Rebased PNaCl localmods in LLVM to 223109

include/llvm/Bitcode/NaCl/NaClBitstreamReader.h

-//===- BitstreamReader.h - Low-level bitstream reader interface -*- C++ -*-===//
+//===- NaClBitstreamReader.h -----------------------------------*- C++ -*-===//
+//     Low-level bitstream reader interface
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 // This header defines the BitstreamReader class.  This class can be used to
 // read an arbitrary bitstream, regardless of its contents.
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef LLVM_BITCODE_BITSTREAMREADER_H
-#define LLVM_BITCODE_BITSTREAMREADER_H
+#ifndef LLVM_BITCODE_NACL_NACLBITSTREAMREADER_H
+#define LLVM_BITCODE_NACL_NACLBITSTREAMREADER_H
 
-#include "llvm/Bitcode/BitCodes.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/Bitcode/NaCl/NaClLLVMBitCodes.h"
 #include "llvm/Support/Endian.h"
 #include "llvm/Support/StreamingMemoryObject.h"
 #include <climits>
-#include <string>
 #include <vector>
 
 namespace llvm {
 
 class Deserializer;
 
-/// This class is used to read from an LLVM bitcode stream, maintaining
-/// information that is global to decoding the entire file. While a file is
-/// being read, multiple cursors can be independently advanced or skipped around
-/// within the file.  These are represented by the BitstreamCursor class.
-class BitstreamReader {
+/// This class is used to read from a NaCl bitcode wire format stream,
+/// maintaining information that is global to decoding the entire file.
+/// While a file is being read, multiple cursors can be independently
+/// advanced or skipped around within the file.  These are represented by
+/// the NaClBitstreamCursor class.
+class NaClBitstreamReader {
 public:
   /// 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<IntrusiveRefCntPtr<BitCodeAbbrev>> Abbrevs;
-    std::string Name;
-
-    std::vector<std::pair<unsigned, std::string> > RecordNames;
+    std::vector<NaClBitCodeAbbrev*> Abbrevs;
   };
 private:
   std::unique_ptr<MemoryObject> BitcodeBytes;
 
   std::vector<BlockInfo> BlockInfoRecords;
 
-  /// This is set to true if we don't care about the block/record name
-  /// information in the BlockInfo block. Only llvm-bcanalyzer uses this.
-  bool IgnoreBlockInfoNames;
+  /// \brief Holds the offset of the first byte after the header.
+  size_t InitialAddress;
 
-  BitstreamReader(const BitstreamReader&) LLVM_DELETED_FUNCTION;
-  void operator=(const BitstreamReader&) LLVM_DELETED_FUNCTION;
+  NaClBitstreamReader(const NaClBitstreamReader&) LLVM_DELETED_FUNCTION;
+  void operator=(const NaClBitstreamReader&) LLVM_DELETED_FUNCTION;
 public:
-  BitstreamReader() : IgnoreBlockInfoNames(true) {
-  }
+  NaClBitstreamReader() : InitialAddress(0) {}
 
-  BitstreamReader(const unsigned char *Start, const unsigned char *End)
-      : IgnoreBlockInfoNames(true) {
+  NaClBitstreamReader(const unsigned char *Start, const unsigned char *End,
+                      size_t MyInitialAddress=0) {
+    InitialAddress = MyInitialAddress;
     init(Start, End);
   }
 
-  BitstreamReader(MemoryObject *bytes) : IgnoreBlockInfoNames(true) {
-    BitcodeBytes.reset(bytes);
-  }
-
-  BitstreamReader(BitstreamReader &&Other) {
-    *this = std::move(Other);
-  }
-
-  BitstreamReader &operator=(BitstreamReader &&Other) {
-    BitcodeBytes = std::move(Other.BitcodeBytes);
-    // Explicitly swap block info, so that nothing gets destroyed twice.
-    std::swap(BlockInfoRecords, Other.BlockInfoRecords);
-    IgnoreBlockInfoNames = Other.IgnoreBlockInfoNames;
-    return *this;
+  NaClBitstreamReader(MemoryObject *Bytes, size_t MyInitialAddress=0)
+      : InitialAddress(MyInitialAddress) {
+    BitcodeBytes.reset(Bytes);
   }
 
   void init(const unsigned char *Start, const unsigned char *End) {
     assert(((End-Start) & 3) == 0 &&"Bitcode stream not a multiple of 4 bytes");
     BitcodeBytes.reset(getNonStreamedMemoryObject(Start, End));
   }
 
