1 //===- InstrProfReader.h - Instrumented profiling readers -------*- C++ -*-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file contains support for reading profiling data for instrumentation
10 // based PGO and coverage.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #ifndef LLVM_PROFILEDATA_INSTRPROFREADER_H
15 #define LLVM_PROFILEDATA_INSTRPROFREADER_H
16 
17 #include "llvm/ADT/ArrayRef.h"
18 #include "llvm/ADT/StringRef.h"
19 #include "llvm/IR/ProfileSummary.h"
20 #include "llvm/Object/BuildID.h"
21 #include "llvm/ProfileData/InstrProf.h"
22 #include "llvm/ProfileData/InstrProfCorrelator.h"
23 #include "llvm/ProfileData/MemProf.h"
24 #include "llvm/Support/Endian.h"
25 #include "llvm/Support/Error.h"
26 #include "llvm/Support/LineIterator.h"
27 #include "llvm/Support/MathExtras.h"
28 #include "llvm/Support/MemoryBuffer.h"
29 #include "llvm/Support/OnDiskHashTable.h"
30 #include "llvm/Support/SwapByteOrder.h"
31 #include <algorithm>
32 #include <cassert>
33 #include <cstddef>
34 #include <cstdint>
35 #include <iterator>
36 #include <memory>
37 #include <utility>
38 #include <vector>
39 
40 namespace llvm {
41 
42 class InstrProfReader;
43 
44 namespace vfs {
45 class FileSystem;
46 } // namespace vfs
47 
48 /// A file format agnostic iterator over profiling data.
49 template <class record_type = NamedInstrProfRecord,
50           class reader_type = InstrProfReader>
51 class InstrProfIterator {
52 public:
53   using iterator_category = std::input_iterator_tag;
54   using value_type = record_type;
55   using difference_type = std::ptrdiff_t;
56   using pointer = value_type *;
57   using reference = value_type &;
58 
59 private:
60   reader_type *Reader = nullptr;
61   value_type Record;
62 
increment()63   void increment() {
64     if (Error E = Reader->readNextRecord(Record)) {
65       // Handle errors in the reader.
66       InstrProfError::take(std::move(E));
67       *this = InstrProfIterator();
68     }
69   }
70 
71 public:
72   InstrProfIterator() = default;
InstrProfIterator(reader_type * Reader)73   InstrProfIterator(reader_type *Reader) : Reader(Reader) { increment(); }
74 
75   InstrProfIterator &operator++() {
76     increment();
77     return *this;
78   }
79   bool operator==(const InstrProfIterator &RHS) const {
80     return Reader == RHS.Reader;
81   }
82   bool operator!=(const InstrProfIterator &RHS) const {
83     return Reader != RHS.Reader;
84   }
85   value_type &operator*() { return Record; }
86   value_type *operator->() { return &Record; }
87 };
88 
89 /// Base class and interface for reading profiling data of any known instrprof
90 /// format. Provides an iterator over NamedInstrProfRecords.
91 class InstrProfReader {
92   instrprof_error LastError = instrprof_error::success;
93   std::string LastErrorMsg;
94 
95 public:
96   InstrProfReader() = default;
97   virtual ~InstrProfReader() = default;
98 
99   /// Read the header.  Required before reading first record.
100   virtual Error readHeader() = 0;
101 
102   /// Read a single record.
103   virtual Error readNextRecord(NamedInstrProfRecord &Record) = 0;
104 
105   /// Read a list of binary ids.
readBinaryIds(std::vector<llvm::object::BuildID> & BinaryIds)106   virtual Error readBinaryIds(std::vector<llvm::object::BuildID> &BinaryIds) {
107     return success();
108   }
109 
110   /// Print binary ids.
printBinaryIds(raw_ostream & OS)111   virtual Error printBinaryIds(raw_ostream &OS) { return success(); };
112 
113   /// Iterator over profile data.
begin()114   InstrProfIterator<> begin() { return InstrProfIterator<>(this); }
end()115   InstrProfIterator<> end() { return InstrProfIterator<>(); }
116 
117   /// Return the profile version.
118   virtual uint64_t getVersion() const = 0;
119 
120   virtual bool isIRLevelProfile() const = 0;
121 
122   virtual bool hasCSIRLevelProfile() const = 0;
123 
124   virtual bool instrEntryBBEnabled() const = 0;
125 
126   /// Return true if the profile has single byte counters representing coverage.
