[PATCH] D38478: Use the first instruction's count to estimate the funciton's entry frequency.

[Teresa Johnson via Phabricator via llvm-commits]

tejohnson added a comment.

Started looking at, but have a few initial questions.



================
Comment at: include/llvm/ProfileData/SampleProf.h:303
+  /// profile without referring to potentially inaccurate debug info, this
+  /// gives more accurate profile data and is preferred for standalong symbols.
   uint64_t getHeadSamples() const { return TotalHeadSamples; }
----------------
s/standalong/standalone/


================
Comment at: include/llvm/ProfileData/SampleProf.h:304
+  /// gives more accurate profile data and is preferred for standalong symbols.
   uint64_t getHeadSamples() const { return TotalHeadSamples; }
 
----------------
Should this be used somewhere in this change since it is preferred?


================
Comment at: include/llvm/ProfileData/SampleProf.h:311
+        (CallsiteSamples.empty() ||
+         BodySamples.begin()->first < CallsiteSamples.begin()->first))
+      return BodySamples.begin()->second.getSamples();
----------------
What is this line checking?


================
Comment at: include/llvm/ProfileData/SampleProf.h:316
+      for (const auto &N_FS : CallsiteSamples.begin()->second)
+        T += N_FS.second.getEntrySamples();
+      return T;
----------------
What do these samples correspond to?


https://reviews.llvm.org/D38478