   MemoryObject &getBitcodeBytes() { return *BitcodeBytes; }
 
-  /// This is called by clients that want block/record name information.
-  void CollectBlockInfoNames() { IgnoreBlockInfoNames = false; }
-  bool isIgnoringBlockInfoNames() { return IgnoreBlockInfoNames; }
+  ~NaClBitstreamReader() {
+    // Free the BlockInfoRecords.
+    while (!BlockInfoRecords.empty()) {
+      BlockInfo &Info = BlockInfoRecords.back();
+      // Free blockinfo abbrev info.
+      for (unsigned i = 0, e = static_cast<unsigned>(Info.Abbrevs.size());
+           i != e; ++i)
+        Info.Abbrevs[i]->dropRef();
+      BlockInfoRecords.pop_back();
+    }
+  }
+
+  /// \brief Returns the initial address (after the header) of the input stream.
+  size_t getInitialAddress() const {
+    return InitialAddress;
+  }
 
   //===--------------------------------------------------------------------===//
   // Block Manipulation
   //===--------------------------------------------------------------------===//
 
   /// Return true if we've already read and processed the block info block for
   /// this Bitstream. We only process it for the first cursor that walks over
   /// it.
   bool hasBlockInfoRecords() const { return !BlockInfoRecords.empty(); }
 
@@ skipping 13 matching lines @@
 
   BlockInfo &getOrCreateBlockInfo(unsigned BlockID) {
     if (const BlockInfo *BI = getBlockInfo(BlockID))
       return *const_cast<BlockInfo*>(BI);
 
     // Otherwise, add a new record.
     BlockInfoRecords.push_back(BlockInfo());
     BlockInfoRecords.back().BlockID = BlockID;
     return BlockInfoRecords.back();
   }
-
-  /// Takes block info from the other bitstream reader.
-  ///
-  /// This is a "take" operation because BlockInfo records are non-trivial, and
-  /// indeed rather expensive.
-  void takeBlockInfo(BitstreamReader &&Other) {
-    assert(!hasBlockInfoRecords());
-    BlockInfoRecords = std::move(Other.BlockInfoRecords);
-  }
 };
 
 /// When advancing through a bitstream cursor, each advance can discover a few
 /// different kinds of entries:
-struct BitstreamEntry {
+struct NaClBitstreamEntry {
   enum {
     Error,    // Malformed bitcode was found.
     EndBlock, // We've reached the end of the current block, (or the end of the
               // file, which is treated like a series of EndBlock records.
     SubBlock, // This is the start of a new subblock of a specific ID.
     Record    // This is a record with a specific AbbrevID.
   } Kind;
 
   unsigned ID;
 
-  static BitstreamEntry getError() {
-    BitstreamEntry E; E.Kind = Error; return E;
+  static NaClBitstreamEntry getError() {
+    NaClBitstreamEntry E; E.Kind = Error; return E;
   }
-  static BitstreamEntry getEndBlock() {
-    BitstreamEntry E; E.Kind = EndBlock; return E;
+  static NaClBitstreamEntry getEndBlock() {
+    NaClBitstreamEntry E; E.Kind = EndBlock; return E;
   }
-  static BitstreamEntry getSubBlock(unsigned ID) {
-    BitstreamEntry E; E.Kind = SubBlock; E.ID = ID; return E;
+  static NaClBitstreamEntry getSubBlock(unsigned ID) {
+    NaClBitstreamEntry E; E.Kind = SubBlock; E.ID = ID; return E;
   }
-  static BitstreamEntry getRecord(unsigned AbbrevID) {
-    BitstreamEntry E; E.Kind = Record; E.ID = AbbrevID; return E;
+  static NaClBitstreamEntry getRecord(unsigned AbbrevID) {
+    NaClBitstreamEntry E; E.Kind = Record; E.ID = AbbrevID; return E;
   }
 };
 