127   virtual bool hasSingleByteCoverage() const = 0;
128 
129   /// Return true if the profile only instruments function entries.
130   virtual bool functionEntryOnly() const = 0;
131 
132   /// Return true if profile includes a memory profile.
133   virtual bool hasMemoryProfile() const = 0;
134 
135   /// Return true if this has a temporal profile.
136   virtual bool hasTemporalProfile() const = 0;
137 
138   /// Returns a BitsetEnum describing the attributes of the profile. To check
139   /// individual attributes prefer using the helpers above.
140   virtual InstrProfKind getProfileKind() const = 0;
141 
142   /// Return the PGO symtab. There are three different readers:
143   /// Raw, Text, and Indexed profile readers. The first two types
144   /// of readers are used only by llvm-profdata tool, while the indexed
145   /// profile reader is also used by llvm-cov tool and the compiler (
146   /// backend or frontend). Since creating PGO symtab can create
147   /// significant runtime and memory overhead (as it touches data
148   /// for the whole program), InstrProfSymtab for the indexed profile
149   /// reader should be created on demand and it is recommended to be
150   /// only used for dumping purpose with llvm-proftool, not with the
151   /// compiler.
152   virtual InstrProfSymtab &getSymtab() = 0;
153 
154   /// Compute the sum of counts and return in Sum.
155   void accumulateCounts(CountSumOrPercent &Sum, bool IsCS);
156 
157 protected:
158   std::unique_ptr<InstrProfSymtab> Symtab;
159   /// A list of temporal profile traces.
160   SmallVector<TemporalProfTraceTy> TemporalProfTraces;
161   /// The total number of temporal profile traces seen.
162   uint64_t TemporalProfTraceStreamSize = 0;
163 
164   /// Set the current error and return same.
165   Error error(instrprof_error Err, const std::string &ErrMsg = "") {
166     LastError = Err;
167     LastErrorMsg = ErrMsg;
168     if (Err == instrprof_error::success)
169       return Error::success();
170     return make_error<InstrProfError>(Err, ErrMsg);
171   }
172 
error(Error && E)173   Error error(Error &&E) {
174     handleAllErrors(std::move(E), [&](const InstrProfError &IPE) {
175       LastError = IPE.get();
176       LastErrorMsg = IPE.getMessage();
177     });
178     return make_error<InstrProfError>(LastError, LastErrorMsg);
179   }
180 
181   /// Clear the current error and return a successful one.
success()182   Error success() { return error(instrprof_error::success); }
183 
184 public:
185   /// Return true if the reader has finished reading the profile data.
isEOF()186   bool isEOF() { return LastError == instrprof_error::eof; }
187 
188   /// Return true if the reader encountered an error reading profiling data.
hasError()189   bool hasError() { return LastError != instrprof_error::success && !isEOF(); }
190 
191   /// Get the current error.
getError()192   Error getError() {
193     if (hasError())
194       return make_error<InstrProfError>(LastError, LastErrorMsg);
195     return Error::success();
196   }
197 
198   /// Factory method to create an appropriately typed reader for the given
199   /// instrprof file.
200   static Expected<std::unique_ptr<InstrProfReader>>
201   create(const Twine &Path, vfs::FileSystem &FS,
202          const InstrProfCorrelator *Correlator = nullptr,
203          std::function<void(Error)> Warn = nullptr);
204 
205   static Expected<std::unique_ptr<InstrProfReader>>
206   create(std::unique_ptr<MemoryBuffer> Buffer,
207          const InstrProfCorrelator *Correlator = nullptr,
208          std::function<void(Error)> Warn = nullptr);
209 
210   /// \param Weight for raw profiles use this as the temporal profile trace
211   ///               weight
212   /// \returns a list of temporal profile traces.
213   virtual SmallVector<TemporalProfTraceTy> &
214   getTemporalProfTraces(std::optional<uint64_t> Weight = {}) {
215     // For non-raw profiles we ignore the input weight and instead use the
216     // weights already in the traces.
217     return TemporalProfTraces;
218   }
219   /// \returns the total number of temporal profile traces seen.
getTemporalProfTraceStreamSize()220   uint64_t getTemporalProfTraceStreamSize() {
221     return TemporalProfTraceStreamSize;
222   }
223 };
224 
225 /// Reader for the simple text based instrprof format.
226 ///
227 /// This format is a simple text format that's suitable for test data. Records
228 /// are separated by one or more blank lines, and record fields are separated by
229 /// new lines.
