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/StringRef.h"
20 #include "llvm/IR/ProfileSummary.h"
21 #include "llvm/ProfileData/InstrProf.h"
22 #include "llvm/Support/Endian.h"
23 #include "llvm/Support/Error.h"
24 #include "llvm/Support/LineIterator.h"
25 #include "llvm/Support/MemoryBuffer.h"
26 #include "llvm/Support/OnDiskHashTable.h"
27 #include "llvm/Support/SwapByteOrder.h"
28 #include <algorithm>
29 #include <cassert>
30 #include <cstddef>
31 #include <cstdint>
32 #include <iterator>
33 #include <memory>
34 #include <utility>
35 #include <vector>
36
37 namespace llvm {
38
39 class InstrProfReader;
40
41 /// A file format agnostic iterator over profiling data.
42 class InstrProfIterator : public std::iterator<std::input_iterator_tag,
43 NamedInstrProfRecord> {
44 InstrProfReader *Reader = nullptr;
45 value_type Record;
46
47 void Increment();
48
49 public:
50 InstrProfIterator() = default;
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 value_type &operator*() { return Record; }
57 value_type *operator->() { return &Record; }
58 };
59
60 /// Base class and interface for reading profiling data of any known instrprof
61 /// format. Provides an iterator over NamedInstrProfRecords.
62 class InstrProfReader {
63 instrprof_error LastError = instrprof_error::success;
64
65 public:
66 InstrProfReader() = default;
67 virtual ~InstrProfReader() = default;
68
69 /// Read the header. Required before reading first record.
70 virtual Error readHeader() = 0;
71
72 /// Read a single record.
73 virtual Error readNextRecord(NamedInstrProfRecord &Record) = 0;
74
75 /// Iterator over profile data.
76 InstrProfIterator begin() { return InstrProfIterator(this); }
77 InstrProfIterator end() { return InstrProfIterator(); }
78
79 virtual bool isIRLevelProfile() const = 0;
80
81 /// Return the PGO symtab. There are three different readers:
82 /// Raw, Text, and Indexed profile readers. The first two types
83 /// of readers are used only by llvm-profdata tool, while the indexed
84 /// profile reader is also used by llvm-cov tool and the compiler (
85 /// backend or frontend). Since creating PGO symtab can create
86 /// significant runtime and memory overhead (as it touches data
87 /// for the whole program), InstrProfSymtab for the indexed profile
88 /// reader should be created on demand and it is recommended to be
89 /// only used for dumping purpose with llvm-proftool, not with the
90 /// compiler.
getStreamer()91 virtual InstrProfSymtab &getSymtab() = 0;
92
93 protected:
94 std::unique_ptr<InstrProfSymtab> Symtab;
95
96 /// Set the current error and return same.
97 Error error(instrprof_error Err) {
98 LastError = Err;
99 if (Err == instrprof_error::success)
100 return Error::success();
101 return make_error<InstrProfError>(Err);
102 }
103
104 Error error(Error &&E) { return error(InstrProfError::take(std::move(E))); }
105
106 /// Clear the current error and return a successful one.
107 Error success() { return error(instrprof_error::success); }
108
109 public:
110 /// Return true if the reader has finished reading the profile data.
111 bool isEOF() { return LastError == instrprof_error::eof; }
112
113 /// Return true if the reader encountered an error reading profiling data.
114 bool hasError() { return LastError != instrprof_error::success && !isEOF(); }
115
116 /// Get the current error.
117 Error getError() {
118 if (hasError())
119 return make_error<InstrProfError>(LastError);
120 return Error::success();
121 }
122
123 /// Factory method to create an appropriately typed reader for the given
124 /// instrprof file.
125 static Expected<std::unique_ptr<InstrProfReader>> create(const Twine &Path);
126
127 static Expected<std::unique_ptr<InstrProfReader>>
128 create(std::unique_ptr<MemoryBuffer> Buffer);
129 };
130
131 /// Reader for the simple text based instrprof format.
132 ///
133 /// This format is a simple text format that's suitable for test data. Records
134 /// are separated by one or more blank lines, and record fields are separated by
135 /// new lines.
136 ///
137 /// Each record consists of a function name, a function hash, a number of
138 /// counters, and then each counter value, in that order.
139 class TextInstrProfReader : public InstrProfReader {
140 private:
141 /// The profile data file contents.
142 std::unique_ptr<MemoryBuffer> DataBuffer;
143 /// Iterator over the profile data.
