1 //===- CoverageMapping.h - Code coverage mapping support --------*- 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 // Code coverage mapping data is generated by clang and read by
10 // llvm-cov to show code coverage statistics for a file.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #ifndef LLVM_PROFILEDATA_COVERAGE_COVERAGEMAPPING_H
15 #define LLVM_PROFILEDATA_COVERAGE_COVERAGEMAPPING_H
16
17 #include "llvm/ADT/ArrayRef.h"
18 #include "llvm/ADT/DenseMap.h"
19 #include "llvm/ADT/DenseSet.h"
20 #include "llvm/ADT/Hashing.h"
21 #include "llvm/ADT/StringRef.h"
22 #include "llvm/ADT/iterator.h"
23 #include "llvm/ADT/iterator_range.h"
24 #include "llvm/Object/BuildID.h"
25 #include "llvm/ProfileData/InstrProf.h"
26 #include "llvm/Support/Alignment.h"
27 #include "llvm/Support/Compiler.h"
28 #include "llvm/Support/Debug.h"
29 #include "llvm/Support/Endian.h"
30 #include "llvm/Support/Error.h"
31 #include "llvm/Support/raw_ostream.h"
32 #include <cassert>
33 #include <cstdint>
34 #include <iterator>
35 #include <memory>
36 #include <string>
37 #include <system_error>
38 #include <tuple>
39 #include <utility>
40 #include <vector>
41
42 namespace llvm {
43
44 class IndexedInstrProfReader;
45
46 namespace object {
47 class BuildIDFetcher;
48 } // namespace object
49
50 namespace coverage {
51
52 class CoverageMappingReader;
53 struct CoverageMappingRecord;
54
55 enum class coveragemap_error {
56 success = 0,
57 eof,
58 no_data_found,
59 unsupported_version,
60 truncated,
61 malformed,
62 decompression_failed,
63 invalid_or_missing_arch_specifier
64 };
65
66 const std::error_category &coveragemap_category();
67
make_error_code(coveragemap_error E)68 inline std::error_code make_error_code(coveragemap_error E) {
69 return std::error_code(static_cast<int>(E), coveragemap_category());
70 }
71
72 class CoverageMapError : public ErrorInfo<CoverageMapError> {
73 public:
CoverageMapError(coveragemap_error Err)74 CoverageMapError(coveragemap_error Err) : Err(Err) {
75 assert(Err != coveragemap_error::success && "Not an error");
76 }
77
78 std::string message() const override;
79
log(raw_ostream & OS)80 void log(raw_ostream &OS) const override { OS << message(); }
81
convertToErrorCode()82 std::error_code convertToErrorCode() const override {
83 return make_error_code(Err);
84 }
85
get()86 coveragemap_error get() const { return Err; }
87
88 static char ID;
89
90 private:
91 coveragemap_error Err;
92 };
93
94 /// A Counter is an abstract value that describes how to compute the
95 /// execution count for a region of code using the collected profile count data.
96 struct Counter {
97 /// The CounterExpression kind (Add or Subtract) is encoded in bit 0 next to
98 /// the CounterKind. This means CounterKind has to leave bit 0 free.
99 enum CounterKind { Zero, CounterValueReference, Expression };
100 static const unsigned EncodingTagBits = 2;
101 static const unsigned EncodingTagMask = 0x3;
102 static const unsigned EncodingCounterTagAndExpansionRegionTagBits =
103 EncodingTagBits + 1;
104
105 private:
106 CounterKind Kind = Zero;
107 unsigned ID = 0;
108
CounterCounter109 Counter(CounterKind Kind, unsigned ID) : Kind(Kind), ID(ID) {}
110
111 public:
112 Counter() = default;
113
getKindCounter114 CounterKind getKind() const { return Kind; }
115
isZeroCounter116 bool isZero() const { return Kind == Zero; }
117
isExpressionCounter118 bool isExpression() const { return Kind == Expression; }
119
getCounterIDCounter120 unsigned getCounterID() const { return ID; }
121
getExpressionIDCounter122 unsigned getExpressionID() const { return ID; }
123
124 friend bool operator==(const Counter &LHS, const Counter &RHS) {
125 return LHS.Kind == RHS.Kind && LHS.ID == RHS.ID;
126 }
127
128 friend bool operator!=(const Counter &LHS, const Counter &RHS) {
129 return !(LHS == RHS);
130 }
131
132 friend bool operator<(const Counter &LHS, const Counter &RHS) {
133 return std::tie(LHS.Kind, LHS.ID) < std::tie(RHS.Kind, RHS.ID);
134 }
135
136 /// Return the counter that represents the number zero.
getZeroCounter137 static Counter getZero() { return Counter(); }
138
139 /// Return the counter that corresponds to a specific profile counter.
getCounterCounter140 static Counter getCounter(unsigned CounterId) {
141 return Counter(CounterValueReference, CounterId);
142 }
143
144 /// Return the counter that corresponds to a specific addition counter
145 /// expression.
getExpressionCounter146 static Counter getExpression(unsigned ExpressionId) {
147 return Counter(Expression, ExpressionId);
148 }
149 };
150
151 /// A Counter expression is a value that represents an arithmetic operation
152 /// with two counters.
153 struct CounterExpression {
154 enum ExprKind { Subtract, Add };
155 ExprKind Kind;
156 Counter LHS, RHS;
157
CounterExpressionCounterExpression158 CounterExpression(ExprKind Kind, Counter LHS, Counter RHS)
159 : Kind(Kind), LHS(LHS), RHS(RHS) {}
160 };
161
162 /// A Counter expression builder is used to construct the counter expressions.
163 /// It avoids unnecessary duplication and simplifies algebraic expressions.
164 class CounterExpressionBuilder {
165 /// A list of all the counter expressions
166 std::vector<CounterExpression> Expressions;
167
168 /// A lookup table for the index of a given expression.
