[Coverage] Store compilation dir separately in coverage mapping

We currently always store absolute filenames in coverage mapping.  This
is problematic for several reasons. It poses a problem for distributed
compilation as source location might vary across machines.  We are also
duplicating the path prefix potentially wasting space.

This change modifies how we store filenames in coverage mapping. Rather
than absolute paths, it stores the compilation directory and file paths
as given to the compiler, either relative or absolute. Later when
reading the coverage mapping information, we recombine relative paths
with the working directory. This approach is similar to handling
ofDW_AT_comp_dir in DWARF.

Finally, we also provide a new option, -fprofile-compilation-dir akin
to -fdebug-compilation-dir which can be used to manually override the
compilation directory which is useful in distributed compilation cases.

Differential Revision: https://reviews.llvm.org/D95753

clang/include/clang/Basic/CodeGenOptions.h

@@ -172,6 +172,9 @@ class CodeGenOptions : public CodeGenOptionsBase {
   /// The string to embed in debug information as the current working directory.
   std::string DebugCompilationDir;
 
+  /// The string to embed in coverage mapping as the current working directory.
+  std::string ProfileCompilationDir;
+
   /// The string to embed in the debug information for the compile unit, if
   /// non-empty.
   std::string DwarfDebugFlags;

clang/include/clang/Driver/Options.td

@@ -1103,6 +1103,10 @@ def fdebug_compilation_dir_EQ : Joined<["-"], "fdebug-compilation-dir=">,
 def fdebug_compilation_dir : Separate<["-"], "fdebug-compilation-dir">,
     Group<f_Group>, Flags<[CC1Option, CC1AsOption, CoreOption]>,
     Alias<fdebug_compilation_dir_EQ>;
+def fprofile_compilation_dir_EQ : Joined<["-"], "fprofile-compilation-dir=">,
+    Group<f_Group>, Flags<[CC1Option, CC1AsOption, CoreOption]>,
+    HelpText<"The compilation directory to embed in the coverage mapping.">,
+    MarshallingInfoString<CodeGenOpts<"ProfileCompilationDir">>;
 defm debug_info_for_profiling : BoolFOption<"debug-info-for-profiling",
   CodeGenOpts<"DebugInfoForProfiling">, DefaultFalse,
   PosFlag<SetTrue, [CC1Option], "Emit extra debug info to make sample profile more accurate">,

clang/lib/CodeGen/CoverageMappingGen.cpp

@@ -1606,9 +1606,17 @@ CoverageMappingModuleGen::CoverageMappingModuleGen(
   ProfilePrefixMap = CGM.getCodeGenOpts().ProfilePrefixMap;
 }
 
+std::string CoverageMappingModuleGen::getCurrentDirname() {
+  if (!CGM.getCodeGenOpts().ProfileCompilationDir.empty())
+    return CGM.getCodeGenOpts().ProfileCompilationDir;
+
+  SmallString<256> CWD;
+  llvm::sys::fs::current_path(CWD);
+  return CWD.str().str();
+}
+
 std::string CoverageMappingModuleGen::normalizeFilename(StringRef Filename) {
   llvm::SmallString<256> Path(Filename);
-  llvm::sys::fs::make_absolute(Path);
   llvm::sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
   for (const auto &Entry : ProfilePrefixMap) {
     if (llvm::sys::path::replace_path_prefix(Path, Entry.first, Entry.second))
@@ -1689,18 +1697,17 @@ void CoverageMappingModuleGen::addFunctionMappingRecord(
     // also processed by the CoverageMappingWriter which performs
     // additional minimization operations such as reducing the number of
     // expressions.
+    llvm::SmallVector<std::string, 16> FilenameStrs;
     std::vector<StringRef> Filenames;
     std::vector<CounterExpression> Expressions;
     std::vector<CounterMappingRegion> Regions;
-    llvm::SmallVector<std::string, 16> FilenameStrs;
-    llvm::SmallVector<StringRef, 16> FilenameRefs;
-    FilenameStrs.resize(FileEntries.size());
-    FilenameRefs.resize(FileEntries.size());
+    FilenameStrs.resize(FileEntries.size() + 1);
+    FilenameStrs[0] = normalizeFilename(getCurrentDirname());
     for (const auto &Entry : FileEntries) {
       auto I = Entry.second;
       FilenameStrs[I] = normalizeFilename(Entry.first->getName());
-      FilenameRefs[I] = FilenameStrs[I];
     }
+    ArrayRef<std::string> FilenameRefs = llvm::makeArrayRef(FilenameStrs);
     RawCoverageMappingReader Reader(CoverageMapping, FilenameRefs, Filenames,
                                     Expressions, Regions);
     if (Reader.read())
@@ -1717,19 +1724,18 @@ void CoverageMappingModuleGen::emit() {
 
