[BOLT] Support profile density with basic samples (#137644)

For profile with LBR samples, binary function profile density is
computed as a ratio of executed bytes to function size in bytes.

For profile with IP samples, use the size of basic block containing the
sample IP as a numerator.

Test Plan: updated perf_test.test

Author: aaupov

bolt/include/bolt/Profile/DataReader.h

@@ -246,6 +246,9 @@ struct FuncSampleData {
   /// 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;
 

bolt/lib/Profile/DataAggregator.cpp

@@ -565,15 +565,14 @@ 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.RawBranchCount = FBD->getNumExecutedBranches();
+    } else if (FuncSampleData *FSD = getFuncSampleData(BF.getNames())) {
+      BF.markProfiled(BinaryFunction::PF_SAMPLE);
+      BF.RawBranchCount = FSD->getSamples();
     }
   }
 
@@ -630,10 +629,18 @@ StringRef DataAggregator::getLocationName(const BinaryFunction &Func,
 
 bool DataAggregator::doSample(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()) {

bolt/lib/Profile/DataReader.cpp

@@ -128,6 +128,13 @@ uint64_t FuncSampleData::getSamples(uint64_t Start, uint64_t End) const {
   return Result;
 }
 
+uint64_t FuncSampleData::getSamples() const {
+  uint64_t Result = 0;
+  for (const SampleInfo &I : Data)
+    Result += I.Hits;
+  return Result;
+}
+
 void FuncSampleData::bumpCount(uint64_t Offset, uint64_t Count) {
   auto Iter = Index.find(Offset);
   if (Iter == Index.end()) {

bolt/test/perf2bolt/perf_test.test

@@ -8,6 +8,7 @@ RUN: perf2bolt %t -p=%t2 -o %t3 -nl -ignore-build-id 2>&1 | FileCheck %s
 
 CHECK-NOT: PERF2BOLT-ERROR
 CHECK-NOT: !! WARNING !! This high mismatch ratio indicates the input binary is probably not the same binary used during profiling collection.
+CHECK: BOLT-INFO: Functions with density >= {{.*}} account for 99.00% total sample counts.
 
 RUN: %clang %S/Inputs/perf_test.c -no-pie -fuse-ld=lld -o %t4
 RUN: perf record -Fmax -e cycles:u -o %t5 -- %t4