[𝘀𝗽𝗿] changes to main this commit is based on

Created using spr 1.3.4

[skip ci]

Author: aaupov

bolt/include/bolt/Core/BinaryFunction.h

@@ -388,6 +388,10 @@ class BinaryFunction {
   /// The profile data for the number of times the function was executed.
   uint64_t ExecutionCount{COUNT_NO_PROFILE};
 
+  /// Profile data for the number of times this function was entered from
+  /// external code (DSO, JIT, etc).
+  uint64_t ExternEntryCount{0};
+
   /// Profile match ratio.
   float ProfileMatchRatio{0.0f};
 
@@ -1877,6 +1881,10 @@ class BinaryFunction {
     return *this;
   }
 
+  /// Set the profile data for the number of times the function was entered from
+  /// external code (DSO/JIT).
+  void setExternEntryCount(uint64_t Count) { ExternEntryCount = Count; }
+
   /// Adjust execution count for the function by a given \p Count. The value
   /// \p Count will be subtracted from the current function count.
   ///
@@ -1904,6 +1912,10 @@ class BinaryFunction {
   /// Return COUNT_NO_PROFILE if there's no profile info.
   uint64_t getExecutionCount() const { return ExecutionCount; }
 
+  /// Return the profile information about the number of times the function was
+  /// entered from external code (DSO/JIT).
+  uint64_t getExternEntryCount() const { return ExternEntryCount; }
+
   /// Return the raw profile information about the number of branch
   /// executions corresponding to this function.
   uint64_t getRawSampleCount() const { return RawSampleCount; }

bolt/include/bolt/Profile/BoltAddressTranslation.h

@@ -98,12 +98,13 @@ class BoltAddressTranslation {
                      bool IsBranchSrc) const;
 
   /// Use the map keys containing basic block addresses to infer fall-throughs
-  /// taken in the path started at FirstLBR.To and ending at SecondLBR.From.
+  /// taken in the path starting at \p From and ending at \p To.
   /// Return std::nullopt if trace is invalid or the list of fall-throughs
   /// otherwise.
   std::optional<FallthroughListTy> getFallthroughsInTrace(uint64_t FuncAddress,
                                                           uint64_t From,
-                                                          uint64_t To) const;
+                                                          uint64_t To,
+                                                          bool IsReturn) const;
 
   /// If available, fetch the address of the hot part linked to the cold part
   /// at \p Address. Return 0 otherwise.

bolt/include/bolt/Profile/DataAggregator.h

@@ -78,6 +78,13 @@ class DataAggregator : public DataReader {
   static bool checkPerfDataMagic(StringRef FileName);
 
 private:
+  struct LBREntry {
+    uint64_t From;
+    uint64_t To;
+    bool Mispred;
+  };
+  friend raw_ostream &operator<<(raw_ostream &OS, const LBREntry &);
+
   struct PerfBranchSample {
     SmallVector<LBREntry, 32> LBR;
   };
@@ -92,35 +99,43 @@ class DataAggregator : public DataReader {
     uint64_t Addr;
   };
 
+  /// Container for the unit of branch data.
+  /// Backwards compatible with legacy use for branches and fall-throughs:
+  /// - if \p Branch is FT_ONLY or FT_EXTERNAL_ORIGIN, the trace only contains
+  ///   fall-through data,
+  /// - if \p To is EXTERNAL, the trace only contains branch data.
   struct Trace {
+    static constexpr const uint64_t EXTERNAL = 0ULL;
+    static constexpr const uint64_t FT_ONLY = -1ULL;
+    static constexpr const uint64_t FT_EXTERNAL_ORIGIN = -2ULL;
+
+    uint64_t Branch;
     uint64_t From;
     uint64_t To;
-    Trace(uint64_t From, uint64_t To) : From(From), To(To) {}
     bool operator==(const Trace &Other) const {
-      return From == Other.From && To == Other.To;
+      return Branch == Other.Branch && From == Other.From && To == Other.To;
     }
   };
+  friend raw_ostream &operator<<(raw_ostream &OS, const Trace &);
 
   struct TraceHash {
     size_t operator()(const Trace &L) const {
-      return std::hash<uint64_t>()(L.From << 32 | L.To);
+      return llvm::hash_combine(L.Branch, L.From, L.To);
     }
   };
 