230 ///
231 /// Each record consists of a function name, a function hash, a number of
232 /// counters, and then each counter value, in that order.
233 class TextInstrProfReader : public InstrProfReader {
234 private:
235   /// The profile data file contents.
236   std::unique_ptr<MemoryBuffer> DataBuffer;
237   /// Iterator over the profile data.
238   line_iterator Line;
239   /// The attributes of the current profile.
240   InstrProfKind ProfileKind = InstrProfKind::Unknown;
241 
242   Error readValueProfileData(InstrProfRecord &Record);
243 
244   Error readTemporalProfTraceData();
245 
246 public:
TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_)247   TextInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer_)
248       : DataBuffer(std::move(DataBuffer_)), Line(*DataBuffer, true, '#') {}
249   TextInstrProfReader(const TextInstrProfReader &) = delete;
250   TextInstrProfReader &operator=(const TextInstrProfReader &) = delete;
251 
252   /// Return true if the given buffer is in text instrprof format.
253   static bool hasFormat(const MemoryBuffer &Buffer);
254 
255   // Text format does not have version, so return 0.
getVersion()256   uint64_t getVersion() const override { return 0; }
257 
isIRLevelProfile()258   bool isIRLevelProfile() const override {
259     return static_cast<bool>(ProfileKind & InstrProfKind::IRInstrumentation);
260   }
261 
hasCSIRLevelProfile()262   bool hasCSIRLevelProfile() const override {
263     return static_cast<bool>(ProfileKind & InstrProfKind::ContextSensitive);
264   }
265 
instrEntryBBEnabled()266   bool instrEntryBBEnabled() const override {
267     return static_cast<bool>(ProfileKind &
268                              InstrProfKind::FunctionEntryInstrumentation);
269   }
270 
hasSingleByteCoverage()271   bool hasSingleByteCoverage() const override {
272     return static_cast<bool>(ProfileKind & InstrProfKind::SingleByteCoverage);
273   }
274 
functionEntryOnly()275   bool functionEntryOnly() const override {
276     return static_cast<bool>(ProfileKind & InstrProfKind::FunctionEntryOnly);
277   }
278 
hasMemoryProfile()279   bool hasMemoryProfile() const override {
280     // TODO: Add support for text format memory profiles.
281     return false;
282   }
283 
hasTemporalProfile()284   bool hasTemporalProfile() const override {
285     return static_cast<bool>(ProfileKind & InstrProfKind::TemporalProfile);
286   }
287 
getProfileKind()288   InstrProfKind getProfileKind() const override { return ProfileKind; }
289 
290   /// Read the header.
291   Error readHeader() override;
292 
293   /// Read a single record.
294   Error readNextRecord(NamedInstrProfRecord &Record) override;
295 
getSymtab()296   InstrProfSymtab &getSymtab() override {
297     assert(Symtab);
298     return *Symtab;
299   }
300 };
301 
302 /// Reader for the raw instrprof binary format from runtime.
303 ///
304 /// This format is a raw memory dump of the instrumentation-based profiling data
305 /// from the runtime.  It has no index.
306 ///
307 /// Templated on the unsigned type whose size matches pointers on the platform
308 /// that wrote the profile.
309 template <class IntPtrT>
310 class RawInstrProfReader : public InstrProfReader {
311 private:
312   /// The profile data file contents.
313   std::unique_ptr<MemoryBuffer> DataBuffer;
314   /// If available, this hold the ProfileData array used to correlate raw
315   /// instrumentation data to their functions.
316   const InstrProfCorrelatorImpl<IntPtrT> *Correlator;
317   /// A list of timestamps paired with a function name reference.
318   std::vector<std::pair<uint64_t, uint64_t>> TemporalProfTimestamps;
319   bool ShouldSwapBytes;
320   // The value of the version field of the raw profile data header. The lower 32
321   // bits specifies the format version and the most significant 32 bits specify
322   // the variant types of the profile.