169 DenseMap<CounterExpression, unsigned> ExpressionIndices;
170
171 /// Return the counter which corresponds to the given expression.
172 ///
173 /// If the given expression is already stored in the builder, a counter
174 /// that references that expression is returned. Otherwise, the given
175 /// expression is added to the builder's collection of expressions.
176 Counter get(const CounterExpression &E);
177
178 /// Represents a term in a counter expression tree.
179 struct Term {
180 unsigned CounterID;
181 int Factor;
182
TermTerm183 Term(unsigned CounterID, int Factor)
184 : CounterID(CounterID), Factor(Factor) {}
185 };
186
187 /// Gather the terms of the expression tree for processing.
188 ///
189 /// This collects each addition and subtraction referenced by the counter into
190 /// a sequence that can be sorted and combined to build a simplified counter
191 /// expression.
192 void extractTerms(Counter C, int Sign, SmallVectorImpl<Term> &Terms);
193
194 /// Simplifies the given expression tree
195 /// by getting rid of algebraically redundant operations.
196 Counter simplify(Counter ExpressionTree);
197
198 public:
getExpressions()199 ArrayRef<CounterExpression> getExpressions() const { return Expressions; }
200
201 /// Return a counter that represents the expression that adds LHS and RHS.
202 Counter add(Counter LHS, Counter RHS, bool Simplify = true);
203
204 /// Return a counter that represents the expression that subtracts RHS from
205 /// LHS.
206 Counter subtract(Counter LHS, Counter RHS, bool Simplify = true);
207 };
208
209 using LineColPair = std::pair<unsigned, unsigned>;
210
211 /// A Counter mapping region associates a source range with a specific counter.
212 struct CounterMappingRegion {
213 enum RegionKind {
214 /// A CodeRegion associates some code with a counter
215 CodeRegion,
216
217 /// An ExpansionRegion represents a file expansion region that associates
218 /// a source range with the expansion of a virtual source file, such as
219 /// for a macro instantiation or #include file.
220 ExpansionRegion,
221
222 /// A SkippedRegion represents a source range with code that was skipped
223 /// by a preprocessor or similar means.
224 SkippedRegion,
225
226 /// A GapRegion is like a CodeRegion, but its count is only set as the
227 /// line execution count when its the only region in the line.
228 GapRegion,
229
230 /// A BranchRegion represents leaf-level boolean expressions and is
231 /// associated with two counters, each representing the number of times the
232 /// expression evaluates to true or false.
233 BranchRegion
234 };
235
236 /// Primary Counter that is also used for Branch Regions (TrueCount).
237 Counter Count;
238
239 /// Secondary Counter used for Branch Regions (FalseCount).
240 Counter FalseCount;
241
242 unsigned FileID, ExpandedFileID;
243 unsigned LineStart, ColumnStart, LineEnd, ColumnEnd;
244 RegionKind Kind;
245
CounterMappingRegionCounterMappingRegion246 CounterMappingRegion(Counter Count, unsigned FileID, unsigned ExpandedFileID,
247 unsigned LineStart, unsigned ColumnStart,
248 unsigned LineEnd, unsigned ColumnEnd, RegionKind Kind)
249 : Count(Count), FileID(FileID), ExpandedFileID(ExpandedFileID),
250 LineStart(LineStart), ColumnStart(ColumnStart), LineEnd(LineEnd),
251 ColumnEnd(ColumnEnd), Kind(Kind) {}
252
CounterMappingRegionCounterMappingRegion253 CounterMappingRegion(Counter Count, Counter FalseCount, unsigned FileID,
254 unsigned ExpandedFileID, unsigned LineStart,
255 unsigned ColumnStart, unsigned LineEnd,
256 unsigned ColumnEnd, RegionKind Kind)
257 : Count(Count), FalseCount(FalseCount), FileID(FileID),
258 ExpandedFileID(ExpandedFileID), LineStart(LineStart),
259 ColumnStart(ColumnStart), LineEnd(LineEnd), ColumnEnd(ColumnEnd),
260 Kind(Kind) {}
261
262 static CounterMappingRegion
makeRegionCounterMappingRegion263 makeRegion(Counter Count, unsigned FileID, unsigned LineStart,
264 unsigned ColumnStart, unsigned LineEnd, unsigned ColumnEnd) {
265 return CounterMappingRegion(Count, FileID, 0, LineStart, ColumnStart,
266 LineEnd, ColumnEnd, CodeRegion);
267 }
268
269 static CounterMappingRegion
makeExpansionCounterMappingRegion270 makeExpansion(unsigned FileID, unsigned ExpandedFileID, unsigned LineStart,
271 unsigned ColumnStart, unsigned LineEnd, unsigned ColumnEnd) {
272 return CounterMappingRegion(Counter(), FileID, ExpandedFileID, LineStart,
273 ColumnStart, LineEnd, ColumnEnd,
274 ExpansionRegion);
275 }
276
277 static CounterMappingRegion
makeSkippedCounterMappingRegion278 makeSkipped(unsigned FileID, unsigned LineStart, unsigned ColumnStart,
279 unsigned LineEnd, unsigned ColumnEnd) {
280 return CounterMappingRegion(Counter(), FileID, 0, LineStart, ColumnStart,
281 LineEnd, ColumnEnd, SkippedRegion);
282 }
283
284 static CounterMappingRegion
makeGapRegionCounterMappingRegion285 makeGapRegion(Counter Count, unsigned FileID, unsigned LineStart,
286 unsigned ColumnStart, unsigned LineEnd, unsigned ColumnEnd) {
287 return CounterMappingRegion(Count, FileID, 0, LineStart, ColumnStart,
288 LineEnd, (1U << 31) | ColumnEnd, GapRegion);
289 }
290
291 static CounterMappingRegion
makeBranchRegionCounterMappingRegion292 makeBranchRegion(Counter Count, Counter FalseCount, unsigned FileID,
293 unsigned LineStart, unsigned ColumnStart, unsigned LineEnd,
294 unsigned ColumnEnd) {
295 return CounterMappingRegion(Count, FalseCount, FileID, 0, LineStart,
296 ColumnStart, LineEnd, ColumnEnd, BranchRegion);
297 }
298
startLocCounterMappingRegion299 inline LineColPair startLoc() const {
300 return LineColPair(LineStart, ColumnStart);
301 }
302
endLocCounterMappingRegion303 inline LineColPair endLoc() const { return LineColPair(LineEnd, ColumnEnd); }
304 };
305
306 /// Associates a source range with an execution count.