+/// Models default view of a bitcode record.
+typedef SmallVector<uint64_t, 8> NaClBitcodeRecordVector;
+
+/// Class NaClAbbrevListener is used to allow instances of class
+/// NaClBitcodeParser to listen to record details when processing
+/// abbreviations. The major reason for using a listener is that the
+/// NaCl bitcode reader would require a major rewrite (including the
+/// introduction of more overhead) if we were to lift abbreviations up
+/// to the bitcode reader. That is, not only would we have to lift the
+/// block processing up into the readers (i.e. many blocks in
+/// NaClBitcodeReader and NaClBitcodeParser), but add many new API's
+/// to allow the readers to update internals of the bit stream reader
+/// appropriately.
+class NaClAbbrevListener {
+  NaClAbbrevListener(const NaClAbbrevListener&) LLVM_DELETED_FUNCTION;
+  void operator=(const NaClAbbrevListener&) LLVM_DELETED_FUNCTION;
+public:
+  NaClAbbrevListener() {}
+  virtual ~NaClAbbrevListener() {}
+
+  /// Called to process the read abbreviation.
+  virtual void ProcessAbbreviation(NaClBitCodeAbbrev *Abbrev,
+                                   bool IsLocal) = 0;
+
+  /// Called after entering block. NumWords is the number of words
+  /// in the block.
+  virtual void BeginBlockInfoBlock(unsigned NumWords) = 0;
+
+  /// Called if a naclbitc::BLOCKINFO_CODE_SETBID record is found in
+  /// NaClBitstreamCursor::ReadBlockInfoBlock.
+  virtual void SetBID() = 0;
+
+  /// Called just before an EndBlock record is processed by
+  /// NaClBitstreamCursor::ReadBlockInfoBlock
+  virtual void EndBlockInfoBlock() = 0;
+
+  /// The values of the bitcode record associated with the called
+  /// virtual function.
+  NaClBitcodeRecordVector Values;
+
+  /// Start bit for current record being processed in
+  /// NaClBitstreamCursor::ReadBlockInfoBlock.
+  uint64_t StartBit;
+};
+
 /// This represents a position within a bitcode file. There may be multiple
-/// independent cursors reading within one bitstream, each maintaining their own
-/// local state.
+/// independent cursors reading within one bitstream, each maintaining their
+/// own local state.
 ///
-/// Unlike iterators, BitstreamCursors are heavy-weight objects that should not
-/// be passed by value.
-class BitstreamCursor {
+/// Unlike iterators, NaClBitstreamCursors are heavy-weight objects
+/// that should not be passed by value.
+class NaClBitstreamCursor {
   friend class Deserializer;
-  BitstreamReader *BitStream;
+  NaClBitstreamReader *BitStream;
   size_t NextChar;
 
-  // The size of the bicode. 0 if we don't know it yet.
+  // The size of the bitcode. 0 if we don't know it yet.
   size_t Size;
 
   /// This is the current data we have pulled from the stream but have not
   /// returned to the client. This is specifically and intentionally defined to
   /// follow the word size of the host machine for efficiency. We use word_t in
   /// places that are aware of this to make it perfectly explicit what is going
   /// on.
   typedef size_t word_t;
   word_t CurWord;
 
-  /// This is the number of bits in CurWord that are valid. This is always from
-  /// [0...bits_of(size_t)-1] inclusive.
+  /// This is the number of bits in CurWord that are valid. This
+  /// is always from [0...bits_of(word_t)-1] inclusive.
   unsigned BitsInCurWord;
 
