BitcodeReader.cpp revision e84bcb922de726309c299ecba4ec3d14d5f67c41
1//===- BitcodeReader.cpp - Internal BitcodeReader implementation ----------===// 2// 3// The LLVM Compiler Infrastructure 4// 5// This file was developed by Chris Lattner and is distributed under 6// the University of Illinois Open Source License. See LICENSE.TXT for details. 7// 8//===----------------------------------------------------------------------===// 9// 10// This header defines the BitcodeReader class. 11// 12//===----------------------------------------------------------------------===// 13 14#include "BitcodeReader.h" 15#include "llvm/Bitcode/BitstreamReader.h" 16#include "llvm/DerivedTypes.h" 17#include "llvm/Module.h" 18#include "llvm/ADT/SmallString.h" 19using namespace llvm; 20 21/// ConvertToString - Convert a string from a record into an std::string, return 22/// true on failure. 23template<typename StrTy> 24static bool ConvertToString(SmallVector<uint64_t, 64> &Record, unsigned Idx, 25 StrTy &Result) { 26 if (Record.size() < Idx+1 || Record.size() < Record[Idx]+Idx+1) 27 return true; 28 29 for (unsigned i = 0, e = Record[Idx]; i != e; ++i) 30 Result += (char)Record[Idx+i+1]; 31 return false; 32} 33 34static GlobalValue::LinkageTypes GetDecodedLinkage(unsigned Val) { 35 switch (Val) { 36 default: // Map unknown/new linkages to external 37 case 0: return GlobalValue::ExternalLinkage; 38 case 1: return GlobalValue::WeakLinkage; 39 case 2: return GlobalValue::AppendingLinkage; 40 case 3: return GlobalValue::InternalLinkage; 41 case 4: return GlobalValue::LinkOnceLinkage; 42 case 5: return GlobalValue::DLLImportLinkage; 43 case 6: return GlobalValue::DLLExportLinkage; 44 case 7: return GlobalValue::ExternalWeakLinkage; 45 } 46} 47 48static GlobalValue::VisibilityTypes GetDecodedVisibility(unsigned Val) { 49 switch (Val) { 50 default: // Map unknown visibilities to default. 51 case 0: return GlobalValue::DefaultVisibility; 52 case 1: return GlobalValue::HiddenVisibility; 53 } 54} 55 56 57const Type *BitcodeReader::getTypeByID(unsigned ID, bool isTypeTable) { 58 // If the TypeID is in range, return it. 59 if (ID < TypeList.size()) 60 return TypeList[ID].get(); 61 if (!isTypeTable) return 0; 62 63 // The type table allows forward references. Push as many Opaque types as 64 // needed to get up to ID. 65 while (TypeList.size() <= ID) 66 TypeList.push_back(OpaqueType::get()); 67 return TypeList.back().get(); 68} 69 70 71bool BitcodeReader::ParseTypeTable(BitstreamReader &Stream) { 72 if (Stream.EnterSubBlock()) 73 return Error("Malformed block record"); 74 75 if (!TypeList.empty()) 76 return Error("Multiple TYPE_BLOCKs found!"); 77 78 SmallVector<uint64_t, 64> Record; 79 unsigned NumRecords = 0; 80 81 // Read all the records for this type table. 82 while (1) { 83 unsigned Code = Stream.ReadCode(); 84 if (Code == bitc::END_BLOCK) { 85 if (NumRecords != TypeList.size()) 86 return Error("Invalid type forward reference in TYPE_BLOCK"); 87 return Stream.ReadBlockEnd(); 88 } 89 90 if (Code == bitc::ENTER_SUBBLOCK) { 91 // No known subblocks, always skip them. 92 Stream.ReadSubBlockID(); 93 if (Stream.SkipBlock()) 94 return Error("Malformed block record"); 95 continue; 96 } 97 98 if (Code == bitc::DEFINE_ABBREV) { 99 Stream.ReadAbbrevRecord(); 100 continue; 101 } 102 103 // Read a record. 