LLVM  9.0.0svn
BitstreamReader.h
Go to the documentation of this file.
1 //===- BitstreamReader.h - Low-level bitstream reader interface -*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This header defines the BitstreamReader class. This class can be used to
10 // read an arbitrary bitstream, regardless of its contents.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #ifndef LLVM_BITCODE_BITSTREAMREADER_H
15 #define LLVM_BITCODE_BITSTREAMREADER_H
16 
17 #include "llvm/ADT/ArrayRef.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/Bitcode/BitCodes.h"
20 #include "llvm/Support/Endian.h"
24 #include <algorithm>
25 #include <cassert>
26 #include <climits>
27 #include <cstddef>
28 #include <cstdint>
29 #include <memory>
30 #include <string>
31 #include <utility>
32 #include <vector>
33 
34 namespace llvm {
35 
36 /// This class maintains the abbreviations read from a block info block.
38 public:
39  /// This contains information emitted to BLOCKINFO_BLOCK blocks. These
40  /// describe abbreviations that all blocks of the specified ID inherit.
41  struct BlockInfo {
42  unsigned BlockID;
43  std::vector<std::shared_ptr<BitCodeAbbrev>> Abbrevs;
44  std::string Name;
45  std::vector<std::pair<unsigned, std::string>> RecordNames;
46  };
47 
48 private:
49  std::vector<BlockInfo> BlockInfoRecords;
50 
51 public:
52  /// If there is block info for the specified ID, return it, otherwise return
53  /// null.
54  const BlockInfo *getBlockInfo(unsigned BlockID) const {
55  // Common case, the most recent entry matches BlockID.
56  if (!BlockInfoRecords.empty() && BlockInfoRecords.back().BlockID == BlockID)
57  return &BlockInfoRecords.back();
58 
59  for (unsigned i = 0, e = static_cast<unsigned>(BlockInfoRecords.size());
60  i != e; ++i)
61  if (BlockInfoRecords[i].BlockID == BlockID)
62  return &BlockInfoRecords[i];
63  return nullptr;
64  }
65 
67  if (const BlockInfo *BI = getBlockInfo(BlockID))
68  return *const_cast<BlockInfo*>(BI);
69 
70  // Otherwise, add a new record.
71  BlockInfoRecords.emplace_back();
72  BlockInfoRecords.back().BlockID = BlockID;
73  return BlockInfoRecords.back();
74  }
75 };
76 
77 /// This represents a position within a bitstream. There may be multiple
78 /// independent cursors reading within one bitstream, each maintaining their
79 /// own local state.
81  ArrayRef<uint8_t> BitcodeBytes;
82  size_t NextChar = 0;
83 
84 public:
85  /// This is the current data we have pulled from the stream but have not
86  /// returned to the client. This is specifically and intentionally defined to
87  /// follow the word size of the host machine for efficiency. We use word_t in
88  /// places that are aware of this to make it perfectly explicit what is going
89  /// on.
90  using word_t = size_t;
91 
92 private:
93  word_t CurWord = 0;
94 
95  /// This is the number of bits in CurWord that are valid. This is always from
96  /// [0...bits_of(size_t)-1] inclusive.
97  unsigned BitsInCurWord = 0;
98 
99 public:
100  static const size_t MaxChunkSize = sizeof(word_t) * 8;
101 
102  SimpleBitstreamCursor() = default;
104  : BitcodeBytes(BitcodeBytes) {}
105  explicit SimpleBitstreamCursor(StringRef BitcodeBytes)
106  : BitcodeBytes(reinterpret_cast<const uint8_t *>(BitcodeBytes.data()),
107  BitcodeBytes.size()) {}
108  explicit SimpleBitstreamCursor(MemoryBufferRef BitcodeBytes)
109  : SimpleBitstreamCursor(BitcodeBytes.getBuffer()) {}
110 
111  bool canSkipToPos(size_t pos) const {
112  // pos can be skipped to if it is a valid address or one byte past the end.
113  return pos <= BitcodeBytes.size();
114  }
115 
116  bool AtEndOfStream() {
117  return BitsInCurWord == 0 && BitcodeBytes.size() <= NextChar;
118  }
119 
120  /// Return the bit # of the bit we are reading.
121  uint64_t GetCurrentBitNo() const {
122  return NextChar*CHAR_BIT - BitsInCurWord;
123  }
124 
125  // Return the byte # of the current bit.
126  uint64_t getCurrentByteNo() const { return GetCurrentBitNo() / 8; }
127 
128  ArrayRef<uint8_t> getBitcodeBytes() const { return BitcodeBytes; }
129 
130  /// Reset the stream to the specified bit number.
