1 //=-- CoverageMapping.cpp - Code coverage mapping support ---------*- 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 clang's and llvm's instrumentation based
11 // code coverage.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "llvm/ProfileData/CoverageMapping.h"
16 #include "llvm/ADT/DenseMap.h"
17 #include "llvm/ADT/Optional.h"
18 #include "llvm/ADT/SmallBitVector.h"
19 #include "llvm/ProfileData/CoverageMappingReader.h"
20 #include "llvm/ProfileData/InstrProfReader.h"
21 #include "llvm/Support/Debug.h"
22 #include "llvm/Support/Errc.h"
23 #include "llvm/Support/ErrorHandling.h"
24 #include "llvm/Support/ManagedStatic.h"
25 #include "llvm/Support/Path.h"
26 #include "llvm/Support/raw_ostream.h"
27 
28 using namespace llvm;
29 using namespace coverage;
30 
31 #define DEBUG_TYPE "coverage-mapping"
32 
get(const CounterExpression & E)33 Counter CounterExpressionBuilder::get(const CounterExpression &E) {
34   auto It = ExpressionIndices.find(E);
35   if (It != ExpressionIndices.end())
36     return Counter::getExpression(It->second);
37   unsigned I = Expressions.size();
38   Expressions.push_back(E);
39   ExpressionIndices[E] = I;
40   return Counter::getExpression(I);
41 }
42 
extractTerms(Counter C,int Sign,SmallVectorImpl<std::pair<unsigned,int>> & Terms)43 void CounterExpressionBuilder::extractTerms(
44     Counter C, int Sign, SmallVectorImpl<std::pair<unsigned, int>> &Terms) {
45   switch (C.getKind()) {
46   case Counter::Zero:
47     break;
48   case Counter::CounterValueReference:
49     Terms.push_back(std::make_pair(C.getCounterID(), Sign));
50     break;
51   case Counter::Expression:
52     const auto &E = Expressions[C.getExpressionID()];
53     extractTerms(E.LHS, Sign, Terms);
54     extractTerms(E.RHS, E.Kind == CounterExpression::Subtract ? -Sign : Sign,
55                  Terms);
56     break;
57   }
58 }
59 
simplify(Counter ExpressionTree)60 Counter CounterExpressionBuilder::simplify(Counter ExpressionTree) {
61   // Gather constant terms.
62   llvm::SmallVector<std::pair<unsigned, int>, 32> Terms;
63   extractTerms(ExpressionTree, +1, Terms);
64 
65   // If there are no terms, this is just a zero. The algorithm below assumes at
66   // least one term.
67   if (Terms.size() == 0)
68     return Counter::getZero();
69 
70   // Group the terms by counter ID.
71   std::sort(Terms.begin(), Terms.end(),
72             [](const std::pair<unsigned, int> &LHS,
73                const std::pair<unsigned, int> &RHS) {
74     return LHS.first < RHS.first;
75   });
76 
77   // Combine terms by counter ID to eliminate counters that sum to zero.
78   auto Prev = Terms.begin();
79   for (auto I = Prev + 1, E = Terms.end(); I != E; ++I) {
80     if (I->first == Prev->first) {
81       Prev->second += I->second;
82       continue;
83     }
84     ++Prev;
85     *Prev = *I;
86   }
87   Terms.erase(++Prev, Terms.end());
88 
89   Counter C;
90   // Create additions. We do this before subtractions to avoid constructs like
91   // ((0 - X) + Y), as opposed to (Y - X).
92   for (auto Term : Terms) {
93     if (Term.second <= 0)
94       continue;
95     for (int I = 0; I < Term.second; ++I)
96       if (C.isZero())
97         C = Counter::getCounter(Term.first);
98       else
99         C = get(CounterExpression(CounterExpression::Add, C,
100                                   Counter::getCounter(Term.first)));
101   }
102 
103   // Create subtractions.