104 Record.clear(); 105 const Type *ResultTy = 0; 106 switch (Stream.ReadRecord(Code, Record)) { 107 default: // Default behavior: unknown type. 108 ResultTy = 0; 109 break; 110 case bitc::TYPE_CODE_NUMENTRY: // TYPE_CODE_NUMENTRY: [numentries] 111 // TYPE_CODE_NUMENTRY contains a count of the number of types in the 112 // type list. This allows us to reserve space. 113 if (Record.size() < 1) 114 return Error("Invalid TYPE_CODE_NUMENTRY record"); 115 TypeList.reserve(Record[0]); 116 continue; 117 case bitc::TYPE_CODE_META: // TYPE_CODE_META: [metacode]... 118 // No metadata supported yet. 119 if (Record.size() < 1) 120 return Error("Invalid TYPE_CODE_META record"); 121 continue; 122 123 case bitc::TYPE_CODE_VOID: // VOID 124 ResultTy = Type::VoidTy; 125 break; 126 case bitc::TYPE_CODE_FLOAT: // FLOAT 127 ResultTy = Type::FloatTy; 128 break; 129 case bitc::TYPE_CODE_DOUBLE: // DOUBLE 130 ResultTy = Type::DoubleTy; 131 break; 132 case bitc::TYPE_CODE_LABEL: // LABEL 133 ResultTy = Type::LabelTy; 134 break; 135 case bitc::TYPE_CODE_OPAQUE: // OPAQUE 136 ResultTy = 0; 137 break; 138 case bitc::TYPE_CODE_INTEGER: // INTEGER: [width] 139 if (Record.size() < 1) 140 return Error("Invalid Integer type record"); 141 142 ResultTy = IntegerType::get(Record[0]); 143 break; 144 case bitc::TYPE_CODE_POINTER: // POINTER: [pointee type] 145 if (Record.size() < 1) 146 return Error("Invalid POINTER type record"); 147 ResultTy = PointerType::get(getTypeByID(Record[0], true)); 148 break; 149 case bitc::TYPE_CODE_FUNCTION: { 150 // FUNCTION: [vararg, retty, #pararms, paramty N] 151 if (Record.size() < 3 || Record.size() < Record[2]+3) 152 return Error("Invalid FUNCTION type record"); 153 std::vector<const Type*> ArgTys; 154 for (unsigned i = 0, e = Record[2]; i != e; ++i) 155 ArgTys.push_back(getTypeByID(Record[3+i], true)); 156 157 // FIXME: PARAM TYS. 158 ResultTy = FunctionType::get(getTypeByID(Record[1], true), ArgTys, 159 Record[0]); 160 break; 161 } 162 case bitc::TYPE_CODE_STRUCT: { // STRUCT: [ispacked, #elts, eltty x N] 163 if (Record.size() < 2 || Record.size() < Record[1]+2) 164 return Error("Invalid STRUCT type record"); 165 std::vector<const Type*> EltTys; 166 for (unsigned i = 0, e = Record[1]; i != e; ++i) 167 EltTys.push_back(getTypeByID(Record[2+i], true)); 168 ResultTy = StructType::get(EltTys, Record[0]); 169 break; 170 } 171 case bitc::TYPE_CODE_ARRAY: // ARRAY: [numelts, eltty] 172 if (Record.size() < 2) 173 return Error("Invalid ARRAY type record"); 174 ResultTy = ArrayType::get(getTypeByID(Record[1], true), Record[0]); 175 break; 176 case bitc::TYPE_CODE_VECTOR: // VECTOR: [numelts, eltty] 177 if (Record.size() < 2) 178 return Error("Invalid VECTOR type record"); 179 ResultTy = VectorType::get(getTypeByID(Record[1], true), Record[0]); 180 break; 181 } 182 183 if (NumRecords == TypeList.size()) { 184 // If this is a new type slot, just append it. 185 TypeList.push_back(ResultTy ? ResultTy : OpaqueType::get()); 186 ++NumRecords; 187 } else if (ResultTy == 0) { 188 // Otherwise, this was forward referenced, so an opaque type was created, 189 // but the result type is actually just an opaque. Leave the one we 190 // created previously. 191 ++NumRecords; 192 } else { 193 // Otherwise, this was forward referenced, so an opaque type was created. 