsigned)Op.getEncodingData()));
+        break;
+      case BitCodeAbbrevOp::VBR:
+        Vals.push_back(ReadVBR64((unsigned)Op.getEncodingData()));
+        break;
+      case BitCodeAbbrevOp::Char6:
+        Vals.push_back(BitCodeAbbrevOp::DecodeChar6(Read(6)));
+        break;
+      }
+    }
+  }
+public:
   unsigned ReadRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals) {
     if (AbbrevID == bitc::UNABBREV_RECORD) {
       unsigned Code = ReadVBR(6);
@@ -251,30 +360,29 @@ public:
       return Code;
     }
     
-    unsigned AbbrevNo = AbbrevID-bitc::FIRST_ABBREV;
+    unsigned AbbrevNo = AbbrevID-bitc::FIRST_APPLICATION_ABBREV;
     assert(AbbrevNo < CurAbbrevs.size() && "Invalid abbrev #!");
     BitCodeAbbrev *Abbv = CurAbbrevs[AbbrevNo];
 
     for (unsigned i = 0, e = Abbv->getNumOperandInfos(); i != e; ++i) {
       const BitCodeAbbrevOp &Op = Abbv->getOperandInfo(i);
-      if (Op.isLiteral()) {
-        // If the abbrev specifies the literal value to use, use it.
-        Vals.push_back(Op.getLiteralValue());
+      if (Op.isLiteral() || Op.getEncoding() != BitCodeAbbrevOp::Array) {
+        ReadAbbreviatedField(Op, Vals);
       } else {
-        // Decode the value as we are commanded.
-        switch (Op.getEncoding()) {
-        default: assert(0 && "Unknown encoding!");
-        case BitCodeAbbrevOp::FixedWidth:
-          Vals.push_back(Read(Op.getEncodingData()));
-          break;
-        case BitCodeAbbrevOp::VBR:
-          Vals.push_back(ReadVBR64(Op.getEncodingData()));
-          break;
-        }
+        // 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)
+          ReadAbbreviatedField(EltEnc, Vals);
       }
     }
     
-    unsigned Code = Vals[0];
+    unsigned Code = (unsigned)Vals[0];
     Vals.erase(Vals.begin());
     return Code;
   }
@@ -287,21 +395,79 @@ public:
     BitCodeAbbrev *Abbv = new BitCodeAbbrev();
     unsigned NumOpInfo = ReadVBR(5);
     for (unsigned i = 0; i != NumOpInfo; ++i) {
-      bool IsLiteral = Read(1);
+      bool IsLiteral = Read(1) ? true : false;
       if (IsLiteral) {
         Abbv->Add(BitCodeAbbrevOp(ReadVBR64(8)));
         continue;
       }
 
       BitCodeAbbrevOp::Encoding E = (BitCodeAbbrevOp::Encoding)Read(3);
-      if (BitCodeAbbrevOp::hasEncodingData(E)) {
+      if (BitCodeAbbrevOp::hasEncodingData(E))
         Abbv->Add(BitCodeAbbrevOp(E, ReadVBR64(5)));
-      } else {
-        assert(0 && "unimp");
-      }
+      else
+        Abbv->Add(BitCodeAbbrevOp(E));
     }
     CurAbbrevs.push_back(Abbv);
   }
+  
+  //===--------------------------------------------------------------------===//
+  // BlockInfo Block Reading
+  //===--------------------------------------------------------------------===//
+  
+private:  
+  BlockInfo &getOrCreateBlockInfo(unsigned BlockID) {
+    if (BlockInfo *BI = getBlockInfo(BlockID))
+      return *BI;
+    
+    // Otherwise, add a new record.
+    BlockInfoRecords.push_back(BlockInfo());
+    BlockInfoRecords.back().BlockID = BlockID;
+    return BlockInfoRecords.back();
+  }
+  
+public:
+    
+  bool ReadBlockInfoBlock() {
+    if (EnterSubBlock(bitc::BLOCKINFO_BLOCK_ID)) return true;
+
+    SmallVector<uint64_t, 64> Record;
+    BlockInfo *CurBlockInfo = 0;
+    
+    // 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;
+      }
+
+      // Read abbrev records, associate them with CurBID.
+      if (Code == bitc::DEFINE_ABBREV) {
+        if (!CurBlockInfo) return true;
+        ReadAbbrevRecord();
+        
+        // ReadAbbrevRecord installs the abbrev in CurAbbrevs.  Move it to the
+        // appropriate BlockInfo.
+        BitCodeAbbrev *Abbv = CurAbbrevs.back();
+        CurAbbrevs.pop_back();
+        CurBlockInfo->Abbrevs.push_back(Abbv);
+        continue;
+      }
+
+      // Read a record.
+      Record.clear();
+      switch (ReadRecord(Code, Record)) {
+      default: break;  // Default behavior, ignore unknown content.
+      case bitc::BLOCKINFO_CODE_SETBID:
+        if (Record.size() < 1) return true;
+        CurBlockInfo = &getOrCreateBlockInfo((unsigned)Record[0]);
+        break;
+      }
+    }      
+  }
 };
 
 } // End llvm namespace