   // Create the filenames and merge them with coverage mappings
   llvm::SmallVector<std::string, 16> FilenameStrs;
-  llvm::SmallVector<StringRef, 16> FilenameRefs;
-  FilenameStrs.resize(FileEntries.size());
-  FilenameRefs.resize(FileEntries.size());
+  FilenameStrs.resize(FileEntries.size() + 1);
+  // The first filename is the current working directory.
+  FilenameStrs[0] = getCurrentDirname();
   for (const auto &Entry : FileEntries) {
     auto I = Entry.second;
     FilenameStrs[I] = normalizeFilename(Entry.first->getName());
-    FilenameRefs[I] = FilenameStrs[I];
   }
 
   std::string Filenames;
   {
     llvm::raw_string_ostream OS(Filenames);
-    CoverageFilenamesSectionWriter(FilenameRefs).write(OS);
+    CoverageFilenamesSectionWriter(FilenameStrs).write(OS);
   }
   auto *FilenamesVal =
       llvm::ConstantDataArray::getString(Ctx, Filenames, false);
@@ -1787,7 +1793,7 @@ unsigned CoverageMappingModuleGen::getFileID(const FileEntry *File) {
   auto It = FileEntries.find(File);
   if (It != FileEntries.end())
     return It->second;
-  unsigned FileID = FileEntries.size();
+  unsigned FileID = FileEntries.size() + 1;
   FileEntries.insert(std::make_pair(File, FileID));
   return FileID;
 }

clang/lib/CodeGen/CoverageMappingGen.h

@@ -95,6 +95,7 @@ class CoverageMappingModuleGen {
   std::vector<FunctionInfo> FunctionRecords;
   std::map<std::string, std::string> ProfilePrefixMap;
 
+  std::string getCurrentDirname();
   std::string normalizeFilename(StringRef Filename);
 
   /// Emit a function record.

clang/lib/Driver/ToolChains/Clang.cpp

@@ -858,6 +858,13 @@ static void addPGOAndCoverageFlags(const ToolChain &TC, Compilation &C,
     CmdArgs.push_back("-fcoverage-mapping");
   }
 
+  if (Arg *A = Args.getLastArg(options::OPT_fprofile_compilation_dir_EQ)) {
+    A->render(Args, CmdArgs);
+  } else if (llvm::ErrorOr<std::string> CWD =
+                 D.getVFS().getCurrentWorkingDirectory()) {
+    Args.MakeArgString("-fprofile-compilation-dir=" + *CWD);
+  }
+
   if (Args.hasArg(options::OPT_fprofile_exclude_files_EQ)) {
     auto *Arg = Args.getLastArg(options::OPT_fprofile_exclude_files_EQ);
     if (!Args.hasArg(options::OPT_coverage))

clang/test/CodeGen/profile-compilation-dir.c

@@ -0,0 +1,7 @@
+// RUN: mkdir -p %t.dir && cd %t.dir
+// RUN: cp %s rel.c
+// RUN: %clang_cc1 -fprofile-instrument=clang -fprofile-compilation-dir=/nonsense -fcoverage-mapping -emit-llvm -mllvm -enable-name-compression=false rel.c -o - | FileCheck -check-prefix=CHECK-NONSENSE %s
+
+// CHECK-NONSENSE: nonsense
+
+void f() {}

