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/ErrorOr.h"
23#include "llvm/Support/LineIterator.h"
24#include "llvm/Support/MemoryBuffer.h"
25#include "llvm/Support/OnDiskHashTable.h"
26#include <iterator>
27
28namespace llvm {
29
30class InstrProfReader;
31
32/// Profiling information for a single function.
33struct InstrProfRecord {
34  InstrProfRecord() {}
35  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.
43class InstrProfIterator : public std::iterator<std::input_iterator_tag,
44                                               InstrProfRecord> {
45  InstrProfReader *Reader;
46  InstrProfRecord Record;
47
48  void Increment();
49public:
50  InstrProfIterator() : Reader(nullptr) {}
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.
62class InstrProfReader {
63  std::error_code LastError;
64
65public:
66  InstrProfReader() : LastError(instrprof_error::success) {}
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.
74  InstrProfIterator begin() { return InstrProfIterator(this); }
75  InstrProfIterator end() { return InstrProfIterator(); }
76
77protected:
78  /// Set the current std::error_code and return same.
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.
85  std::error_code success() { return error(instrprof_error::success); }
86
87public:
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; }
94
95  /// Factory method to create an appropriately typed reader for the given
96  /// instrprof file.
97  static ErrorOr<std::unique_ptr<InstrProfReader>> create(std::string Path);
98
99  static ErrorOr<std::unique_ptr<InstrProfReader>>
100  create(std::unique_ptr<MemoryBuffer> Buffer);
101};
102
103/// Reader for the simple text based instrprof format.
104///
105/// This format is a simple text format that's suitable for test data. Records
106/// are separated by one or more blank lines, and record fields are separated by
107/// new lines.
108///
109/// Each record consists of a function name, a function hash, a number of
110/// counters, and then each counter value, in that order.
111class TextInstrProfReader : public InstrProfReader {
112private:
113  /// The profile data file contents.
114  std::unique_ptr<MemoryBuffer> DataBuffer;
115  /// Iterator over the profile data.
116  line_iterator Line;
117  /// The current set of counter values.
118  std::vector<uint64_t> Counts;
119
120  TextInstrProfReader(const TextInstrProfReader &) = delete;
121  TextInstrProfReader &operator=(const TextInstrProfReader &) = delete;
122public:
123  TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_)
124      : DataBuffer(std::move(DataBuffer_)), Line(*DataBuffer, true, '#') {}
125
126  /// Read the header.
127  std::error_code readHeader() override { return success(); }
128  /// Read a single record.
129  std::error_code readNextRecord(InstrProfRecord &Record) override;
130};
131
132/// Reader for the raw instrprof binary format from runtime.
133///
134/// This format is a raw memory dump of the instrumentation-baed profiling data
135/// from the runtime.  It has no index.
136///
137/// Templated on the unsigned type whose size matches pointers on the platform
138/// that wrote the profile.
139template <class IntPtrT>
140class RawInstrProfReader : public InstrProfReader {
141private:
142  /// The profile data file contents.
143  std::unique_ptr<MemoryBuffer> DataBuffer;
144  /// The current set of counter values.
145  std::vector<uint64_t> Counts;
146  struct ProfileData {
147    const uint32_t NameSize;
148    const uint32_t NumCounters;
149    const uint64_t FuncHash;
150    const IntPtrT NamePtr;
151    const IntPtrT CounterPtr;
152  };
153  struct RawHeader {
154    const uint64_t Magic;
155    const uint64_t Version;
156    const uint64_t DataSize;
157    const uint64_t CountersSize;
158    const uint64_t NamesSize;
159    const uint64_t CountersDelta;
160    const uint64_t NamesDelta;
161  };
162
163  bool ShouldSwapBytes;
164  uint64_t CountersDelta;
165  uint64_t NamesDelta;
166  const ProfileData *Data;
167  const ProfileData *DataEnd;
168  const uint64_t *CountersStart;
169  const char *NamesStart;
170  const char *ProfileEnd;
171
172  RawInstrProfReader(const RawInstrProfReader &) = delete;
173  RawInstrProfReader &operator=(const RawInstrProfReader &) = delete;
174public:
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
182private:
183  std::error_code readNextHeader(const char *CurrentPos);
184  std::error_code readHeader(const RawHeader &Header);
185  template <class IntT>
186  IntT swap(IntT Int) const {
187    return ShouldSwapBytes ? sys::getSwappedBytes(Int) : Int;
188  }
189  const uint64_t *getCounter(IntPtrT CounterPtr) const {
190    ptrdiff_t Offset = (swap(CounterPtr) - CountersDelta) / sizeof(uint64_t);
191    return CountersStart + Offset;
192  }
193  const char *getName(IntPtrT NamePtr) const {
194    ptrdiff_t Offset = (swap(NamePtr) - NamesDelta) / sizeof(char);
195    return NamesStart + Offset;
196  }
197};
198
199typedef RawInstrProfReader<uint32_t> RawInstrProfReader32;
200typedef RawInstrProfReader<uint64_t> RawInstrProfReader64;
201
202namespace IndexedInstrProf {
203enum class HashT : uint32_t;
204}
205
206/// Trait for lookups into the on-disk hash table for the binary instrprof
207/// format.
