cleanup

Created using spr 1.3.4

Author: aaupov

bolt/include/bolt/Profile/DataAggregator.h

@@ -101,27 +101,26 @@ class DataAggregator : public DataReader {
 
   /// 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.
+  /// - if \p Branch is FT_ONLY or FT_EXTERNAL_ORIGIN, the trace only
+  ///   contains fall-through data,
+  /// - if \p To is BR_ONLY, the trace only contains branch data.
   struct Trace {
     static constexpr const uint64_t EXTERNAL = 0ULL;
+    static constexpr const uint64_t BR_ONLY = -1ULL;
     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;
-    bool operator==(const Trace &Other) const {
-      return Branch == Other.Branch && From == Other.From && To == Other.To;
-    }
+    auto tie() const { return std::tie(Branch, From, To); }
+    bool operator==(const Trace &Other) const { return tie() == Other.tie(); }
+    bool operator<(const Trace &Other) const { return tie() < Other.tie(); }
   };
   friend raw_ostream &operator<<(raw_ostream &OS, const Trace &);
 
   struct TraceHash {
-    size_t operator()(const Trace &L) const {
-      return llvm::hash_combine(L.Branch, L.From, L.To);
-    }
+    size_t operator()(const Trace &L) const { return hash_combine(L.tie()); }
   };
 
   struct TakenBranchInfo {
@@ -531,7 +530,7 @@ inline raw_ostream &operator<<(raw_ostream &OS,
     OS << Twine::utohexstr(T.Branch) << " -> ";
   }
   OS << Twine::utohexstr(T.From);
-  if (T.To)
+  if (T.To != DataAggregator::Trace::BR_ONLY)
     OS << " ... " << Twine::utohexstr(T.To);
   return OS;
 }

bolt/lib/Profile/DataAggregator.cpp

@@ -514,6 +514,10 @@ Error DataAggregator::preprocessProfile(BinaryContext &BC) {
   deleteTempFiles();
 
 heatmap:
+  // Sort parsed traces for faster processing.
+  if (!opts::BasicAggregation)
+    llvm::sort(Traces, llvm::less_first());
+
   if (!opts::HeatmapMode)
     return Error::success();
 
@@ -1283,15 +1287,15 @@ std::error_code DataAggregator::parseAggregatedLBREntry() {
   }
 
   if (Type == BRANCH) {
-    Addr[2] = Location(Trace::EXTERNAL);
+    Addr[2] = Location(Trace::BR_ONLY);
   }
 
   Trace T{Addr[0]->Offset, Addr[1]->Offset, Addr[2]->Offset};
   TakenBranchInfo TI{(uint64_t)Count, (uint64_t)Mispreds};
 
   Traces.emplace_back(T, TI);
 
-  if (Addr[2]->Offset)
+  if (Addr[2]->Offset != Trace::BR_ONLY)
     NumTraces += Count;
   NumTotalSamples += Count;
 
@@ -1305,7 +1309,7 @@ bool DataAggregator::ignoreKernelInterrupt(LBREntry &LBR) const {
 
 std::error_code DataAggregator::printLBRHeatMap() {
   outs() << "PERF2BOLT: parse branch events...\n";
-  NamedRegionTimer T("parseBranch", "Parsing branch events", TimerGroupName,
+  NamedRegionTimer T("buildHeatmap", "Building heatmap", TimerGroupName,
                      TimerGroupDesc, opts::TimeAggregator);
 
   if (BC->IsLinuxKernel) {
@@ -1342,7 +1346,7 @@ std::error_code DataAggregator::printLBRHeatMap() {
   for (const auto &[PC, Hits] : BasicSamples)
     HM.registerAddress(PC, Hits);
   for (const auto &[Trace, Info] : Traces)
-    if (Trace.To)
+    if (Trace.To != Trace::BR_ONLY)
       HM.registerAddressRange(Trace.From, Trace.To, Info.TakenCount);
 
   if (HM.getNumInvalidRanges())
@@ -1540,7 +1544,7 @@ void DataAggregator::processBranchEvents() {
                      TimerGroupName, TimerGroupDesc, opts::TimeAggregator);
 
   for (const auto &[Trace, Info] : Traces) {
-    if (Trace.To)
+    if (Trace.To != Trace::BR_ONLY)
       doTrace(Trace, Info.TakenCount);
     if (Trace.Branch != Trace::FT_ONLY &&
         Trace.Branch != Trace::FT_EXTERNAL_ORIGIN)