Line data Source code
1 : //===- BitstreamReader.h - Low-level bitstream reader interface -*- C++ -*-===//
2 : //
3 : // The LLVM Compiler Infrastructure
4 : //
5 : // This file is distributed under the University of Illinois Open Source
6 : // License. See LICENSE.TXT for details.
7 : //
8 : //===----------------------------------------------------------------------===//
9 : //
10 : // This header defines the BitstreamReader class. This class can be used to
11 : // read an arbitrary bitstream, regardless of its contents.
12 : //
13 : //===----------------------------------------------------------------------===//
14 :
15 : #ifndef LLVM_BITCODE_BITSTREAMREADER_H
16 : #define LLVM_BITCODE_BITSTREAMREADER_H
17 :
18 : #include "llvm/ADT/ArrayRef.h"
19 : #include "llvm/ADT/SmallVector.h"
20 : #include "llvm/Bitcode/BitCodes.h"
21 : #include "llvm/Support/Endian.h"
22 : #include "llvm/Support/ErrorHandling.h"
23 : #include "llvm/Support/MathExtras.h"
24 : #include "llvm/Support/MemoryBuffer.h"
25 : #include <algorithm>
26 : #include <cassert>
27 : #include <climits>
28 : #include <cstddef>
29 : #include <cstdint>
30 : #include <memory>
31 : #include <string>
32 : #include <utility>
33 : #include <vector>
34 :
35 : namespace llvm {
36 :
37 : /// This class maintains the abbreviations read from a block info block.
38 12451 : class BitstreamBlockInfo {
39 : public:
40 : /// This contains information emitted to BLOCKINFO_BLOCK blocks. These
41 : /// describe abbreviations that all blocks of the specified ID inherit.
42 : struct BlockInfo {
43 : unsigned BlockID;
44 : std::vector<std::shared_ptr<BitCodeAbbrev>> Abbrevs;
45 : std::string Name;
46 : std::vector<std::pair<unsigned, std::string>> RecordNames;
47 : };
48 :
49 : private:
50 : std::vector<BlockInfo> BlockInfoRecords;
51 :
52 : public:
53 : /// If there is block info for the specified ID, return it, otherwise return
54 : /// null.
55 : const BlockInfo *getBlockInfo(unsigned BlockID) const {
56 : // Common case, the most recent entry matches BlockID.
57 93038 : if (!BlockInfoRecords.empty() && BlockInfoRecords.back().BlockID == BlockID)
58 : return &BlockInfoRecords.back();
59 :
60 204637 : for (unsigned i = 0, e = static_cast<unsigned>(BlockInfoRecords.size());
61 204637 : i != e; ++i)
62 308540 : if (BlockInfoRecords[i].BlockID == BlockID)
63 : return &BlockInfoRecords[i];
64 : return nullptr;
65 : }
66 :
67 12576 : BlockInfo &getOrCreateBlockInfo(unsigned BlockID) {
68 12576 : if (const BlockInfo *BI = getBlockInfo(BlockID))
69 : return *const_cast<BlockInfo*>(BI);
70 :
71 : // Otherwise, add a new record.
72 12444 : BlockInfoRecords.emplace_back();
73 12444 : BlockInfoRecords.back().BlockID = BlockID;
74 12444 : return BlockInfoRecords.back();
75 : }
76 : };
77 :
78 : /// This represents a position within a bitstream. There may be multiple
79 : /// independent cursors reading within one bitstream, each maintaining their
80 : /// own local state.
81 : class SimpleBitstreamCursor {
82 : ArrayRef<uint8_t> BitcodeBytes;
83 : size_t NextChar = 0;
84 :
85 : public:
86 : /// This is the current data we have pulled from the stream but have not
87 : /// returned to the client. This is specifically and intentionally defined to
88 : /// follow the word size of the host machine for efficiency. We use word_t in
89 : /// places that are aware of this to make it perfectly explicit what is going
90 : /// on.
91 : using word_t = size_t;
92 :
93 : private:
94 : word_t CurWord = 0;
95 :
96 : /// This is the number of bits in CurWord that are valid. This is always from
97 : /// [0...bits_of(size_t)-1] inclusive.
