1 //=-- InstrProfReader.h - Instrumented profiling readers ----------*- C++ -*-=//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file contains support for reading profiling data for instrumentation
11 // based PGO and coverage.
13 //===----------------------------------------------------------------------===//
15 #ifndef LLVM_PROFILEDATA_INSTRPROFREADER_H
16 #define LLVM_PROFILEDATA_INSTRPROFREADER_H
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/ErrorOr.h"
23 #include "llvm/Support/LineIterator.h"
24 #include "llvm/Support/MemoryBuffer.h"
25 #include "llvm/Support/OnDiskHashTable.h"
30 class InstrProfReader;
32 /// Profiling information for a single function.
33 struct InstrProfRecord {
35 InstrProfRecord(StringRef Name, uint64_t Hash, ArrayRef<uint64_t> Counts)
36 : Name(Name), Hash(Hash), Counts(Counts) {}
39 ArrayRef<uint64_t> Counts;
42 /// A file format agnostic iterator over profiling data.
43 class InstrProfIterator : public std::iterator<std::input_iterator_tag,
45 InstrProfReader *Reader;
46 InstrProfRecord Record;
50 InstrProfIterator() : Reader(nullptr) {}
51 InstrProfIterator(InstrProfReader *Reader) : Reader(Reader) { Increment(); }
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; }
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;
66 InstrProfReader() : LastError(instrprof_error::success) {}
67 virtual ~InstrProfReader() {}
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.
74 InstrProfIterator begin() { return InstrProfIterator(this); }
75 InstrProfIterator end() { return InstrProfIterator(); }
78 /// Set the current std::error_code and return same.
79 std::error_code error(std::error_code EC) {
84 /// Clear the current error code and return a successful one.
85 std::error_code success() { return error(instrprof_error::success); }
88 /// Return true if the reader has finished reading the profile data.
89 bool isEOF() { return LastError == instrprof_error::eof; }
90 /// Return true if the reader encountered an error reading profiling data.
91 bool hasError() { return LastError && !isEOF(); }
92 /// Get the current error code.
93 std::error_code getError() { return LastError; }
95 /// Factory method to create an appropriately typed reader for the given
97 static ErrorOr<std::unique_ptr<InstrProfReader>> create(std::string Path);
100 /// Reader for the simple text based instrprof format.
102 /// This format is a simple text format that's suitable for test data. Records
103 /// are separated by one or more blank lines, and record fields are separated by
106 /// Each record consists of a function name, a function hash, a number of
107 /// counters, and then each counter value, in that order.
108 class TextInstrProfReader : public InstrProfReader {
110 /// The profile data file contents.
111 std::unique_ptr<MemoryBuffer> DataBuffer;
112 /// Iterator over the profile data.
114 /// The current set of counter values.
115 std::vector<uint64_t> Counts;
117 TextInstrProfReader(const TextInstrProfReader &) LLVM_DELETED_FUNCTION;
118 TextInstrProfReader &operator=(const TextInstrProfReader &)
119 LLVM_DELETED_FUNCTION;
121 TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_)
122 : DataBuffer(std::move(DataBuffer_)), Line(*DataBuffer, true, '#') {}
125 std::error_code readHeader() override { return success(); }
126 /// Read a single record.
127 std::error_code readNextRecord(InstrProfRecord &Record) override;
130 /// Reader for the raw instrprof binary format from runtime.
132 /// This format is a raw memory dump of the instrumentation-baed profiling data
133 /// from the runtime. It has no index.
135 /// Templated on the unsigned type whose size matches pointers on the platform
136 /// that wrote the profile.
137 template <class IntPtrT>
138 class RawInstrProfReader : public InstrProfReader {
140 /// The profile data file contents.
141 std::unique_ptr<MemoryBuffer> DataBuffer;
142 /// The current set of counter values.
