1 //=-- InstrProfReader.h - Instrumented profiling readers ----------*- 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 file contains support for reading profiling data for instrumentation 11 // based PGO and coverage. 12 // 13 //===----------------------------------------------------------------------===// 14 15 #ifndef LLVM_PROFILEDATA_INSTRPROFREADER_H 16 #define LLVM_PROFILEDATA_INSTRPROFREADER_H 17 18 #include "llvm/ADT/ArrayRef.h" 19 #include "llvm/ADT/StringExtras.h" 20 #include "llvm/ProfileData/InstrProf.h" 21 #include "llvm/Support/EndianStream.h" 22 #include "llvm/Support/LineIterator.h" 23 #include "llvm/Support/MemoryBuffer.h" 24 #include "llvm/Support/OnDiskHashTable.h" 25 #include "llvm/Support/raw_ostream.h" 26 #include <iterator> 27 28 namespace llvm { 29 30 class InstrProfReader; 31 32 /// A file format agnostic iterator over profiling data. 33 class InstrProfIterator : public std::iterator<std::input_iterator_tag, 34 InstrProfRecord> { 35 InstrProfReader *Reader; 36 InstrProfRecord Record; 37 38 void Increment(); 39 public: 40 InstrProfIterator() : Reader(nullptr) {} 41 InstrProfIterator(InstrProfReader *Reader) : Reader(Reader) { Increment(); } 42 43 InstrProfIterator &operator++() { Increment(); return *this; } 44 bool operator==(const InstrProfIterator &RHS) { return Reader == RHS.Reader; } 45 bool operator!=(const InstrProfIterator &RHS) { return Reader != RHS.Reader; } 46 InstrProfRecord &operator*() { return Record; } 47 InstrProfRecord *operator->() { return &Record; } 48 }; 49 50 /// Base class and interface for reading profiling data of any known instrprof 51 /// format. Provides an iterator over InstrProfRecords. 52 class InstrProfReader { 53 instrprof_error LastError; 54 55 public: 56 InstrProfReader() : LastError(instrprof_error::success), Symtab() {} 57 virtual ~InstrProfReader() {} 58 59 /// Read the header. Required before reading first record. 60 virtual Error readHeader() = 0; 61 /// Read a single record. 62 virtual Error readNextRecord(InstrProfRecord &Record) = 0; 63 /// Iterator over profile data. 64 InstrProfIterator begin() { return InstrProfIterator(this); } 65 InstrProfIterator end() { return InstrProfIterator(); } 66 virtual bool isIRLevelProfile() const = 0; 67 68 /// Return the PGO symtab. There are three different readers: 69 /// Raw, Text, and Indexed profile readers. The first two types 70 /// of readers are used only by llvm-profdata tool, while the indexed 71 /// profile reader is also used by llvm-cov tool and the compiler ( 72 /// backend or frontend). Since creating PGO symtab can create 73 /// significant runtime and memory overhead (as it touches data 74 /// for the whole program), InstrProfSymtab for the indexed profile 75 /// reader should be created on demand and it is recommended to be 76 /// only used for dumping purpose with llvm-proftool, not with the 77 /// compiler. 78 virtual InstrProfSymtab &getSymtab() = 0; 79 80 protected: 81 std::unique_ptr<InstrProfSymtab> Symtab; 82 /// Set the current error and return same. 83 Error error(instrprof_error Err) { 84 LastError = Err; 85 if (Err == instrprof_error::success) 86 return Error::success(); 87 return make_error<InstrProfError>(Err); 88 } 89 Error error(Error E) { return error(InstrProfError::take(std::move(E))); } 90 91 /// Clear the current error and return a successful one. 92 Error success() { return error(instrprof_error::success); } 93 94 public: 95 /// Return true if the reader has finished reading the profile data. 96 bool isEOF() { return LastError == instrprof_error::eof; } 97 /// Return true if the reader encountered an error reading profiling data. 98 bool hasError() { return LastError != instrprof_error::success && !isEOF(); } 99 /// Get the current error. 