-  // This is the declared size of code values used for the current block, in
-  // bits.
-  unsigned CurCodeSize;
+  /// This is the declared size of code values used for the current
+  /// block, in bits.
+  NaClBitcodeSelectorAbbrev CurCodeSize;
 
-  /// Abbrevs installed at in this block.
-  std::vector<IntrusiveRefCntPtr<BitCodeAbbrev>> CurAbbrevs;
+  /// Abbrevs installed in this block.
+  std::vector<NaClBitCodeAbbrev*> CurAbbrevs;
 
   struct Block {
-    unsigned PrevCodeSize;
-    std::vector<IntrusiveRefCntPtr<BitCodeAbbrev>> PrevAbbrevs;
-    explicit Block(unsigned PCS) : PrevCodeSize(PCS) {}
+    NaClBitcodeSelectorAbbrev PrevCodeSize;
+    std::vector<NaClBitCodeAbbrev*> PrevAbbrevs;
+    Block() : PrevCodeSize() {}
+    explicit Block(const NaClBitcodeSelectorAbbrev& PCS)
+        : PrevCodeSize(PCS) {}
   };
 
   /// This tracks the codesize of parent blocks.
   SmallVector<Block, 8> BlockScope;
 
+  NaClBitstreamCursor(const NaClBitstreamCursor &) LLVM_DELETED_FUNCTION;
+  NaClBitstreamCursor &operator=(const NaClBitstreamCursor &) LLVM_DELETED_FUNCTION;
 
 public:
-  BitstreamCursor() { init(nullptr); }
+  NaClBitstreamCursor() { init(nullptr); }
 
-  explicit BitstreamCursor(BitstreamReader &R) { init(&R); }
+  explicit NaClBitstreamCursor(NaClBitstreamReader &R) { init(&R); }
 
-  void init(BitstreamReader *R) {
+  void init(NaClBitstreamReader *R) {
     freeState();
-
     BitStream = R;
-    NextChar = 0;
+    NextChar = (BitStream == nullptr) ? 0 : BitStream->getInitialAddress();
     Size = 0;
     BitsInCurWord = 0;
-    CurCodeSize = 2;
+  }
+
+  ~NaClBitstreamCursor() {
+    freeState();
   }
 
   void freeState();
 
   bool canSkipToPos(size_t pos) const {
     // pos can be skipped to if it is a valid address or one byte past the end.
     return pos == 0 || BitStream->getBitcodeBytes().isValidAddress(
         static_cast<uint64_t>(pos - 1));
   }
 
   bool AtEndOfStream() {
     if (BitsInCurWord != 0)
       return false;
     if (Size != 0)
       return Size == NextChar;
     fillCurWord();
     return BitsInCurWord == 0;
   }
 
   /// Return the number of bits used to encode an abbrev #.
-  unsigned getAbbrevIDWidth() const { return CurCodeSize; }
+  unsigned getAbbrevIDWidth() const { return CurCodeSize.NumBits; }
 
   /// Return the bit # of the bit we are reading.
   uint64_t GetCurrentBitNo() const {
     return NextChar*CHAR_BIT - BitsInCurWord;
   }
 
-  BitstreamReader *getBitStreamReader() {
+  NaClBitstreamReader *getBitStreamReader() {
     return BitStream;
   }
-  const BitstreamReader *getBitStreamReader() const {
+  const NaClBitstreamReader *getBitStreamReader() const {
     return BitStream;
   }
 
   /// Flags that modify the behavior of advance().
   enum {
     /// If this flag is used, the advance() method does not automatically pop
     /// the block scope when the end of a block is reached.
     AF_DontPopBlockAtEnd = 1,
 
     /// If this flag is used, abbrev entries are returned just like normal
     /// records.
     AF_DontAutoprocessAbbrevs = 2
   };
 