104   for (auto Term : Terms) {
105     if (Term.second >= 0)
106       continue;
107     for (int I = 0; I < -Term.second; ++I)
108       C = get(CounterExpression(CounterExpression::Subtract, C,
109                                 Counter::getCounter(Term.first)));
110   }
111   return C;
112 }
113 
add(Counter LHS,Counter RHS)114 Counter CounterExpressionBuilder::add(Counter LHS, Counter RHS) {
115   return simplify(get(CounterExpression(CounterExpression::Add, LHS, RHS)));
116 }
117 
subtract(Counter LHS,Counter RHS)118 Counter CounterExpressionBuilder::subtract(Counter LHS, Counter RHS) {
119   return simplify(
120       get(CounterExpression(CounterExpression::Subtract, LHS, RHS)));
121 }
122 
dump(const Counter & C,llvm::raw_ostream & OS) const123 void CounterMappingContext::dump(const Counter &C,
124                                  llvm::raw_ostream &OS) const {
125   switch (C.getKind()) {
126   case Counter::Zero:
127     OS << '0';
128     return;
129   case Counter::CounterValueReference:
130     OS << '#' << C.getCounterID();
131     break;
132   case Counter::Expression: {
133     if (C.getExpressionID() >= Expressions.size())
134       return;
135     const auto &E = Expressions[C.getExpressionID()];
136     OS << '(';
137     dump(E.LHS, OS);
138     OS << (E.Kind == CounterExpression::Subtract ? " - " : " + ");
139     dump(E.RHS, OS);
140     OS << ')';
141     break;
142   }
143   }
144   if (CounterValues.empty())
145     return;
146   ErrorOr<int64_t> Value = evaluate(C);
147   if (!Value)
148     return;
149   OS << '[' << *Value << ']';
150 }
151 
evaluate(const Counter & C) const152 ErrorOr<int64_t> CounterMappingContext::evaluate(const Counter &C) const {
153   switch (C.getKind()) {
154   case Counter::Zero:
155     return 0;
156   case Counter::CounterValueReference:
157     if (C.getCounterID() >= CounterValues.size())
158       return make_error_code(errc::argument_out_of_domain);
159     return CounterValues[C.getCounterID()];
160   case Counter::Expression: {
161     if (C.getExpressionID() >= Expressions.size())
162       return make_error_code(errc::argument_out_of_domain);
163     const auto &E = Expressions[C.getExpressionID()];
164     ErrorOr<int64_t> LHS = evaluate(E.LHS);
165     if (!LHS)
166       return LHS;
167     ErrorOr<int64_t> RHS = evaluate(E.RHS);
168     if (!RHS)
169       return RHS;
170     return E.Kind == CounterExpression::Subtract ? *LHS - *RHS : *LHS + *RHS;
171   }
172   }
173   llvm_unreachable("Unhandled CounterKind");
174 }
175 
skipOtherFiles()176 void FunctionRecordIterator::skipOtherFiles() {
177   while (Current != Records.end() && !Filename.empty() &&
178          Filename != Current->Filenames[0])
179     ++Current;
180   if (Current == Records.end())
181     *this = FunctionRecordIterator();
182 }
183 
184 ErrorOr<std::unique_ptr<CoverageMapping>>
load(CoverageMappingReader & CoverageReader,IndexedInstrProfReader & ProfileReader)185 CoverageMapping::load(CoverageMappingReader &CoverageReader,
186                       IndexedInstrProfReader &ProfileReader) {
187   auto Coverage = std::unique_ptr<CoverageMapping>(new CoverageMapping());
188 
189   std::vector<uint64_t> Counts;
190   for (const auto &Record : CoverageReader) {
191     CounterMappingContext Ctx(Record.Expressions);
192 
193     Counts.clear();
194     if (std::error_code EC = ProfileReader.getFunctionCounts(
195             Record.FunctionName, Record.FunctionHash, Counts)) {
196       if (EC == instrprof_error::hash_mismatch) {
197         Coverage->MismatchedFunctionCount++;
198         continue;
199       } else if (EC != instrprof_error::unknown_function)
200         return EC;
201       Counts.assign(Record.MappingRegions.size(), 0);
202     }
203     Ctx.setCounts(Counts);
204 
205     assert(!Record.MappingRegions.empty() && "Function has no regions");
206 
207     StringRef OrigFuncName = Record.FunctionName;
208     if (!Record.Filenames.empty())
209       OrigFuncName =
210           getFuncNameWithoutPrefix(OrigFuncName, Record.Filenames[0]);
211     FunctionRecord Function(OrigFuncName, Record.Filenames);
212     for (const auto &Region : Record.MappingRegions) {
213       ErrorOr<int64_t> ExecutionCount = Ctx.evaluate(Region.Count);
214       if (!ExecutionCount)
215         break;
216       Function.pushRegion(Region, *ExecutionCount);
217     }
218     if (Function.CountedRegions.size() != Record.MappingRegions.size()) {
219       Coverage->MismatchedFunctionCount++;
220       continue;
221     }
222 
223     Coverage->Functions.push_back(std::move(Function));
224   }
225 
226   return std::move(Coverage);
227 }
228 
229 ErrorOr<std::unique_ptr<CoverageMapping>>
load(StringRef ObjectFilename,StringRef ProfileFilename,StringRef Arch)230 CoverageMapping::load(StringRef ObjectFilename, StringRef ProfileFilename,
231                       StringRef Arch) {
232   auto CounterMappingBuff = MemoryBuffer::getFileOrSTDIN(ObjectFilename);
233   if (std::error_code EC = CounterMappingBuff.getError())
234     return EC;
235   auto CoverageReaderOrErr =
236       BinaryCoverageReader::create(CounterMappingBuff.get(), Arch);
237   if (std::error_code EC = CoverageReaderOrErr.getError())
238     return EC;
239   auto CoverageReader = std::move(CoverageReaderOrErr.get());
240   auto ProfileReaderOrErr = IndexedInstrProfReader::create(ProfileFilename);
241   if (auto EC = ProfileReaderOrErr.getError())
242     return EC;
243   auto ProfileReader = std::move(ProfileReaderOrErr.get());
244   return load(*CoverageReader, *ProfileReader);
245 }
246 
247 namespace {
248 /// \brief Distributes functions into instantiation sets.