194 // Resolve the opaque type to the real type now. 195 assert(NumRecords < TypeList.size() && "Typelist imbalance"); 196 const OpaqueType *OldTy = cast<OpaqueType>(TypeList[NumRecords++].get()); 197 198 // Don't directly push the new type on the Tab. Instead we want to replace 199 // the opaque type we previously inserted with the new concrete value. The 200 // refinement from the abstract (opaque) type to the new type causes all 201 // uses of the abstract type to use the concrete type (NewTy). This will 202 // also cause the opaque type to be deleted. 203 const_cast<OpaqueType*>(OldTy)->refineAbstractTypeTo(ResultTy); 204 205 // This should have replaced the old opaque type with the new type in the 206 // value table... or with a preexisting type that was already in the system. 207 // Let's just make sure it did. 208 assert(TypeList[NumRecords-1].get() != OldTy && 209 "refineAbstractType didn't work!"); 210 } 211 } 212} 213 214 215bool BitcodeReader::ParseTypeSymbolTable(BitstreamReader &Stream) { 216 if (Stream.EnterSubBlock()) 217 return Error("Malformed block record"); 218 219 SmallVector<uint64_t, 64> Record; 220 221 // Read all the records for this type table. 222 std::string TypeName; 223 while (1) { 224 unsigned Code = Stream.ReadCode(); 225 if (Code == bitc::END_BLOCK) 226 return Stream.ReadBlockEnd(); 227 228 if (Code == bitc::ENTER_SUBBLOCK) { 229 // No known subblocks, always skip them. 230 Stream.ReadSubBlockID(); 231 if (Stream.SkipBlock()) 232 return Error("Malformed block record"); 233 continue; 234 } 235 236 if (Code == bitc::DEFINE_ABBREV) { 237 Stream.ReadAbbrevRecord(); 238 continue; 239 } 240 241 // Read a record. 242 Record.clear(); 243 switch (Stream.ReadRecord(Code, Record)) { 244 default: // Default behavior: unknown type. 245 break; 246 case bitc::TST_CODE_ENTRY: // TST_ENTRY: [typeid, namelen, namechar x N] 247 if (ConvertToString(Record, 1, TypeName)) 248 return Error("Invalid TST_ENTRY record"); 249 unsigned TypeID = Record[0]; 250 if (TypeID >= TypeList.size()) 251 return Error("Invalid Type ID in TST_ENTRY record"); 252 253 TheModule->addTypeName(TypeName, TypeList[TypeID].get()); 254 TypeName.clear(); 255 break; 256 } 257 } 258} 259 260bool BitcodeReader::ParseValueSymbolTable(BitstreamReader &Stream) { 261 if (Stream.EnterSubBlock()) 262 return Error("Malformed block record"); 263 264 SmallVector<uint64_t, 64> Record; 265 266 // Read all the records for this value table. 267 SmallString<128> ValueName; 268 while (1) { 269 unsigned Code = Stream.ReadCode(); 270 if (Code == bitc::END_BLOCK) 271 return Stream.ReadBlockEnd(); 272 273 if (Code == bitc::ENTER_SUBBLOCK) { 274 // No known subblocks, always skip them. 275 Stream.ReadSubBlockID(); 276 if (Stream.SkipBlock()) 277 return Error("Malformed block record"); 278 continue; 279 } 280 281 if (Code == bitc::DEFINE_ABBREV) { 282 Stream.ReadAbbrevRecord(); 283 continue; 284 } 285 286 // Read a record. 287 Record.clear(); 288 switch (Stream.ReadRecord(Code, Record)) { 289 default: // Default behavior: unknown type. 290 break; 291 case bitc::TST_CODE_ENTRY: // VST_ENTRY: [valueid, namelen, namechar x N] 292 if (ConvertToString(Record, 1, ValueName)) 293 return Error("Invalid TST_ENTRY record"); 294 unsigned ValueID = Record[0]; 295 if (ValueID >= ValueList.size()) 296 return Error("Invalid Value ID in VST_ENTRY record"); 297 Value *V = ValueList[ValueID]; 298 299 V->setName(&ValueName[0], ValueName.size()); 300 ValueName.clear(); 301 break; 302 } 303 } 304} 305 306 307bool BitcodeReader::ParseModule(BitstreamReader &Stream, 308 const std::string &ModuleID) { 309 // Reject multiple MODULE_BLOCK's in a single bitstream. 