[BOLT][NFC] Disambiguate sample as basic sample

bolt/include/bolt/Core/BinaryFunction.h

@@ -143,7 +143,7 @@ class BinaryFunction {
   enum {
     PF_NONE = 0,     /// No profile.
     PF_LBR = 1,      /// Profile is based on last branch records.
-    PF_SAMPLE = 2,   /// Non-LBR sample-based profile.
+    PF_IP = 2,       /// Non-LBR sample-based profile.
     PF_MEMEVENT = 4, /// Profile has mem events.
   };
 
@@ -392,7 +392,7 @@ class BinaryFunction {
   float ProfileMatchRatio{0.0f};
 
   /// Raw branch count for this function in the profile.
-  uint64_t RawBranchCount{0};
+  uint64_t RawSampleCount{0};
 
   /// Dynamically executed function bytes, used for density computation.
   uint64_t SampleCountInBytes{0};
@@ -1893,11 +1893,11 @@ class BinaryFunction {
 
   /// Return the raw profile information about the number of branch
   /// executions corresponding to this function.
-  uint64_t getRawBranchCount() const { return RawBranchCount; }
+  uint64_t getRawSampleCount() const { return RawSampleCount; }
 
   /// Set the profile data about the number of branch executions corresponding
   /// to this function.
-  void setRawBranchCount(uint64_t Count) { RawBranchCount = Count; }
+  void setRawSampleCount(uint64_t Count) { RawSampleCount = Count; }
 
   /// Return the number of dynamically executed bytes, from raw perf data.
   uint64_t getSampleCountInBytes() const { return SampleCountInBytes; }

bolt/include/bolt/Profile/DataAggregator.h

@@ -257,7 +257,8 @@ class DataAggregator : public DataReader {
 
   /// Semantic actions - parser hooks to interpret parsed perf samples
   /// Register a sample (non-LBR mode), i.e. a new hit at \p Address
-  bool doSample(BinaryFunction &Func, const uint64_t Address, uint64_t Count);
+  bool doBasicSample(BinaryFunction &Func, const uint64_t Address,
+                     uint64_t Count);
 
   /// Register an intraprocedural branch \p Branch.
   bool doIntraBranch(BinaryFunction &Func, uint64_t From, uint64_t To,

bolt/include/bolt/Profile/DataReader.h

@@ -212,15 +212,16 @@ struct FuncMemData {
 /// Similar to BranchInfo, but instead of recording from-to address (an edge),
 /// it records the address of a perf event and the number of times samples hit
 /// this address.
-struct SampleInfo {
+struct BasicSampleInfo {
   Location Loc;
   int64_t Hits;
 
-  SampleInfo(Location Loc, int64_t Hits) : Loc(std::move(Loc)), Hits(Hits) {}
+  BasicSampleInfo(Location Loc, int64_t Hits)
+      : Loc(std::move(Loc)), Hits(Hits) {}
 
-  bool operator==(const SampleInfo &RHS) const { return Loc == RHS.Loc; }
+  bool operator==(const BasicSampleInfo &RHS) const { return Loc == RHS.Loc; }
 
-  bool operator<(const SampleInfo &RHS) const {
+  bool operator<(const BasicSampleInfo &RHS) const {
     if (Loc < RHS.Loc)
       return true;
 
@@ -229,23 +230,26 @@ struct SampleInfo {
 
   void print(raw_ostream &OS) const;
 
-  void mergeWith(const SampleInfo &SI);
+  void mergeWith(const BasicSampleInfo &SI);
 };
 
 /// Helper class to store samples recorded in the address space of a given
 /// function, analogous to FuncBranchData but for samples instead of branches.
-struct FuncSampleData {
-  typedef std::vector<SampleInfo> ContainerTy;
+struct FuncBasicSampleData {
+  typedef std::vector<BasicSampleInfo> ContainerTy;
 
   StringRef Name;
   ContainerTy Data;
 
-  FuncSampleData(StringRef Name, ContainerTy Data)
+  FuncBasicSampleData(StringRef Name, ContainerTy Data)
       : Name(Name), Data(std::move(Data)) {}
 
   /// Get the number of samples recorded in [Start, End)
   uint64_t getSamples(uint64_t Start, uint64_t End) const;
 
+  /// Returns the total number of samples recorded in this function.
+  uint64_t getSamples() const;
+
   /// Aggregation helper
   DenseMap<uint64_t, size_t> Index;
 
@@ -305,7 +309,7 @@ class DataReader : public ProfileReaderBase {
   /// The last step is to infer edge counts based on BB execution count. Note
   /// this is the opposite of the LBR way, where we infer BB execution count
   /// based on edge counts.
-  void readSampleData(BinaryFunction &BF);
+  void readBasicSampleData(BinaryFunction &BF);
 
   /// Convert function-level branch data into instruction annotations.
   void convertBranchData(BinaryFunction &BF) const;
@@ -379,7 +383,8 @@ class DataReader : public ProfileReaderBase {
   /// Return mem data matching one of the names in \p FuncNames.
   FuncMemData *getMemDataForNames(const std::vector<StringRef> &FuncNames);
 