249 ///
250 /// An instantiation set is a collection of functions that have the same source
251 /// code, ie, template functions specializations.
252 class FunctionInstantiationSetCollector {
253   typedef DenseMap<std::pair<unsigned, unsigned>,
254                    std::vector<const FunctionRecord *>> MapT;
255   MapT InstantiatedFunctions;
256 
257 public:
insert(const FunctionRecord & Function,unsigned FileID)258   void insert(const FunctionRecord &Function, unsigned FileID) {
259     auto I = Function.CountedRegions.begin(), E = Function.CountedRegions.end();
260     while (I != E && I->FileID != FileID)
261       ++I;
262     assert(I != E && "function does not cover the given file");
263     auto &Functions = InstantiatedFunctions[I->startLoc()];
264     Functions.push_back(&Function);
265   }
266 
begin()267   MapT::iterator begin() { return InstantiatedFunctions.begin(); }
268 
end()269   MapT::iterator end() { return InstantiatedFunctions.end(); }
270 };
271 
272 class SegmentBuilder {
273   std::vector<CoverageSegment> Segments;
274   SmallVector<const CountedRegion *, 8> ActiveRegions;
275 
276   /// Start a segment with no count specified.
startSegment(unsigned Line,unsigned Col)277   void startSegment(unsigned Line, unsigned Col) {
278     DEBUG(dbgs() << "Top level segment at " << Line << ":" << Col << "\n");
279     Segments.emplace_back(Line, Col, /*IsRegionEntry=*/false);
280   }
281 
282   /// Start a segment with the given Region's count.
startSegment(unsigned Line,unsigned Col,bool IsRegionEntry,const CountedRegion & Region)283   void startSegment(unsigned Line, unsigned Col, bool IsRegionEntry,
284                     const CountedRegion &Region) {
285     if (Segments.empty())
286       Segments.emplace_back(Line, Col, IsRegionEntry);
287     CoverageSegment S = Segments.back();
288     // Avoid creating empty regions.
289     if (S.Line != Line || S.Col != Col) {
290       Segments.emplace_back(Line, Col, IsRegionEntry);
291       S = Segments.back();
292     }
293     DEBUG(dbgs() << "Segment at " << Line << ":" << Col);
294     // Set this region's count.
295     if (Region.Kind != coverage::CounterMappingRegion::SkippedRegion) {
296       DEBUG(dbgs() << " with count " << Region.ExecutionCount);
297       Segments.back().setCount(Region.ExecutionCount);
298     }
299     DEBUG(dbgs() << "\n");
300   }
301 
302   /// Start a segment for the given region.
startSegment(const CountedRegion & Region)303   void startSegment(const CountedRegion &Region) {
304     startSegment(Region.LineStart, Region.ColumnStart, true, Region);
305   }
306 
307   /// Pop the top region off of the active stack, starting a new segment with
308   /// the containing Region's count.
popRegion()309   void popRegion() {
310     const CountedRegion *Active = ActiveRegions.back();
311     unsigned Line = Active->LineEnd, Col = Active->ColumnEnd;
312     ActiveRegions.pop_back();
313     if (ActiveRegions.empty())
314       startSegment(Line, Col);
315     else
316       startSegment(Line, Col, false, *ActiveRegions.back());
317   }
318 
319 public:
320   /// Build a list of CoverageSegments from a sorted list of Regions.
buildSegments(ArrayRef<CountedRegion> Regions)321   std::vector<CoverageSegment> buildSegments(ArrayRef<CountedRegion> Regions) {
322     const CountedRegion *PrevRegion = nullptr;
323     for (const auto &Region : Regions) {
324       // Pop any regions that end before this one starts.