323   uint64_t Version;
324   uint64_t CountersDelta;
325   uint64_t BitmapDelta;
326   uint64_t NamesDelta;
327   const RawInstrProf::ProfileData<IntPtrT> *Data;
328   const RawInstrProf::ProfileData<IntPtrT> *DataEnd;
329   const RawInstrProf::VTableProfileData<IntPtrT> *VTableBegin = nullptr;
330   const RawInstrProf::VTableProfileData<IntPtrT> *VTableEnd = nullptr;
331   const char *CountersStart;
332   const char *CountersEnd;
333   const char *BitmapStart;
334   const char *BitmapEnd;
335   const char *NamesStart;
336   const char *NamesEnd;
337   const char *VNamesStart = nullptr;
338   const char *VNamesEnd = nullptr;
339   // After value profile is all read, this pointer points to
340   // the header of next profile data (if exists)
341   const uint8_t *ValueDataStart;
342   uint32_t ValueKindLast;
343   uint32_t CurValueDataSize;
344   std::vector<llvm::object::BuildID> BinaryIds;
345 
346   std::function<void(Error)> Warn;
347 
348   /// Maxium counter value 2^56.
349   static const uint64_t MaxCounterValue = (1ULL << 56);
350 
351 public:
RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer,const InstrProfCorrelator * Correlator,std::function<void (Error)> Warn)352   RawInstrProfReader(std::unique_ptr<MemoryBuffer> DataBuffer,
353                      const InstrProfCorrelator *Correlator,
354                      std::function<void(Error)> Warn)
355       : DataBuffer(std::move(DataBuffer)),
356         Correlator(dyn_cast_or_null<const InstrProfCorrelatorImpl<IntPtrT>>(
357             Correlator)),
358         Warn(Warn) {}
359   RawInstrProfReader(const RawInstrProfReader &) = delete;
360   RawInstrProfReader &operator=(const RawInstrProfReader &) = delete;
361 
362   static bool hasFormat(const MemoryBuffer &DataBuffer);
363   Error readHeader() override;
364   Error readNextRecord(NamedInstrProfRecord &Record) override;
365   Error readBinaryIds(std::vector<llvm::object::BuildID> &BinaryIds) override;
366   Error printBinaryIds(raw_ostream &OS) override;
367 
getVersion()368   uint64_t getVersion() const override { return Version; }
369 
isIRLevelProfile()370   bool isIRLevelProfile() const override {
371     return (Version & VARIANT_MASK_IR_PROF) != 0;
372   }
373 
hasCSIRLevelProfile()374   bool hasCSIRLevelProfile() const override {
375     return (Version & VARIANT_MASK_CSIR_PROF) != 0;
376   }
377 
instrEntryBBEnabled()378   bool instrEntryBBEnabled() const override {
379     return (Version & VARIANT_MASK_INSTR_ENTRY) != 0;
380   }
381 
hasSingleByteCoverage()382   bool hasSingleByteCoverage() const override {
383     return (Version & VARIANT_MASK_BYTE_COVERAGE) != 0;
384   }
385 
functionEntryOnly()386   bool functionEntryOnly() const override {
387     return (Version & VARIANT_MASK_FUNCTION_ENTRY_ONLY) != 0;
388   }
389 
hasMemoryProfile()390   bool hasMemoryProfile() const override {
391     // Memory profiles have a separate raw format, so this should never be set.
392     assert(!(Version & VARIANT_MASK_MEMPROF));
393     return false;
394   }
395 
hasTemporalProfile()396   bool hasTemporalProfile() const override {
397     return (Version & VARIANT_MASK_TEMPORAL_PROF) != 0;
398   }
399 
400   /// Returns a BitsetEnum describing the attributes of the raw instr profile.
401   InstrProfKind getProfileKind() const override;
402 
getSymtab()403   InstrProfSymtab &getSymtab() override {
404     assert(Symtab.get());
405     return *Symtab.get();
406   }
407 
408   SmallVector<TemporalProfTraceTy> &
409   getTemporalProfTraces(std::optional<uint64_t> Weight = {}) override;
410 
411 private:
412   Error createSymtab(InstrProfSymtab &Symtab);
413   Error readNextHeader(const char *CurrentPos);
414   Error readHeader(const RawInstrProf::Header &Header);
415 
swap(IntT Int)416   template <class IntT> IntT swap(IntT Int) const {
417     return ShouldSwapBytes ? llvm::byteswap(Int) : Int;
418   }
419 
getDataEndianness()420   llvm::endianness getDataEndianness() const {
421     if (!ShouldSwapBytes)
422       return llvm::endianness::native;
423     if (llvm::endianness::native == llvm::endianness::little)
424       return llvm::endianness::big;
425     else
426       return llvm::endianness::little;
427   }
428 
getNumPaddingBytes(uint64_t SizeInBytes)429   inline uint8_t getNumPaddingBytes(uint64_t SizeInBytes) {
430     return 7 & (sizeof(uint64_t) - SizeInBytes % sizeof(uint64_t));
431   }
432 
433   Error readName(NamedInstrProfRecord &Record);
434   Error readFuncHash(NamedInstrProfRecord &Record);
435   Error readRawCounts(InstrProfRecord &Record);
436   Error readRawBitmapBytes(InstrProfRecord &Record);
437   Error readValueProfilingData(InstrProfRecord &Record);
atEnd()438   bool atEnd() const { return Data == DataEnd; }
439 
advanceData()440   void advanceData() {
441     // `CountersDelta` is a constant zero when using debug info correlation.