-  FuncSampleData *getFuncSampleData(const std::vector<StringRef> &FuncNames);
+  FuncBasicSampleData *
+  getFuncBasicSampleData(const std::vector<StringRef> &FuncNames);
 
   /// Return a vector of all FuncBranchData matching the list of names.
   /// Internally use fuzzy matching to match special names like LTO-generated
@@ -422,7 +427,7 @@ class DataReader : public ProfileReaderBase {
   }
 
   using NamesToBranchesMapTy = std::map<StringRef, FuncBranchData>;
-  using NamesToSamplesMapTy = std::map<StringRef, FuncSampleData>;
+  using NamesToBasicSamplesMapTy = std::map<StringRef, FuncBasicSampleData>;
   using NamesToMemEventsMapTy = std::map<StringRef, FuncMemData>;
   using FuncsToBranchesMapTy =
       std::unordered_map<const BinaryFunction *, FuncBranchData *>;
@@ -471,7 +476,7 @@ class DataReader : public ProfileReaderBase {
     return parseLocation(EndChar, EndNl, true);
   }
   ErrorOr<BranchInfo> parseBranchInfo();
-  ErrorOr<SampleInfo> parseSampleInfo();
+  ErrorOr<BasicSampleInfo> parseSampleInfo();
   ErrorOr<MemInfo> parseMemInfo();
   ErrorOr<bool> maybeParseNoLBRFlag();
   ErrorOr<bool> maybeParseBATFlag();
@@ -485,7 +490,7 @@ class DataReader : public ProfileReaderBase {
   unsigned Line{0};
   unsigned Col{0};
   NamesToBranchesMapTy NamesToBranches;
-  NamesToSamplesMapTy NamesToSamples;
+  NamesToBasicSamplesMapTy NamesToBasicSamples;
   NamesToMemEventsMapTy NamesToMemEvents;
   FuncsToBranchesMapTy FuncsToBranches;
   FuncsToMemDataMapTy FuncsToMemData;

bolt/include/bolt/Profile/ProfileYAMLMapping.h

@@ -231,7 +231,7 @@ LLVM_YAML_STRONG_TYPEDEF(uint16_t, PROFILE_PF)
 template <> struct ScalarBitSetTraits<PROFILE_PF> {
   static void bitset(IO &io, PROFILE_PF &value) {
     io.bitSetCase(value, "lbr", BinaryFunction::PF_LBR);
-    io.bitSetCase(value, "sample", BinaryFunction::PF_SAMPLE);
+    io.bitSetCase(value, "sample", BinaryFunction::PF_IP);
     io.bitSetCase(value, "memevent", BinaryFunction::PF_MEMEVENT);
   }
 };

bolt/lib/Core/BinaryFunction.cpp

@@ -473,7 +473,7 @@ void BinaryFunction::print(raw_ostream &OS, std::string Annotation) {
     OS << "\n  Image       : 0x" << Twine::utohexstr(getImageAddress());
   if (ExecutionCount != COUNT_NO_PROFILE) {
     OS << "\n  Exec Count  : " << ExecutionCount;
-    OS << "\n  Branch Count: " << RawBranchCount;
+    OS << "\n  Sample Count: " << RawSampleCount;
     OS << "\n  Profile Acc : " << format("%.1f%%", ProfileMatchRatio * 100.0f);
   }
 

bolt/lib/Passes/BinaryPasses.cpp

@@ -1445,7 +1445,7 @@ Error PrintProgramStats::runOnFunctions(BinaryContext &BC) {
     if (!Function.hasProfile())
       continue;
 
-    uint64_t SampleCount = Function.getRawBranchCount();
+    uint64_t SampleCount = Function.getRawSampleCount();
     TotalSampleCount += SampleCount;
 
     if (Function.hasValidProfile()) {

bolt/lib/Passes/MCF.cpp

@@ -458,15 +458,15 @@ void EstimateEdgeCounts::runOnFunction(BinaryFunction &BF) {
 Error EstimateEdgeCounts::runOnFunctions(BinaryContext &BC) {
   if (llvm::none_of(llvm::make_second_range(BC.getBinaryFunctions()),
                     [](const BinaryFunction &BF) {
-                      return BF.getProfileFlags() == BinaryFunction::PF_SAMPLE;
+                      return BF.getProfileFlags() == BinaryFunction::PF_IP;
                     }))
     return Error::success();
 
   ParallelUtilities::WorkFuncTy WorkFun = [&](BinaryFunction &BF) {
     runOnFunction(BF);
   };
   ParallelUtilities::PredicateTy SkipFunc = [&](const BinaryFunction &BF) {
-    return BF.getProfileFlags() != BinaryFunction::PF_SAMPLE;
+    return BF.getProfileFlags() != BinaryFunction::PF_IP;
   };
 