131  void JumpToBit(uint64_t BitNo) {
132  size_t ByteNo = size_t(BitNo/8) & ~(sizeof(word_t)-1);
133  unsigned WordBitNo = unsigned(BitNo & (sizeof(word_t)*8-1));
134  assert(canSkipToPos(ByteNo) && "Invalid location");
135 
136  // Move the cursor to the right word.
137  NextChar = ByteNo;
138  BitsInCurWord = 0;
139 
140  // Skip over any bits that are already consumed.
141  if (WordBitNo)
142  Read(WordBitNo);
143  }
144 
145  /// Get a pointer into the bitstream at the specified byte offset.
146  const uint8_t *getPointerToByte(uint64_t ByteNo, uint64_t NumBytes) {
147  return BitcodeBytes.data() + ByteNo;
148  }
149 
150  /// Get a pointer into the bitstream at the specified bit offset.
151  ///
152  /// The bit offset must be on a byte boundary.
153  const uint8_t *getPointerToBit(uint64_t BitNo, uint64_t NumBytes) {
154  assert(!(BitNo % 8) && "Expected bit on byte boundary");
155  return getPointerToByte(BitNo / 8, NumBytes);
156  }
157 
158  void fillCurWord() {
159  if (NextChar >= BitcodeBytes.size())
160  report_fatal_error("Unexpected end of file");
161 
162  // Read the next word from the stream.
163  const uint8_t *NextCharPtr = BitcodeBytes.data() + NextChar;
164  unsigned BytesRead;
165  if (BitcodeBytes.size() >= NextChar + sizeof(word_t)) {
166  BytesRead = sizeof(word_t);
167  CurWord =
168  support::endian::read<word_t, support::little, support::unaligned>(
169  NextCharPtr);
170  } else {
171  // Short read.
172  BytesRead = BitcodeBytes.size() - NextChar;
173  CurWord = 0;
174  for (unsigned B = 0; B != BytesRead; ++B)
175  CurWord |= uint64_t(NextCharPtr[B]) << (B * 8);
176  }
177  NextChar += BytesRead;
178  BitsInCurWord = BytesRead * 8;
179  }
180 
181  word_t Read(unsigned NumBits) {
182  static const unsigned BitsInWord = MaxChunkSize;
183 
184  assert(NumBits && NumBits <= BitsInWord &&
185  "Cannot return zero or more than BitsInWord bits!");
186 
187  static const unsigned Mask = sizeof(word_t) > 4 ? 0x3f : 0x1f;
188 
189  // If the field is fully contained by CurWord, return it quickly.
190  if (BitsInCurWord >= NumBits) {
191  word_t R = CurWord & (~word_t(0) >> (BitsInWord - NumBits));
192 
193  // Use a mask to avoid undefined behavior.
194  CurWord >>= (NumBits & Mask);
195 
196  BitsInCurWord -= NumBits;
197  return R;
198  }
199 
200  word_t R = BitsInCurWord ? CurWord : 0;
201  unsigned BitsLeft = NumBits - BitsInCurWord;
202 
203  fillCurWord();
204 
205  // If we run out of data, abort.
206  if (BitsLeft > BitsInCurWord)
207  report_fatal_error("Unexpected end of file");
208 
209  word_t R2 = CurWord & (~word_t(0) >> (BitsInWord - BitsLeft));
210 
211  // Use a mask to avoid undefined behavior.
212  CurWord >>= (BitsLeft & Mask);
213 
214  BitsInCurWord -= BitsLeft;
215 
216  R |= R2 << (NumBits - BitsLeft);
217 
218  return R;
219  }
220 
221  uint32_t ReadVBR(unsigned NumBits) {
222  uint32_t Piece = Read(NumBits);
223  if ((Piece & (1U << (NumBits-1))) == 0)
224  return Piece;
225 
226  uint32_t Result = 0;
227  unsigned NextBit = 0;
228  while (true) {
229  Result |= (Piece & ((1U << (NumBits-1))-1)) << NextBit;
230 
231  if ((Piece & (1U << (NumBits-1))) == 0)
232  return Result;
233 
234  NextBit += NumBits-1;
235  Piece = Read(NumBits);
236  }
237  }
238 
239  // Read a VBR that may have a value up to 64-bits in size. The chunk size of
240  // the VBR must still be <= 32 bits though.
