xref: /llvm-project/llvm/lib/ProfileData/Coverage/CoverageMappingWriter.cpp (revision ea61750c35a11140e9245bd9cbeb383c37f6e031)
1 //===- CoverageMappingWriter.cpp - Code coverage mapping writer -----------===//
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 writing coverage mapping data for
10 // instrumentation based coverage.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/ProfileData/InstrProf.h"
15 #include "llvm/ProfileData/Coverage/CoverageMappingWriter.h"
16 #include "llvm/ADT/ArrayRef.h"
17 #include "llvm/ADT/SmallVector.h"
18 #include "llvm/Support/Compression.h"
19 #include "llvm/Support/LEB128.h"
20 #include "llvm/Support/raw_ostream.h"
21 #include <algorithm>
22 #include <cassert>
23 #include <limits>
24 #include <vector>
25 
26 using namespace llvm;
27 using namespace coverage;
28 
29 CoverageFilenamesSectionWriter::CoverageFilenamesSectionWriter(
30     ArrayRef<std::string> Filenames)
31     : Filenames(Filenames) {
32 #ifndef NDEBUG
33   StringSet<> NameSet;
34   for (StringRef Name : Filenames)
35     assert(NameSet.insert(Name).second && "Duplicate filename");
36 #endif
37 }
38 
39 void CoverageFilenamesSectionWriter::write(raw_ostream &OS, bool Compress) {
40   std::string FilenamesStr;
41   {
42     raw_string_ostream FilenamesOS{FilenamesStr};
43     for (const auto &Filename : Filenames) {
44       encodeULEB128(Filename.size(), FilenamesOS);
45       FilenamesOS << Filename;
46     }
47   }
48 
49   SmallString<128> CompressedStr;
50   bool doCompression = Compress && compression::zlib::isAvailable() &&
51                        DoInstrProfNameCompression;
52   if (doCompression)
53     compression::zlib::compress(FilenamesStr, CompressedStr,
54                                 compression::zlib::BestSizeCompression);
55 
56   // ::= <num-filenames>
57   //     <uncompressed-len>
58   //     <compressed-len-or-zero>
59   //     (<compressed-filenames> | <uncompressed-filenames>)
60   encodeULEB128(Filenames.size(), OS);
61   encodeULEB128(FilenamesStr.size(), OS);
62   encodeULEB128(doCompression ? CompressedStr.size() : 0U, OS);
63   OS << (doCompression ? CompressedStr.str() : StringRef(FilenamesStr));
64 }
65 
66 namespace {
67 
68 /// Gather only the expressions that are used by the mapping
69 /// regions in this function.
70 class CounterExpressionsMinimizer {
71   ArrayRef<CounterExpression> Expressions;
72   SmallVector<CounterExpression, 16> UsedExpressions;
73   std::vector<unsigned> AdjustedExpressionIDs;
74 
75 public:
76   CounterExpressionsMinimizer(ArrayRef<CounterExpression> Expressions,
77                               ArrayRef<CounterMappingRegion> MappingRegions)
78       : Expressions(Expressions) {
79     AdjustedExpressionIDs.resize(Expressions.size(), 0);
80     for (const auto &I : MappingRegions) {
81       mark(I.Count);
82       mark(I.FalseCount);
83     }
84     for (const auto &I : MappingRegions) {
85       gatherUsed(I.Count);
86       gatherUsed(I.FalseCount);
87     }
88   }
89 
90   void mark(Counter C) {
91     if (!C.isExpression())
92       return;
93     unsigned ID = C.getExpressionID();
94     AdjustedExpressionIDs[ID] = 1;
95     mark(Expressions[ID].LHS);
96     mark(Expressions[ID].RHS);
97   }
98 
99   void gatherUsed(Counter C) {
100     if (!C.isExpression() || !AdjustedExpressionIDs[C.getExpressionID()])
101       return;
102     AdjustedExpressionIDs[C.getExpressionID()] = UsedExpressions.size();
103     const auto &E = Expressions[C.getExpressionID()];
104     UsedExpressions.push_back(E);
105     gatherUsed(E.LHS);
106     gatherUsed(E.RHS);
107   }
108 
109   ArrayRef<CounterExpression> getExpressions() const { return UsedExpressions; }
110 
111   /// Adjust the given counter to correctly transition from the old
112   /// expression ids to the new expression ids.
113   Counter adjust(Counter C) const {
114     if (C.isExpression())
115       C = Counter::getExpression(AdjustedExpressionIDs[C.getExpressionID()]);
116     return C;
117   }
118 };
119 
120 } // end anonymous namespace
121 
122 /// Encode the counter.
123 ///
124 /// The encoding uses the following format:
125 /// Low 2 bits - Tag:
126 ///   Counter::Zero(0) - A Counter with kind Counter::Zero
127 ///   Counter::CounterValueReference(1) - A counter with kind
128 ///     Counter::CounterValueReference
129 ///   Counter::Expression(2) + CounterExpression::Subtract(0) -
130 ///     A counter with kind Counter::Expression and an expression
131 ///     with kind CounterExpression::Subtract
132 ///   Counter::Expression(2) + CounterExpression::Add(1) -
133 ///     A counter with kind Counter::Expression and an expression
134 ///     with kind CounterExpression::Add
135 /// Remaining bits - Counter/Expression ID.