307 struct CountedRegion : public CounterMappingRegion {
308 uint64_t ExecutionCount;
309 uint64_t FalseExecutionCount;
310 bool Folded;
311
CountedRegionCountedRegion312 CountedRegion(const CounterMappingRegion &R, uint64_t ExecutionCount)
313 : CounterMappingRegion(R), ExecutionCount(ExecutionCount),
314 FalseExecutionCount(0), Folded(false) {}
315
CountedRegionCountedRegion316 CountedRegion(const CounterMappingRegion &R, uint64_t ExecutionCount,
317 uint64_t FalseExecutionCount)
318 : CounterMappingRegion(R), ExecutionCount(ExecutionCount),
319 FalseExecutionCount(FalseExecutionCount), Folded(false) {}
320 };
321
322 /// A Counter mapping context is used to connect the counters, expressions
323 /// and the obtained counter values.
324 class CounterMappingContext {
325 ArrayRef<CounterExpression> Expressions;
326 ArrayRef<uint64_t> CounterValues;
327
328 public:
329 CounterMappingContext(ArrayRef<CounterExpression> Expressions,
330 ArrayRef<uint64_t> CounterValues = std::nullopt)
Expressions(Expressions)331 : Expressions(Expressions), CounterValues(CounterValues) {}
332
setCounts(ArrayRef<uint64_t> Counts)333 void setCounts(ArrayRef<uint64_t> Counts) { CounterValues = Counts; }
334
335 void dump(const Counter &C, raw_ostream &OS) const;
dump(const Counter & C)336 void dump(const Counter &C) const { dump(C, dbgs()); }
337
338 /// Return the number of times that a region of code associated with this
339 /// counter was executed.
340 Expected<int64_t> evaluate(const Counter &C) const;
341
342 unsigned getMaxCounterID(const Counter &C) const;
343 };
344
345 /// Code coverage information for a single function.
346 struct FunctionRecord {
347 /// Raw function name.
348 std::string Name;
349 /// Mapping from FileID (i.e. vector index) to filename. Used to support
350 /// macro expansions within a function in which the macro and function are
351 /// defined in separate files.
352 ///
353 /// TODO: Uniquing filenames across all function records may be a performance
354 /// optimization.
355 std::vector<std::string> Filenames;
356 /// Regions in the function along with their counts.
357 std::vector<CountedRegion> CountedRegions;
358 /// Branch Regions in the function along with their counts.
359 std::vector<CountedRegion> CountedBranchRegions;
360 /// The number of times this function was executed.
361 uint64_t ExecutionCount = 0;
362
FunctionRecordFunctionRecord363 FunctionRecord(StringRef Name, ArrayRef<StringRef> Filenames)
364 : Name(Name), Filenames(Filenames.begin(), Filenames.end()) {}
365
366 FunctionRecord(FunctionRecord &&FR) = default;
367 FunctionRecord &operator=(FunctionRecord &&) = default;
368
pushRegionFunctionRecord369 void pushRegion(CounterMappingRegion Region, uint64_t Count,
370 uint64_t FalseCount) {
371 if (Region.Kind == CounterMappingRegion::BranchRegion) {
372 CountedBranchRegions.emplace_back(Region, Count, FalseCount);
373 // If both counters are hard-coded to zero, then this region represents a
374 // constant-folded branch.
375 if (Region.Count.isZero() && Region.FalseCount.isZero())
376 CountedBranchRegions.back().Folded = true;
377 return;
378 }
379 if (CountedRegions.empty())
380 ExecutionCount = Count;
381 CountedRegions.emplace_back(Region, Count, FalseCount);
382 }
383 };
384
385 /// Iterator over Functions, optionally filtered to a single file.
386 class FunctionRecordIterator
387 : public iterator_facade_base<FunctionRecordIterator,
388 std::forward_iterator_tag, FunctionRecord> {
389 ArrayRef<FunctionRecord> Records;
390 ArrayRef<FunctionRecord>::iterator Current;
391 StringRef Filename;
392
393 /// Skip records whose primary file is not \c Filename.
394 void skipOtherFiles();
395
396 public:
397 FunctionRecordIterator(ArrayRef<FunctionRecord> Records_,
398 StringRef Filename = "")
Records(Records_)399 : Records(Records_), Current(Records.begin()), Filename(Filename) {
400 skipOtherFiles();
401 }
402
FunctionRecordIterator()403 FunctionRecordIterator() : Current(Records.begin()) {}
404
405 bool operator==(const FunctionRecordIterator &RHS) const {
406 return Current == RHS.Current && Filename == RHS.Filename;
407 }
408
409 const FunctionRecord &operator*() const { return *Current; }
410
411 FunctionRecordIterator &operator++() {
412 assert(Current != Records.end() && "incremented past end");
413 ++Current;
414 skipOtherFiles();
415 return *this;
416 }
417 };
418
419 /// Coverage information for a macro expansion or #included file.
420 ///
421 /// When covered code has pieces that can be expanded for more detail, such as a
422 /// preprocessor macro use and its definition, these are represented as
423 /// expansions whose coverage can be looked up independently.
424 struct ExpansionRecord {
425 /// The abstract file this expansion covers.
426 unsigned FileID;
427 /// The region that expands to this record.
428 const CountedRegion &Region;
429 /// Coverage for the expansion.