325       while (!ActiveRegions.empty() &&
326              ActiveRegions.back()->endLoc() <= Region.startLoc())
327         popRegion();
328       if (PrevRegion && PrevRegion->startLoc() == Region.startLoc() &&
329           PrevRegion->endLoc() == Region.endLoc()) {
330         if (Region.Kind == coverage::CounterMappingRegion::CodeRegion)
331           Segments.back().addCount(Region.ExecutionCount);
332       } else {
333         // Add this region to the stack.
334         ActiveRegions.push_back(&Region);
335         startSegment(Region);
336       }
337       PrevRegion = &Region;
338     }
339     // Pop any regions that are left in the stack.
340     while (!ActiveRegions.empty())
341       popRegion();
342     return Segments;
343   }
344 };
345 }
346 
getUniqueSourceFiles() const347 std::vector<StringRef> CoverageMapping::getUniqueSourceFiles() const {
348   std::vector<StringRef> Filenames;
349   for (const auto &Function : getCoveredFunctions())
350     Filenames.insert(Filenames.end(), Function.Filenames.begin(),
351                      Function.Filenames.end());
352   std::sort(Filenames.begin(), Filenames.end());
353   auto Last = std::unique(Filenames.begin(), Filenames.end());
354   Filenames.erase(Last, Filenames.end());
355   return Filenames;
356 }
357 
gatherFileIDs(StringRef SourceFile,const FunctionRecord & Function)358 static SmallBitVector gatherFileIDs(StringRef SourceFile,
359                                     const FunctionRecord &Function) {
360   SmallBitVector FilenameEquivalence(Function.Filenames.size(), false);
361   for (unsigned I = 0, E = Function.Filenames.size(); I < E; ++I)
362     if (SourceFile == Function.Filenames[I])
363       FilenameEquivalence[I] = true;
364   return FilenameEquivalence;
365 }
366 
findMainViewFileID(StringRef SourceFile,const FunctionRecord & Function)367 static Optional<unsigned> findMainViewFileID(StringRef SourceFile,
368                                              const FunctionRecord &Function) {
369   SmallBitVector IsNotExpandedFile(Function.Filenames.size(), true);
370   SmallBitVector FilenameEquivalence = gatherFileIDs(SourceFile, Function);
371   for (const auto &CR : Function.CountedRegions)
372     if (CR.Kind == CounterMappingRegion::ExpansionRegion &&
373         FilenameEquivalence[CR.FileID])
374       IsNotExpandedFile[CR.ExpandedFileID] = false;
375   IsNotExpandedFile &= FilenameEquivalence;
376   int I = IsNotExpandedFile.find_first();
377   if (I == -1)
378     return None;
379   return I;
380 }
381 
findMainViewFileID(const FunctionRecord & Function)382 static Optional<unsigned> findMainViewFileID(const FunctionRecord &Function) {
383   SmallBitVector IsNotExpandedFile(Function.Filenames.size(), true);
384   for (const auto &CR : Function.CountedRegions)
385     if (CR.Kind == CounterMappingRegion::ExpansionRegion)
386       IsNotExpandedFile[CR.ExpandedFileID] = false;
387   int I = IsNotExpandedFile.find_first();
388   if (I == -1)
389     return None;
390   return I;
391 }
392 
393 /// Sort a nested sequence of regions from a single file.
sortNestedRegions(It First,It Last)394 template <class It> static void sortNestedRegions(It First, It Last) {
395   std::sort(First, Last,
396             [](const CountedRegion &LHS, const CountedRegion &RHS) {
397     if (LHS.startLoc() == RHS.startLoc())
398       // When LHS completely contains RHS, we sort LHS first.