241  uint64_t ReadVBR64(unsigned NumBits) {
242  uint32_t Piece = Read(NumBits);
243  if ((Piece & (1U << (NumBits-1))) == 0)
244  return uint64_t(Piece);
245 
246  uint64_t Result = 0;
247  unsigned NextBit = 0;
248  while (true) {
249  Result |= uint64_t(Piece & ((1U << (NumBits-1))-1)) << NextBit;
250 
251  if ((Piece & (1U << (NumBits-1))) == 0)
252  return Result;
253 
254  NextBit += NumBits-1;
255  Piece = Read(NumBits);
256  }
257  }
258 
260  // If word_t is 64-bits and if we've read less than 32 bits, just dump
261  // the bits we have up to the next 32-bit boundary.
262  if (sizeof(word_t) > 4 &&
263  BitsInCurWord >= 32) {
264  CurWord >>= BitsInCurWord-32;
265  BitsInCurWord = 32;
266  return;
267  }
268 
269  BitsInCurWord = 0;
270  }
271 
272  /// Skip to the end of the file.
273  void skipToEnd() { NextChar = BitcodeBytes.size(); }
274 };
275 
276 /// When advancing through a bitstream cursor, each advance can discover a few
277 /// different kinds of entries:
279  enum {
280  Error, // Malformed bitcode was found.
281  EndBlock, // We've reached the end of the current block, (or the end of the
282  // file, which is treated like a series of EndBlock records.
283  SubBlock, // This is the start of a new subblock of a specific ID.
284  Record // This is a record with a specific AbbrevID.
285  } Kind;
286 
287  unsigned ID;
288 
290  BitstreamEntry E; E.Kind = Error; return E;
291  }
292 
294  BitstreamEntry E; E.Kind = EndBlock; return E;
295  }
296 
297  static BitstreamEntry getSubBlock(unsigned ID) {
298  BitstreamEntry E; E.Kind = SubBlock; E.ID = ID; return E;
299  }
300 
301  static BitstreamEntry getRecord(unsigned AbbrevID) {
302  BitstreamEntry E; E.Kind = Record; E.ID = AbbrevID; return E;
303  }
304 };
305 
306 /// This represents a position within a bitcode file, implemented on top of a
307 /// SimpleBitstreamCursor.
308 ///
309 /// Unlike iterators, BitstreamCursors are heavy-weight objects that should not
310 /// be passed by value.
312  // This is the declared size of code values used for the current block, in
313  // bits.
314  unsigned CurCodeSize = 2;
315 
316  /// Abbrevs installed at in this block.
317  std::vector<std::shared_ptr<BitCodeAbbrev>> CurAbbrevs;
318 
319  struct Block {
320  unsigned PrevCodeSize;
321  std::vector<std::shared_ptr<BitCodeAbbrev>> PrevAbbrevs;
322 
323  explicit Block(unsigned PCS) : PrevCodeSize(PCS) {}
324  };
325 
326  /// This tracks the codesize of parent blocks.
327  SmallVector<Block, 8> BlockScope;
328 
329  BitstreamBlockInfo *BlockInfo = nullptr;
330 
331 public:
332  static const size_t MaxChunkSize = sizeof(word_t) * 8;
333 
334  BitstreamCursor() = default;
335  explicit BitstreamCursor(ArrayRef<uint8_t> BitcodeBytes)
336  : SimpleBitstreamCursor(BitcodeBytes) {}
337  explicit BitstreamCursor(StringRef BitcodeBytes)
338  : SimpleBitstreamCursor(BitcodeBytes) {}
339  explicit BitstreamCursor(MemoryBufferRef BitcodeBytes)
340  : SimpleBitstreamCursor(BitcodeBytes) {}
341 
353 
354  /// Return the number of bits used to encode an abbrev #.
355  unsigned getAbbrevIDWidth() const { return CurCodeSize; }
356 
357  /// Flags that modify the behavior of advance().
358  enum {
359  /// If this flag is used, the advance() method does not automatically pop
360  /// the block scope when the end of a block is reached.
361  AF_DontPopBlockAtEnd = 1,
362 
363  /// If this flag is used, abbrev entries are returned just like normal
364  /// records.
365  AF_DontAutoprocessAbbrevs = 2
366  };
367 
368  /// Advance the current bitstream, returning the next entry in the stream.
369  BitstreamEntry advance(unsigned Flags = 0) {
370  while (true) {
371  if (AtEndOfStream())
372  return BitstreamEntry::getError();
373 
374  unsigned Code = ReadCode();
375  if (Code == bitc::END_BLOCK) {
376  // Pop the end of the block unless Flags tells us not to.