430 const FunctionRecord &Function;
431
ExpansionRecordExpansionRecord432 ExpansionRecord(const CountedRegion &Region,
433 const FunctionRecord &Function)
434 : FileID(Region.ExpandedFileID), Region(Region), Function(Function) {}
435 };
436
437 /// The execution count information starting at a point in a file.
438 ///
439 /// A sequence of CoverageSegments gives execution counts for a file in format
440 /// that's simple to iterate through for processing.
441 struct CoverageSegment {
442 /// The line where this segment begins.
443 unsigned Line;
444 /// The column where this segment begins.
445 unsigned Col;
446 /// The execution count, or zero if no count was recorded.
447 uint64_t Count;
448 /// When false, the segment was uninstrumented or skipped.
449 bool HasCount;
450 /// Whether this enters a new region or returns to a previous count.
451 bool IsRegionEntry;
452 /// Whether this enters a gap region.
453 bool IsGapRegion;
454
CoverageSegmentCoverageSegment455 CoverageSegment(unsigned Line, unsigned Col, bool IsRegionEntry)
456 : Line(Line), Col(Col), Count(0), HasCount(false),
457 IsRegionEntry(IsRegionEntry), IsGapRegion(false) {}
458
459 CoverageSegment(unsigned Line, unsigned Col, uint64_t Count,
460 bool IsRegionEntry, bool IsGapRegion = false,
461 bool IsBranchRegion = false)
LineCoverageSegment462 : Line(Line), Col(Col), Count(Count), HasCount(true),
463 IsRegionEntry(IsRegionEntry), IsGapRegion(IsGapRegion) {}
464
465 friend bool operator==(const CoverageSegment &L, const CoverageSegment &R) {
466 return std::tie(L.Line, L.Col, L.Count, L.HasCount, L.IsRegionEntry,
467 L.IsGapRegion) == std::tie(R.Line, R.Col, R.Count,
468 R.HasCount, R.IsRegionEntry,
469 R.IsGapRegion);
470 }
471 };
472
473 /// An instantiation group contains a \c FunctionRecord list, such that each
474 /// record corresponds to a distinct instantiation of the same function.
475 ///
476 /// Note that it's possible for a function to have more than one instantiation
477 /// (consider C++ template specializations or static inline functions).
478 class InstantiationGroup {
479 friend class CoverageMapping;
480
481 unsigned Line;
482 unsigned Col;
483 std::vector<const FunctionRecord *> Instantiations;
484
InstantiationGroup(unsigned Line,unsigned Col,std::vector<const FunctionRecord * > Instantiations)485 InstantiationGroup(unsigned Line, unsigned Col,
486 std::vector<const FunctionRecord *> Instantiations)
487 : Line(Line), Col(Col), Instantiations(std::move(Instantiations)) {}
488
489 public:
490 InstantiationGroup(const InstantiationGroup &) = delete;
491 InstantiationGroup(InstantiationGroup &&) = default;
492
493 /// Get the number of instantiations in this group.
size()494 size_t size() const { return Instantiations.size(); }
495
496 /// Get the line where the common function was defined.
getLine()497 unsigned getLine() const { return Line; }
498
499 /// Get the column where the common function was defined.
getColumn()500 unsigned getColumn() const { return Col; }
501
502 /// Check if the instantiations in this group have a common mangled name.
hasName()503 bool hasName() const {
504 for (unsigned I = 1, E = Instantiations.size(); I < E; ++I)
505 if (Instantiations[I]->Name != Instantiations[0]->Name)
506 return false;
507 return true;
508 }
509
510 /// Get the common mangled name for instantiations in this group.
getName()511 StringRef getName() const {
512 assert(hasName() && "Instantiations don't have a shared name");
513 return Instantiations[0]->Name;
514 }
515
516 /// Get the total execution count of all instantiations in this group.
getTotalExecutionCount()517 uint64_t getTotalExecutionCount() const {
518 uint64_t Count = 0;
519 for (const FunctionRecord *F : Instantiations)
520 Count += F->ExecutionCount;
521 return Count;
522 }
523
524 /// Get the instantiations in this group.
getInstantiations()525 ArrayRef<const FunctionRecord *> getInstantiations() const {
526 return Instantiations;
527 }
528 };
529
530 /// Coverage information to be processed or displayed.
531 ///
532 /// This represents the coverage of an entire file, expansion, or function. It
533 /// provides a sequence of CoverageSegments to iterate through, as well as the
534 /// list of expansions that can be further processed.
535 class CoverageData {
536 friend class CoverageMapping;
537
538 std::string Filename;
539 std::vector<CoverageSegment> Segments;
540 std::vector<ExpansionRecord> Expansions;
541 std::vector<CountedRegion> BranchRegions;
542
543 public:
544 CoverageData() = default;
545
CoverageData(StringRef Filename)546 CoverageData(StringRef Filename) : Filename(Filename) {}
547
548 /// Get the name of the file this data covers.
getFilename()549 StringRef getFilename() const { return Filename; }
550
551 /// Get an iterator over the coverage segments for this object. The segments
552 /// are guaranteed to be uniqued and sorted by location.
begin()553 std::vector<CoverageSegment>::const_iterator begin() const {
554 return Segments.begin();
555 }
556
end()557 std::vector<CoverageSegment>::const_iterator end() const {
558 return Segments.end();
559 }
560
empty()561 bool empty() const { return Segments.empty(); }
562
563 /// Expansions that can be further processed.
getExpansions()564 ArrayRef<ExpansionRecord> getExpansions() const { return Expansions; }
565
566 /// Branches that can be further processed.
getBranches()567 ArrayRef<CountedRegion> getBranches() const { return BranchRegions; }
568 };
569
570 /// The mapping of profile information to coverage data.
571 ///
572 /// This is the main interface to get coverage information, using a profile to
573 /// fill out execution counts.