-  struct FTInfo {
-    uint64_t InternCount{0};
-    uint64_t ExternCount{0};
-  };
-
   struct TakenBranchInfo {
     uint64_t TakenCount{0};
     uint64_t MispredCount{0};
   };
 
   /// Intermediate storage for profile data. We save the results of parsing
   /// and use them later for processing and assigning profile.
-  std::unordered_map<Trace, TakenBranchInfo, TraceHash> BranchLBRs;
-  std::unordered_map<Trace, FTInfo, TraceHash> FallthroughLBRs;
+  std::unordered_map<Trace, TakenBranchInfo, TraceHash> TraceMap;
+  std::vector<std::pair<Trace, TakenBranchInfo>> Traces;
+  /// Pre-populated addresses of returns, coming from pre-aggregated data or
+  /// disassembly. Used to disambiguate call-continuation fall-throughs.
+  std::unordered_set<uint64_t> Returns;
   std::unordered_map<uint64_t, uint64_t> BasicSamples;
   std::vector<PerfMemSample> MemSamples;
 
@@ -193,8 +208,8 @@ class DataAggregator : public DataReader {
   /// Return a vector of offsets corresponding to a trace in a function
   /// if the trace is valid, std::nullopt otherwise.
   std::optional<SmallVector<std::pair<uint64_t, uint64_t>, 16>>
-  getFallthroughsInTrace(BinaryFunction &BF, const LBREntry &First,
-                         const LBREntry &Second, uint64_t Count = 1) const;
+  getFallthroughsInTrace(BinaryFunction &BF, const Trace &Trace, uint64_t Count,
+                         bool IsReturn) const;
 
   /// Record external entry into the function \p BF.
   ///
@@ -255,11 +270,10 @@ class DataAggregator : public DataReader {
                      uint64_t Mispreds);
 
   /// Register a \p Branch.
-  bool doBranch(uint64_t From, uint64_t To, uint64_t Count, uint64_t Mispreds);
+  bool doBranch(const Trace &Trace, uint64_t Count, uint64_t Mispreds);
 
   /// Register a trace between two LBR entries supplied in execution order.
-  bool doTrace(const LBREntry &First, const LBREntry &Second,
-               uint64_t Count = 1);
+  bool doTrace(const Trace &Trace, uint64_t Count);
 
   /// Parser helpers
   /// Return false if we exhausted our parser buffer and finished parsing
@@ -476,7 +490,6 @@ class DataAggregator : public DataReader {
 
   /// Debugging dump methods
   void dump() const;
-  void dump(const LBREntry &LBR) const;
   void dump(const PerfBranchSample &Sample) const;
   void dump(const PerfMemSample &Sample) const;
 
@@ -504,6 +517,27 @@ class DataAggregator : public DataReader {
 
   friend class YAMLProfileWriter;
 };
+
+inline raw_ostream &operator<<(raw_ostream &OS,
+                               const DataAggregator::LBREntry &L) {
+  OS << formatv("{0:x} -> {1:x}/{2}", L.From, L.To, L.Mispred ? 'M' : 'P');
+  return OS;
+}
+
+inline raw_ostream &operator<<(raw_ostream &OS,
+                               const DataAggregator::Trace &T) {
+  switch (T.Branch) {
+  case DataAggregator::Trace::FT_ONLY:
+  case DataAggregator::Trace::FT_EXTERNAL_ORIGIN:
+    break;
+  default:
+    OS << Twine::utohexstr(T.Branch) << " -> ";
+  }
+  OS << Twine::utohexstr(T.From);
+  if (T.To)
+    OS << " ... " << Twine::utohexstr(T.To);
+  return OS;
+}
 } // namespace bolt
 } // namespace llvm
 

bolt/include/bolt/Profile/DataReader.h

@@ -32,18 +32,6 @@ namespace bolt {
 
 class BinaryFunction;
 
-struct LBREntry {
-  uint64_t From;
-  uint64_t To;
-  bool Mispred;
-};
-
-inline raw_ostream &operator<<(raw_ostream &OS, const LBREntry &LBR) {
-  OS << "0x" << Twine::utohexstr(LBR.From) << " -> 0x"
-     << Twine::utohexstr(LBR.To);
-  return OS;
-}
-
 struct Location {
   bool IsSymbol;
   StringRef Name;
@@ -109,6 +97,9 @@ struct FuncBranchData {
   /// Total execution count for the function.
   int64_t ExecutionCount{0};
 
