BitstreamReader.h revision dc60fc1aa710812be17c5b01f34ef251c87b42b4
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/OwningPtr.h" 19#include "llvm/Bitcode/BitCodes.h" 20#include "llvm/Support/Endian.h" 21#include "llvm/Support/StreamableMemoryObject.h" 22#include <climits> 23#include <string> 24#include <vector> 25 26namespace llvm { 27 28 class Deserializer; 29 30/// BitstreamReader - This class is used to read from an LLVM bitcode stream, 31/// maintaining information that is global to decoding the entire file. While 32/// a file is being read, multiple cursors can be independently advanced or 33/// skipped around within the file. These are represented by the 34/// BitstreamCursor class. 35class BitstreamReader { 36public: 37 /// BlockInfo - This contains information emitted to BLOCKINFO_BLOCK blocks. 38 /// These describe abbreviations that all blocks of the specified ID inherit. 39 struct BlockInfo { 40 unsigned BlockID; 41 std::vector<BitCodeAbbrev*> Abbrevs; 42 std::string Name; 43 44 std::vector<std::pair<unsigned, std::string> > RecordNames; 45 }; 46private: 47 OwningPtr<StreamableMemoryObject> BitcodeBytes; 48 49 std::vector<BlockInfo> BlockInfoRecords; 50 51 /// IgnoreBlockInfoNames - This is set to true if we don't care about the 52 /// block/record name information in the BlockInfo block. Only llvm-bcanalyzer 53 /// uses this. 54 bool IgnoreBlockInfoNames; 55 56 BitstreamReader(const BitstreamReader&) LLVM_DELETED_FUNCTION; 57 void operator=(const BitstreamReader&) LLVM_DELETED_FUNCTION; 58public: 59 BitstreamReader() : IgnoreBlockInfoNames(true) { 60 } 61 62 BitstreamReader(const unsigned char *Start, const unsigned char *End) { 63 IgnoreBlockInfoNames = true; 64 init(Start, End); 65 } 66 67 BitstreamReader(StreamableMemoryObject *bytes) { 68 BitcodeBytes.reset(bytes); 69 } 70 71 void init(const unsigned char *Start, const unsigned char *End) { 72 assert(((End-Start) & 3) == 0 &&"Bitcode stream not a multiple of 4 bytes"); 73 BitcodeBytes.reset(getNonStreamedMemoryObject(Start, End)); 74 } 75 76 StreamableMemoryObject &getBitcodeBytes() { return *BitcodeBytes; } 77 78 ~BitstreamReader() { 79 // Free the BlockInfoRecords. 80 while (!BlockInfoRecords.empty()) { 81 BlockInfo &Info = BlockInfoRecords.back(); 82 // Free blockinfo abbrev info. 83 for (unsigned i = 0, e = static_cast<unsigned>(Info.Abbrevs.size()); 84 i != e; ++i) 85 Info.Abbrevs[i]->dropRef(); 86 BlockInfoRecords.pop_back(); 87 } 88 } 89 90 /// CollectBlockInfoNames - This is called by clients that want block/record 91 /// name information. 92 void CollectBlockInfoNames() { IgnoreBlockInfoNames = false; } 93 bool isIgnoringBlockInfoNames() { return IgnoreBlockInfoNames; } 94 95 //===--------------------------------------------------------------------===// 96 // Block Manipulation 97 //===--------------------------------------------------------------------===// 98 99 /// hasBlockInfoRecords - Return true if we've already read and processed the 100 /// block info block for this Bitstream. We only process it for the first 101 /// cursor that walks over it. 102 bool hasBlockInfoRecords() const { return !BlockInfoRecords.empty(); } 103 104 /// getBlockInfo - If there is block info for the specified ID, return it, 105 /// otherwise return null. 106 const BlockInfo *getBlockInfo(unsigned BlockID) const { 107 // Common case, the most recent entry matches BlockID. 