574 class CoverageMapping {
575 DenseMap<size_t, DenseSet<size_t>> RecordProvenance;
576 std::vector<FunctionRecord> Functions;
577 DenseMap<size_t, SmallVector<unsigned, 0>> FilenameHash2RecordIndices;
578 std::vector<std::pair<std::string, uint64_t>> FuncHashMismatches;
579
580 CoverageMapping() = default;
581
582 // Load coverage records from readers.
583 static Error loadFromReaders(
584 ArrayRef<std::unique_ptr<CoverageMappingReader>> CoverageReaders,
585 IndexedInstrProfReader &ProfileReader, CoverageMapping &Coverage);
586
587 // Load coverage records from file.
588 static Error
589 loadFromFile(StringRef Filename, StringRef Arch, StringRef CompilationDir,
590 IndexedInstrProfReader &ProfileReader, CoverageMapping &Coverage,
591 bool &DataFound,
592 SmallVectorImpl<object::BuildID> *FoundBinaryIDs = nullptr);
593
594 /// Add a function record corresponding to \p Record.
595 Error loadFunctionRecord(const CoverageMappingRecord &Record,
596 IndexedInstrProfReader &ProfileReader);
597
598 /// Look up the indices for function records which are at least partially
599 /// defined in the specified file. This is guaranteed to return a superset of
600 /// such records: extra records not in the file may be included if there is
601 /// a hash collision on the filename. Clients must be robust to collisions.
602 ArrayRef<unsigned>
603 getImpreciseRecordIndicesForFilename(StringRef Filename) const;
604
605 public:
606 CoverageMapping(const CoverageMapping &) = delete;
607 CoverageMapping &operator=(const CoverageMapping &) = delete;
608
609 /// Load the coverage mapping using the given readers.
610 static Expected<std::unique_ptr<CoverageMapping>>
611 load(ArrayRef<std::unique_ptr<CoverageMappingReader>> CoverageReaders,
612 IndexedInstrProfReader &ProfileReader);
613
614 /// Load the coverage mapping from the given object files and profile. If
615 /// \p Arches is non-empty, it must specify an architecture for each object.
616 /// Ignores non-instrumented object files unless all are not instrumented.
617 static Expected<std::unique_ptr<CoverageMapping>>
618 load(ArrayRef<StringRef> ObjectFilenames, StringRef ProfileFilename,
619 ArrayRef<StringRef> Arches = std::nullopt, StringRef CompilationDir = "",
620 const object::BuildIDFetcher *BIDFetcher = nullptr);
621
622 /// The number of functions that couldn't have their profiles mapped.
623 ///
624 /// This is a count of functions whose profile is out of date or otherwise
625 /// can't be associated with any coverage information.
getMismatchedCount()626 unsigned getMismatchedCount() const { return FuncHashMismatches.size(); }
627
628 /// A hash mismatch occurs when a profile record for a symbol does not have
629 /// the same hash as a coverage mapping record for the same symbol. This
630 /// returns a list of hash mismatches, where each mismatch is a pair of the
631 /// symbol name and its coverage mapping hash.
getHashMismatches()632 ArrayRef<std::pair<std::string, uint64_t>> getHashMismatches() const {
633 return FuncHashMismatches;
634 }
635
636 /// Returns a lexicographically sorted, unique list of files that are
637 /// covered.
638 std::vector<StringRef> getUniqueSourceFiles() const;
639
640 /// Get the coverage for a particular file.
641 ///
642 /// The given filename must be the name as recorded in the coverage
643 /// information. That is, only names returned from getUniqueSourceFiles will
644 /// yield a result.
645 CoverageData getCoverageForFile(StringRef Filename) const;
646
647 /// Get the coverage for a particular function.
648 CoverageData getCoverageForFunction(const FunctionRecord &Function) const;
649
650 /// Get the coverage for an expansion within a coverage set.
651 CoverageData getCoverageForExpansion(const ExpansionRecord &Expansion) const;
652
653 /// Gets all of the functions covered by this profile.
getCoveredFunctions()654 iterator_range<FunctionRecordIterator> getCoveredFunctions() const {
655 return make_range(FunctionRecordIterator(Functions),
656 FunctionRecordIterator());
657 }
658
659 /// Gets all of the functions in a particular file.
660 iterator_range<FunctionRecordIterator>
getCoveredFunctions(StringRef Filename)661 getCoveredFunctions(StringRef Filename) const {
662 return make_range(FunctionRecordIterator(Functions, Filename),
663 FunctionRecordIterator());
664 }
665
666 /// Get the list of function instantiation groups in a particular file.
667 ///
668 /// Every instantiation group in a program is attributed to exactly one file:
669 /// the file in which the definition for the common function begins.
670 std::vector<InstantiationGroup>
671 getInstantiationGroups(StringRef Filename) const;
672 };
673
674 /// Coverage statistics for a single line.
675 class LineCoverageStats {
676 uint64_t ExecutionCount;
677 bool HasMultipleRegions;
678 bool Mapped;
679 unsigned Line;
680 ArrayRef<const CoverageSegment *> LineSegments;
681 const CoverageSegment *WrappedSegment;
682
683 friend class LineCoverageIterator;
684 LineCoverageStats() = default;
685
686 public:
687 LineCoverageStats(ArrayRef<const CoverageSegment *> LineSegments,
688 const CoverageSegment *WrappedSegment, unsigned Line);
689
getExecutionCount()690 uint64_t getExecutionCount() const { return ExecutionCount; }
691
hasMultipleRegions()692 bool hasMultipleRegions() const { return HasMultipleRegions; }
693
isMapped()694 bool isMapped() const { return Mapped; }
695
getLine()696 unsigned getLine() const { return Line; }
697
getLineSegments()698 ArrayRef<const CoverageSegment *> getLineSegments() const {
699 return LineSegments;
700 }
701
getWrappedSegment()702 const CoverageSegment *getWrappedSegment() const { return WrappedSegment; }
703 };
704
705 /// An iterator over the \c LineCoverageStats objects for lines described by
706 /// a \c CoverageData instance.
