#ifndef LLVM_BITCODE_BITSTREAMREADER_H
#define LLVM_BITCODE_BITSTREAMREADER_H
-#include "llvm/ADT/OwningPtr.h"
#include "llvm/Bitcode/BitCodes.h"
#include "llvm/Support/Endian.h"
-#include "llvm/Support/StreamableMemoryObject.h"
+#include "llvm/Support/StreamingMemoryObject.h"
#include <climits>
#include <string>
#include <vector>
namespace llvm {
- class Deserializer;
-
-/// BitstreamReader - 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.
+/// 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 {
public:
- /// BlockInfo - This contains information emitted to BLOCKINFO_BLOCK blocks.
- /// These describe abbreviations that all blocks of the specified ID inherit.
+ /// 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<IntrusiveRefCntPtr<BitCodeAbbrev>> Abbrevs;
std::string Name;
std::vector<std::pair<unsigned, std::string> > RecordNames;
};
private:
- OwningPtr<StreamableMemoryObject> BitcodeBytes;
+ std::unique_ptr<MemoryObject> BitcodeBytes;
std::vector<BlockInfo> BlockInfoRecords;
- /// IgnoreBlockInfoNames - 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.
+ /// 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;
- BitstreamReader(const BitstreamReader&) LLVM_DELETED_FUNCTION;
- void operator=(const BitstreamReader&) LLVM_DELETED_FUNCTION;
+ BitstreamReader(const BitstreamReader&) = delete;
+ void operator=(const BitstreamReader&) = delete;
public:
BitstreamReader() : IgnoreBlockInfoNames(true) {
}
- BitstreamReader(const unsigned char *Start, const unsigned char *End) {
- IgnoreBlockInfoNames = true;
+ BitstreamReader(const unsigned char *Start, const unsigned char *End)
+ : IgnoreBlockInfoNames(true) {
init(Start, End);
}
- BitstreamReader(StreamableMemoryObject *bytes) {
- BitcodeBytes.reset(bytes);
+ BitstreamReader(std::unique_ptr<MemoryObject> BitcodeBytes)
+ : BitcodeBytes(std::move(BitcodeBytes)), IgnoreBlockInfoNames(true) {}
+
+ 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;
}
void init(const unsigned char *Start, const unsigned char *End) {
BitcodeBytes.reset(getNonStreamedMemoryObject(Start, End));
}
- StreamableMemoryObject &getBitcodeBytes() { return *BitcodeBytes; }
+ MemoryObject &getBitcodeBytes() { return *BitcodeBytes; }
- ~BitstreamReader() {
- // 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();
- }
- }
-
- /// CollectBlockInfoNames - This is called by clients that want block/record
- /// name information.
+ /// This is called by clients that want block/record name information.
void CollectBlockInfoNames() { IgnoreBlockInfoNames = false; }
bool isIgnoringBlockInfoNames() { return IgnoreBlockInfoNames; }
// Block Manipulation
//===--------------------------------------------------------------------===//
- /// hasBlockInfoRecords - 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.
+ /// 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(); }
- /// getBlockInfo - If there is block info for the specified ID, return it,
- /// otherwise return null.
+ /// If there is block info for the specified ID, return it, otherwise return
+ /// null.
const BlockInfo *getBlockInfo(unsigned BlockID) const {
// Common case, the most recent entry matches BlockID.
if (!BlockInfoRecords.empty() && BlockInfoRecords.back().BlockID == BlockID)
i != e; ++i)
if (BlockInfoRecords[i].BlockID == BlockID)
return &BlockInfoRecords[i];
- return 0;
+ return nullptr;
}
BlockInfo &getOrCreateBlockInfo(unsigned BlockID) {
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);
+ }
};
-
-/// BitstreamEntry - When advancing through a bitstream cursor, each advance can
-/// discover a few different kinds of entries:
-/// 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.
-///
+/// When advancing through a bitstream cursor, each advance can discover a few
+/// different kinds of entries:
struct BitstreamEntry {
enum {
- Error,
- EndBlock,
- SubBlock,
- Record
+ 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() {
}
};
-
-/// BitstreamCursor - This represents a position within a bitcode file. There
-/// may be multiple independent cursors reading within one bitstream, each
-/// maintaining their own local state.
+/// This represents a position within a bitcode file. There may be multiple
+/// 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 {
- friend class Deserializer;
BitstreamReader *BitStream;
size_t NextChar;
- /// CurWord - This is the current data we have pulled from the stream but have
- /// not returned to the client.
- uint32_t CurWord;
+ // The size of the bicode. 0 if we don't know it yet.
+ size_t Size;
- /// BitsInCurWord - This is the number of bits in CurWord that are valid. This
- /// is always from [0...31] inclusive.