399       return RHS.endLoc() < LHS.endLoc();
400     return LHS.startLoc() < RHS.startLoc();
401   });
402 }
403 
isExpansion(const CountedRegion & R,unsigned FileID)404 static bool isExpansion(const CountedRegion &R, unsigned FileID) {
405   return R.Kind == CounterMappingRegion::ExpansionRegion && R.FileID == FileID;
406 }
407 
getCoverageForFile(StringRef Filename)408 CoverageData CoverageMapping::getCoverageForFile(StringRef Filename) {
409   CoverageData FileCoverage(Filename);
410   std::vector<coverage::CountedRegion> Regions;
411 
412   for (const auto &Function : Functions) {
413     auto MainFileID = findMainViewFileID(Filename, Function);
414     if (!MainFileID)
415       continue;
416     auto FileIDs = gatherFileIDs(Filename, Function);
417     for (const auto &CR : Function.CountedRegions)
418       if (FileIDs.test(CR.FileID)) {
419         Regions.push_back(CR);
420         if (isExpansion(CR, *MainFileID))
421           FileCoverage.Expansions.emplace_back(CR, Function);
422       }
423   }
424 
425   sortNestedRegions(Regions.begin(), Regions.end());
426   DEBUG(dbgs() << "Emitting segments for file: " << Filename << "\n");
427   FileCoverage.Segments = SegmentBuilder().buildSegments(Regions);
428 
429   return FileCoverage;
430 }
431 
432 std::vector<const FunctionRecord *>
getInstantiations(StringRef Filename)433 CoverageMapping::getInstantiations(StringRef Filename) {
434   FunctionInstantiationSetCollector InstantiationSetCollector;
435   for (const auto &Function : Functions) {
436     auto MainFileID = findMainViewFileID(Filename, Function);
437     if (!MainFileID)
438       continue;
439     InstantiationSetCollector.insert(Function, *MainFileID);
440   }
441 
442   std::vector<const FunctionRecord *> Result;
443   for (const auto &InstantiationSet : InstantiationSetCollector) {
444     if (InstantiationSet.second.size() < 2)
445       continue;
446     Result.insert(Result.end(), InstantiationSet.second.begin(),
447                   InstantiationSet.second.end());
448   }
449   return Result;
450 }
451 
452 CoverageData
getCoverageForFunction(const FunctionRecord & Function)453 CoverageMapping::getCoverageForFunction(const FunctionRecord &Function) {
454   auto MainFileID = findMainViewFileID(Function);
455   if (!MainFileID)
456     return CoverageData();
457 
458   CoverageData FunctionCoverage(Function.Filenames[*MainFileID]);
459   std::vector<coverage::CountedRegion> Regions;
460   for (const auto &CR : Function.CountedRegions)
461     if (CR.FileID == *MainFileID) {
462       Regions.push_back(CR);
463       if (isExpansion(CR, *MainFileID))
464         FunctionCoverage.Expansions.emplace_back(CR, Function);
465     }
466 
467   sortNestedRegions(Regions.begin(), Regions.end());
468   DEBUG(dbgs() << "Emitting segments for function: " << Function.Name << "\n");
469   FunctionCoverage.Segments = SegmentBuilder().buildSegments(Regions);
470 
471   return FunctionCoverage;
472 }
473 
474 CoverageData
getCoverageForExpansion(const ExpansionRecord & Expansion)475 CoverageMapping::getCoverageForExpansion(const ExpansionRecord &Expansion) {
476   CoverageData ExpansionCoverage(
477       Expansion.Function.Filenames[Expansion.FileID]);
478   std::vector<coverage::CountedRegion> Regions;
479   for (const auto &CR : Expansion.Function.CountedRegions)
480     if (CR.FileID == Expansion.FileID) {
481       Regions.push_back(CR);
482       if (isExpansion(CR, Expansion.FileID))
483         ExpansionCoverage.Expansions.emplace_back(CR, Expansion.Function);
484     }
485 
486   sortNestedRegions(Regions.begin(), Regions.end());
487   DEBUG(dbgs() << "Emitting segments for expansion of file " << Expansion.FileID
488                << "\n");
489   ExpansionCoverage.Segments = SegmentBuilder().buildSegments(Regions);
490 
491   return ExpansionCoverage;
492 }
493 
494 namespace {
495 class CoverageMappingErrorCategoryType : public std::error_category {
name() const496   const char *name() const LLVM_NOEXCEPT override { return "llvm.coveragemap"; }
message(int IE) const497   std::string message(int IE) const override {
498     auto E = static_cast<coveragemap_error>(IE);
499     switch (E) {
500     case coveragemap_error::success:
501       return "Success";
502     case coveragemap_error::eof:
503       return "End of File";
504     case coveragemap_error::no_data_found:
505       return "No coverage data found";
506     case coveragemap_error::unsupported_version:
507       return "Unsupported coverage format version";
508     case coveragemap_error::truncated:
509       return "Truncated coverage data";
510     case coveragemap_error::malformed:
511       return "Malformed coverage data";
512     }
513     llvm_unreachable("A value of coveragemap_error has no message.");
514   }
515 };
516 }
517 
518 static ManagedStatic<CoverageMappingErrorCategoryType> ErrorCategory;
519 
coveragemap_category()520 const std::error_category &llvm::coveragemap_category() {
521   return *ErrorCategory;
522 }
523