707 class LineCoverageIterator
708 : public iterator_facade_base<LineCoverageIterator,
709 std::forward_iterator_tag,
710 const LineCoverageStats> {
711 public:
LineCoverageIterator(const CoverageData & CD)712 LineCoverageIterator(const CoverageData &CD)
713 : LineCoverageIterator(CD, CD.begin()->Line) {}
714
LineCoverageIterator(const CoverageData & CD,unsigned Line)715 LineCoverageIterator(const CoverageData &CD, unsigned Line)
716 : CD(CD), WrappedSegment(nullptr), Next(CD.begin()), Ended(false),
717 Line(Line) {
718 this->operator++();
719 }
720
721 bool operator==(const LineCoverageIterator &R) const {
722 return &CD == &R.CD && Next == R.Next && Ended == R.Ended;
723 }
724
725 const LineCoverageStats &operator*() const { return Stats; }
726
727 LineCoverageIterator &operator++();
728
getEnd()729 LineCoverageIterator getEnd() const {
730 auto EndIt = *this;
731 EndIt.Next = CD.end();
732 EndIt.Ended = true;
733 return EndIt;
734 }
735
736 private:
737 const CoverageData &CD;
738 const CoverageSegment *WrappedSegment;
739 std::vector<CoverageSegment>::const_iterator Next;
740 bool Ended;
741 unsigned Line;
742 SmallVector<const CoverageSegment *, 4> Segments;
743 LineCoverageStats Stats;
744 };
745
746 /// Get a \c LineCoverageIterator range for the lines described by \p CD.
747 static inline iterator_range<LineCoverageIterator>
getLineCoverageStats(const coverage::CoverageData & CD)748 getLineCoverageStats(const coverage::CoverageData &CD) {
749 auto Begin = LineCoverageIterator(CD);
750 auto End = Begin.getEnd();
751 return make_range(Begin, End);
752 }
753
754 // Coverage mappping data (V2) has the following layout:
755 // IPSK_covmap:
756 // [CoverageMapFileHeader]
757 // [ArrayStart]
758 // [CovMapFunctionRecordV2]
759 // [CovMapFunctionRecordV2]
760 // ...
761 // [ArrayEnd]
762 // [Encoded Filenames and Region Mapping Data]
763 //
764 // Coverage mappping data (V3) has the following layout:
765 // IPSK_covmap:
766 // [CoverageMapFileHeader]
767 // [Encoded Filenames]
768 // IPSK_covfun:
769 // [ArrayStart]
770 // odr_name_1: [CovMapFunctionRecordV3]
771 // odr_name_2: [CovMapFunctionRecordV3]
772 // ...
773 // [ArrayEnd]
774 //
775 // Both versions of the coverage mapping format encode the same information,
776 // but the V3 format does so more compactly by taking advantage of linkonce_odr
777 // semantics (it allows exactly 1 function record per name reference).
778
779 /// This namespace defines accessors shared by different versions of coverage
780 /// mapping records.
781 namespace accessors {
782
783 /// Return the structural hash associated with the function.
784 template <class FuncRecordTy, support::endianness Endian>
getFuncHash(const FuncRecordTy * Record)785 uint64_t getFuncHash(const FuncRecordTy *Record) {
786 return support::endian::byte_swap<uint64_t, Endian>(Record->FuncHash);
787 }
788
789 /// Return the coverage map data size for the function.
790 template <class FuncRecordTy, support::endianness Endian>
getDataSize(const FuncRecordTy * Record)791 uint64_t getDataSize(const FuncRecordTy *Record) {
792 return support::endian::byte_swap<uint32_t, Endian>(Record->DataSize);
793 }
794
795 /// Return the function lookup key. The value is considered opaque.
796 template <class FuncRecordTy, support::endianness Endian>
getFuncNameRef(const FuncRecordTy * Record)797 uint64_t getFuncNameRef(const FuncRecordTy *Record) {
798 return support::endian::byte_swap<uint64_t, Endian>(Record->NameRef);
799 }
800
801 /// Return the PGO name of the function. Used for formats in which the name is
802 /// a hash.
803 template <class FuncRecordTy, support::endianness Endian>
getFuncNameViaRef(const FuncRecordTy * Record,InstrProfSymtab & ProfileNames,StringRef & FuncName)804 Error getFuncNameViaRef(const FuncRecordTy *Record,
805 InstrProfSymtab &ProfileNames, StringRef &FuncName) {
806 uint64_t NameRef = getFuncNameRef<FuncRecordTy, Endian>(Record);
807 FuncName = ProfileNames.getFuncName(NameRef);
808 return Error::success();
809 }
810
811 /// Read coverage mapping out-of-line, from \p MappingBuf. This is used when the
812 /// coverage mapping is attached to the file header, instead of to the function
813 /// record.
814 template <class FuncRecordTy, support::endianness Endian>
getCoverageMappingOutOfLine(const FuncRecordTy * Record,const char * MappingBuf)815 StringRef getCoverageMappingOutOfLine(const FuncRecordTy *Record,
816 const char *MappingBuf) {
817 return {MappingBuf, size_t(getDataSize<FuncRecordTy, Endian>(Record))};
818 }
819
820 /// Advance to the next out-of-line coverage mapping and its associated
821 /// function record.