310 if (TheModule) 311 return Error("Multiple MODULE_BLOCKs in same stream"); 312 313 if (Stream.EnterSubBlock()) 314 return Error("Malformed block record"); 315 316 // Otherwise, create the module. 317 TheModule = new Module(ModuleID); 318 319 SmallVector<uint64_t, 64> Record; 320 std::vector<std::string> SectionTable; 321 322 // Read all the records for this module. 323 while (!Stream.AtEndOfStream()) { 324 unsigned Code = Stream.ReadCode(); 325 if (Code == bitc::END_BLOCK) { 326 if (!GlobalInits.empty()) 327 return Error("Malformed global initializer set"); 328 return Stream.ReadBlockEnd(); 329 } 330 331 if (Code == bitc::ENTER_SUBBLOCK) { 332 switch (Stream.ReadSubBlockID()) { 333 default: // Skip unknown content. 334 if (Stream.SkipBlock()) 335 return Error("Malformed block record"); 336 break; 337 case bitc::TYPE_BLOCK_ID: 338 if (ParseTypeTable(Stream)) 339 return true; 340 break; 341 case bitc::TYPE_SYMTAB_BLOCK_ID: 342 if (ParseTypeSymbolTable(Stream)) 343 return true; 344 break; 345 case bitc::VALUE_SYMTAB_BLOCK_ID: 346 if (ParseValueSymbolTable(Stream)) 347 return true; 348 break; 349 } 350 continue; 351 } 352 353 if (Code == bitc::DEFINE_ABBREV) { 354 Stream.ReadAbbrevRecord(); 355 continue; 356 } 357 358 // Read a record. 359 switch (Stream.ReadRecord(Code, Record)) { 360 default: break; // Default behavior, ignore unknown content. 361 case bitc::MODULE_CODE_VERSION: // VERSION: [version#] 362 if (Record.size() < 1) 363 return Error("Malformed MODULE_CODE_VERSION"); 364 // Only version #0 is supported so far. 365 if (Record[0] != 0) 366 return Error("Unknown bitstream version!"); 367 break; 368 case bitc::MODULE_CODE_TRIPLE: { // TRIPLE: [strlen, strchr x N] 369 std::string S; 370 if (ConvertToString(Record, 0, S)) 371 return Error("Invalid MODULE_CODE_TRIPLE record"); 372 TheModule->setTargetTriple(S); 373 break; 374 } 375 case bitc::MODULE_CODE_DATALAYOUT: { // DATALAYOUT: [strlen, strchr x N] 376 std::string S; 377 if (ConvertToString(Record, 0, S)) 378 return Error("Invalid MODULE_CODE_DATALAYOUT record"); 379 TheModule->setDataLayout(S); 380 break; 381 } 382 case bitc::MODULE_CODE_ASM: { // ASM: [strlen, strchr x N] 383 std::string S; 384 if (ConvertToString(Record, 0, S)) 385 return Error("Invalid MODULE_CODE_ASM record"); 386 TheModule->setModuleInlineAsm(S); 387 break; 388 } 389 case bitc::MODULE_CODE_DEPLIB: { // DEPLIB: [strlen, strchr x N] 390 std::string S; 391 if (ConvertToString(Record, 0, S)) 392 return Error("Invalid MODULE_CODE_DEPLIB record"); 393 TheModule->addLibrary(S); 394 break; 395 } 396 case bitc::MODULE_CODE_SECTIONNAME: { // SECTIONNAME: [strlen, strchr x N] 397 std::string S; 398 if (ConvertToString(Record, 0, S)) 399 return Error("Invalid MODULE_CODE_SECTIONNAME record"); 400 SectionTable.push_back(S); 401 break; 402 } 403 // GLOBALVAR: [type, isconst, initid, 404 // linkage, alignment, section, visibility, threadlocal] 405 case bitc::MODULE_CODE_GLOBALVAR: { 406 if (Record.size() < 6) 407 return Error("Invalid MODULE_CODE_GLOBALVAR record"); 408 const Type *Ty = getTypeByID(Record[0]); 409 if (!isa<PointerType>(Ty)) 410 return Error("Global not a pointer type!"); 411 Ty = cast<PointerType>(Ty)->getElementType(); 412 413 bool