100 Error getError() { 101 if (hasError()) 102 return make_error<InstrProfError>(LastError); 103 return Error::success(); 104 } 105 106 /// Factory method to create an appropriately typed reader for the given 107 /// instrprof file. 108 static Expected<std::unique_ptr<InstrProfReader>> create(const Twine &Path); 109 110 static Expected<std::unique_ptr<InstrProfReader>> 111 create(std::unique_ptr<MemoryBuffer> Buffer); 112 }; 113 114 /// Reader for the simple text based instrprof format. 115 /// 116 /// This format is a simple text format that's suitable for test data. Records 117 /// are separated by one or more blank lines, and record fields are separated by 118 /// new lines. 119 /// 120 /// Each record consists of a function name, a function hash, a number of 121 /// counters, and then each counter value, in that order. 122 class TextInstrProfReader : public InstrProfReader { 123 private: 124 /// The profile data file contents. 125 std::unique_ptr<MemoryBuffer> DataBuffer; 126 /// Iterator over the profile data. 127 line_iterator Line; 128 bool IsIRLevelProfile; 129 130 TextInstrProfReader(const TextInstrProfReader &) = delete; 131 TextInstrProfReader &operator=(const TextInstrProfReader &) = delete; 132 Error readValueProfileData(InstrProfRecord &Record); 133 134 public: 135 TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_) 136 : DataBuffer(std::move(DataBuffer_)), Line(*DataBuffer, true, '#'), 137 IsIRLevelProfile(false) {} 138 139 /// Return true if the given buffer is in text instrprof format. 140 static bool hasFormat(const MemoryBuffer &Buffer); 141 142 bool isIRLevelProfile() const override { return IsIRLevelProfile; } 143 144 /// Read the header. 145 Error readHeader() override; 146 /// Read a single record. 147 Error readNextRecord(InstrProfRecord &Record) override; 148 149 InstrProfSymtab &getSymtab() override { 150 assert(Symtab.get()); 151 return *Symtab.get(); 152 } 153 }; 154 155 /// Reader for the raw instrprof binary format from runtime. 156 /// 157 /// This format is a raw memory dump of the instrumentation-baed profiling data 158 /// from the runtime. It has no index. 159 /// 160 /// Templated on the unsigned type whose size matches pointers on the platform 161 /// that wrote the profile. 162 template <class IntPtrT> 163 class RawInstrProfReader : public InstrProfReader { 164 private: 165 /// The profile data file contents. 166 std::unique_ptr<MemoryBuffer> DataBuffer; 167 bool ShouldSwapBytes; 168 // The value of the version field of the raw profile data header. The lower 56 169 // bits specifies the format version and the most significant 8 bits specify 170 // the variant types of the profile. 171 uint64_t Version; 172 uint64_t CountersDelta; 173 uint64_t NamesDelta; 174 const RawInstrProf::ProfileData<IntPtrT> *Data; 175 const RawInstrProf::ProfileData<IntPtrT> *DataEnd; 176 const uint64_t *CountersStart; 177 const char *NamesStart; 178 uint64_t NamesSize; 179 // After value profile is all read, this pointer points to 180 // the header of next profile data (if exists) 181 const uint8_t *ValueDataStart; 182 uint32_t ValueKindLast; 183 uint32_t CurValueDataSize; 184 185 InstrProfRecord::ValueMapType FunctionPtrToNameMap; 186 187 RawInstrProfReader(const RawInstrProfReader &) = delete; 188 RawInstrProfReader &operator=(const RawInstrProfReader &) = delete; 189 public: 190 RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer) 191 : DataBuffer(std::move(DataBuffer)) { } 192 193 static bool hasFormat(const MemoryBuffer &DataBuffer); 194 Error readHeader() override; 195 Error readNextRecord(InstrProfRecord &Record) override; 196 bool isIRLevelProfile() const override { 197 return (Version & VARIANT_MASK_IR_PROF) != 0; 198 } 199 200 InstrProfSymtab &getSymtab() override { 201 assert(Symtab.get()); 202 return *Symtab.get(); 203 } 204 205 private: 206 Error createSymtab(InstrProfSymtab &Symtab); 207 Error readNextHeader(const char *CurrentPos); 208 Error readHeader(const RawInstrProf::Header &Header); 209 template <class IntT> IntT swap(IntT Int) const { 210 return ShouldSwapBytes ? sys::getSwappedBytes(Int) : Int; 211 } 212 support::endianness getDataEndianness() const { 213 