[BOLT] computing raw branch count for yaml profiles

[spupyrev]

Summary

Function.RawBranchCount is initialized for fdata profile but not for yaml one.
The diff adds the computation of the field for yaml profiles

[Amir]

Is that primarily needed for "X out of Y samples in the binary (Z%) belong to functions with invalid (possibly stale) profile" log message? Can you add a test using yaml profile, e.g. based on bolt/test/X86/pre-aggregated-perf.test?

[spupyrev]

In D144211#4157099, @Amir wrote:

Is that primarily needed for "X out of Y samples in the binary (Z%) belong to functions with invalid (possibly stale) profile" log message? Can you add a test using yaml profile, e.g. based on bolt/test/X86/pre-aggregated-perf.test?

I noticed a few (minor) differences between how yaml and fdata profiles are parsed and processed -- this is one of them and can easily be fixed. I will add a test.

[spupyrev]

• re-enabled test in branch-data.test
• added a check to the test verifying the new flag

[Amir]

LGTM with small nits

bolt/lib/Core/BinaryFunction.cpp

474

bolt/lib/Profile/YAMLProfileReader.cpp

87–91

bolt/test/X86/branch-data.test

18

[spupyrev]

comments

bolt/include/bolt/Core/BinaryFunction.h

   ///
   /// Return COUNT_NO_PROFILE if there's no profile info.
   uint64_t getExecutionCount() const { return ExecutionCount; }
 
   /// Return the raw profile information about the number of branch
   /// executions corresponding to this function.
   uint64_t getRawBranchCount() const { return RawBranchCount; }
 
+  /// Set the profile data about the number of branch executions corresponding
+  /// to this function.
+  void setRawBranchCount(uint64_t Count) { RawBranchCount = Count; }
+
   /// Return the execution count for functions with known profile.
   /// Return 0 if the function has no profile.
   uint64_t getKnownExecutionCount() const {
     return ExecutionCount == COUNT_NO_PROFILE ? 0 : ExecutionCount;
   }
 
   /// Return original LSDA address for the function or NULL.
   uint64_t getLSDAAddress() const { return LSDAAddress; }

bolt/lib/Core/BinaryFunction.cpp

if (!Layout.block_empty()) {
     for (const BinaryBasicBlock *BB : Layout.blocks())
       OS << LS << BB->getName();
   }
   if (getImageAddress())
     OS << "\n  Image       : 0x" << Twine::utohexstr(getImageAddress());
   if (ExecutionCount != COUNT_NO_PROFILE) {
     OS << "\n  Exec Count  : " << ExecutionCount;
     OS << "\n  Profile Acc : " << format("%.1f%%", ProfileMatchRatio * 100.0f);
+    OS << "\n  RawBranchCount : " << RawBranchCount;
   }
 
   if (opts::PrintDynoStats && !getLayout().block_empty()) {
     OS << '\n';
     DynoStats dynoStats = getDynoStats(*this);
     OS << dynoStats;
   }
 

bolt/lib/Profile/YAMLProfileReader.cpp

bool YAMLProfileReader::parseFunctionProfile(
   uint64_t MismatchedBlocks = 0;
   uint64_t MismatchedCalls = 0;
   uint64_t MismatchedEdges = 0;
 
   uint64_t FunctionExecutionCount = 0;
 
   BF.setExecutionCount(YamlBF.ExecCount);
 
+  uint64_t FuncRawBranchCount = 0;
+  for (const yaml::bolt::BinaryBasicBlockProfile &YamlBB : YamlBF.Blocks) {
+    for (const yaml::bolt::SuccessorInfo &YamlSI : YamlBB.Successors) {
+      FuncRawBranchCount += YamlSI.Count;
+    }
+  }
+  BF.setRawBranchCount(FuncRawBranchCount);
+
   if (!opts::IgnoreHash && YamlBF.Hash != BF.computeHash(/*UseDFS=*/true)) {
     if (opts::Verbosity >= 1)
       errs() << "BOLT-WARNING: function hash mismatch\n";
     ProfileMatched = false;
   }
 
   if (YamlBF.NumBasicBlocks != BF.size()) {
     if (opts::Verbosity >= 1)

bolt/test/X86/branch-data.test

 # Checks that llvm-bolt is able to read data generated by perf2bolt and update
 # the CFG edges accordingly with absolute number of branches and mispredictions.
 # Also checks that llvm-bolt disassembler and CFG builder is working properly.
 
 RUN: yaml2obj %p/Inputs/blarge.yaml &> %t.exe
-RUN: llvm-bolt %t.exe -o /dev/null --data %p/Inputs/blarge.fdata --print-cfg
+RUN: llvm-bolt %t.exe -o /dev/null --data %p/Inputs/blarge.fdata --print-cfg | FileCheck %s
 
 CHECK:    Binary Function "usqrt"
 CHECK:      State       : CFG constructed
 CHECK:      Address     : 0x401170
 CHECK:      Size        : 0x43
-CHECK:      MaxSize     : 0x50
-CHECK:      Offset      : 0x1170
+CHECK:      MaxSize     : 0x43
+CHECK:      Offset      : 0xcb0
 CHECK:      Section     : .text
 CHECK:      IsSimple    : 1
 CHECK:      BB Count    : 5
 CHECK:      Exec Count  : 199
+CHECK:      RawBranchCount  : 7689
 CHECK:    }
 CHECK:    .LBB{{.*}}
 CHECK:      Exec Count : 199
 CHECK:        {{.*}}:   movl    $0x20, %r8d
 CHECK:        {{.*}}:   xorl    %eax, %eax
 CHECK:        {{.*}}:   xorl    %edx, %edx
 CHECK:      Successors: .Ltmp{{.*}} (mispreds: 0, count: 0)
 CHECK:    .Ltmp{{.*}}