108 if (!BlockInfoRecords.empty() && BlockInfoRecords.back().BlockID == BlockID) 109 return &BlockInfoRecords.back(); 110 111 for (unsigned i = 0, e = static_cast<unsigned>(BlockInfoRecords.size()); 112 i != e; ++i) 113 if (BlockInfoRecords[i].BlockID == BlockID) 114 return &BlockInfoRecords[i]; 115 return 0; 116 } 117 118 BlockInfo &getOrCreateBlockInfo(unsigned BlockID) { 119 if (const BlockInfo *BI = getBlockInfo(BlockID)) 120 return *const_cast<BlockInfo*>(BI); 121 122 // Otherwise, add a new record. 123 BlockInfoRecords.push_back(BlockInfo()); 124 BlockInfoRecords.back().BlockID = BlockID; 125 return BlockInfoRecords.back(); 126 } 127}; 128 129 130/// BitstreamEntry - When advancing through a bitstream cursor, each advance can 131/// discover a few different kinds of entries: 132/// Error - Malformed bitcode was found. 133/// EndBlock - We've reached the end of the current block, (or the end of the 134/// file, which is treated like a series of EndBlock records. 135/// SubBlock - This is the start of a new subblock of a specific ID. 136/// Record - This is a record with a specific AbbrevID. 137/// 138struct BitstreamEntry { 139 enum { 140 Error, 141 EndBlock, 142 SubBlock, 143 Record 144 } Kind; 145 146 unsigned ID; 147 148 static BitstreamEntry getError() { 149 BitstreamEntry E; E.Kind = Error; return E; 150 } 151 static BitstreamEntry getEndBlock() { 152 BitstreamEntry E; E.Kind = EndBlock; return E; 153 } 154 static BitstreamEntry getSubBlock(unsigned ID) { 155 BitstreamEntry E; E.Kind = SubBlock; E.ID = ID; return E; 156 } 157 static BitstreamEntry getRecord(unsigned AbbrevID) { 158 BitstreamEntry E; E.Kind = Record; E.ID = AbbrevID; return E; 159 } 160}; 161 162/// BitstreamCursor - This represents a position within a bitcode file. There 163/// may be multiple independent cursors reading within one bitstream, each 164/// maintaining their own local state. 165/// 166/// Unlike iterators, BitstreamCursors are heavy-weight objects that should not 167/// be passed by value. 168class BitstreamCursor { 169 friend class Deserializer; 170 BitstreamReader *BitStream; 171 size_t NextChar; 172 173 /// CurWord - This is the current data we have pulled from the stream but have 174 /// not returned to the client. 175 uint32_t CurWord; 176 177 /// BitsInCurWord - This is the number of bits in CurWord that are valid. This 178 /// is always from [0...31] inclusive. 179 unsigned BitsInCurWord; 180 181 // CurCodeSize - This is the declared size of code values used for the current 182 // block, in bits. 183 unsigned CurCodeSize; 184 185 /// CurAbbrevs - Abbrevs installed at in this block. 186 std::vector<BitCodeAbbrev*> CurAbbrevs; 187 188 struct Block { 189 unsigned PrevCodeSize; 190 std::vector<BitCodeAbbrev*> PrevAbbrevs; 191 explicit Block(unsigned PCS) : PrevCodeSize(PCS) {} 192 }; 193 194 /// BlockScope - This tracks the codesize of parent blocks. 