377  if (!(Flags & AF_DontPopBlockAtEnd) && ReadBlockEnd())
378  return BitstreamEntry::getError();
380  }
381 
382  if (Code == bitc::ENTER_SUBBLOCK)
383  return BitstreamEntry::getSubBlock(ReadSubBlockID());
384 
385  if (Code == bitc::DEFINE_ABBREV &&
386  !(Flags & AF_DontAutoprocessAbbrevs)) {
387  // We read and accumulate abbrev's, the client can't do anything with
388  // them anyway.
389  ReadAbbrevRecord();
390  continue;
391  }
392 
393  return BitstreamEntry::getRecord(Code);
394  }
395  }
396 
397  /// This is a convenience function for clients that don't expect any
398  /// subblocks. This just skips over them automatically.
400  while (true) {
401  // If we found a normal entry, return it.
402  BitstreamEntry Entry = advance(Flags);
403  if (Entry.Kind != BitstreamEntry::SubBlock)
404  return Entry;
405 
406  // If we found a sub-block, just skip over it and check the next entry.
407  if (SkipBlock())
408  return BitstreamEntry::getError();
409  }
410  }
411 
412  unsigned ReadCode() {
413  return Read(CurCodeSize);
414  }
415 
416  // Block header:
417  // [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen]
418 
419  /// Having read the ENTER_SUBBLOCK code, read the BlockID for the block.
420  unsigned ReadSubBlockID() {
421  return ReadVBR(bitc::BlockIDWidth);
422  }
423 
424  /// Having read the ENTER_SUBBLOCK abbrevid and a BlockID, skip over the body
425  /// of this block. If the block record is malformed, return true.
426  bool SkipBlock() {
427  // Read and ignore the codelen value. Since we are skipping this block, we
428  // don't care what code widths are used inside of it.
429  ReadVBR(bitc::CodeLenWidth);
430  SkipToFourByteBoundary();
431  size_t NumFourBytes = Read(bitc::BlockSizeWidth);
432 
433  // Check that the block wasn't partially defined, and that the offset isn't
434  // bogus.
435  size_t SkipTo = GetCurrentBitNo() + NumFourBytes*4*8;
436  if (AtEndOfStream() || !canSkipToPos(SkipTo/8))
437  return true;
438 
439  JumpToBit(SkipTo);
440  return false;
441  }
442 
443  /// Having read the ENTER_SUBBLOCK abbrevid, enter the block, and return true
444  /// if the block has an error.
445  bool EnterSubBlock(unsigned BlockID, unsigned *NumWordsP = nullptr);
446 
447  bool ReadBlockEnd() {
448  if (BlockScope.empty()) return true;
449 
450  // Block tail:
451  // [END_BLOCK, <align4bytes>]
452  SkipToFourByteBoundary();
453 
454  popBlockScope();
455  return false;
456  }
457 
458 private:
459  void popBlockScope() {
460  CurCodeSize = BlockScope.back().PrevCodeSize;
461 
462  CurAbbrevs = std::move(BlockScope.back().PrevAbbrevs);
463  BlockScope.pop_back();
464  }
465 
466  //===--------------------------------------------------------------------===//
467  // Record Processing
468  //===--------------------------------------------------------------------===//
469 
470 public:
471  /// Return the abbreviation for the specified AbbrevId.
472  const BitCodeAbbrev *getAbbrev(unsigned AbbrevID) {
473  unsigned AbbrevNo = AbbrevID - bitc::FIRST_APPLICATION_ABBREV;
474  if (AbbrevNo >= CurAbbrevs.size())
475  report_fatal_error("Invalid abbrev number");
476  return CurAbbrevs[AbbrevNo].get();
477  }
478 
479  /// Read the current record and discard it, returning the code for the record.
480  unsigned skipRecord(unsigned AbbrevID);
481 
482  unsigned readRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
483  StringRef *Blob = nullptr);
484 
485  //===--------------------------------------------------------------------===//
486  // Abbrev Processing
487  //===--------------------------------------------------------------------===//
488  void ReadAbbrevRecord();
489 
490  /// Read and return a block info block from the bitstream. If an error was
491  /// encountered, return None.
492  ///
493  /// \param ReadBlockInfoNames Whether to read block/record name information in
494  /// the BlockInfo block. Only llvm-bcanalyzer uses this.
496  ReadBlockInfoBlock(bool ReadBlockInfoNames = false);
497 
498  /// Set the block info to be used by this BitstreamCursor to interpret
499  /// abbreviated records.