+ /// 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.
unsigned BitsInCurWord;
- // CurCodeSize - This is the declared size of code values used for the current
- // block, in bits.
+ // This is the declared size of code values used for the current block, in
+ // bits.
unsigned CurCodeSize;
- /// CurAbbrevs - Abbrevs installed at in this block.
- std::vector<BitCodeAbbrev*> CurAbbrevs;
+ /// Abbrevs installed at in this block.
+ std::vector<IntrusiveRefCntPtr<BitCodeAbbrev>> CurAbbrevs;
struct Block {
unsigned PrevCodeSize;
- std::vector<BitCodeAbbrev*> PrevAbbrevs;
+ std::vector<IntrusiveRefCntPtr<BitCodeAbbrev>> PrevAbbrevs;
explicit Block(unsigned PCS) : PrevCodeSize(PCS) {}
};
- /// BlockScope - This tracks the codesize of parent blocks.
+ /// This tracks the codesize of parent blocks.
SmallVector<Block, 8> BlockScope;
-
+
public:
- BitstreamCursor() : BitStream(0), NextChar(0) {
- }
- BitstreamCursor(const BitstreamCursor &RHS) : BitStream(0), NextChar(0) {
- operator=(RHS);
- }
+ BitstreamCursor() { init(nullptr); }
- explicit BitstreamCursor(BitstreamReader &R) : BitStream(&R) {
- NextChar = 0;
- CurWord = 0;
- BitsInCurWord = 0;
- CurCodeSize = 2;
- }
+ explicit BitstreamCursor(BitstreamReader &R) { init(&R); }
- void init(BitstreamReader &R) {
+ void init(BitstreamReader *R) {
freeState();
- BitStream = &R;
+ BitStream = R;
NextChar = 0;
- CurWord = 0;
+ Size = 0;
BitsInCurWord = 0;
CurCodeSize = 2;
}
- ~BitstreamCursor() {
- freeState();
- }
-
- void operator=(const BitstreamCursor &RHS);
-
void freeState();
-
- bool isEndPos(size_t pos) {
- return BitStream->getBitcodeBytes().isObjectEnd(static_cast<uint64_t>(pos));
- }
bool canSkipToPos(size_t pos) const {
// pos can be skipped to if it is a valid address or one byte past the end.
static_cast<uint64_t>(pos - 1));
}
- unsigned char getByte(size_t pos) {
- uint8_t byte = -1;
- BitStream->getBitcodeBytes().readByte(pos, &byte);
- return byte;
- }
-
- uint32_t getWord(size_t pos) {
- uint8_t buf[4] = { 0xFF, 0xFF, 0xFF, 0xFF };
- BitStream->getBitcodeBytes().readBytes(pos, sizeof(buf), buf, NULL);
- return *reinterpret_cast<support::ulittle32_t *>(buf);
- }
-
bool AtEndOfStream() {
- return isEndPos(NextChar) && BitsInCurWord == 0;
+ if (BitsInCurWord != 0)
+ return false;
+ if (Size != 0)
+ return Size == NextChar;
+ fillCurWord();
+ return BitsInCurWord == 0;
}
- /// getAbbrevIDWidth - Return the number of bits used to encode an abbrev #.
+ /// Return the number of bits used to encode an abbrev #.
unsigned getAbbrevIDWidth() const { return CurCodeSize; }
- /// GetCurrentBitNo - Return the bit # of the bit we are reading.
+ /// Return the bit # of the bit we are reading.
uint64_t GetCurrentBitNo() const {
return NextChar*CHAR_BIT - BitsInCurWord;
}
return BitStream;
}
-
- /// advance - Advance the current bitstream, returning the next entry in the
- /// stream.
- BitstreamEntry advance() {
+ /// 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) {
while (1) {
unsigned Code = ReadCode();
if (Code == bitc::END_BLOCK) {
- if (ReadBlockEnd())
+ // Pop the end of the block unless Flags tells us not to.
+ if (!(Flags & AF_DontPopBlockAtEnd) && ReadBlockEnd())
return BitstreamEntry::getError();
return BitstreamEntry::getEndBlock();
}
-
+
if (Code == bitc::ENTER_SUBBLOCK)
return BitstreamEntry::getSubBlock(ReadSubBlockID());
-
- if (Code == bitc::DEFINE_ABBREV) {
+
+ if (Code == bitc::DEFINE_ABBREV &&
+ !(Flags & AF_DontAutoprocessAbbrevs)) {
// We read and accumulate abbrev's, the client can't do anything with
// them anyway.
ReadAbbrevRecord();
}
}
- /// advanceSkippingSubblocks - This is a convenience function for clients that
- /// don't expect any subblocks. This just skips over them automatically.
