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_INSTRPROF_READER_H_ 16 #define LLVM_PROFILEDATA_INSTRPROF_READER_H_ 17 18 #include "llvm/ADT/ArrayRef.h" 19 #include "llvm/ADT/StringExtras.h" 20 #include "llvm/ProfileData/InstrProf.h" 21 #include "llvm/Support/LineIterator.h" 22 #include "llvm/Support/MemoryBuffer.h" 23 #include "llvm/Support/EndianStream.h" 24 #include "llvm/Support/OnDiskHashTable.h" 25 26 #include <iterator> 27 28 namespace llvm { 29 30 class InstrProfReader; 31 32 /// Profiling information for a single function. 33 struct InstrProfRecord { InstrProfRecordInstrProfRecord34 InstrProfRecord() {} InstrProfRecordInstrProfRecord35 InstrProfRecord(StringRef Name, uint64_t Hash, ArrayRef<uint64_t> Counts) 36 : Name(Name), Hash(Hash), Counts(Counts) {} 37 StringRef Name; 38 uint64_t Hash; 39 ArrayRef<uint64_t> Counts; 40 }; 41 42 /// A file format agnostic iterator over profiling data. 43 class InstrProfIterator : public std::iterator<std::input_iterator_tag, 44 InstrProfRecord> { 45 InstrProfReader *Reader; 46 InstrProfRecord Record; 47 48 void Increment(); 49 public: InstrProfIterator()50 InstrProfIterator() : Reader(nullptr) {} InstrProfIterator(InstrProfReader * Reader)51 InstrProfIterator(InstrProfReader *Reader) : Reader(Reader) { Increment(); } 52 53 InstrProfIterator &operator++() { Increment(); return *this; } 54 bool operator==(const InstrProfIterator &RHS) { return Reader == RHS.Reader; } 55 bool operator!=(const InstrProfIterator &RHS) { return Reader != RHS.Reader; } 56 InstrProfRecord &operator*() { return Record; } 57 InstrProfRecord *operator->() { return &Record; } 58 }; 59 60 /// Base class and interface for reading profiling data of any known instrprof 61 /// format. Provides an iterator over InstrProfRecords. 62 class InstrProfReader { 63 std::error_code LastError; 64 65 public: InstrProfReader()66 InstrProfReader() : LastError(instrprof_error::success) {} ~InstrProfReader()67 virtual ~InstrProfReader() {} 68 69 /// Read the header. Required before reading first record. 70 virtual std::error_code readHeader() = 0; 71 /// Read a single record. 72 virtual std::error_code readNextRecord(InstrProfRecord &Record) = 0; 73 /// Iterator over profile data. begin()74 InstrProfIterator begin() { return InstrProfIterator(this); } end()75 InstrProfIterator end() { return InstrProfIterator(); } 76 77 protected: 78 /// Set the current std::error_code and return same. error(std::error_code EC)79 std::error_code error(std::error_code EC) { 80 LastError = EC; 81 return EC; 82 } 83 84 /// Clear the current error code and return a successful one. success()85 std::error_code success() { return error(instrprof_error::success); } 86 87 public: 88 /// Return true if the reader has finished reading the profile data. isEOF()89 bool isEOF() { return LastError == instrprof_error::eof; } 90 /// Return true if the reader encountered an error reading profiling data. hasError()91 bool hasError() { return LastError && !isEOF(); } 92 /// Get the current error code. getError()93 std::error_code getError() { return LastError; } 94 95 /// Factory method to create an appropriately typed reader for the given 96 /// instrprof file. 97 static std::error_code create(std::string Path, 98 std::unique_ptr<InstrProfReader> &Result); 99 }; 100 101 /// Reader for the simple text based instrprof format. 102 /// 103 /// This format is a simple text format that's suitable for test data. Records 104 /// are separated by one or more blank lines, and record fields are separated by 105 /// new lines. 106 /// 107 /// Each record consists of a function name, a function hash, a number of 108 /// counters, and then each counter value, in that order. 109 class TextInstrProfReader : public InstrProfReader { 110 private: 111 /// The profile data file contents. 