isConstant = Record[1]; 414 GlobalValue::LinkageTypes Linkage = GetDecodedLinkage(Record[3]); 415 unsigned Alignment = (1 << Record[4]) >> 1; 416 std::string Section; 417 if (Record[5]) { 418 if (Record[5]-1 >= SectionTable.size()) 419 return Error("Invalid section ID"); 420 Section = SectionTable[Record[5]-1]; 421 } 422 GlobalValue::VisibilityTypes Visibility = GlobalValue::DefaultVisibility; 423 if (Record.size() >= 6) Visibility = GetDecodedVisibility(Record[6]); 424 bool isThreadLocal = false; 425 if (Record.size() >= 7) isThreadLocal = Record[7]; 426 427 GlobalVariable *NewGV = 428 new GlobalVariable(Ty, isConstant, Linkage, 0, "", TheModule); 429 NewGV->setAlignment(Alignment); 430 if (!Section.empty()) 431 NewGV->setSection(Section); 432 NewGV->setVisibility(Visibility); 433 NewGV->setThreadLocal(isThreadLocal); 434 435 ValueList.push_back(NewGV); 436 437 // Remember which value to use for the global initializer. 438 if (unsigned InitID = Record[2]) 439 GlobalInits.push_back(std::make_pair(NewGV, InitID-1)); 440 break; 441 } 442 // FUNCTION: [type, callingconv, isproto, linkage, alignment, section, 443 // visibility] 444 case bitc::MODULE_CODE_FUNCTION: { 445 if (Record.size() < 7) 446 return Error("Invalid MODULE_CODE_FUNCTION record"); 447 const Type *Ty = getTypeByID(Record[0]); 448 if (!isa<PointerType>(Ty)) 449 return Error("Function not a pointer type!"); 450 const FunctionType *FTy = 451 dyn_cast<FunctionType>(cast<PointerType>(Ty)->getElementType()); 452 if (!FTy) 453 return Error("Function not a pointer to function type!"); 454 455 Function *Func = new Function(FTy, GlobalValue::ExternalLinkage, 456 "", TheModule); 457 458 Func->setCallingConv(Record[1]); 459 Func->setLinkage(GetDecodedLinkage(Record[3])); 460 Func->setAlignment((1 << Record[4]) >> 1); 461 if (Record[5]) { 462 if (Record[5]-1 >= SectionTable.size()) 463 return Error("Invalid section ID"); 464 Func->setSection(SectionTable[Record[5]-1]); 465 } 466 Func->setVisibility(GetDecodedVisibility(Record[6])); 467 468 ValueList.push_back(Func); 469 // TODO: remember initializer/global pair for later substitution. 470 break; 471 } 472 } 473 Record.clear(); 474 } 475 476 return Error("Premature end of bitstream"); 477} 478 479 480bool BitcodeReader::ParseBitcode(unsigned char *Buf, unsigned Length, 481 const std::string &ModuleID) { 482 TheModule = 0; 483 484 if (Length & 3) 485 return Error("Bitcode stream should be a multiple of 4 bytes in length"); 486 487 BitstreamReader Stream(Buf, Buf+Length); 488 489 // Sniff for the signature. 490 if (Stream.Read(8) != 'B' || 491 Stream.Read(8) != 'C' || 492 Stream.Read(4) != 0x0 || 493 Stream.Read(4) != 0xC || 494 Stream.Read(4) != 0xE || 495 Stream.Read(4) != 0xD) 496 return Error("Invalid bitcode signature"); 497 498 // We expect a number of well-defined blocks, though we don't necessarily 499 // need to understand them all. 500 while (!Stream.AtEndOfStream()) { 501 unsigned Code = Stream.ReadCode(); 502 503 if (Code != bitc::ENTER_SUBBLOCK) 504 return Error("Invalid record at top-level"); 505 506 unsigned BlockID = Stream.ReadSubBlockID(); 507 508 // We only know the MODULE subblock ID. 509 if (BlockID == bitc::MODULE_BLOCK_ID) { 510 if (ParseModule(Stream, ModuleID)) 511 return true; 512 } else if (Stream.SkipBlock()) { 513 return Error("Malformed block record"); 514 } 515 } 516 517 return false; 518} 519