195 SmallVector<Block, 8> BlockScope; 196 197 198public: 199 BitstreamCursor() : BitStream(0), NextChar(0) { 200 } 201 BitstreamCursor(const BitstreamCursor &RHS) : BitStream(0), NextChar(0) { 202 operator=(RHS); 203 } 204 205 explicit BitstreamCursor(BitstreamReader &R) : BitStream(&R) { 206 NextChar = 0; 207 CurWord = 0; 208 BitsInCurWord = 0; 209 CurCodeSize = 2; 210 } 211 212 void init(BitstreamReader &R) { 213 freeState(); 214 215 BitStream = &R; 216 NextChar = 0; 217 CurWord = 0; 218 BitsInCurWord = 0; 219 CurCodeSize = 2; 220 } 221 222 ~BitstreamCursor() { 223 freeState(); 224 } 225 226 void operator=(const BitstreamCursor &RHS); 227 228 void freeState(); 229 230 bool isEndPos(size_t pos) { 231 return BitStream->getBitcodeBytes().isObjectEnd(static_cast<uint64_t>(pos)); 232 } 233 234 bool canSkipToPos(size_t pos) const { 235 // pos can be skipped to if it is a valid address or one byte past the end. 236 return pos == 0 || BitStream->getBitcodeBytes().isValidAddress( 237 static_cast<uint64_t>(pos - 1)); 238 } 239 240 uint32_t getWord(size_t pos) { 241 uint8_t buf[4] = { 0xFF, 0xFF, 0xFF, 0xFF }; 242 BitStream->getBitcodeBytes().readBytes(pos, sizeof(buf), buf, NULL); 243 return *reinterpret_cast<support::ulittle32_t *>(buf); 244 } 245 246 bool AtEndOfStream() { 247 return BitsInCurWord == 0 && isEndPos(NextChar); 248 } 249 250 /// getAbbrevIDWidth - Return the number of bits used to encode an abbrev #. 251 unsigned getAbbrevIDWidth() const { return CurCodeSize; } 252 253 /// GetCurrentBitNo - Return the bit # of the bit we are reading. 254 uint64_t GetCurrentBitNo() const { 255 return NextChar*CHAR_BIT - BitsInCurWord; 256 } 257 258 BitstreamReader *getBitStreamReader() { 259 return BitStream; 260 } 261 const BitstreamReader *getBitStreamReader() const { 262 return BitStream; 263 } 264 265 /// Flags that modify the behavior of advance(). 266 enum { 267 /// AF_DontPopBlockAtEnd - If this flag is used, the advance() method does 268 /// not automatically pop the block scope when the end of a block is 269 /// reached. 270 AF_DontPopBlockAtEnd = 1, 271 272 /// AF_DontAutoprocessAbbrevs - If this flag is used, abbrev entries are 273 /// returned just like normal records. 274 AF_DontAutoprocessAbbrevs = 2 275 }; 276 277 /// advance - Advance the current bitstream, returning the next entry in the 278 /// stream. 279 BitstreamEntry advance(unsigned Flags = 0) { 280 while (1) { 281 unsigned Code = ReadCode(); 282 if (Code == bitc::END_BLOCK) { 283 // Pop the end of the block unless Flags tells us not to. 284 if (!(Flags & AF_DontPopBlockAtEnd) && ReadBlockEnd()) 285 return BitstreamEntry::getError(); 286 return BitstreamEntry::getEndBlock(); 287 } 288 289 if (Code == bitc::ENTER_SUBBLOCK) 290 return BitstreamEntry::getSubBlock(ReadSubBlockID()); 291 292 if (Code == bitc::DEFINE_ABBREV && 293 !(Flags & AF_DontAutoprocessAbbrevs)) { 294 // We read and accumulate abbrev's, the client can't do anything with 295 // them anyway. 296 ReadAbbrevRecord(); 297 continue; 298 } 299 300 return BitstreamEntry::getRecord(Code); 301 } 302 } 303 304 /// advanceSkippingSubblocks - This is a convenience function for clients that 305 /// don't expect any subblocks. This just skips over them automatically. 