clang/test/CoverageMapping/abspath.cpp

@@ -1,15 +1,19 @@
 // RUN: %clang_cc1 -mllvm -emptyline-comment-coverage=false -fprofile-instrument=clang -fcoverage-mapping -mllvm -enable-name-compression=false -emit-llvm -main-file-name abspath.cpp %S/Inputs/../abspath.cpp -o - | FileCheck -check-prefix=RMDOTS %s
 
-// RMDOTS: @__llvm_coverage_mapping = {{.*}}"\01
+// RMDOTS: @__llvm_coverage_mapping = {{.*}}"\02
 // RMDOTS-NOT: Inputs
 // RMDOTS: "
 
 // RUN: mkdir -p %t/test && cd %t/test
 // RUN: echo "void f1() {}" > f1.c
-// RUN: %clang_cc1 -mllvm -emptyline-comment-coverage=false -fprofile-instrument=clang -fcoverage-mapping -mllvm -enable-name-compression=false -emit-llvm -main-file-name abspath.cpp ../test/f1.c -o - | FileCheck -check-prefix=RELPATH %s
+// RUN: %clang_cc1 -mllvm -emptyline-comment-coverage=false -fprofile-instrument=clang -fcoverage-mapping -mllvm -enable-name-compression=false -emit-llvm -main-file-name abspath.cpp %t/test/f1.c -o - | FileCheck -check-prefix=ABSPATH %s
 
-// RELPATH: @__llvm_coverage_mapping = {{.*}}"\01
-// RELPATH: {{[/\\].*(/|\\\\)test(/|\\\\)f1}}.c
+// RELPATH: @__llvm_coverage_mapping = {{.*}}"\02
+// RELPATH: {{..(/|\\\\)test(/|\\\\)f1}}.c
 // RELPATH: "
 
+// ABSPATH: @__llvm_coverage_mapping = {{.*}}"\02
+// ABSPATH: {{[/\\].*(/|\\\\)test(/|\\\\)f1}}.c
+// ABSPATH: "
+
 void f1() {}

clang/test/Profile/profile-prefix-map.c

@@ -5,10 +5,14 @@
 // RUN: echo "void f1() {}" > %t/root/nested/profile-prefix-map.c
 // RUN: cd %t/root
 
-// RUN: %clang_cc1 -fprofile-instrument=clang -fcoverage-mapping -emit-llvm -mllvm -enable-name-compression=false -main-file-name profile-prefix-map.c nested/profile-prefix-map.c -o - | FileCheck --check-prefix=ABSOLUTE %s
+// RUN: %clang_cc1 -fprofile-instrument=clang -fcoverage-mapping -emit-llvm -mllvm -enable-name-compression=false -main-file-name profile-prefix-map.c %t/root/nested/profile-prefix-map.c -o - | FileCheck --check-prefix=ABSOLUTE %s
 //
-// ABSOLUTE: @__llvm_coverage_mapping = {{.*"\\01.*root.*nested.*profile-prefix-map\.c}}
+// ABSOLUTE: @__llvm_coverage_mapping = {{.*"\\02.*root.*nested.*profile-prefix-map\.c}}
 