support::endianness HostEndian = getHostEndianness(); 214 if (!ShouldSwapBytes) 215 return HostEndian; 216 if (HostEndian == support::little) 217 return support::big; 218 else 219 return support::little; 220 } 221 222 inline uint8_t getNumPaddingBytes(uint64_t SizeInBytes) { 223 return 7 & (sizeof(uint64_t) - SizeInBytes % sizeof(uint64_t)); 224 } 225 Error readName(InstrProfRecord &Record); 226 Error readFuncHash(InstrProfRecord &Record); 227 Error readRawCounts(InstrProfRecord &Record); 228 Error readValueProfilingData(InstrProfRecord &Record); 229 bool atEnd() const { return Data == DataEnd; } 230 void advanceData() { 231 Data++; 232 ValueDataStart += CurValueDataSize; 233 } 234 const char *getNextHeaderPos() const { 235 assert(atEnd()); 236 return (const char *)ValueDataStart; 237 } 238 239 const uint64_t *getCounter(IntPtrT CounterPtr) const { 240 ptrdiff_t Offset = (swap(CounterPtr) - CountersDelta) / sizeof(uint64_t); 241 return CountersStart + Offset; 242 } 243 StringRef getName(uint64_t NameRef) const { 244 return Symtab->getFuncName(swap(NameRef)); 245 } 246 }; 247 248 typedef RawInstrProfReader<uint32_t> RawInstrProfReader32; 249 typedef RawInstrProfReader<uint64_t> RawInstrProfReader64; 250 251 namespace IndexedInstrProf { 252 enum class HashT : uint32_t; 253 } 254 255 /// Trait for lookups into the on-disk hash table for the binary instrprof 256 /// format. 257 class InstrProfLookupTrait { 258 std::vector<InstrProfRecord> DataBuffer; 259 IndexedInstrProf::HashT HashType; 260 unsigned FormatVersion; 261 // Endianness of the input value profile data. 262 // It should be LE by default, but can be changed 263 // for testing purpose. 264 support::endianness ValueProfDataEndianness; 265 266 public: 267 InstrProfLookupTrait(IndexedInstrProf::HashT HashType, unsigned FormatVersion) 268 : HashType(HashType), FormatVersion(FormatVersion), 269 ValueProfDataEndianness(support::little) {} 270 271 typedef ArrayRef<InstrProfRecord> data_type; 272 273 typedef StringRef internal_key_type; 274 typedef StringRef external_key_type; 275 typedef uint64_t hash_value_type; 276 typedef uint64_t offset_type; 277 278 static bool EqualKey(StringRef A, StringRef B) { return A == B; } 279 static StringRef GetInternalKey(StringRef K) { return K; } 280 static StringRef GetExternalKey(StringRef K) { return K; } 281 282 hash_value_type ComputeHash(StringRef K); 283 284 static std::pair<offset_type, offset_type> 285 ReadKeyDataLength(const unsigned char *&D) { 286 using namespace support; 287 offset_type KeyLen = endian::readNext<offset_type, little, unaligned>(D); 288 offset_type DataLen = endian::readNext<offset_type, little, unaligned>(D); 289 return std::make_pair(KeyLen, DataLen); 290 } 291 292 StringRef ReadKey(const unsigned char *D, offset_type N) { 293 return StringRef((const char *)D, N); 294 } 295 296 bool readValueProfilingData(const unsigned char *&D, 297 const unsigned char *const End); 298 data_type ReadData(StringRef K, const unsigned char *D, offset_type N); 299 300 // Used for testing purpose only. 301 void setValueProfDataEndianness(support::endianness Endianness) { 302 ValueProfDataEndianness = Endianness; 303 } 304 }; 305 306 struct InstrProfReaderIndexBase { 307 // Read all the profile records with the same key pointed to the current 308 // iterator. 309 virtual Error getRecords(ArrayRef<InstrProfRecord> &Data) = 0; 310 // Read all the profile records with the key equal to FuncName 311 virtual Error getRecords(StringRef FuncName, 312 ArrayRef<InstrProfRecord> &Data) = 0; 313 virtual void advanceToNextKey() = 0; 314 virtual bool atEnd() const = 0; 315 virtual void setValueProfDataEndianness(support::endianness Endianness) = 0; 316 virtual ~InstrProfReaderIndexBase() {} 317 virtual uint64_t getVersion() const = 0; 318 virtual bool isIRLevelProfile() const = 0; 319 virtual void populateSymtab(InstrProfSymtab &) = 0; 320 }; 321 322 typedef OnDiskIterableChainedHashTable<InstrProfLookupTrait> 323 OnDiskHashTableImplV3; 