306 BitstreamEntry advanceSkippingSubblocks(unsigned Flags = 0) { 307 while (1) { 308 // If we found a normal entry, return it. 309 BitstreamEntry Entry = advance(Flags); 310 if (Entry.Kind != BitstreamEntry::SubBlock) 311 return Entry; 312 313 // If we found a sub-block, just skip over it and check the next entry. 314 if (SkipBlock()) 315 return BitstreamEntry::getError(); 316 } 317 } 318 319 /// JumpToBit - Reset the stream to the specified bit number. 320 void JumpToBit(uint64_t BitNo) { 321 uintptr_t ByteNo = uintptr_t(BitNo/8) & ~3; 322 uintptr_t WordBitNo = uintptr_t(BitNo) & 31; 323 assert(canSkipToPos(ByteNo) && "Invalid location"); 324 325 // Move the cursor to the right word. 326 NextChar = ByteNo; 327 BitsInCurWord = 0; 328 CurWord = 0; 329 330 // Skip over any bits that are already consumed. 331 if (WordBitNo) 332 Read(static_cast<unsigned>(WordBitNo)); 333 } 334 335 336 uint32_t Read(unsigned NumBits) { 337 assert(NumBits <= 32 && "Cannot return more than 32 bits!"); 338 // If the field is fully contained by CurWord, return it quickly. 339 if (BitsInCurWord >= NumBits) { 340 uint32_t R = CurWord & ((1U << NumBits)-1); 341 CurWord >>= NumBits; 342 BitsInCurWord -= NumBits; 343 return R; 344 } 345 346 // If we run out of data, stop at the end of the stream. 347 if (isEndPos(NextChar)) { 348 CurWord = 0; 349 BitsInCurWord = 0; 350 return 0; 351 } 352 353 unsigned R = CurWord; 354 355 // Read the next word from the stream. 356 CurWord = getWord(NextChar); 357 NextChar += 4; 358 359 // Extract NumBits-BitsInCurWord from what we just read. 360 unsigned BitsLeft = NumBits-BitsInCurWord; 361 362 // Be careful here, BitsLeft is in the range [1..32] inclusive. 363 R |= (CurWord & (~0U >> (32-BitsLeft))) << BitsInCurWord; 364 365 // BitsLeft bits have just been used up from CurWord. 366 if (BitsLeft != 32) 367 CurWord >>= BitsLeft; 368 else 369 CurWord = 0; 370 BitsInCurWord = 32-BitsLeft; 371 return R; 372 } 373 374 uint64_t Read64(unsigned NumBits) { 375 if (NumBits <= 32) return Read(NumBits); 376 377 uint64_t V = Read(32); 378 return V | (uint64_t)Read(NumBits-32) << 32; 379 } 380 381 uint32_t ReadVBR(unsigned NumBits) { 382 uint32_t Piece = Read(NumBits); 383 if ((Piece & (1U << (NumBits-1))) == 0) 384 return Piece; 385 386 uint32_t Result = 0; 387 unsigned NextBit = 0; 388 while (1) { 389 Result |= (Piece & ((1U << (NumBits-1))-1)) << NextBit; 390 391 if ((Piece & (1U << (NumBits-1))) == 0) 392 return Result; 393 394 NextBit += NumBits-1; 395 Piece = Read(NumBits); 396 } 397 } 398 399 // ReadVBR64 - Read a VBR that may have a value up to 64-bits in size. The 400 // chunk size of the VBR must still be <= 32 bits though. 401 uint64_t ReadVBR64(unsigned NumBits) { 402 uint32_t Piece = Read(NumBits); 403 if ((Piece & (1U << (NumBits-1))) == 0) 404 return uint64_t(Piece); 405 406 uint64_t Result = 0; 407 unsigned NextBit = 0; 408 while (1) { 409 Result |= uint64_t(Piece & ((1U << (NumBits-1))-1)) << NextBit; 410 411 if ((Piece & (1U << (NumBits-1))) == 0) 412 return Result; 413 414 NextBit += NumBits-1; 415 Piece = Read(NumBits); 416 } 417 } 418 419 void SkipToFourByteBoundary() { 420 BitsInCurWord = 0; 421 CurWord = 0; 422 } 423 424 unsigned ReadCode() { 425 return Read(CurCodeSize); 426 } 427 428 429 // Block header: 430 // [ENTER_SUBBLOCK, blockid, newcodelen, <align4bytes>, blocklen] 431 432 /// ReadSubBlockID - Having read the ENTER_SUBBLOCK code, read the BlockID for 433 /// the block. 