98 : unsigned BitsInCurWord = 0;
99 :
100 : public:
101 : static const size_t MaxChunkSize = sizeof(word_t) * 8;
102 :
103 40549 : SimpleBitstreamCursor() = default;
104 : explicit SimpleBitstreamCursor(ArrayRef<uint8_t> BitcodeBytes)
105 5920 : : BitcodeBytes(BitcodeBytes) {}
106 : explicit SimpleBitstreamCursor(StringRef BitcodeBytes)
107 18098 : : BitcodeBytes(reinterpret_cast<const uint8_t *>(BitcodeBytes.data()),
108 18098 : BitcodeBytes.size()) {}
109 : explicit SimpleBitstreamCursor(MemoryBufferRef BitcodeBytes)
110 : : SimpleBitstreamCursor(BitcodeBytes.getBuffer()) {}
111 :
112 : bool canSkipToPos(size_t pos) const {
113 : // pos can be skipped to if it is a valid address or one byte past the end.
114 330008 : return pos <= BitcodeBytes.size();
115 : }
116 :
117 : bool AtEndOfStream() {
118 2942043 : return BitsInCurWord == 0 && BitcodeBytes.size() <= NextChar;
119 : }
120 :
121 : /// Return the bit # of the bit we are reading.
122 0 : uint64_t GetCurrentBitNo() const {
123 1498152 : return NextChar*CHAR_BIT - BitsInCurWord;
124 : }
125 :
126 : // Return the byte # of the current bit.
127 23251 : uint64_t getCurrentByteNo() const { return GetCurrentBitNo() / 8; }
128 :
129 0 : ArrayRef<uint8_t> getBitcodeBytes() const { return BitcodeBytes; }
130 :
131 : /// Reset the stream to the specified bit number.
132 : void JumpToBit(uint64_t BitNo) {
133 1592693 : size_t ByteNo = size_t(BitNo/8) & ~(sizeof(word_t)-1);
134 1592693 : unsigned WordBitNo = unsigned(BitNo & (sizeof(word_t)*8-1));
135 : assert(canSkipToPos(ByteNo) && "Invalid location");
136 :
137 : // Move the cursor to the right word.
138 1592695 : NextChar = ByteNo;
139 1587044 : BitsInCurWord = 0;
140 :
141 : // Skip over any bits that are already consumed.
142 1592693 : if (WordBitNo)
143 1348316 : Read(WordBitNo);
144 : }
145 :
146 : /// Get a pointer into the bitstream at the specified byte offset.
147 0 : const uint8_t *getPointerToByte(uint64_t ByteNo, uint64_t NumBytes) {
148 229823 : return BitcodeBytes.data() + ByteNo;
149 : }
150 :
151 : /// Get a pointer into the bitstream at the specified bit offset.
152 : ///
153 : /// The bit offset must be on a byte boundary.
154 0 : const uint8_t *getPointerToBit(uint64_t BitNo, uint64_t NumBytes) {
155 : assert(!(BitNo % 8) && "Expected bit on byte boundary");
156 229821 : return getPointerToByte(BitNo / 8, NumBytes);
157 : }
158 :
159 6649145 : void fillCurWord() {
160 6649145 : if (NextChar >= BitcodeBytes.size())
161 1 : report_fatal_error("Unexpected end of file");
162 :
163 : // Read the next word from the stream.
164 6649144 : const uint8_t *NextCharPtr = BitcodeBytes.data() + NextChar;
165 : unsigned BytesRead;
166 6649144 : if (BitcodeBytes.size() >= NextChar + sizeof(word_t)) {
167 : BytesRead = sizeof(word_t);
168 6636250 : CurWord =
169 : support::endian::read<word_t, support::little, support::unaligned>(
170 : NextCharPtr);
171 : } else {
172 : // Short read.
173 12894 : BytesRead = BitcodeBytes.size() - NextChar;
174 12894 : CurWord = 0;
175 64470 : for (unsigned B = 0; B != BytesRead; ++B)
176 51576 : CurWord |= uint64_t(NextCharPtr[B]) << (B * 8);
177 : }
178 6649144 : NextChar += BytesRead;
179 6649144 : BitsInCurWord = BytesRead * 8;
180 6649144 : }
181 :
182 61241257 : word_t Read(unsigned NumBits) {
183 : static const unsigned BitsInWord = MaxChunkSize;
184 :
185 : assert(NumBits && NumBits <= BitsInWord &&
186 : "Cannot return zero or more than BitsInWord bits!");
187 :
188 : static const unsigned Mask = sizeof(word_t) > 4 ? 0x3f : 0x1f;
189 :
190 : // If the field is fully contained by CurWord, return it quickly.
191 61241257 : if (BitsInCurWord >= NumBits) {
192 54592112 : word_t R = CurWord & (~word_t(0) >> (BitsInWord - NumBits));
193 :
194 : // Use a mask to avoid undefined behavior.