-      /// Advance the current bitstream, returning the next entry in the stream.
-      BitstreamEntry advance(unsigned Flags = 0) {
+  /// Advance the current bitstream, returning the next entry in the stream.
+  /// Use the given abbreviation listener (if provided).
+  NaClBitstreamEntry advance(unsigned Flags, NaClAbbrevListener *Listener) {
     while (1) {
       unsigned Code = ReadCode();
-      if (Code == bitc::END_BLOCK) {
+      if (Code == naclbitc::END_BLOCK) {
         // Pop the end of the block unless Flags tells us not to.
         if (!(Flags & AF_DontPopBlockAtEnd) && ReadBlockEnd())
-          return BitstreamEntry::getError();
-        return BitstreamEntry::getEndBlock();
+          return NaClBitstreamEntry::getError();
+        return NaClBitstreamEntry::getEndBlock();
       }
 
-      if (Code == bitc::ENTER_SUBBLOCK)
-        return BitstreamEntry::getSubBlock(ReadSubBlockID());
+      if (Code == naclbitc::ENTER_SUBBLOCK)
+        return NaClBitstreamEntry::getSubBlock(ReadSubBlockID());
 
-      if (Code == bitc::DEFINE_ABBREV &&
+      if (Code == naclbitc::DEFINE_ABBREV &&
           !(Flags & AF_DontAutoprocessAbbrevs)) {
         // We read and accumulate abbrev's, the client can't do anything with
         // them anyway.
-        ReadAbbrevRecord();
+        ReadAbbrevRecord(true, Listener);
         continue;
       }
 
-      return BitstreamEntry::getRecord(Code);
+      return NaClBitstreamEntry::getRecord(Code);
     }
   }
 
   /// This is a convenience function for clients that don't expect any
   /// subblocks. This just skips over them automatically.
-  BitstreamEntry advanceSkippingSubblocks(unsigned Flags = 0) {
+  NaClBitstreamEntry advanceSkippingSubblocks(unsigned Flags = 0) {
     while (1) {
       // If we found a normal entry, return it.
-      BitstreamEntry Entry = advance(Flags);
-      if (Entry.Kind != BitstreamEntry::SubBlock)
+      NaClBitstreamEntry Entry = advance(Flags, 0);
+      if (Entry.Kind != NaClBitstreamEntry::SubBlock)
         return Entry;
 
       // If we found a sub-block, just skip over it and check the next entry.
       if (SkipBlock())
-        return BitstreamEntry::getError();
+        return NaClBitstreamEntry::getError();
     }
   }
 
   /// Reset the stream to the specified bit number.
   void JumpToBit(uint64_t BitNo) {
     uintptr_t ByteNo = uintptr_t(BitNo/8) & ~(sizeof(word_t)-1);
     unsigned WordBitNo = unsigned(BitNo & (sizeof(word_t)*8-1));
     assert(canSkipToPos(ByteNo) && "Invalid location");
 
     // Move the cursor to the right word.
@@ skipping 114 matching lines @@
       CurWord >>= BitsInCurWord-32;
       BitsInCurWord = 32;
       return;
     }
 
     BitsInCurWord = 0;
   }
 public:
 
   unsigned ReadCode() {
-    return Read(CurCodeSize);
+    return CurCodeSize.IsFixed
+        ? Read(CurCodeSize.NumBits)
+        : ReadVBR(CurCodeSize.NumBits);
   }
 
-
   // Block header:
   //    [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen]
 
   /// Having read the ENTER_SUBBLOCK code, read the BlockID for the block.
   unsigned ReadSubBlockID() {
-    return ReadVBR(bitc::BlockIDWidth);
+    return ReadVBR(naclbitc::BlockIDWidth);
   }
 