- BitstreamEntry advanceSkippingSubblocks() {
+ /// This is a convenience function for clients that don't expect any
+ /// subblocks. This just skips over them automatically.
+ BitstreamEntry advanceSkippingSubblocks(unsigned Flags = 0) {
while (1) {
// If we found a normal entry, return it.
- BitstreamEntry Entry = advance();
+ BitstreamEntry Entry = advance(Flags);
if (Entry.Kind != BitstreamEntry::SubBlock)
return Entry;
-
+
// If we found a sub-block, just skip over it and check the next entry.
if (SkipBlock())
return BitstreamEntry::getError();
}
}
- /// JumpToBit - Reset the stream to the specified bit number.
+ /// 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;
+ size_t ByteNo = size_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.
NextChar = ByteNo;
BitsInCurWord = 0;
- CurWord = 0;
// Skip over any bits that are already consumed.
if (WordBitNo)
- Read(static_cast<unsigned>(WordBitNo));
+ Read(WordBitNo);
+ }
+
+ void fillCurWord() {
+ if (Size != 0 && NextChar >= Size)
+ report_fatal_error("Unexpected end of file");
+
+ // Read the next word from the stream.
+ uint8_t Array[sizeof(word_t)] = {0};
+
+ uint64_t BytesRead =
+ BitStream->getBitcodeBytes().readBytes(Array, sizeof(Array), NextChar);
+
+ // If we run out of data, stop at the end of the stream.
+ if (BytesRead == 0) {
+ Size = NextChar;
+ return;
+ }
+
+ CurWord =
+ support::endian::read<word_t, support::little, support::unaligned>(
+ Array);
+ NextChar += BytesRead;
+ BitsInCurWord = BytesRead * 8;
}
+ word_t Read(unsigned NumBits) {
+ static const unsigned BitsInWord = sizeof(word_t) * 8;
+
+ assert(NumBits && NumBits <= BitsInWord &&
+ "Cannot return zero or more than BitsInWord bits!");
+
+ static const unsigned Mask = sizeof(word_t) > 4 ? 0x3f : 0x1f;
- uint32_t Read(unsigned NumBits) {
- assert(NumBits <= 32 && "Cannot return more than 32 bits!");
// If the field is fully contained by CurWord, return it quickly.
if (BitsInCurWord >= NumBits) {
- uint32_t R = CurWord & ((1U << NumBits)-1);
- CurWord >>= NumBits;
+ word_t R = CurWord & (~word_t(0) >> (BitsInWord - NumBits));
+
+ // Use a mask to avoid undefined behavior.
+ CurWord >>= (NumBits & Mask);
+
BitsInCurWord -= NumBits;
return R;
}
+ word_t R = BitsInCurWord ? CurWord : 0;
+ unsigned BitsLeft = NumBits - BitsInCurWord;
+
+ fillCurWord();
+
// If we run out of data, stop at the end of the stream.
- if (isEndPos(NextChar)) {
- CurWord = 0;
- BitsInCurWord = 0;
+ if (BitsLeft > BitsInCurWord)
return 0;
- }
- unsigned R = CurWord;
+ word_t R2 = CurWord & (~word_t(0) >> (BitsInWord - BitsLeft));
- // Read the next word from the stream.
- CurWord = getWord(NextChar);
- NextChar += 4;
+ // Use a mask to avoid undefined behavior.
+ CurWord >>= (BitsLeft & Mask);
- // Extract NumBits-BitsInCurWord from what we just read.
- unsigned BitsLeft = NumBits-BitsInCurWord;
+ BitsInCurWord -= BitsLeft;
- // Be careful here, BitsLeft is in the range [1..32] inclusive.
- R |= (CurWord & (~0U >> (32-BitsLeft))) << BitsInCurWord;
+ R |= R2 << (NumBits - BitsLeft);
- // BitsLeft bits have just been used up from CurWord.
- if (BitsLeft != 32)
- CurWord >>= BitsLeft;
- else
- CurWord = 0;
- BitsInCurWord = 32-BitsLeft;
return R;
}
- uint64_t Read64(unsigned NumBits) {
- if (NumBits <= 32) return Read(NumBits);
-
- uint64_t V = Read(32);
- return V | (uint64_t)Read(NumBits-32) << 32;
- }
-
uint32_t ReadVBR(unsigned NumBits) {
uint32_t Piece = Read(NumBits);
if ((Piece & (1U << (NumBits-1))) == 0)
}
}
- // ReadVBR64 - Read a VBR that may have a value up to 64-bits in size. The
- // chunk size of the VBR must still be <= 32 bits though.