-// RUN: %clang_cc1 -fprofile-instrument=clang -fcoverage-mapping -emit-llvm -mllvm -enable-name-compression=false -main-file-name profile-prefix-map.c nested/profile-prefix-map.c -fprofile-prefix-map=%/t/root=. -o - | FileCheck --check-prefix=PROFILE-PREFIX-MAP %s --implicit-check-not=root
+// RUN: %clang_cc1 -fprofile-instrument=clang -fcoverage-mapping -emit-llvm -mllvm -enable-name-compression=false -main-file-name profile-prefix-map.c ../root/nested/profile-prefix-map.c -o - | FileCheck --check-prefix=RELATIVE %s
 //
-// PROFILE-PREFIX-MAP: @__llvm_coverage_mapping = {{.*"\\01[^/]*}}.{{/|\\+}}nested{{.*profile-prefix-map\.c}}
+// RELATIVE: @__llvm_coverage_mapping = {{.*"\\02.*}}..{{/|\\+}}root{{/|\\+}}nested{{.*profile-prefix-map\.c}}
+
+// RUN: %clang_cc1 -fprofile-instrument=clang -fcoverage-mapping -emit-llvm -mllvm -enable-name-compression=false -main-file-name profile-prefix-map.c %t/root/nested/profile-prefix-map.c -fprofile-prefix-map=%/t/root=. -o - | FileCheck --check-prefix=PROFILE-PREFIX-MAP %s
+// RUN: %clang_cc1 -fprofile-instrument=clang -fcoverage-mapping -emit-llvm -mllvm -enable-name-compression=false -main-file-name profile-prefix-map.c ../root/nested/profile-prefix-map.c -fprofile-prefix-map=../root=. -o - | FileCheck --check-prefix=PROFILE-PREFIX-MAP %s
+// PROFILE-PREFIX-MAP: @__llvm_coverage_mapping = {{.*"\\02.*}}.{{/|\\+}}nested{{.*profile-prefix-map\.c}}

compiler-rt/include/profile/InstrProfData.inc

@@ -649,7 +649,7 @@ serializeValueProfDataFrom(ValueProfRecordClosure *Closure,
 /* Indexed profile format version (start from 1). */
 #define INSTR_PROF_INDEX_VERSION 7
 /* Coverage mapping format version (start from 0). */
-#define INSTR_PROF_COVMAP_VERSION 4
+#define INSTR_PROF_COVMAP_VERSION 5
 
 /* Profile version is always of type uint64_t. Reserve the upper 8 bits in the
  * version for other variants of profile. We set the lowest bit of the upper 8

llvm/docs/CoverageMappingFormat.rst

@@ -266,7 +266,16 @@ too deeply).
    [32 x i8] c"..." ; Encoded data (dissected later)
   }, section "__llvm_covmap", align 8
 
-The current version of the format is version 5. There is one difference from version 4:
+The current version of the format is version 6.
+
+There is one difference between versions 6 and 5:
+
+* The first entry in the filename list is the compilation directory. When the
+  filename is relative, the compilation directory is combined with the relative
+  path to get an absolute path. This can reduce size by omitting the duplicate
+  prefix in filenames.
+
+There is one difference between versions 5 and 4:
 
 * The notion of branch region has been introduced along with a corresponding
   region kind.  Branch regions encode two counters, one to track how many

llvm/include/llvm/ProfileData/Coverage/CoverageMapping.h

@@ -996,7 +996,10 @@ enum CovMapVersion {
   Version4 = 3,
   // Branch regions referring to two counters are added
   Version5 = 4,
-  // The current version is Version5.
+  // Compilation directory is stored separately and combined with relative
+  // filenames to produce an absolute file path.
+  Version6 = 5,
+  // The current version is Version6.
   CurrentVersion = INSTR_PROF_COVMAP_VERSION
 };
 

llvm/include/llvm/ProfileData/Coverage/CoverageMappingReader.h

@@ -125,14 +125,14 @@ class RawCoverageMappingDummyChecker : public RawCoverageReader {
 