112 std::unique_ptr<MemoryBuffer> DataBuffer; 113 /// Iterator over the profile data. 114 line_iterator Line; 115 /// The current set of counter values. 116 std::vector<uint64_t> Counts; 117 118 TextInstrProfReader(const TextInstrProfReader &) LLVM_DELETED_FUNCTION; 119 TextInstrProfReader &operator=(const TextInstrProfReader &) 120 LLVM_DELETED_FUNCTION; 121 public: TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_)122 TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_) 123 : DataBuffer(std::move(DataBuffer_)), Line(*DataBuffer, '#') {} 124 125 /// Read the header. readHeader()126 std::error_code readHeader() override { return success(); } 127 /// Read a single record. 128 std::error_code readNextRecord(InstrProfRecord &Record) override; 129 }; 130 131 /// Reader for the raw instrprof binary format from runtime. 132 /// 133 /// This format is a raw memory dump of the instrumentation-baed profiling data 134 /// from the runtime. It has no index. 135 /// 136 /// Templated on the unsigned type whose size matches pointers on the platform 137 /// that wrote the profile. 138 template <class IntPtrT> 139 class RawInstrProfReader : public InstrProfReader { 140 private: 141 /// The profile data file contents. 142 std::unique_ptr<MemoryBuffer> DataBuffer; 143 /// The current set of counter values. 144 std::vector<uint64_t> Counts; 145 struct ProfileData { 146 const uint32_t NameSize; 147 const uint32_t NumCounters; 148 const uint64_t FuncHash; 149 const IntPtrT NamePtr; 150 const IntPtrT CounterPtr; 151 }; 152 struct RawHeader { 153 const uint64_t Magic; 154 const uint64_t Version; 155 const uint64_t DataSize; 156 const uint64_t CountersSize; 157 const uint64_t NamesSize; 158 const uint64_t CountersDelta; 159 const uint64_t NamesDelta; 160 }; 161 162 bool ShouldSwapBytes; 163 uint64_t CountersDelta; 164 uint64_t NamesDelta; 165 const ProfileData *Data; 166 const ProfileData *DataEnd; 167 const uint64_t *CountersStart; 168 const char *NamesStart; 169 const char *ProfileEnd; 170 171 RawInstrProfReader(const RawInstrProfReader &) LLVM_DELETED_FUNCTION; 172 RawInstrProfReader &operator=(const RawInstrProfReader &) 173 LLVM_DELETED_FUNCTION; 174 public: RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)175 RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer) 176 : DataBuffer(std::move(DataBuffer)) { } 177 178 static bool hasFormat(const MemoryBuffer &DataBuffer); 179 std::error_code readHeader() override; 180 std::error_code readNextRecord(InstrProfRecord &Record) override; 181 182 private: 183 std::error_code readNextHeader(const char *CurrentPos); 184 std::error_code readHeader(const RawHeader &Header); 185 template <class IntT> swap(IntT Int)186 IntT swap(IntT Int) const { 187 return ShouldSwapBytes ? sys::getSwappedBytes(Int) : Int; 188 } getCounter(IntPtrT CounterPtr)189 const uint64_t *getCounter(IntPtrT CounterPtr) const { 190 ptrdiff_t Offset = (swap(CounterPtr) - CountersDelta) / sizeof(uint64_t); 191 return CountersStart + Offset; 192 } getName(IntPtrT NamePtr)193 const char *getName(IntPtrT NamePtr) const { 194 ptrdiff_t Offset = (swap(NamePtr) - NamesDelta) / sizeof(char); 195 return NamesStart + Offset; 196 } 197 }; 198 199 typedef RawInstrProfReader<uint32_t> RawInstrProfReader32; 200 typedef RawInstrProfReader<uint64_t> RawInstrProfReader64; 201 202 namespace IndexedInstrProf { 203 enum class HashT : uint32_t; 204 } 205 206 /// Trait for lookups into the on-disk hash table for the binary instrprof 207 /// format. 