+ // Read a VBR that may have a value up to 64-bits in size. The chunk size of
+ // the VBR must still be <= 32 bits though.
uint64_t ReadVBR64(unsigned NumBits) {
+ assert(NumBits <= 64 && "VBR can only be up to 64 bits in size.");
uint32_t Piece = Read(NumBits);
if ((Piece & (1U << (NumBits-1))) == 0)
return uint64_t(Piece);
}
}
- void SkipToWord() {
+private:
+ void SkipToFourByteBoundary() {
+ // If word_t is 64-bits and if we've read less than 32 bits, just dump
+ // the bits we have up to the next 32-bit boundary.
+ if (sizeof(word_t) > 4 &&
+ BitsInCurWord >= 32) {
+ CurWord >>= BitsInCurWord-32;
+ BitsInCurWord = 32;
+ return;
+ }
+
BitsInCurWord = 0;
- CurWord = 0;
}
+public:
unsigned ReadCode() {
return Read(CurCodeSize);
// Block header:
// [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen]
- /// ReadSubBlockID - Having read the ENTER_SUBBLOCK code, read the BlockID for
- /// the block.
+ /// Having read the ENTER_SUBBLOCK code, read the BlockID for the block.
unsigned ReadSubBlockID() {
return ReadVBR(bitc::BlockIDWidth);
}
- /// SkipBlock - Having read the ENTER_SUBBLOCK abbrevid and a BlockID, skip
- /// over the body of this block. If the block record is malformed, return
- /// true.
+ /// 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);
- SkipToWord();
- unsigned NumWords = Read(bitc::BlockSizeWidth);
+ SkipToFourByteBoundary();
+ unsigned NumFourBytes = Read(bitc::BlockSizeWidth);
// Check that the block wasn't partially defined, and that the offset isn't
// bogus.
- size_t SkipTo = NextChar + NumWords*4;
- if (AtEndOfStream() || !canSkipToPos(SkipTo))
+ size_t SkipTo = GetCurrentBitNo() + NumFourBytes*4*8;
+ if (AtEndOfStream() || !canSkipToPos(SkipTo/8))
return true;
- NextChar = SkipTo;
+ JumpToBit(SkipTo);
return false;
}
- /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter
- /// the block, and return true if the block has an error.
- bool EnterSubBlock(unsigned BlockID, unsigned *NumWordsP = 0);
-
+ /// Having read the ENTER_SUBBLOCK abbrevid, enter the block, and return true
+ /// if the block has an error.
+ bool EnterSubBlock(unsigned BlockID, unsigned *NumWordsP = nullptr);
+
bool ReadBlockEnd() {
if (BlockScope.empty()) return true;
// Block tail:
// [END_BLOCK, <align4bytes>]
- SkipToWord();
+ SkipToFourByteBoundary();
popBlockScope();
return false;
void popBlockScope() {
CurCodeSize = BlockScope.back().PrevCodeSize;
- // 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);
+ CurAbbrevs = std::move(BlockScope.back().PrevAbbrevs);
BlockScope.pop_back();
}
- //===--------------------------------------------------------------------===//
+ //===--------------------------------------------------------------------===//
// Record Processing
//===--------------------------------------------------------------------===//
-private:
- void 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());
- }
-
- void 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()) {
- default: llvm_unreachable("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:
-
- /// getAbbrev - Return the abbreviation for the specified AbbrevId.
+ /// Return the abbreviation for the specified AbbrevId.
const BitCodeAbbrev *getAbbrev(unsigned AbbrevID) {
- unsigned AbbrevNo = AbbrevID-bitc::FIRST_APPLICATION_ABBREV;
- assert(AbbrevNo < CurAbbrevs.size() && "Invalid abbrev #!");
- return CurAbbrevs[AbbrevNo];
+ unsigned AbbrevNo = AbbrevID - bitc::FIRST_APPLICATION_ABBREV;
+ if (AbbrevNo >= CurAbbrevs.size())
+ report_fatal_error("Invalid abbrev number");
+ return CurAbbrevs[AbbrevNo].get();
}
- unsigned ReadRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
- const char **BlobStart = 0, unsigned *BlobLen = 0);
-
- unsigned ReadRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
- const char *&BlobStart, unsigned &BlobLen) {
- return ReadRecord(AbbrevID, Vals, &BlobStart, &BlobLen);
- }
+ /// Read the current record and discard it.
+ void skipRecord(unsigned AbbrevID);
+ unsigned readRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
+ StringRef *Blob = nullptr);
//===--------------------------------------------------------------------===//
// Abbrev Processing
//===--------------------------------------------------------------------===//
void ReadAbbrevRecord();
-
+
bool ReadBlockInfoBlock();
};