822 template <class FuncRecordTy, support::endianness Endian>
823 std::pair<const char *, const FuncRecordTy *>
advanceByOneOutOfLine(const FuncRecordTy * Record,const char * MappingBuf)824 advanceByOneOutOfLine(const FuncRecordTy *Record, const char *MappingBuf) {
825 return {MappingBuf + getDataSize<FuncRecordTy, Endian>(Record), Record + 1};
826 }
827
828 } // end namespace accessors
829
830 LLVM_PACKED_START
831 template <class IntPtrT>
832 struct CovMapFunctionRecordV1 {
833 using ThisT = CovMapFunctionRecordV1<IntPtrT>;
834
835 #define COVMAP_V1
836 #define COVMAP_FUNC_RECORD(Type, LLVMType, Name, Init) Type Name;
837 #include "llvm/ProfileData/InstrProfData.inc"
838 #undef COVMAP_V1
839 CovMapFunctionRecordV1() = delete;
840
getFuncHashCovMapFunctionRecordV1841 template <support::endianness Endian> uint64_t getFuncHash() const {
842 return accessors::getFuncHash<ThisT, Endian>(this);
843 }
844
getDataSizeCovMapFunctionRecordV1845 template <support::endianness Endian> uint64_t getDataSize() const {
846 return accessors::getDataSize<ThisT, Endian>(this);
847 }
848
849 /// Return function lookup key. The value is consider opaque.
getFuncNameRefCovMapFunctionRecordV1850 template <support::endianness Endian> IntPtrT getFuncNameRef() const {
851 return support::endian::byte_swap<IntPtrT, Endian>(NamePtr);
852 }
853
854 /// Return the PGO name of the function.
855 template <support::endianness Endian>
getFuncNameCovMapFunctionRecordV1856 Error getFuncName(InstrProfSymtab &ProfileNames, StringRef &FuncName) const {
857 IntPtrT NameRef = getFuncNameRef<Endian>();
858 uint32_t NameS = support::endian::byte_swap<uint32_t, Endian>(NameSize);
859 FuncName = ProfileNames.getFuncName(NameRef, NameS);
860 if (NameS && FuncName.empty())
861 return make_error<CoverageMapError>(coveragemap_error::malformed);
862 return Error::success();
863 }
864
865 template <support::endianness Endian>
866 std::pair<const char *, const ThisT *>
advanceByOneCovMapFunctionRecordV1867 advanceByOne(const char *MappingBuf) const {
868 return accessors::advanceByOneOutOfLine<ThisT, Endian>(this, MappingBuf);
869 }
870
getFilenamesRefCovMapFunctionRecordV1871 template <support::endianness Endian> uint64_t getFilenamesRef() const {
872 llvm_unreachable("V1 function format does not contain a filenames ref");
873 }
874
875 template <support::endianness Endian>
getCoverageMappingCovMapFunctionRecordV1876 StringRef getCoverageMapping(const char *MappingBuf) const {
877 return accessors::getCoverageMappingOutOfLine<ThisT, Endian>(this,
878 MappingBuf);
879 }
880 };
881
882 struct CovMapFunctionRecordV2 {
883 using ThisT = CovMapFunctionRecordV2;
884
885 #define COVMAP_V2
886 #define COVMAP_FUNC_RECORD(Type, LLVMType, Name, Init) Type Name;
887 #include "llvm/ProfileData/InstrProfData.inc"
888 #undef COVMAP_V2
889 CovMapFunctionRecordV2() = delete;
890
getFuncHashCovMapFunctionRecordV2891 template <support::endianness Endian> uint64_t getFuncHash() const {
892 return accessors::getFuncHash<ThisT, Endian>(this);
893 }
894
getDataSizeCovMapFunctionRecordV2895 template <support::endianness Endian> uint64_t getDataSize() const {
896 return accessors::getDataSize<ThisT, Endian>(this);
897 }
898
getFuncNameRefCovMapFunctionRecordV2899 template <support::endianness Endian> uint64_t getFuncNameRef() const {
900 return accessors::getFuncNameRef<ThisT, Endian>(this);
901 }
902
903 template <support::endianness Endian>
getFuncNameCovMapFunctionRecordV2904 Error getFuncName(InstrProfSymtab &ProfileNames, StringRef &FuncName) const {
905 return accessors::getFuncNameViaRef<ThisT, Endian>(this, ProfileNames,
906 FuncName);
907 }
908
909 template <support::endianness Endian>
910 std::pair<const char *, const ThisT *>
advanceByOneCovMapFunctionRecordV2911 advanceByOne(const char *MappingBuf) const {
912 return accessors::advanceByOneOutOfLine<ThisT, Endian>(this, MappingBuf);
913 }
914
getFilenamesRefCovMapFunctionRecordV2915 template <support::endianness Endian> uint64_t getFilenamesRef() const {
916 llvm_unreachable("V2 function format does not contain a filenames ref");
917 }
918
919 template <support::endianness Endian>
getCoverageMappingCovMapFunctionRecordV2920 StringRef getCoverageMapping(const char *MappingBuf) const {
921 return accessors::getCoverageMappingOutOfLine<ThisT, Endian>(this,
922 MappingBuf);
923 }
924 };
925
926 struct CovMapFunctionRecordV3 {
927 using ThisT = CovMapFunctionRecordV3;
928
929 #define COVMAP_V3
930 #define COVMAP_FUNC_RECORD(Type, LLVMType, Name, Init) Type Name;
931 #include "llvm/ProfileData/InstrProfData.inc"
932 #undef COVMAP_V3
933 CovMapFunctionRecordV3() = delete;
934
getFuncHashCovMapFunctionRecordV3935 template <support::endianness Endian> uint64_t getFuncHash() const {
936 return accessors::getFuncHash<ThisT, Endian>(this);
937 }
938
getDataSizeCovMapFunctionRecordV3939 template <support::endianness Endian> uint64_t getDataSize() const {
940 return accessors::getDataSize<ThisT, Endian>(this);
941 }
942
getFuncNameRefCovMapFunctionRecordV3943 template <support::endianness Endian> uint64_t getFuncNameRef() const {
944 return accessors::getFuncNameRef<ThisT, Endian>(this);
945 }
946
947 template <support::endianness Endian>
getFuncNameCovMapFunctionRecordV3948 Error getFuncName(InstrProfSymtab &ProfileNames, StringRef &FuncName) const {
949 return accessors::getFuncNameViaRef<ThisT, Endian>(this, ProfileNames,
950 FuncName);
951 }
952
953 /// Get the filename set reference.
getFilenamesRefCovMapFunctionRecordV3954 template <support::endianness Endian> uint64_t getFilenamesRef() const {
955 return support::endian::byte_swap<uint64_t, Endian>(FilenamesRef);
956 }
957
958 /// Read the inline coverage mapping. Ignore the buffer parameter, it is for
959 /// out-of-line coverage mapping data only.