144 line_iterator Line;
145 bool IsIRLevelProfile = false;
146
147 Error readValueProfileData(InstrProfRecord &Record);
148
149 public:
150 TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_)
151 : DataBuffer(std::move(DataBuffer_)), Line(*DataBuffer, true, '#') {}
152 TextInstrProfReader(const TextInstrProfReader &) = delete;
153 TextInstrProfReader &operator=(const TextInstrProfReader &) = delete;
154
155 /// Return true if the given buffer is in text instrprof format.
156 static bool hasFormat(const MemoryBuffer &Buffer);
157
158 bool isIRLevelProfile() const override { return IsIRLevelProfile; }
159
160 /// Read the header.
161 Error readHeader() override;
162
163 /// Read a single record.
164 Error readNextRecord(NamedInstrProfRecord &Record) override;
165
166 InstrProfSymtab &getSymtab() override {
167 assert(Symtab.get());
168 return *Symtab.get();
169 }
170 };
171
172 /// Reader for the raw instrprof binary format from runtime.
173 ///
174 /// This format is a raw memory dump of the instrumentation-baed profiling data
175 /// from the runtime. It has no index.
176 ///
177 /// Templated on the unsigned type whose size matches pointers on the platform
178 /// that wrote the profile.
179 template <class IntPtrT>
180 class RawInstrProfReader : public InstrProfReader {
181 private:
182 /// The profile data file contents.
183 std::unique_ptr<MemoryBuffer> DataBuffer;
184 bool ShouldSwapBytes;
185 // The value of the version field of the raw profile data header. The lower 56
186 // bits specifies the format version and the most significant 8 bits specify
187 // the variant types of the profile.
188 uint64_t Version;
189 uint64_t CountersDelta;
190 uint64_t NamesDelta;
191 const RawInstrProf::ProfileData<IntPtrT> *Data;
192 const RawInstrProf::ProfileData<IntPtrT> *DataEnd;
193 const uint64_t *CountersStart;
194 const char *NamesStart;
195 uint64_t NamesSize;
196 // After value profile is all read, this pointer points to
197 // the header of next profile data (if exists)
198 const uint8_t *ValueDataStart;
199 uint32_t ValueKindLast;
200 uint32_t CurValueDataSize;
201
202 public:
203 RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)
204 : DataBuffer(std::move(DataBuffer)) {}
205 RawInstrProfReader(const RawInstrProfReader &) = delete;
206 RawInstrProfReader &operator=(const RawInstrProfReader &) = delete;
207
208 static bool hasFormat(const MemoryBuffer &DataBuffer);
209 Error readHeader() override;
210 Error readNextRecord(NamedInstrProfRecord &Record) override;
211
212 bool isIRLevelProfile() const override {
213 return (Version & VARIANT_MASK_IR_PROF) != 0;
214 }
215
216 InstrProfSymtab &getSymtab() override {
217 assert(Symtab.get());
218 return *Symtab.get();
219 }
220
221 private:
222 Error createSymtab(InstrProfSymtab &Symtab);
223 Error readNextHeader(const char *CurrentPos);
224 Error readHeader(const RawInstrProf::Header &Header);
225
getCurrentWinFrameInfo()226 template <class IntT> IntT swap(IntT Int) const {
227 return ShouldSwapBytes ? sys::getSwappedBytes(Int) : Int;
228 }
229
230 support::endianness getDataEndianness() const {
231 support::endianness HostEndian = getHostEndianness();
232 if (!ShouldSwapBytes)
233 return HostEndian;
234 if (HostEndian == support::little)
235 return support::big;
236 else
237 return support::little;
238 }
239
240 inline uint8_t getNumPaddingBytes(uint64_t SizeInBytes) {
241 return 7 & (sizeof(uint64_t) - SizeInBytes % sizeof(uint64_t));
setTargetStreamer(MCTargetStreamer * TS)242 }
243
244 Error readName(NamedInstrProfRecord &Record);
245 Error readFuncHash(NamedInstrProfRecord &Record);
246 Error readRawCounts(InstrProfRecord &Record);
247 Error readValueProfilingData(InstrProfRecord &Record);
248 bool atEnd() const { return Data == DataEnd; }
249
getContext()250 void advanceData() {
251 Data++;
252 ValueDataStart += CurValueDataSize;
253 }
setUseAssemblerInfoForParsing(bool v)254
255 const char *getNextHeaderPos() const {
256 assert(atEnd());
257 return (const char *)ValueDataStart;
258 }
259
260 const uint64_t *getCounter(IntPtrT CounterPtr) const {
getNumFrameInfos()261 ptrdiff_t Offset = (swap(CounterPtr) - CountersDelta) / sizeof(uint64_t);
getDwarfFrameInfos()262 return CountersStart + Offset;
263 }
264
265 StringRef getName(uint64_t NameRef) const {
266 return Symtab->getFuncName(swap(NameRef));
267 }
getNumWinFrameInfos()268 };
getWinFrameInfos()269
270 using RawInstrProfReader32 = RawInstrProfReader<uint32_t>;
271 using RawInstrProfReader64 = RawInstrProfReader<uint64_t>;
272
273 namespace IndexedInstrProf {
274
275 enum class HashT : uint32_t;
276
277 } // end namespace IndexedInstrProf
278
279 /// Trait for lookups into the on-disk hash table for the binary instrprof
isVerboseAsm()280 /// format.