143 std::vector<uint64_t> Counts;
145 const uint32_t NameSize;
146 const uint32_t NumCounters;
147 const uint64_t FuncHash;
148 const IntPtrT NamePtr;
149 const IntPtrT CounterPtr;
152 const uint64_t Magic;
153 const uint64_t Version;
154 const uint64_t DataSize;
155 const uint64_t CountersSize;
156 const uint64_t NamesSize;
157 const uint64_t CountersDelta;
158 const uint64_t NamesDelta;
161 bool ShouldSwapBytes;
162 uint64_t CountersDelta;
164 const ProfileData *Data;
165 const ProfileData *DataEnd;
166 const uint64_t *CountersStart;
167 const char *NamesStart;
168 const char *ProfileEnd;
170 RawInstrProfReader(const RawInstrProfReader &) LLVM_DELETED_FUNCTION;
171 RawInstrProfReader &operator=(const RawInstrProfReader &)
172 LLVM_DELETED_FUNCTION;
174 RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)
175 : DataBuffer(std::move(DataBuffer)) { }
177 static bool hasFormat(const MemoryBuffer &DataBuffer);
178 std::error_code readHeader() override;
179 std::error_code readNextRecord(InstrProfRecord &Record) override;
182 std::error_code readNextHeader(const char *CurrentPos);
183 std::error_code readHeader(const RawHeader &Header);
184 template <class IntT>
185 IntT swap(IntT Int) const {
186 return ShouldSwapBytes ? sys::getSwappedBytes(Int) : Int;
188 const uint64_t *getCounter(IntPtrT CounterPtr) const {
189 ptrdiff_t Offset = (swap(CounterPtr) - CountersDelta) / sizeof(uint64_t);
190 return CountersStart + Offset;
192 const char *getName(IntPtrT NamePtr) const {
193 ptrdiff_t Offset = (swap(NamePtr) - NamesDelta) / sizeof(char);
194 return NamesStart + Offset;
198 typedef RawInstrProfReader<uint32_t> RawInstrProfReader32;
199 typedef RawInstrProfReader<uint64_t> RawInstrProfReader64;
201 namespace IndexedInstrProf {
202 enum class HashT : uint32_t;
205 /// Trait for lookups into the on-disk hash table for the binary instrprof
207 class InstrProfLookupTrait {
208 std::vector<uint64_t> DataBuffer;
209 IndexedInstrProf::HashT HashType;
211 InstrProfLookupTrait(IndexedInstrProf::HashT HashType) : HashType(HashType) {}
214 data_type(StringRef Name, ArrayRef<uint64_t> Data)
215 : Name(Name), Data(Data) {}
217 ArrayRef<uint64_t> Data;
219 typedef StringRef internal_key_type;
220 typedef StringRef external_key_type;
221 typedef uint64_t hash_value_type;
222 typedef uint64_t offset_type;
224 static bool EqualKey(StringRef A, StringRef B) { return A == B; }
225 static StringRef GetInternalKey(StringRef K) { return K; }
227 hash_value_type ComputeHash(StringRef K);
229 static std::pair<offset_type, offset_type>
230 ReadKeyDataLength(const unsigned char *&D) {
231 using namespace support;
232 offset_type KeyLen = endian::readNext<offset_type, little, unaligned>(D);
233 offset_type DataLen = endian::readNext<offset_type, little, unaligned>(D);
234 return std::make_pair(KeyLen, DataLen);
237 StringRef ReadKey(const unsigned char *D, offset_type N) {
238 return StringRef((const char *)D, N);
241 data_type ReadData(StringRef K, const unsigned char *D, offset_type N) {
243 if (N % sizeof(uint64_t))
244 // The data is corrupt, don't try to read it.
245 return data_type("", DataBuffer);
247 using namespace support;
248 // We just treat the data as opaque here. It's simpler to handle in
249 // IndexedInstrProfReader.
250 unsigned NumEntries = N / sizeof(uint64_t);
251 DataBuffer.reserve(NumEntries);
252 for (unsigned I = 0; I < NumEntries; ++I)
253 DataBuffer.push_back(endian::readNext<uint64_t, little, unaligned>(D));
254 return data_type(K, DataBuffer);
257 typedef OnDiskIterableChainedHashTable<InstrProfLookupTrait>
258 InstrProfReaderIndex;
260 /// Reader for the indexed binary instrprof format.
261 class IndexedInstrProfReader : public InstrProfReader {
263 /// The profile data file contents.
264 std::unique_ptr<MemoryBuffer> DataBuffer;
265 /// The index into the profile data.
266 std::unique_ptr<InstrProfReaderIndex> Index;
267 /// Iterator over the profile data.
268 InstrProfReaderIndex::data_iterator RecordIterator;
269 /// Offset into our current data set.
270 size_t CurrentOffset;
271 /// The file format version of the profile data.
272 uint64_t FormatVersion;
273 /// The maximal execution count among all functions.
274 uint64_t MaxFunctionCount;
276 IndexedInstrProfReader(const IndexedInstrProfReader &) LLVM_DELETED_FUNCTION;
277 IndexedInstrProfReader &operator=(const IndexedInstrProfReader &)
278 LLVM_DELETED_FUNCTION;
280 IndexedInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)
281 : DataBuffer(std::move(DataBuffer)), Index(nullptr), CurrentOffset(0) {}
283 /// Return true if the given buffer is in an indexed instrprof format.
284 static bool hasFormat(const MemoryBuffer &DataBuffer);
286 /// Read the file header.
287 std::error_code readHeader() override;
288 /// Read a single record.
289 std::error_code readNextRecord(InstrProfRecord &Record) override;
291 /// Fill Counts with the profile data for the given function name.
292 std::error_code getFunctionCounts(StringRef FuncName, uint64_t FuncHash,
293 std::vector<uint64_t> &Counts);
294 /// Return the maximum of all known function counts.
295 uint64_t getMaximumFunctionCount() { return MaxFunctionCount; }
297 /// Factory method to create an indexed reader.
298 static std::error_code
299 create(std::string Path, std::unique_ptr<IndexedInstrProfReader> &Result);
302 } // end namespace llvm