195 54592112 : CurWord >>= (NumBits & Mask);
196 :
197 54592112 : BitsInCurWord -= NumBits;
198 54592112 : return R;
199 : }
200 :
201 6649145 : word_t R = BitsInCurWord ? CurWord : 0;
202 6649145 : unsigned BitsLeft = NumBits - BitsInCurWord;
203 :
204 6649145 : fillCurWord();
205 :
206 : // If we run out of data, abort.
207 6649145 : if (BitsLeft > BitsInCurWord)
208 0 : report_fatal_error("Unexpected end of file");
209 :
210 6649145 : word_t R2 = CurWord & (~word_t(0) >> (BitsInWord - BitsLeft));
211 :
212 : // Use a mask to avoid undefined behavior.
213 6649145 : CurWord >>= (BitsLeft & Mask);
214 :
215 6649145 : BitsInCurWord -= BitsLeft;
216 :
217 6649145 : R |= R2 << (NumBits - BitsLeft);
218 :
219 6649145 : return R;
220 : }
221 :
222 4604856 : uint32_t ReadVBR(unsigned NumBits) {
223 4604856 : uint32_t Piece = Read(NumBits);
224 4604860 : if ((Piece & (1U << (NumBits-1))) == 0)
225 : return Piece;
226 :
227 : uint32_t Result = 0;
228 : unsigned NextBit = 0;
229 : while (true) {
230 3809665 : Result |= (Piece & ((1U << (NumBits-1))-1)) << NextBit;
231 :
232 2533420 : if ((Piece & (1U << (NumBits-1))) == 0)
233 1257175 : return Result;
234 :
235 1276245 : NextBit += NumBits-1;
236 1276245 : Piece = Read(NumBits);
237 : }
238 : }
239 :
240 : // Read a VBR that may have a value up to 64-bits in size. The chunk size of
241 : // the VBR must still be <= 32 bits though.
242 27414759 : uint64_t ReadVBR64(unsigned NumBits) {
243 27414759 : uint32_t Piece = Read(NumBits);
244 27414757 : if ((Piece & (1U << (NumBits-1))) == 0)
245 17832871 : return uint64_t(Piece);
246 :
247 : uint64_t Result = 0;
248 : unsigned NextBit = 0;
249 : while (true) {
250 34724428 : Result |= uint64_t(Piece & ((1U << (NumBits-1))-1)) << NextBit;
251 :
252 22153157 : if ((Piece & (1U << (NumBits-1))) == 0)
253 9581892 : return Result;
254 :
255 12571265 : NextBit += NumBits-1;
256 12571265 : Piece = Read(NumBits);
257 : }
258 : }
259 :
260 0 : void SkipToFourByteBoundary() {
261 : // If word_t is 64-bits and if we've read less than 32 bits, just dump
262 : // the bits we have up to the next 32-bit boundary.
263 595347 : if (sizeof(word_t) > 4 &&
264 595347 : BitsInCurWord >= 32) {
265 270169 : CurWord >>= BitsInCurWord-32;
266 270169 : BitsInCurWord = 32;
267 0 : return;
268 : }
269 :
270 325178 : BitsInCurWord = 0;
271 : }
272 :
273 : /// Skip to the end of the file.
274 0 : void skipToEnd() { NextChar = BitcodeBytes.size(); }
275 : };
276 :
277 : /// When advancing through a bitstream cursor, each advance can discover a few
278 : /// different kinds of entries:
279 : struct BitstreamEntry {
280 : enum {
281 : Error, // Malformed bitcode was found.
282 : EndBlock, // We've reached the end of the current block, (or the end of the
283 : // file, which is treated like a series of EndBlock records.
284 : SubBlock, // This is the start of a new subblock of a specific ID.
285 : Record // This is a record with a specific AbbrevID.
286 : } Kind;
287 :
288 : unsigned ID;
289 :
290 : static BitstreamEntry getError() {
291 : BitstreamEntry E; E.Kind = Error; return E;
292 : }
293 :
294 : static BitstreamEntry getEndBlock() {
295 : BitstreamEntry E; E.Kind = EndBlock; return E;
296 : }
297 :
298 : static BitstreamEntry getSubBlock(unsigned ID) {
299 : BitstreamEntry E; E.Kind = SubBlock; E.ID = ID; return E;
300 : }
301 :
302 : static BitstreamEntry getRecord(unsigned AbbrevID) {
303 : BitstreamEntry E; E.Kind = Record; E.ID = AbbrevID; return E;
304 : }
305 : };
306 :
307 : /// This represents a position within a bitcode file, implemented on top of a
308 : /// SimpleBitstreamCursor.