   /// Having read the ENTER_SUBBLOCK abbrevid and a BlockID, skip over the body
   /// of this block. If the block record is malformed, return true.
   bool SkipBlock() {
     // Read and ignore the codelen value.  Since we are skipping this block, we
     // don't care what code widths are used inside of it.
-    ReadVBR(bitc::CodeLenWidth);
+    ReadVBR(naclbitc::CodeLenWidth);
     SkipToFourByteBoundary();
-    unsigned NumFourBytes = Read(bitc::BlockSizeWidth);
+    unsigned NumFourBytes = Read(naclbitc::BlockSizeWidth);
 
     // Check that the block wasn't partially defined, and that the offset isn't
     // bogus.
     size_t SkipTo = GetCurrentBitNo() + NumFourBytes*4*8;
     if (AtEndOfStream() || !canSkipToPos(SkipTo/8))
       return true;
 
     JumpToBit(SkipTo);
     return false;
   }
@@ skipping 11 matching lines @@
 
     popBlockScope();
     return false;
   }
 
 private:
 
   void popBlockScope() {
     CurCodeSize = BlockScope.back().PrevCodeSize;
 
-    CurAbbrevs = std::move(BlockScope.back().PrevAbbrevs);
+    // Delete abbrevs from popped scope.
+    for (unsigned i = 0, e = static_cast<unsigned>(CurAbbrevs.size());
+         i != e; ++i)
+      CurAbbrevs[i]->dropRef();
+
+    BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
     BlockScope.pop_back();
   }
 
   //===--------------------------------------------------------------------===//
   // Record Processing
   //===--------------------------------------------------------------------===//
 
+private:
+  void skipAbbreviatedField(const NaClBitCodeAbbrevOp &Op);
+
+  // Reads the next Value using the abbreviation Op. Returns true only
+  // if Op is an array (and sets Value to the number of elements in the
+  // array).
+  inline bool readRecordAbbrevField(const NaClBitCodeAbbrevOp &Op,
+                                    uint64_t &Value);
+
+  // Reads and returns the next value using the abbreviation Op,
+  // assuming Op appears after an array abbreviation.
+  inline uint64_t readArrayAbbreviatedField(const NaClBitCodeAbbrevOp &Op);
+
+  // Reads the array abbreviation Op, NumArrayElements times, putting
+  // the read values in Vals.
+  inline void readArrayAbbrev(const NaClBitCodeAbbrevOp &Op,
+                              unsigned NumArrayElements,
+                              SmallVectorImpl<uint64_t> &Vals);
+
 public:
 
   /// Return the abbreviation for the specified AbbrevId.
-  const BitCodeAbbrev *getAbbrev(unsigned AbbrevID) {
-    unsigned AbbrevNo = AbbrevID-bitc::FIRST_APPLICATION_ABBREV;
+  const NaClBitCodeAbbrev *getAbbrev(unsigned AbbrevID) const {
+    unsigned AbbrevNo = AbbrevID-naclbitc::FIRST_APPLICATION_ABBREV;
     assert(AbbrevNo < CurAbbrevs.size() && "Invalid abbrev #!");
-    return CurAbbrevs[AbbrevNo].get();
+    return CurAbbrevs[AbbrevNo];
   }
 
   /// Read the current record and discard it.
   void skipRecord(unsigned AbbrevID);
-
-  unsigned readRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
-                      StringRef *Blob = nullptr);
+  
+  unsigned readRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals);
 
   //===--------------------------------------------------------------------===//
   // Abbrev Processing
   //===--------------------------------------------------------------------===//
-  void ReadAbbrevRecord();
+  // IsLocal indicates where the abbreviation occurs. If it is in the
+  // BlockInfo block, IsLocal is false. In all other cases, IsLocal is
+  // true.
+  void ReadAbbrevRecord(bool IsLocal,
+                        NaClAbbrevListener *Listener);
 
-  bool ReadBlockInfoBlock();
+  // Skips over an abbreviation record. Duplicates code of ReadAbbrevRecord,
+  // except that no abbreviation is built.
+  void SkipAbbrevRecord();
+  
+  bool ReadBlockInfoBlock(NaClAbbrevListener *Listener);
 };
 
 } // End llvm namespace
 
 #endif