281 class InstrProfLookupTrait {
282 std::vector<NamedInstrProfRecord> DataBuffer;
283 IndexedInstrProf::HashT HashType;
284 unsigned FormatVersion;
285 // Endianness of the input value profile data.
286 // It should be LE by default, but can be changed
287 // for testing purpose.
288 support::endianness ValueProfDataEndianness = support::little;
289
290 public:
291 InstrProfLookupTrait(IndexedInstrProf::HashT HashType, unsigned FormatVersion)
292 : HashType(HashType), FormatVersion(FormatVersion) {}
293
294 using data_type = ArrayRef<NamedInstrProfRecord>;
295
296 using internal_key_type = StringRef;
297 using external_key_type = StringRef;
298 using hash_value_type = uint64_t;
299 using offset_type = uint64_t;
300
301 static bool EqualKey(StringRef A, StringRef B) { return A == B; }
302 static StringRef GetInternalKey(StringRef K) { return K; }
303 static StringRef GetExternalKey(StringRef K) { return K; }
304
305 hash_value_type ComputeHash(StringRef K);
306
307 static std::pair<offset_type, offset_type>
308 ReadKeyDataLength(const unsigned char *&D) {
309 using namespace support;
310
311 offset_type KeyLen = endian::readNext<offset_type, little, unaligned>(D);
312 offset_type DataLen = endian::readNext<offset_type, little, unaligned>(D);
313 return std::make_pair(KeyLen, DataLen);
314 }
315
316 StringRef ReadKey(const unsigned char *D, offset_type N) {
317 return StringRef((const char *)D, N);
318 }
319
320 bool readValueProfilingData(const unsigned char *&D,
321 const unsigned char *const End);
322 data_type ReadData(StringRef K, const unsigned char *D, offset_type N);
323
324 // Used for testing purpose only.
325 void setValueProfDataEndianness(support::endianness Endianness) {
326 ValueProfDataEndianness = Endianness;
327 }
328 };
329
330 struct InstrProfReaderIndexBase {
331 virtual ~InstrProfReaderIndexBase() = default;
getCurrentSection()332
333 // Read all the profile records with the same key pointed to the current
334 // iterator.
335 virtual Error getRecords(ArrayRef<NamedInstrProfRecord> &Data) = 0;
336
getCurrentSectionOnly()337 // Read all the profile records with the key equal to FuncName
338 virtual Error getRecords(StringRef FuncName,
339 ArrayRef<NamedInstrProfRecord> &Data) = 0;
getPreviousSection()340 virtual void advanceToNextKey() = 0;
341 virtual bool atEnd() const = 0;
342 virtual void setValueProfDataEndianness(support::endianness Endianness) = 0;
343 virtual uint64_t getVersion() const = 0;
344 virtual bool isIRLevelProfile() const = 0;
345 virtual Error populateSymtab(InstrProfSymtab &) = 0;
346 };
347
GetSymbolOrder(const MCSymbol * Sym)348 using OnDiskHashTableImplV3 =
349 OnDiskIterableChainedHashTable<InstrProfLookupTrait>;
350
351 template <typename HashTableImpl>
352 class InstrProfReaderIndex : public InstrProfReaderIndexBase {
353 private:
354 std::unique_ptr<HashTableImpl> HashTable;
355 typename HashTableImpl::data_iterator RecordIterator;
356 uint64_t FormatVersion;
357
358 public:
PushSection()359 InstrProfReaderIndex(const unsigned char *Buckets,
360 const unsigned char *const Payload,
361 const unsigned char *const Base,
362 IndexedInstrProf::HashT HashType, uint64_t Version);
363 ~InstrProfReaderIndex() override = default;
364
365 Error getRecords(ArrayRef<NamedInstrProfRecord> &Data) override;
366 Error getRecords(StringRef FuncName,
367 ArrayRef<NamedInstrProfRecord> &Data) override;
PopSection()368 void advanceToNextKey() override { RecordIterator++; }
369
370 bool atEnd() const override {
371 return RecordIterator == HashTable->data_end();
372 }
373
374 void setValueProfDataEndianness(support::endianness Endianness) override {
375 HashTable->getInfoObj().setValueProfDataEndianness(Endianness);
376 }
377
378 uint64_t getVersion() const override { return GET_VERSION(FormatVersion); }
379
380 bool isIRLevelProfile() const override {
381 return (FormatVersion & VARIANT_MASK_IR_PROF) != 0;
382 }
SubSection(const MCExpr * Subsection)383
384 Error populateSymtab(InstrProfSymtab &Symtab) override {
385 return Symtab.create(HashTable->keys());
386 }
387 };
388
389 /// Reader for the indexed binary instrprof format.