309 : ///
310 : /// Unlike iterators, BitstreamCursors are heavy-weight objects that should not
311 : /// be passed by value.
312 73773 : class BitstreamCursor : SimpleBitstreamCursor {
313 : // This is the declared size of code values used for the current block, in
314 : // bits.
315 : unsigned CurCodeSize = 2;
316 :
317 : /// Abbrevs installed at in this block.
318 : std::vector<std::shared_ptr<BitCodeAbbrev>> CurAbbrevs;
319 :
320 398682 : struct Block {
321 : unsigned PrevCodeSize;
322 : std::vector<std::shared_ptr<BitCodeAbbrev>> PrevAbbrevs;
323 :
324 142725 : explicit Block(unsigned PCS) : PrevCodeSize(PCS) {}
325 : };
326 :
327 : /// This tracks the codesize of parent blocks.
328 : SmallVector<Block, 8> BlockScope;
329 :
330 : BitstreamBlockInfo *BlockInfo = nullptr;
331 :
332 : public:
333 : static const size_t MaxChunkSize = sizeof(word_t) * 8;
334 :
335 81098 : BitstreamCursor() = default;
336 : explicit BitstreamCursor(ArrayRef<uint8_t> BitcodeBytes)
337 11920 : : SimpleBitstreamCursor(BitcodeBytes) {}
338 : explicit BitstreamCursor(StringRef BitcodeBytes)
339 49503 : : SimpleBitstreamCursor(BitcodeBytes) {}
340 : explicit BitstreamCursor(MemoryBufferRef BitcodeBytes)
341 600 : : SimpleBitstreamCursor(BitcodeBytes) {}
342 :
343 : using SimpleBitstreamCursor::canSkipToPos;
344 : using SimpleBitstreamCursor::AtEndOfStream;
345 : using SimpleBitstreamCursor::getBitcodeBytes;
346 : using SimpleBitstreamCursor::GetCurrentBitNo;
347 : using SimpleBitstreamCursor::getCurrentByteNo;
348 : using SimpleBitstreamCursor::getPointerToByte;
349 : using SimpleBitstreamCursor::JumpToBit;
350 : using SimpleBitstreamCursor::fillCurWord;
351 : using SimpleBitstreamCursor::Read;
352 : using SimpleBitstreamCursor::ReadVBR;
353 : using SimpleBitstreamCursor::ReadVBR64;
354 :
355 : /// Return the number of bits used to encode an abbrev #.
356 0 : unsigned getAbbrevIDWidth() const { return CurCodeSize; }
357 :
358 : /// Flags that modify the behavior of advance().
359 : enum {
360 : /// If this flag is used, the advance() method does not automatically pop
361 : /// the block scope when the end of a block is reached.
362 : AF_DontPopBlockAtEnd = 1,
363 :
364 : /// If this flag is used, abbrev entries are returned just like normal
365 : /// records.
366 : AF_DontAutoprocessAbbrevs = 2
367 : };
368 :
369 : /// Advance the current bitstream, returning the next entry in the stream.
370 2458984 : BitstreamEntry advance(unsigned Flags = 0) {
371 : while (true) {
372 : if (AtEndOfStream())
373 : return BitstreamEntry::getError();
374 :
375 : unsigned Code = ReadCode();
376 2720688 : if (Code == bitc::END_BLOCK) {
377 : // Pop the end of the block unless Flags tells us not to.
378 108350 : if (!(Flags & AF_DontPopBlockAtEnd) && ReadBlockEnd())
379 : return BitstreamEntry::getError();
380 : return BitstreamEntry::getEndBlock();
381 : }
382 :
383 2612338 : if (Code == bitc::ENTER_SUBBLOCK)
384 : return BitstreamEntry::getSubBlock(ReadSubBlockID());
385 :
386 2408503 : if (Code == bitc::DEFINE_ABBREV &&
387 337926 : !(Flags & AF_DontAutoprocessAbbrevs)) {
388 : // We read and accumulate abbrev's, the client can't do anything with
389 : // them anyway.
390 268951 : ReadAbbrevRecord();
391 : continue;
392 : }
393 :
394 : return BitstreamEntry::getRecord(Code);
395 268949 : }
396 : }
397 :
398 : /// This is a convenience function for clients that don't expect any
399 : /// subblocks. This just skips over them automatically.