+  /// Total entry count from external code for the function.
+  uint64_t ExternEntryCount{0};
+
   /// Indicate if the data was used.
   bool Used{false};
 

bolt/include/bolt/Profile/ProfileYAMLMapping.h

@@ -206,6 +206,7 @@ struct BinaryFunctionProfile {
   uint32_t Id{0};
   llvm::yaml::Hex64 Hash{0};
   uint64_t ExecCount{0};
+  uint64_t ExternEntryCount{0};
   std::vector<BinaryBasicBlockProfile> Blocks;
   std::vector<InlineTreeNode> InlineTree;
   bool Used{false};
@@ -218,6 +219,7 @@ template <> struct MappingTraits<bolt::BinaryFunctionProfile> {
     YamlIO.mapRequired("fid", BFP.Id);
     YamlIO.mapRequired("hash", BFP.Hash);
     YamlIO.mapRequired("exec", BFP.ExecCount);
+    YamlIO.mapOptional("extern", BFP.ExternEntryCount, 0);
     YamlIO.mapRequired("nblocks", BFP.NumBasicBlocks);
     YamlIO.mapOptional("blocks", BFP.Blocks,
                        std::vector<bolt::BinaryBasicBlockProfile>());

bolt/lib/Core/BinaryFunction.cpp

@@ -471,6 +471,8 @@ void BinaryFunction::print(raw_ostream &OS, std::string Annotation) {
     OS << "\n  Sample Count: " << RawSampleCount;
     OS << "\n  Profile Acc : " << format("%.1f%%", ProfileMatchRatio * 100.0f);
   }
+  if (ExternEntryCount)
+    OS << "\n  Extern Entry Count: " << ExternEntryCount;
 
   if (opts::PrintDynoStats && !getLayout().block_empty()) {
     OS << '\n';

bolt/lib/Passes/ProfileQualityStats.cpp

@@ -532,6 +532,9 @@ void computeFlowMappings(const BinaryContext &BC, FlowInfo &TotalFlowMap) {
     std::vector<uint64_t> &MaxCountMap = TotalMaxCountMaps[FunctionNum];
     std::vector<uint64_t> &MinCountMap = TotalMinCountMaps[FunctionNum];
 
+    // Record external entry count into CallGraphIncomingFlows
+    CallGraphIncomingFlows[FunctionNum] += Function->getExternEntryCount();
+
     // Update MaxCountMap, MinCountMap, and CallGraphIncomingFlows
     auto recordCall = [&](const BinaryBasicBlock *SourceBB,
                           const MCSymbol *DestSymbol, uint64_t Count,

bolt/lib/Profile/BoltAddressTranslation.cpp

@@ -511,8 +511,8 @@ uint64_t BoltAddressTranslation::translate(uint64_t FuncAddress,
 
 std::optional<BoltAddressTranslation::FallthroughListTy>
 BoltAddressTranslation::getFallthroughsInTrace(uint64_t FuncAddress,
-                                               uint64_t From,
-                                               uint64_t To) const {
+                                               uint64_t From, uint64_t To,
+                                               bool IsReturn) const {
   SmallVector<std::pair<uint64_t, uint64_t>, 16> Res;
 
   // Filter out trivial case
@@ -530,6 +530,12 @@ BoltAddressTranslation::getFallthroughsInTrace(uint64_t FuncAddress,
   auto FromIter = Map.upper_bound(From);
   if (FromIter == Map.begin())
     return Res;
+
+  // For fall-throughs originating at returns, go back one entry to cover call
+  // site.
+  if (IsReturn)
+    --FromIter;
+
   // Skip instruction entries, to create fallthroughs we are only interested in
   // BB boundaries
   do {
@@ -546,7 +552,7 @@ BoltAddressTranslation::getFallthroughsInTrace(uint64_t FuncAddress,
     return Res;
 
   for (auto Iter = FromIter; Iter != ToIter;) {
-    const uint32_t Src = Iter->first;
+    const uint32_t Src = Iter->second >> 1;
     if (Iter->second & BRANCHENTRY) {
       ++Iter;
       continue;
@@ -557,7 +563,7 @@ BoltAddressTranslation::getFallthroughsInTrace(uint64_t FuncAddress,
       ++Iter;
     if (Iter->second & BRANCHENTRY)
       break;
-    Res.emplace_back(Src, Iter->first);
+    Res.emplace_back(Src, Iter->second >> 1);
   }
 
   return Res;