390 class IndexedInstrProfReader : public InstrProfReader {
391 private:
392 /// The profile data file contents.
393 std::unique_ptr<MemoryBuffer> DataBuffer;
394 /// The index into the profile data.
395 std::unique_ptr<InstrProfReaderIndexBase> Index;
396 /// Profile summary data.
397 std::unique_ptr<ProfileSummary> Summary;
398 // Index to the current record in the record array.
399 unsigned RecordIndex;
400
401 // Read the profile summary. Return a pointer pointing to one byte past the
402 // end of the summary data if it exists or the input \c Cur.
403 const unsigned char *readSummary(IndexedInstrProf::ProfVersion Version,
404 const unsigned char *Cur);
405
406 public:
407 IndexedInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer)
408 : DataBuffer(std::move(DataBuffer)), RecordIndex(0) {}
409 IndexedInstrProfReader(const IndexedInstrProfReader &) = delete;
410 IndexedInstrProfReader &operator=(const IndexedInstrProfReader &) = delete;
411
412 /// Return the profile version.
413 uint64_t getVersion() const { return Index->getVersion(); }
414 bool isIRLevelProfile() const override { return Index->isIRLevelProfile(); }
415
416 /// Return true if the given buffer is in an indexed instrprof format.
417 static bool hasFormat(const MemoryBuffer &DataBuffer);
418
419 /// Read the file header.
420 Error readHeader() override;
421 /// Read a single record.
422 Error readNextRecord(NamedInstrProfRecord &Record) override;
423
424 /// Return the NamedInstrProfRecord associated with FuncName and FuncHash
425 Expected<InstrProfRecord> getInstrProfRecord(StringRef FuncName,
426 uint64_t FuncHash);
427
428 /// Fill Counts with the profile data for the given function name.
429 Error getFunctionCounts(StringRef FuncName, uint64_t FuncHash,
430 std::vector<uint64_t> &Counts);
431
432 /// Return the maximum of all known function counts.
433 uint64_t getMaximumFunctionCount() { return Summary->getMaxFunctionCount(); }
434
435 /// Factory method to create an indexed reader.
436 static Expected<std::unique_ptr<IndexedInstrProfReader>>
437 create(const Twine &Path);
438
439 static Expected<std::unique_ptr<IndexedInstrProfReader>>
EmitLinkerOptions(ArrayRef<std::string> Kind)440 create(std::unique_ptr<MemoryBuffer> Buffer);
441
442 // Used for testing purpose only.
EmitDataRegion(MCDataRegionType Kind)443 void setValueProfDataEndianness(support::endianness Endianness) {
444 Index->setValueProfDataEndianness(Endianness);
445 }
EmitVersionMin(MCVersionMinType Type,unsigned Major,unsigned Minor,unsigned Update)446
447 // See description in the base class. This interface is designed
448 // to be used by llvm-profdata (for dumping). Avoid using this when
449 // the client is the compiler.
450 InstrProfSymtab &getSymtab() override;
EmitBuildVersion(unsigned Platform,unsigned Major,unsigned Minor,unsigned Update)451 ProfileSummary &getSummary() { return *(Summary.get()); }
452 };
453
454 } // end namespace llvm
455
456 #endif // LLVM_PROFILEDATA_INSTRPROFREADER_H
457