442     if (!Correlator) {
443       // The initial CountersDelta is the in-memory address difference between
444       // the data and counts sections:
445       // start(__llvm_prf_cnts) - start(__llvm_prf_data)
446       // As we advance to the next record, we maintain the correct CountersDelta
447       // with respect to the next record.
448       CountersDelta -= sizeof(*Data);
449       BitmapDelta -= sizeof(*Data);
450     }
451     Data++;
452     ValueDataStart += CurValueDataSize;
453   }
454 
getNextHeaderPos()455   const char *getNextHeaderPos() const {
456       assert(atEnd());
457       return (const char *)ValueDataStart;
458   }
459 
getName(uint64_t NameRef)460   StringRef getName(uint64_t NameRef) const {
461     return Symtab->getFuncOrVarName(swap(NameRef));
462   }
463 
getCounterTypeSize()464   int getCounterTypeSize() const {
465     return hasSingleByteCoverage() ? sizeof(uint8_t) : sizeof(uint64_t);
466   }
467 };
468 
469 using RawInstrProfReader32 = RawInstrProfReader<uint32_t>;
470 using RawInstrProfReader64 = RawInstrProfReader<uint64_t>;
471 
472 namespace IndexedInstrProf {
473 
474 enum class HashT : uint32_t;
475 
476 } // end namespace IndexedInstrProf
477 
478 /// Trait for lookups into the on-disk hash table for the binary instrprof
479 /// format.
480 class InstrProfLookupTrait {
481   std::vector<NamedInstrProfRecord> DataBuffer;
482   IndexedInstrProf::HashT HashType;
483   unsigned FormatVersion;
484   // Endianness of the input value profile data.
485   // It should be LE by default, but can be changed
486   // for testing purpose.
487   llvm::endianness ValueProfDataEndianness = llvm::endianness::little;
488 
489 public:
InstrProfLookupTrait(IndexedInstrProf::HashT HashType,unsigned FormatVersion)490   InstrProfLookupTrait(IndexedInstrProf::HashT HashType, unsigned FormatVersion)
491       : HashType(HashType), FormatVersion(FormatVersion) {}
492 
493   using data_type = ArrayRef<NamedInstrProfRecord>;
494 
495   using internal_key_type = StringRef;
496   using external_key_type = StringRef;
497   using hash_value_type = uint64_t;
498   using offset_type = uint64_t;
499 
EqualKey(StringRef A,StringRef B)500   static bool EqualKey(StringRef A, StringRef B) { return A == B; }
GetInternalKey(StringRef K)501   static StringRef GetInternalKey(StringRef K) { return K; }
GetExternalKey(StringRef K)502   static StringRef GetExternalKey(StringRef K) { return K; }
503 
504   hash_value_type ComputeHash(StringRef K);
505 
506   static std::pair<offset_type, offset_type>
ReadKeyDataLength(const unsigned char * & D)507   ReadKeyDataLength(const unsigned char *&D) {
508     using namespace support;
509 
510     offset_type KeyLen =
511         endian::readNext<offset_type, llvm::endianness::little, unaligned>(D);
512     offset_type DataLen =
513         endian::readNext<offset_type, llvm::endianness::little, unaligned>(D);
514     return std::make_pair(KeyLen, DataLen);
515   }
516 
ReadKey(const unsigned char * D,offset_type N)517   StringRef ReadKey(const unsigned char *D, offset_type N) {
518     return StringRef((const char *)D, N);
519   }
520 
521   bool readValueProfilingData(const unsigned char *&D,
522                               const unsigned char *const End);
523   data_type ReadData(StringRef K, const unsigned char *D, offset_type N);
524 
525   // Used for testing purpose only.
setValueProfDataEndianness(llvm::endianness Endianness)526   void setValueProfDataEndianness(llvm::endianness Endianness) {
527     ValueProfDataEndianness = Endianness;
528   }
529 };
530 
531 struct InstrProfReaderIndexBase {
532   virtual ~InstrProfReaderIndexBase() = default;
533 
534   // Read all the profile records with the same key pointed to the current
535   // iterator.