434 unsigned ReadSubBlockID() { 435 return ReadVBR(bitc::BlockIDWidth); 436 } 437 438 /// SkipBlock - Having read the ENTER_SUBBLOCK abbrevid and a BlockID, skip 439 /// over the body of this block. If the block record is malformed, return 440 /// true. 441 bool SkipBlock() { 442 // Read and ignore the codelen value. Since we are skipping this block, we 443 // don't care what code widths are used inside of it. 444 ReadVBR(bitc::CodeLenWidth); 445 SkipToFourByteBoundary(); 446 unsigned NumWords = Read(bitc::BlockSizeWidth); 447 448 // Check that the block wasn't partially defined, and that the offset isn't 449 // bogus. 450 size_t SkipTo = NextChar + NumWords*4; 451 if (AtEndOfStream() || !canSkipToPos(SkipTo)) 452 return true; 453 454 NextChar = SkipTo; 455 return false; 456 } 457 458 /// EnterSubBlock - Having read the ENTER_SUBBLOCK abbrevid, enter 459 /// the block, and return true if the block has an error. 460 bool EnterSubBlock(unsigned BlockID, unsigned *NumWordsP = 0); 461 462 bool ReadBlockEnd() { 463 if (BlockScope.empty()) return true; 464 465 // Block tail: 466 // [END_BLOCK, <align4bytes>] 467 SkipToFourByteBoundary(); 468 469 popBlockScope(); 470 return false; 471 } 472 473private: 474 475 void popBlockScope() { 476 CurCodeSize = BlockScope.back().PrevCodeSize; 477 478 // Delete abbrevs from popped scope. 479 for (unsigned i = 0, e = static_cast<unsigned>(CurAbbrevs.size()); 480 i != e; ++i) 481 CurAbbrevs[i]->dropRef(); 482 483 BlockScope.back().PrevAbbrevs.swap(CurAbbrevs); 484 BlockScope.pop_back(); 485 } 486 487 //===--------------------------------------------------------------------===// 488 // Record Processing 489 //===--------------------------------------------------------------------===// 490 491private: 492 void readAbbreviatedLiteral(const BitCodeAbbrevOp &Op, 493 SmallVectorImpl<uint64_t> &Vals); 494 void readAbbreviatedField(const BitCodeAbbrevOp &Op, 495 SmallVectorImpl<uint64_t> &Vals); 496 void skipAbbreviatedField(const BitCodeAbbrevOp &Op); 497 498public: 499 500 /// getAbbrev - Return the abbreviation for the specified AbbrevId. 501 const BitCodeAbbrev *getAbbrev(unsigned AbbrevID) { 502 unsigned AbbrevNo = AbbrevID-bitc::FIRST_APPLICATION_ABBREV; 503 assert(AbbrevNo < CurAbbrevs.size() && "Invalid abbrev #!"); 504 return CurAbbrevs[AbbrevNo]; 505 } 506 507 /// skipRecord - Read the current record and discard it. 508 void skipRecord(unsigned AbbrevID); 509 510 unsigned readRecord(unsigned AbbrevID, SmallVectorImpl<uint64_t> &Vals, 511 StringRef *Blob = 0); 512 513 //===--------------------------------------------------------------------===// 514 // Abbrev Processing 515 //===--------------------------------------------------------------------===// 516 void ReadAbbrevRecord(); 517 518 bool ReadBlockInfoBlock(); 519}; 520 521} // End llvm namespace 522 523#endif 524