136 static unsigned encodeCounter(ArrayRef<CounterExpression> Expressions,
137                               Counter C) {
138   unsigned Tag = unsigned(C.getKind());
139   if (C.isExpression())
140     Tag += Expressions[C.getExpressionID()].Kind;
141   unsigned ID = C.getCounterID();
142   assert(ID <=
143          (std::numeric_limits<unsigned>::max() >> Counter::EncodingTagBits));
144   return Tag | (ID << Counter::EncodingTagBits);
145 }
146 
147 static void writeCounter(ArrayRef<CounterExpression> Expressions, Counter C,
148                          raw_ostream &OS) {
149   encodeULEB128(encodeCounter(Expressions, C), OS);
150 }
151 
152 void CoverageMappingWriter::write(raw_ostream &OS) {
153   // Check that we don't have any bogus regions.
154   assert(all_of(MappingRegions,
155                 [](const CounterMappingRegion &CMR) {
156                   return CMR.startLoc() <= CMR.endLoc();
157                 }) &&
158          "Source region does not begin before it ends");
159 
160   // Sort the regions in an ascending order by the file id and the starting
161   // location. Sort by region kinds to ensure stable order for tests.
162   llvm::stable_sort(MappingRegions, [](const CounterMappingRegion &LHS,
163                                        const CounterMappingRegion &RHS) {
164     if (LHS.FileID != RHS.FileID)
165       return LHS.FileID < RHS.FileID;
166     if (LHS.startLoc() != RHS.startLoc())
167       return LHS.startLoc() < RHS.startLoc();
168     return LHS.Kind < RHS.Kind;
169   });
170 
171   // Write out the fileid -> filename mapping.
172   encodeULEB128(VirtualFileMapping.size(), OS);
173   for (const auto &FileID : VirtualFileMapping)
174     encodeULEB128(FileID, OS);
175 
176   // Write out the expressions.
177   CounterExpressionsMinimizer Minimizer(Expressions, MappingRegions);
178   auto MinExpressions = Minimizer.getExpressions();
179   encodeULEB128(MinExpressions.size(), OS);
180   for (const auto &E : MinExpressions) {
181     writeCounter(MinExpressions, Minimizer.adjust(E.LHS), OS);
182     writeCounter(MinExpressions, Minimizer.adjust(E.RHS), OS);
183   }
184 
185   // Write out the mapping regions.
186   // Split the regions into subarrays where each region in a
187   // subarray has a fileID which is the index of that subarray.
188   unsigned PrevLineStart = 0;
189   unsigned CurrentFileID = ~0U;
190   for (auto I = MappingRegions.begin(), E = MappingRegions.end(); I != E; ++I) {
191     if (I->FileID != CurrentFileID) {
192       // Ensure that all file ids have at least one mapping region.
193       assert(I->FileID == (CurrentFileID + 1));
194       // Find the number of regions with this file id.
195       unsigned RegionCount = 1;
196       for (auto J = I + 1; J != E && I->FileID == J->FileID; ++J)
197         ++RegionCount;
198       // Start a new region sub-array.
199       encodeULEB128(RegionCount, OS);
200 
201       CurrentFileID = I->FileID;
202       PrevLineStart = 0;
203     }
204     Counter Count = Minimizer.adjust(I->Count);
205     Counter FalseCount = Minimizer.adjust(I->FalseCount);
206     switch (I->Kind) {
207     case CounterMappingRegion::CodeRegion:
208     case CounterMappingRegion::GapRegion:
209       writeCounter(MinExpressions, Count, OS);
210       break;
211     case CounterMappingRegion::ExpansionRegion: {
212       assert(Count.isZero());
213       assert(I->ExpandedFileID <=
214              (std::numeric_limits<unsigned>::max() >>
215               Counter::EncodingCounterTagAndExpansionRegionTagBits));
216       // Mark an expansion region with a set bit that follows the counter tag,
217       // and pack the expanded file id into the remaining bits.
218       unsigned EncodedTagExpandedFileID =
219           (1 << Counter::EncodingTagBits) |
220           (I->ExpandedFileID
221            << Counter::EncodingCounterTagAndExpansionRegionTagBits);
222       encodeULEB128(EncodedTagExpandedFileID, OS);
223       break;
224     }
225     case CounterMappingRegion::SkippedRegion:
226       assert(Count.isZero());
227       encodeULEB128(unsigned(I->Kind)
228                         << Counter::EncodingCounterTagAndExpansionRegionTagBits,
229                     OS);
230       break;
231     case CounterMappingRegion::BranchRegion:
232       encodeULEB128(unsigned(I->Kind)
233                         << Counter::EncodingCounterTagAndExpansionRegionTagBits,
234                     OS);
235       writeCounter(MinExpressions, Count, OS);
236       writeCounter(MinExpressions, FalseCount, OS);
237       break;
238     }
239     assert(I->LineStart >= PrevLineStart);
240     encodeULEB128(I->LineStart - PrevLineStart, OS);
241     encodeULEB128(I->ColumnStart, OS);
242     assert(I->LineEnd >= I->LineStart);
243     encodeULEB128(I->LineEnd - I->LineStart, OS);
244     encodeULEB128(I->ColumnEnd, OS);
245     PrevLineStart = I->LineStart;
246   }
247   // Ensure that all file ids have at least one mapping region.
248   assert(CurrentFileID == (VirtualFileMapping.size() - 1));
249 }
250