536   virtual Error getRecords(ArrayRef<NamedInstrProfRecord> &Data) = 0;
537 
538   // Read all the profile records with the key equal to FuncName
539   virtual Error getRecords(StringRef FuncName,
540                                      ArrayRef<NamedInstrProfRecord> &Data) = 0;
541   virtual void advanceToNextKey() = 0;
542   virtual bool atEnd() const = 0;
543   virtual void setValueProfDataEndianness(llvm::endianness Endianness) = 0;
544   virtual uint64_t getVersion() const = 0;
545   virtual bool isIRLevelProfile() const = 0;
546   virtual bool hasCSIRLevelProfile() const = 0;
547   virtual bool instrEntryBBEnabled() const = 0;
548   virtual bool hasSingleByteCoverage() const = 0;
549   virtual bool functionEntryOnly() const = 0;
550   virtual bool hasMemoryProfile() const = 0;
551   virtual bool hasTemporalProfile() const = 0;
552   virtual InstrProfKind getProfileKind() const = 0;
553   virtual Error populateSymtab(InstrProfSymtab &) = 0;
554 };
555 
556 using OnDiskHashTableImplV3 =
557     OnDiskIterableChainedHashTable<InstrProfLookupTrait>;
558 
559 using MemProfRecordHashTable =
560     OnDiskIterableChainedHashTable<memprof::RecordLookupTrait>;
561 using MemProfFrameHashTable =
562     OnDiskIterableChainedHashTable<memprof::FrameLookupTrait>;
563 
564 template <typename HashTableImpl>
565 class InstrProfReaderItaniumRemapper;
566 
567 template <typename HashTableImpl>
568 class InstrProfReaderIndex : public InstrProfReaderIndexBase {
569 private:
570   std::unique_ptr<HashTableImpl> HashTable;
571   typename HashTableImpl::data_iterator RecordIterator;
572   uint64_t FormatVersion;
573 
574   friend class InstrProfReaderItaniumRemapper<HashTableImpl>;
575 
576 public:
577   InstrProfReaderIndex(const unsigned char *Buckets,
578                        const unsigned char *const Payload,
579                        const unsigned char *const Base,
580                        IndexedInstrProf::HashT HashType, uint64_t Version);
581   ~InstrProfReaderIndex() override = default;
582 
583   Error getRecords(ArrayRef<NamedInstrProfRecord> &Data) override;
584   Error getRecords(StringRef FuncName,
585                    ArrayRef<NamedInstrProfRecord> &Data) override;
advanceToNextKey()586   void advanceToNextKey() override { RecordIterator++; }
587 
atEnd()588   bool atEnd() const override {
589     return RecordIterator == HashTable->data_end();
590   }
591 
setValueProfDataEndianness(llvm::endianness Endianness)592   void setValueProfDataEndianness(llvm::endianness Endianness) override {
593     HashTable->getInfoObj().setValueProfDataEndianness(Endianness);
594   }
595 
getVersion()596   uint64_t getVersion() const override { return GET_VERSION(FormatVersion); }
597 
isIRLevelProfile()598   bool isIRLevelProfile() const override {
599     return (FormatVersion & VARIANT_MASK_IR_PROF) != 0;
600   }
601 
hasCSIRLevelProfile()602   bool hasCSIRLevelProfile() const override {
603     return (FormatVersion & VARIANT_MASK_CSIR_PROF) != 0;
604   }
605 
instrEntryBBEnabled()606   bool instrEntryBBEnabled() const override {
607     return (FormatVersion & VARIANT_MASK_INSTR_ENTRY) != 0;
608   }
609 
hasSingleByteCoverage()610   bool hasSingleByteCoverage() const override {
611     return (FormatVersion & VARIANT_MASK_BYTE_COVERAGE) != 0;
612   }
613 
functionEntryOnly()614   bool functionEntryOnly() const override {
615     return (FormatVersion & VARIANT_MASK_FUNCTION_ENTRY_ONLY) != 0;
616   }
617 
hasMemoryProfile()618   bool hasMemoryProfile() const override {
619     return (FormatVersion & VARIANT_MASK_MEMPROF) != 0;
620   }
621 
hasTemporalProfile()622   bool hasTemporalProfile() const override {
623     return (FormatVersion & VARIANT_MASK_TEMPORAL_PROF) != 0;
624   }
625 
626   InstrProfKind getProfileKind() const override;
627 
populateSymtab(InstrProfSymtab & Symtab)628   Error populateSymtab(InstrProfSymtab &Symtab) override {
629     return Symtab.create(HashTable->keys());
630   }
631 };
632 
633 /// Name matcher supporting fuzzy matching of symbol names to names in profiles.