   ParallelUtilities::runOnEachFunction(

bolt/lib/Profile/DataAggregator.cpp

@@ -565,15 +565,15 @@ void DataAggregator::processProfile(BinaryContext &BC) {
   processMemEvents();
 
   // Mark all functions with registered events as having a valid profile.
-  const auto Flags = opts::BasicAggregation ? BinaryFunction::PF_SAMPLE
-                                            : BinaryFunction::PF_LBR;
   for (auto &BFI : BC.getBinaryFunctions()) {
     BinaryFunction &BF = BFI.second;
-    FuncBranchData *FBD = getBranchData(BF);
-    if (FBD || getFuncSampleData(BF.getNames())) {
-      BF.markProfiled(Flags);
-      if (FBD)
-        BF.RawBranchCount = FBD->getNumExecutedBranches();
+    if (FuncBranchData *FBD = getBranchData(BF)) {
+      BF.markProfiled(BinaryFunction::PF_LBR);
+      BF.RawSampleCount = FBD->getNumExecutedBranches();
+    } else if (FuncBasicSampleData *FSD =
+                   getFuncBasicSampleData(BF.getNames())) {
+      BF.markProfiled(BinaryFunction::PF_IP);
+      BF.RawSampleCount = FSD->getSamples();
     }
   }
 
@@ -628,20 +628,28 @@ StringRef DataAggregator::getLocationName(const BinaryFunction &Func,
   return OrigFunc->getOneName();
 }
 
-bool DataAggregator::doSample(BinaryFunction &OrigFunc, uint64_t Address,
-                              uint64_t Count) {
+bool DataAggregator::doBasicSample(BinaryFunction &OrigFunc, uint64_t Address,
+                                   uint64_t Count) {
+  // To record executed bytes, use basic block size as is regardless of BAT.
+  uint64_t BlockSize = 0;
+  if (BinaryBasicBlock *BB = OrigFunc.getBasicBlockContainingOffset(
+          Address - OrigFunc.getAddress()))
+    BlockSize = BB->getOriginalSize();
+
   BinaryFunction *ParentFunc = getBATParentFunction(OrigFunc);
   BinaryFunction &Func = ParentFunc ? *ParentFunc : OrigFunc;
-  if (ParentFunc || (BAT && !BAT->isBATFunction(OrigFunc.getAddress())))
+  if (ParentFunc || (BAT && !BAT->isBATFunction(Func.getAddress())))
     NumColdSamples += Count;
+  // Attach executed bytes to parent function in case of cold fragment.
+  Func.SampleCountInBytes += Count * BlockSize;
 
-  auto I = NamesToSamples.find(Func.getOneName());
-  if (I == NamesToSamples.end()) {
+  auto I = NamesToBasicSamples.find(Func.getOneName());
+  if (I == NamesToBasicSamples.end()) {
     bool Success;
     StringRef LocName = getLocationName(Func, BAT);
-    std::tie(I, Success) = NamesToSamples.insert(
-        std::make_pair(Func.getOneName(),
-                       FuncSampleData(LocName, FuncSampleData::ContainerTy())));
+    std::tie(I, Success) = NamesToBasicSamples.insert(std::make_pair(
+        Func.getOneName(),
+        FuncBasicSampleData(LocName, FuncBasicSampleData::ContainerTy())));
   }
 
   Address -= Func.getAddress();
@@ -1654,7 +1662,7 @@ void DataAggregator::processBasicEvents() {
       continue;
     }
 
-    doSample(*Func, PC, HitCount);
+    doBasicSample(*Func, PC, HitCount);
   }
   outs() << "PERF2BOLT: read " << NumTotalSamples << " samples\n";
 
@@ -2185,9 +2193,9 @@ DataAggregator::writeAggregatedFile(StringRef OutputFilename) const {
       OutFile << " " << Entry.getKey();
     OutFile << "\n";
 
-    for (const auto &KV : NamesToSamples) {
-      const FuncSampleData &FSD = KV.second;
-      for (const SampleInfo &SI : FSD.Data) {
+    for (const auto &KV : NamesToBasicSamples) {
+      const FuncBasicSampleData &FSD = KV.second;
+      for (const BasicSampleInfo &SI : FSD.Data) {
         writeLocation(SI.Loc);
         OutFile << SI.Hits << "\n";
         ++BranchValues;
@@ -2260,8 +2268,8 @@ std::error_code DataAggregator::writeBATYAML(BinaryContext &BC,
   for (const StringMapEntry<std::nullopt_t> &EventEntry : EventNames)
     EventNamesOS << LS << EventEntry.first().str();
 
-  BP.Header.Flags = opts::BasicAggregation ? BinaryFunction::PF_SAMPLE
-                                           : BinaryFunction::PF_LBR;
+  BP.Header.Flags =
+      opts::BasicAggregation ? BinaryFunction::PF_IP : BinaryFunction::PF_LBR;
 
   // Add probe inline tree nodes.
   YAMLProfileWriter::InlineTreeDesc InlineTree;