208class InstrProfLookupTrait {
209  std::vector<uint64_t> DataBuffer;
210  IndexedInstrProf::HashT HashType;
211public:
212  InstrProfLookupTrait(IndexedInstrProf::HashT HashType) : HashType(HashType) {}
213
214  struct data_type {
215    data_type(StringRef Name, ArrayRef<uint64_t> Data)
216        : Name(Name), Data(Data) {}
217    StringRef Name;
218    ArrayRef<uint64_t> Data;
219  };
220  typedef StringRef internal_key_type;
221  typedef StringRef external_key_type;
222  typedef uint64_t hash_value_type;
223  typedef uint64_t offset_type;
224
225  static bool EqualKey(StringRef A, StringRef B) { return A == B; }
226  static StringRef GetInternalKey(StringRef K) { return K; }
227
228  hash_value_type ComputeHash(StringRef K);
229
230  static std::pair<offset_type, offset_type>
231  ReadKeyDataLength(const unsigned char *&D) {
232    using namespace support;
233    offset_type KeyLen = endian::readNext<offset_type, little, unaligned>(D);
234    offset_type DataLen = endian::readNext<offset_type, little, unaligned>(D);
235    return std::make_pair(KeyLen, DataLen);
236  }
237
238  StringRef ReadKey(const unsigned char *D, offset_type N) {
239    return StringRef((const char *)D, N);
240  }
241
242  data_type ReadData(StringRef K, const unsigned char *D, offset_type N) {
243    DataBuffer.clear();
244    if (N % sizeof(uint64_t))
245      // The data is corrupt, don't try to read it.
246      return data_type("", DataBuffer);
247
248    using namespace support;
249    // We just treat the data as opaque here. It's simpler to handle in
250    // IndexedInstrProfReader.
251    unsigned NumEntries = N / sizeof(uint64_t);
252    DataBuffer.reserve(NumEntries);
253    for (unsigned I = 0; I < NumEntries; ++I)
254      DataBuffer.push_back(endian::readNext<uint64_t, little, unaligned>(D));
255    return data_type(K, DataBuffer);
256  }
257};
258typedef OnDiskIterableChainedHashTable<InstrProfLookupTrait>
259    InstrProfReaderIndex;
260
261/// Reader for the indexed binary instrprof format.
262class IndexedInstrProfReader : public InstrProfReader {
263private:
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  /// Offset into our current data set.
271  size_t CurrentOffset;
272  /// The file format version of the profile data.
273  uint64_t FormatVersion;
274  /// The maximal execution count among all functions.
275  uint64_t MaxFunctionCount;
276
277  IndexedInstrProfReader(const IndexedInstrProfReader &) = delete;
278  IndexedInstrProfReader &operator=(const IndexedInstrProfReader &) = delete;
279public:
280  IndexedInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)
281      : DataBuffer(std::move(DataBuffer)), Index(nullptr), CurrentOffset(0) {}
282
283  /// Return true if the given buffer is in an indexed instrprof format.
284  static bool hasFormat(const MemoryBuffer &DataBuffer);
285
286  /// Read the file header.
287  std::error_code readHeader() override;
288  /// Read a single record.
289  std::error_code readNextRecord(InstrProfRecord &Record) override;
290
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; }
296
297  /// Factory method to create an indexed reader.
298  static ErrorOr<std::unique_ptr<IndexedInstrProfReader>>
299  create(std::string Path);
300
301  static ErrorOr<std::unique_ptr<IndexedInstrProfReader>>
302  create(std::unique_ptr<MemoryBuffer> Buffer);
303};
304
305} // end namespace llvm
306
307#endif
308