324 325 template <typename HashTableImpl> 326 class InstrProfReaderIndex : public InstrProfReaderIndexBase { 327 328 private: 329 std::unique_ptr<HashTableImpl> HashTable; 330 typename HashTableImpl::data_iterator RecordIterator; 331 uint64_t FormatVersion; 332 333 public: 334 InstrProfReaderIndex(const unsigned char *Buckets, 335 const unsigned char *const Payload, 336 const unsigned char *const Base, 337 IndexedInstrProf::HashT HashType, uint64_t Version); 338 339 Error getRecords(ArrayRef<InstrProfRecord> &Data) override; 340 Error getRecords(StringRef FuncName, 341 ArrayRef<InstrProfRecord> &Data) override; 342 void advanceToNextKey() override { RecordIterator++; } 343 bool atEnd() const override { 344 return RecordIterator == HashTable->data_end(); 345 } 346 void setValueProfDataEndianness(support::endianness Endianness) override { 347 HashTable->getInfoObj().setValueProfDataEndianness(Endianness); 348 } 349 ~InstrProfReaderIndex() override {} 350 uint64_t getVersion() const override { return GET_VERSION(FormatVersion); } 351 bool isIRLevelProfile() const override { 352 return (FormatVersion & VARIANT_MASK_IR_PROF) != 0; 353 } 354 void populateSymtab(InstrProfSymtab &Symtab) override { 355 Symtab.create(HashTable->keys()); 356 } 357 }; 358 359 /// Reader for the indexed binary instrprof format. 360 class IndexedInstrProfReader : public InstrProfReader { 361 private: 362 /// The profile data file contents. 363 std::unique_ptr<MemoryBuffer> DataBuffer; 364 /// The index into the profile data. 365 std::unique_ptr<InstrProfReaderIndexBase> Index; 366 /// Profile summary data. 367 std::unique_ptr<ProfileSummary> Summary; 368 369 IndexedInstrProfReader(const IndexedInstrProfReader &) = delete; 370 IndexedInstrProfReader &operator=(const IndexedInstrProfReader &) = delete; 371 372 // Read the profile summary. Return a pointer pointing to one byte past the 373 // end of the summary data if it exists or the input \c Cur. 374 const unsigned char *readSummary(IndexedInstrProf::ProfVersion Version, 375 const unsigned char *Cur); 376 377 public: 378 /// Return the profile version. 379 uint64_t getVersion() const { return Index->getVersion(); } 380 bool isIRLevelProfile() const override { return Index->isIRLevelProfile(); } 381 IndexedInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer) 382 : DataBuffer(std::move(DataBuffer)), Index(nullptr) {} 383 384 /// Return true if the given buffer is in an indexed instrprof format. 385 static bool hasFormat(const MemoryBuffer &DataBuffer); 386 387 /// Read the file header. 388 Error readHeader() override; 389 /// Read a single record. 390 Error readNextRecord(InstrProfRecord &Record) override; 391 392 /// Return the pointer to InstrProfRecord associated with FuncName 393 /// and FuncHash 394 Expected<InstrProfRecord> getInstrProfRecord(StringRef FuncName, 395 uint64_t FuncHash); 396 397 /// Fill Counts with the profile data for the given function name. 398 Error getFunctionCounts(StringRef FuncName, uint64_t FuncHash, 399 std::vector<uint64_t> &Counts); 400 401 /// Return the maximum of all known function counts. 402 uint64_t getMaximumFunctionCount() { return Summary->getMaxFunctionCount(); } 403 404 /// Factory method to create an indexed reader. 405 static Expected<std::unique_ptr<IndexedInstrProfReader>> 406 create(const Twine &Path); 407 408 static Expected<std::unique_ptr<IndexedInstrProfReader>> 409 create(std::unique_ptr<MemoryBuffer> Buffer); 410 411 // Used for testing purpose only. 412 void setValueProfDataEndianness(support::endianness Endianness) { 413 Index->setValueProfDataEndianness(Endianness); 414 } 415 416 // See description in the base class. This interface is designed 417 // to be used by llvm-profdata (for dumping). Avoid using this when 418 // the client is the compiler. 419 InstrProfSymtab &getSymtab() override; 420 ProfileSummary &getSummary() { return *(Summary.get()); } 421 }; 422 423 } // end namespace llvm 424 425 #endif 426