208 class InstrProfLookupTrait { 209 std::vector<uint64_t> CountBuffer; 210 IndexedInstrProf::HashT HashType; 211 public: InstrProfLookupTrait(IndexedInstrProf::HashT HashType)212 InstrProfLookupTrait(IndexedInstrProf::HashT HashType) : HashType(HashType) {} 213 214 typedef InstrProfRecord data_type; 215 typedef StringRef internal_key_type; 216 typedef StringRef external_key_type; 217 typedef uint64_t hash_value_type; 218 typedef uint64_t offset_type; 219 EqualKey(StringRef A,StringRef B)220 static bool EqualKey(StringRef A, StringRef B) { return A == B; } GetInternalKey(StringRef K)221 static StringRef GetInternalKey(StringRef K) { return K; } 222 223 hash_value_type ComputeHash(StringRef K); 224 225 static std::pair<offset_type, offset_type> ReadKeyDataLength(const unsigned char * & D)226 ReadKeyDataLength(const unsigned char *&D) { 227 using namespace support; 228 offset_type KeyLen = endian::readNext<offset_type, little, unaligned>(D); 229 offset_type DataLen = endian::readNext<offset_type, little, unaligned>(D); 230 return std::make_pair(KeyLen, DataLen); 231 } 232 ReadKey(const unsigned char * D,offset_type N)233 StringRef ReadKey(const unsigned char *D, offset_type N) { 234 return StringRef((const char *)D, N); 235 } 236 ReadData(StringRef K,const unsigned char * D,offset_type N)237 InstrProfRecord ReadData(StringRef K, const unsigned char *D, offset_type N) { 238 if (N < 2 * sizeof(uint64_t) || N % sizeof(uint64_t)) { 239 // The data is corrupt, don't try to read it. 240 CountBuffer.clear(); 241 return InstrProfRecord("", 0, CountBuffer); 242 } 243 244 using namespace support; 245 246 // The first stored value is the hash. 247 uint64_t Hash = endian::readNext<uint64_t, little, unaligned>(D); 248 // Each counter follows. 249 unsigned NumCounters = N / sizeof(uint64_t) - 1; 250 CountBuffer.clear(); 251 CountBuffer.reserve(NumCounters - 1); 252 for (unsigned I = 0; I < NumCounters; ++I) 253 CountBuffer.push_back(endian::readNext<uint64_t, little, unaligned>(D)); 254 255 return InstrProfRecord(K, Hash, CountBuffer); 256 } 257 }; 258 typedef OnDiskIterableChainedHashTable<InstrProfLookupTrait> 259 InstrProfReaderIndex; 260 261 /// Reader for the indexed binary instrprof format. 262 class IndexedInstrProfReader : public InstrProfReader { 263 private: 264 /// The profile data file contents. 265 std::unique_ptr<MemoryBuffer> DataBuffer; 266 /// The index into the profile data. 267 std::unique_ptr<InstrProfReaderIndex> Index; 268 /// Iterator over the profile data. 269 InstrProfReaderIndex::data_iterator RecordIterator; 270 /// The maximal execution count among all fucntions. 271 uint64_t MaxFunctionCount; 272 273 IndexedInstrProfReader(const IndexedInstrProfReader &) LLVM_DELETED_FUNCTION; 274 IndexedInstrProfReader &operator=(const IndexedInstrProfReader &) 275 LLVM_DELETED_FUNCTION; 276 public: IndexedInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)277 IndexedInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer) 278 : DataBuffer(std::move(DataBuffer)), Index(nullptr), 279 RecordIterator(InstrProfReaderIndex::data_iterator()) {} 280 281 /// Return true if the given buffer is in an indexed instrprof format. 282 static bool hasFormat(const MemoryBuffer &DataBuffer); 283 284 /// Read the file header. 285 std::error_code readHeader() override; 286 /// Read a single record. 287 std::error_code readNextRecord(InstrProfRecord &Record) override; 288 289 /// Fill Counts with the profile data for the given function name. 290 std::error_code getFunctionCounts(StringRef FuncName, uint64_t &FuncHash, 291 std::vector<uint64_t> &Counts); 292 /// Return the maximum of all known function counts. getMaximumFunctionCount()293 uint64_t getMaximumFunctionCount() { return MaxFunctionCount; } 294 295 /// Factory method to create an indexed reader. 296 static std::error_code 297 create(std::string Path, std::unique_ptr<IndexedInstrProfReader> &Result); 298 }; 299 300 } // end namespace llvm 301 302 #endif // LLVM_PROFILEDATA_INSTRPROF_READER_H_ 303