634 class InstrProfReaderRemapper {
635 public:
636   virtual ~InstrProfReaderRemapper() = default;
populateRemappings()637   virtual Error populateRemappings() { return Error::success(); }
638   virtual Error getRecords(StringRef FuncName,
639                            ArrayRef<NamedInstrProfRecord> &Data) = 0;
640 };
641 
642 /// Reader for the indexed binary instrprof format.
643 class IndexedInstrProfReader : public InstrProfReader {
644 private:
645   /// The profile data file contents.
646   std::unique_ptr<MemoryBuffer> DataBuffer;
647   /// The profile remapping file contents.
648   std::unique_ptr<MemoryBuffer> RemappingBuffer;
649   /// The index into the profile data.
650   std::unique_ptr<InstrProfReaderIndexBase> Index;
651   /// The profile remapping file contents.
652   std::unique_ptr<InstrProfReaderRemapper> Remapper;
653   /// Profile summary data.
654   std::unique_ptr<ProfileSummary> Summary;
655   /// Context sensitive profile summary data.
656   std::unique_ptr<ProfileSummary> CS_Summary;
657   /// MemProf profile schema (if available).
658   memprof::MemProfSchema Schema;
659   /// MemProf record profile data on-disk indexed via llvm::md5(FunctionName).
660   std::unique_ptr<MemProfRecordHashTable> MemProfRecordTable;
661   /// MemProf frame profile data on-disk indexed via frame id.
662   std::unique_ptr<MemProfFrameHashTable> MemProfFrameTable;
663   /// VTableNamePtr points to the beginning of compressed vtable names.
664   /// When a symtab is constructed from profiles by llvm-profdata, the list of
665   /// names could be decompressed based on `VTableNamePtr` and
666   /// `CompressedVTableNamesLen`.
667   /// A compiler that reads indexed profiles could construct symtab from module
668   /// IR so it doesn't need the decompressed names.
669   const char *VTableNamePtr = nullptr;
670   /// The length of compressed vtable names.
671   uint64_t CompressedVTableNamesLen = 0;
672   /// Total size of binary ids.
673   uint64_t BinaryIdsSize{0};
674   /// Start address of binary id length and data pairs.
675   const uint8_t *BinaryIdsStart = nullptr;
676 
677   // Index to the current record in the record array.
678   unsigned RecordIndex;
679 
680   // Read the profile summary. Return a pointer pointing to one byte past the
681   // end of the summary data if it exists or the input \c Cur.
682   // \c UseCS indicates whether to use the context-sensitive profile summary.
683   const unsigned char *readSummary(IndexedInstrProf::ProfVersion Version,
684                                    const unsigned char *Cur, bool UseCS);
685 
686 public:
687   IndexedInstrProfReader(
688       std::unique_ptr<MemoryBuffer> DataBuffer,
689       std::unique_ptr<MemoryBuffer> RemappingBuffer = nullptr)
DataBuffer(std::move (DataBuffer))690       : DataBuffer(std::move(DataBuffer)),
691         RemappingBuffer(std::move(RemappingBuffer)), RecordIndex(0) {}
692   IndexedInstrProfReader(const IndexedInstrProfReader &) = delete;
693   IndexedInstrProfReader &operator=(const IndexedInstrProfReader &) = delete;
694 
695   /// Return the profile version.
getVersion()696   uint64_t getVersion() const override { return Index->getVersion(); }
isIRLevelProfile()697   bool isIRLevelProfile() const override { return Index->isIRLevelProfile(); }
hasCSIRLevelProfile()698   bool hasCSIRLevelProfile() const override {
699     return Index->hasCSIRLevelProfile();
700   }
701 
instrEntryBBEnabled()702   bool instrEntryBBEnabled() const override {
703     return Index->instrEntryBBEnabled();
704   }
705 
hasSingleByteCoverage()706   bool hasSingleByteCoverage() const override {
707     return Index->hasSingleByteCoverage();
708   }
709 
functionEntryOnly()710   bool functionEntryOnly() const override { return Index->functionEntryOnly(); }
711 
hasMemoryProfile()712   bool hasMemoryProfile() const override { return Index->hasMemoryProfile(); }
713 
hasTemporalProfile()714   bool hasTemporalProfile() const override {
715     return Index->hasTemporalProfile();
716   }
717 
718   /// Returns a BitsetEnum describing the attributes of the indexed instr
719   /// profile.