500  void setBlockInfo(BitstreamBlockInfo *BI) { BlockInfo = BI; }
501 };
502 
503 } // end llvm namespace
504 
505 #endif // LLVM_BITCODE_BITSTREAMREADER_H
This contains information emitted to BLOCKINFO_BLOCK blocks.
const BitCodeAbbrev * getAbbrev(unsigned AbbrevID)
Return the abbreviation for the specified AbbrevId.
BlockInfo & getOrCreateBlockInfo(unsigned BlockID)
LLVM_ATTRIBUTE_NORETURN void report_fatal_error(Error Err, bool gen_crash_diag=true)
Report a serious error, calling any installed error handler.
Definition: Error.cpp:139
This class represents lattice values for constants.
Definition: AllocatorList.h:23
uint64_t GetCurrentBitNo() const
Return the bit # of the bit we are reading.
const uint8_t * getPointerToByte(uint64_t ByteNo, uint64_t NumBytes)
Get a pointer into the bitstream at the specified byte offset.
BitCodeAbbrev - This class represents an abbreviation record.
Definition: BitCodes.h:165
const uint8_t * getPointerToBit(uint64_t BitNo, uint64_t NumBytes)
Get a pointer into the bitstream at the specified bit offset.
BitstreamCursor(MemoryBufferRef BitcodeBytes)
#define R2(n)
SimpleBitstreamCursor(ArrayRef< uint8_t > BitcodeBytes)
std::vector< std::pair< unsigned, std::string > > RecordNames
void setBlockInfo(BitstreamBlockInfo *BI)
Set the block info to be used by this BitstreamCursor to interpret abbreviated records.
word_t Read(unsigned NumBits)
bool canSkipToPos(size_t pos) const
uint64_t ReadVBR64(unsigned NumBits)
void skipToEnd()
Skip to the end of the file.
uint32_t ReadVBR(unsigned NumBits)
static BitstreamEntry getSubBlock(unsigned ID)
enum llvm::BitstreamEntry::@154 Kind
ArrayRef< uint8_t > getBitcodeBytes() const
BitstreamEntry advanceSkippingSubblocks(unsigned Flags=0)
This is a convenience function for clients that don&#39;t expect any subblocks.
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
uint64_t getCurrentByteNo() const
size_t size() const
size - Get the array size.
Definition: ArrayRef.h:148
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
This represents a position within a bitcode file, implemented on top of a SimpleBitstreamCursor.
const BlockInfo * getBlockInfo(unsigned BlockID) const
If there is block info for the specified ID, return it, otherwise return null.
void JumpToBit(uint64_t BitNo)
Reset the stream to the specified bit number.
unsigned getAbbrevIDWidth() const
Return the number of bits used to encode an abbrev #.
bool SkipBlock()
Having read the ENTER_SUBBLOCK abbrevid and a BlockID, skip over the body of this block...
const T * data() const
Definition: ArrayRef.h:145
std::vector< std::shared_ptr< BitCodeAbbrev > > Abbrevs
SimpleBitstreamCursor(MemoryBufferRef BitcodeBytes)
unsigned ReadSubBlockID()
Having read the ENTER_SUBBLOCK code, read the BlockID for the block.
When advancing through a bitstream cursor, each advance can discover a few different kinds of entries...
auto size(R &&Range, typename std::enable_if< std::is_same< typename std::iterator_traits< decltype(Range.begin())>::iterator_category, std::random_access_iterator_tag >::value, void >::type *=nullptr) -> decltype(std::distance(Range.begin(), Range.end()))
Get the size of a range.
Definition: STLExtras.h:1166
BitstreamCursor(ArrayRef< uint8_t > BitcodeBytes)
This class maintains the abbreviations read from a block info block.
DEFINE_ABBREV - Defines an abbrev for the current block.
Definition: BitCodes.h:52
LLVM_NODISCARD bool empty() const
Definition: SmallVector.h:55
static BitstreamEntry getEndBlock()
static BitstreamEntry getRecord(unsigned AbbrevID)
SimpleBitstreamCursor(StringRef BitcodeBytes)
const unsigned Kind
static BitstreamEntry getError()
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
aarch64 promote const
std::underlying_type< E >::type Mask()
Get a bitmask with 1s in all places up to the high-order bit of E&#39;s largest value.
Definition: BitmaskEnum.h:80
BitstreamEntry advance(unsigned Flags=0)
Advance the current bitstream, returning the next entry in the stream.
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:48
This represents a position within a bitstream.
BitstreamCursor(StringRef BitcodeBytes)