400 1491342 : BitstreamEntry advanceSkippingSubblocks(unsigned Flags = 0) {
401 : while (true) {
402 : // If we found a normal entry, return it.
403 1495109 : BitstreamEntry Entry = advance(Flags);
404 1495109 : if (Entry.Kind != BitstreamEntry::SubBlock)
405 1491342 : return Entry;
406 :
407 : // If we found a sub-block, just skip over it and check the next entry.
408 3767 : if (SkipBlock())
409 : return BitstreamEntry::getError();
410 3767 : }
411 : }
412 :
413 : unsigned ReadCode() {
414 3313571 : return Read(CurCodeSize);
415 : }
416 :
417 : // Block header:
418 : // [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen]
419 :
420 : /// Having read the ENTER_SUBBLOCK code, read the BlockID for the block.
421 : unsigned ReadSubBlockID() {
422 204418 : return ReadVBR(bitc::BlockIDWidth);
423 : }
424 :
425 : /// Having read the ENTER_SUBBLOCK abbrevid and a BlockID, skip over the body
426 : /// of this block. If the block record is malformed, return true.
427 114865 : bool SkipBlock() {
428 : // Read and ignore the codelen value. Since we are skipping this block, we
429 : // don't care what code widths are used inside of it.
430 114865 : ReadVBR(bitc::CodeLenWidth);
431 : SkipToFourByteBoundary();
432 114865 : size_t NumFourBytes = Read(bitc::BlockSizeWidth);
433 :
434 : // Check that the block wasn't partially defined, and that the offset isn't
435 : // bogus.
436 114865 : size_t SkipTo = GetCurrentBitNo() + NumFourBytes*4*8;
437 114865 : if (AtEndOfStream() || !canSkipToPos(SkipTo/8))
438 : return true;
439 :
440 : JumpToBit(SkipTo);
441 : return false;
442 : }
443 :
444 : /// Having read the ENTER_SUBBLOCK abbrevid, enter the block, and return true
445 : /// if the block has an error.
446 : bool EnterSubBlock(unsigned BlockID, unsigned *NumWordsP = nullptr);
447 :
448 107647 : bool ReadBlockEnd() {
449 107647 : if (BlockScope.empty()) return true;
450 :
451 : // Block tail:
452 : // [END_BLOCK, <align4bytes>]
453 : SkipToFourByteBoundary();
454 :
455 107640 : popBlockScope();
456 107640 : return false;
457 : }
458 :
459 : private:
460 107640 : void popBlockScope() {
461 107640 : CurCodeSize = BlockScope.back().PrevCodeSize;
462 :
463 107640 : CurAbbrevs = std::move(BlockScope.back().PrevAbbrevs);
464 : BlockScope.pop_back();
465 107640 : }
466 :
467 : //===--------------------------------------------------------------------===//
468 : // Record Processing
469 : //===--------------------------------------------------------------------===//
470 :
471 : public:
472 : /// Return the abbreviation for the specified AbbrevId.
473 916960 : const BitCodeAbbrev *getAbbrev(unsigned AbbrevID) {
474 916960 : unsigned AbbrevNo = AbbrevID - bitc::FIRST_APPLICATION_ABBREV;
475 1833920 : if (AbbrevNo >= CurAbbrevs.size())
476 0 : report_fatal_error("Invalid abbrev number");
477 916960 : return CurAbbrevs[AbbrevNo].get();
478 : }
479 :
480 : /// Read the current record and discard it, returning the code for the record.
481 : unsigned skipRecord(unsigned AbbrevID);
482 :
483 : unsigned readRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals,
484 : StringRef *Blob = nullptr);
485 :
486 : //===--------------------------------------------------------------------===//
487 : // Abbrev Processing
488 : //===--------------------------------------------------------------------===//
489 : void ReadAbbrevRecord();
490 :
491 : /// Read and return a block info block from the bitstream. If an error was
492 : /// encountered, return None.
493 : ///
494 : /// \param ReadBlockInfoNames Whether to read block/record name information in
495 : /// the BlockInfo block. Only llvm-bcanalyzer uses this.
496 : Optional<BitstreamBlockInfo>
497 : ReadBlockInfoBlock(bool ReadBlockInfoNames = false);
498 :
499 : /// Set the block info to be used by this BitstreamCursor to interpret
500 : /// abbreviated records.
501 4325 : void setBlockInfo(BitstreamBlockInfo *BI) { BlockInfo = BI; }
502 : };
503 :
504 : } // end llvm namespace
505 :
506 : #endif // LLVM_BITCODE_BITSTREAMREADER_H
|