getProfileKind()720   InstrProfKind getProfileKind() const override {
721     return Index->getProfileKind();
722   }
723 
724   /// Return true if the given buffer is in an indexed instrprof format.
725   static bool hasFormat(const MemoryBuffer &DataBuffer);
726 
727   /// Read the file header.
728   Error readHeader() override;
729   /// Read a single record.
730   Error readNextRecord(NamedInstrProfRecord &Record) override;
731 
732   /// Return the NamedInstrProfRecord associated with FuncName and FuncHash.
733   /// When return a hash_mismatch error and MismatchedFuncSum is not nullptr,
734   /// the sum of all counters in the mismatched function will be set to
735   /// MismatchedFuncSum. If there are multiple instances of mismatched
736   /// functions, MismatchedFuncSum returns the maximum. If \c FuncName is not
737   /// found, try to lookup \c DeprecatedFuncName to handle profiles built by
738   /// older compilers.
739   Expected<InstrProfRecord>
740   getInstrProfRecord(StringRef FuncName, uint64_t FuncHash,
741                      StringRef DeprecatedFuncName = "",
742                      uint64_t *MismatchedFuncSum = nullptr);
743 
744   /// Return the memprof record for the function identified by
745   /// llvm::md5(Name).
746   Expected<memprof::MemProfRecord> getMemProfRecord(uint64_t FuncNameHash);
747 
748   /// Fill Counts with the profile data for the given function name.
749   Error getFunctionCounts(StringRef FuncName, uint64_t FuncHash,
750                           std::vector<uint64_t> &Counts);
751 
752   /// Fill Bitmap with the profile data for the given function name.
753   Error getFunctionBitmap(StringRef FuncName, uint64_t FuncHash,
754                           BitVector &Bitmap);
755 
756   /// Return the maximum of all known function counts.
757   /// \c UseCS indicates whether to use the context-sensitive count.
getMaximumFunctionCount(bool UseCS)758   uint64_t getMaximumFunctionCount(bool UseCS) {
759     if (UseCS) {
760       assert(CS_Summary && "No context sensitive profile summary");
761       return CS_Summary->getMaxFunctionCount();
762     } else {
763       assert(Summary && "No profile summary");
764       return Summary->getMaxFunctionCount();
765     }
766   }
767 
768   /// Factory method to create an indexed reader.
769   static Expected<std::unique_ptr<IndexedInstrProfReader>>
770   create(const Twine &Path, vfs::FileSystem &FS,
771          const Twine &RemappingPath = "");
772 
773   static Expected<std::unique_ptr<IndexedInstrProfReader>>
774   create(std::unique_ptr<MemoryBuffer> Buffer,
775          std::unique_ptr<MemoryBuffer> RemappingBuffer = nullptr);
776 
777   // Used for testing purpose only.
setValueProfDataEndianness(llvm::endianness Endianness)778   void setValueProfDataEndianness(llvm::endianness Endianness) {
779     Index->setValueProfDataEndianness(Endianness);
780   }
781 
782   // See description in the base class. This interface is designed
783   // to be used by llvm-profdata (for dumping). Avoid using this when
784   // the client is the compiler.
785   InstrProfSymtab &getSymtab() override;
786 
787   /// Return the profile summary.
788   /// \c UseCS indicates whether to use the context-sensitive summary.
getSummary(bool UseCS)789   ProfileSummary &getSummary(bool UseCS) {
790     if (UseCS) {
791       assert(CS_Summary && "No context sensitive summary");
792       return *CS_Summary;
793     } else {
794       assert(Summary && "No profile summary");
795       return *Summary;
796     }
797   }
798 
799   Error readBinaryIds(std::vector<llvm::object::BuildID> &BinaryIds) override;
800   Error printBinaryIds(raw_ostream &OS) override;
801 };
802 
803 } // end namespace llvm
804 
805 #endif // LLVM_PROFILEDATA_INSTRPROFREADER_H
806