960 template <support::endianness Endian>
getCoverageMappingCovMapFunctionRecordV3961 StringRef getCoverageMapping(const char *) const {
962 return StringRef(&CoverageMapping, getDataSize<Endian>());
963 }
964
965 // Advance to the next inline coverage mapping and its associated function
966 // record. Ignore the out-of-line coverage mapping buffer.
967 template <support::endianness Endian>
968 std::pair<const char *, const CovMapFunctionRecordV3 *>
advanceByOneCovMapFunctionRecordV3969 advanceByOne(const char *) const {
970 assert(isAddrAligned(Align(8), this) && "Function record not aligned");
971 const char *Next = ((const char *)this) + sizeof(CovMapFunctionRecordV3) -
972 sizeof(char) + getDataSize<Endian>();
973 // Each function record has an alignment of 8, so we need to adjust
974 // alignment before reading the next record.
975 Next += offsetToAlignedAddr(Next, Align(8));
976 return {nullptr, reinterpret_cast<const CovMapFunctionRecordV3 *>(Next)};
977 }
978 };
979
980 // Per module coverage mapping data header, i.e. CoverageMapFileHeader
981 // documented above.
982 struct CovMapHeader {
983 #define COVMAP_HEADER(Type, LLVMType, Name, Init) Type Name;
984 #include "llvm/ProfileData/InstrProfData.inc"
getNRecordsCovMapHeader985 template <support::endianness Endian> uint32_t getNRecords() const {
986 return support::endian::byte_swap<uint32_t, Endian>(NRecords);
987 }
988
getFilenamesSizeCovMapHeader989 template <support::endianness Endian> uint32_t getFilenamesSize() const {
990 return support::endian::byte_swap<uint32_t, Endian>(FilenamesSize);
991 }
992
getCoverageSizeCovMapHeader993 template <support::endianness Endian> uint32_t getCoverageSize() const {
994 return support::endian::byte_swap<uint32_t, Endian>(CoverageSize);
995 }
996
getVersionCovMapHeader997 template <support::endianness Endian> uint32_t getVersion() const {
998 return support::endian::byte_swap<uint32_t, Endian>(Version);
999 }
1000 };
1001
1002 LLVM_PACKED_END
1003
1004 enum CovMapVersion {
1005 Version1 = 0,
1006 // Function's name reference from CovMapFuncRecord is changed from raw
1007 // name string pointer to MD5 to support name section compression. Name
1008 // section is also compressed.
1009 Version2 = 1,
1010 // A new interpretation of the columnEnd field is added in order to mark
1011 // regions as gap areas.
1012 Version3 = 2,
1013 // Function records are named, uniqued, and moved to a dedicated section.
1014 Version4 = 3,
1015 // Branch regions referring to two counters are added
1016 Version5 = 4,
1017 // Compilation directory is stored separately and combined with relative
1018 // filenames to produce an absolute file path.
1019 Version6 = 5,
1020 // The current version is Version6.
1021 CurrentVersion = INSTR_PROF_COVMAP_VERSION
1022 };
1023
1024 template <int CovMapVersion, class IntPtrT> struct CovMapTraits {
1025 using CovMapFuncRecordType = CovMapFunctionRecordV3;
1026 using NameRefType = uint64_t;
1027 };
1028
1029 template <class IntPtrT> struct CovMapTraits<CovMapVersion::Version3, IntPtrT> {
1030 using CovMapFuncRecordType = CovMapFunctionRecordV2;
1031 using NameRefType = uint64_t;
1032 };
1033
1034 template <class IntPtrT> struct CovMapTraits<CovMapVersion::Version2, IntPtrT> {
1035 using CovMapFuncRecordType = CovMapFunctionRecordV2;
1036 using NameRefType = uint64_t;
1037 };
1038
1039 template <class IntPtrT> struct CovMapTraits<CovMapVersion::Version1, IntPtrT> {
1040 using CovMapFuncRecordType = CovMapFunctionRecordV1<IntPtrT>;
1041 using NameRefType = IntPtrT;
1042 };
1043
1044 } // end namespace coverage
1045
1046 /// Provide DenseMapInfo for CounterExpression
1047 template<> struct DenseMapInfo<coverage::CounterExpression> {
1048 static inline coverage::CounterExpression getEmptyKey() {
1049 using namespace coverage;
1050
1051 return CounterExpression(CounterExpression::ExprKind::Subtract,
1052 Counter::getCounter(~0U),
1053 Counter::getCounter(~0U));
1054 }
1055
1056 static inline coverage::CounterExpression getTombstoneKey() {
1057 using namespace coverage;
1058
1059 return CounterExpression(CounterExpression::ExprKind::Add,
1060 Counter::getCounter(~0U),
1061 Counter::getCounter(~0U));
1062 }
1063
1064 static unsigned getHashValue(const coverage::CounterExpression &V) {
1065 return static_cast<unsigned>(
1066 hash_combine(V.Kind, V.LHS.getKind(), V.LHS.getCounterID(),
1067 V.RHS.getKind(), V.RHS.getCounterID()));
1068 }
1069
1070 static bool isEqual(const coverage::CounterExpression &LHS,
1071 const coverage::CounterExpression &RHS) {
1072 return LHS.Kind == RHS.Kind && LHS.LHS == RHS.LHS && LHS.RHS == RHS.RHS;
1073 }
1074 };
1075
1076 } // end namespace llvm
1077
1078 #endif // LLVM_PROFILEDATA_COVERAGE_COVERAGEMAPPING_H
1079