 /// Reader for the raw coverage mapping data.
 class RawCoverageMappingReader : public RawCoverageReader {
-  ArrayRef<StringRef> TranslationUnitFilenames;
+  ArrayRef<std::string> &TranslationUnitFilenames;
   std::vector<StringRef> &Filenames;
   std::vector<CounterExpression> &Expressions;
   std::vector<CounterMappingRegion> &MappingRegions;
 
 public:
   RawCoverageMappingReader(StringRef MappingData,
-                           ArrayRef<StringRef> TranslationUnitFilenames,
+                           ArrayRef<std::string> &TranslationUnitFilenames,
                            std::vector<StringRef> &Filenames,
                            std::vector<CounterExpression> &Expressions,
                            std::vector<CounterMappingRegion> &MappingRegions)
@@ -174,10 +174,8 @@ class BinaryCoverageReader : public CoverageMappingReader {
           FilenamesBegin(FilenamesBegin), FilenamesSize(FilenamesSize) {}
   };
 
-  using DecompressedData = std::vector<std::unique_ptr<SmallVector<char, 0>>>;
-
 private:
-  std::vector<StringRef> Filenames;
+  std::vector<std::string> Filenames;
   std::vector<ProfileMappingRecord> MappingRecords;
   InstrProfSymtab ProfileNames;
   size_t CurrentRecord = 0;
@@ -190,10 +188,6 @@ class BinaryCoverageReader : public CoverageMappingReader {
   // D69471, which can split up function records into multiple sections on ELF.
   std::string FuncRecords;
 
-  // Used to tie the lifetimes of decompressed strings to the lifetime of this
-  // BinaryCoverageReader instance.
-  DecompressedData Decompressed;
-
   BinaryCoverageReader(std::string &&FuncRecords)
       : FuncRecords(std::move(FuncRecords)) {}
 
@@ -216,20 +210,20 @@ class BinaryCoverageReader : public CoverageMappingReader {
 
 /// Reader for the raw coverage filenames.
 class RawCoverageFilenamesReader : public RawCoverageReader {
-  std::vector<StringRef> &Filenames;
+  std::vector<std::string> &Filenames;
 
   // Read an uncompressed sequence of filenames.
-  Error readUncompressed(uint64_t NumFilenames);
+  Error readUncompressed(CovMapVersion Version, uint64_t NumFilenames);
 
 public:
-  RawCoverageFilenamesReader(StringRef Data, std::vector<StringRef> &Filenames)
+  RawCoverageFilenamesReader(StringRef Data,
+                             std::vector<std::string> &Filenames)
       : RawCoverageReader(Data), Filenames(Filenames) {}
   RawCoverageFilenamesReader(const RawCoverageFilenamesReader &) = delete;
   RawCoverageFilenamesReader &
   operator=(const RawCoverageFilenamesReader &) = delete;
 
-  Error read(CovMapVersion Version,
-             BinaryCoverageReader::DecompressedData &Decompressed);
+  Error read(CovMapVersion Version);
 };
 
 } // end namespace coverage

llvm/include/llvm/ProfileData/Coverage/CoverageMappingWriter.h

@@ -27,10 +27,10 @@ namespace coverage {
 /// Writer of the filenames section for the instrumentation
 /// based code coverage.
 class CoverageFilenamesSectionWriter {
-  ArrayRef<StringRef> Filenames;
+  ArrayRef<std::string> Filenames;
 
 public:
-  CoverageFilenamesSectionWriter(ArrayRef<StringRef> Filenames);
+  CoverageFilenamesSectionWriter(ArrayRef<std::string> Filenames);
 
   /// Write encoded filenames to the given output stream. If \p Compress is
   /// true, attempt to compress the filenames.

llvm/include/llvm/ProfileData/InstrProfData.inc

@@ -649,7 +649,7 @@ serializeValueProfDataFrom(ValueProfRecordClosure *Closure,
 /* Indexed profile format version (start from 1). */
 #define INSTR_PROF_INDEX_VERSION 7
 /* Coverage mapping format version (start from 0). */
-#define INSTR_PROF_COVMAP_VERSION 4
+#define INSTR_PROF_COVMAP_VERSION 5
 
 /* Profile version is always of type uint64_t. Reserve the upper 8 bits in the
  * version for other variants of profile. We set the lowest bit of the upper 8