[NFC][MLGO] Make logging more robust

1) add some self-diagnosis (when asserts are enabled) to check that all
features have the same nr of entries

2) avoid storing pointers to mutable fields because the proto API
contract doesn't actually guarantee those stay fixed even if no further
mutation of the object occurs.

Differential Revision: https://reviews.llvm.org/D107594

llvm/include/llvm/Analysis/Utils/TFUtils.h

@@ -104,6 +104,9 @@ Optional<TensorSpec> getTensorSpecFromJSON(LLVMContext &Ctx,
 struct LoggedFeatureSpec {
   TensorSpec Spec;
   Optional<std::string> LoggingName;
+  const std::string &getLoggingName() const {
+    return LoggingName ? *LoggingName : Spec.name();
+  }
 };
 
 /// Load the output specs. If SpecFileOverride is not empty, that path is used.
@@ -170,7 +173,9 @@ public:
   // we can consider using bytes.
   char *addEntryAndGetFloatOrInt64Buffer(size_t FeatureID);
 
-  void print(raw_ostream &OS);
+  // Flush the content of the log to the stream, clearing the stored data in the
+  // process.
+  void flush(raw_ostream &OS);
 
 private:
   std::vector<LoggedFeatureSpec> FeatureSpecs;

llvm/lib/Analysis/DevelopmentModeInlineAdvisor.cpp

@@ -377,7 +377,7 @@ void TrainingLogger::logInlineEvent(const InlineEvent &Event,
 void TrainingLogger::print() {
   std::error_code EC;
   raw_fd_ostream OutFile(LogFileName, EC);
-  L->print(OutFile);
+  L->flush(OutFile);
 }
 
 DevelopmentModeMLInlineAdvisor::DevelopmentModeMLInlineAdvisor(

llvm/lib/Analysis/TFUtils.cpp

@@ -262,29 +262,57 @@ private:
 class LoggerDataImpl {
   const std::vector<LoggedFeatureSpec> LoggedFeatureSpecs;
   const TensorSpec RewardSpec;
+  const bool IncludeReward;
 
-  tensorflow::SequenceExample SE;
-  std::vector<tensorflow::FeatureList *> FeatureLists;
-  tensorflow::FeatureList *Reward = nullptr;
+  std::vector<tensorflow::FeatureList> FeatureLists;
+  tensorflow::FeatureList Reward;
+
+  bool isSelfConsistent(const tensorflow::SequenceExample &SE,
+                        size_t NrRecords) const {
+    bool Ret = true;
+    for (const auto &TSpecs : LoggedFeatureSpecs) {
+      const auto &Name = TSpecs.getLoggingName();
+      const auto &FL = SE.feature_lists().feature_list().at(Name).feature();
+      if (NrRecords != static_cast<size_t>(FL.size())) {
+        dbgs() << "[TF-UTILS]: " << Name << " has missing records. Expected "
+               << NrRecords << " got " << FL.size() << "\n";
+        Ret = false;
+      }
+    }
+    if (IncludeReward && static_cast<size_t>(SE.feature_lists()
+                                                 .feature_list()
+                                                 .at(RewardSpec.name())
+                                                 .feature()
+                                                 .size()) != NrRecords) {
+      dbgs() << "[TF-UTILS]: reward is missing records.\n";
+      Ret = false;
+    }
+    return Ret;
+  }
+
+  void transferLog(tensorflow::SequenceExample &SE) {
+    auto *FL = SE.mutable_feature_lists()->mutable_feature_list();
+    if (IncludeReward)
+      (*FL)[RewardSpec.name()].Swap(&Reward);
+    assert(FeatureLists.size() == LoggedFeatureSpecs.size());
+    for (size_t I = 0; I < FeatureLists.size(); ++I) {
+      const auto &LFS = LoggedFeatureSpecs[I];
+      (*FL)[LFS.getLoggingName()].Swap(&FeatureLists[I]);
+    }
+  }
 
 public:
   LoggerDataImpl(const std::vector<LoggedFeatureSpec> &LoggedSpecs,
                  const TensorSpec &RewardSpec, bool IncludeReward)
-      : LoggedFeatureSpecs(LoggedSpecs), RewardSpec(RewardSpec) {
-    auto *FL = SE.mutable_feature_lists()->mutable_feature_list();
-    if (IncludeReward)
-      Reward = &(*FL)[RewardSpec.name()];
-    // Allocate first the map entries, then capture their address. We will not
-    // mutate the set of features after this (i.e. the pointers won't dangle).
-    for (const auto &LFS : LoggedSpecs) {
-      (*FL)[LFS.LoggingName ? *LFS.LoggingName : LFS.Spec.name()] = {};
-    }
-    for (const auto &LFS : LoggedSpecs)
-      FeatureLists.push_back(
-          &(*FL)[LFS.LoggingName ? *LFS.LoggingName : LFS.Spec.name()]);
-  }
+      : LoggedFeatureSpecs(LoggedSpecs), RewardSpec(RewardSpec),
+        IncludeReward(IncludeReward), FeatureLists(LoggedFeatureSpecs.size()) {}
 
-  void print(raw_ostream &OS) {
+  // flush the logged info to a stream and clear the log contents.
+  void flush(raw_ostream &OS) {
+    size_t NrRecords = getNrRecords();
+    tensorflow::SequenceExample SE;
+    transferLog(SE);
+    assert(isSelfConsistent(SE, NrRecords));
     std::string OutStr;
     if (ProtobufTextMode)
       google::protobuf::TextFormat::PrintToString(SE, &OutStr);
@@ -298,14 +326,14 @@ public:
     const auto &Spec = LoggedFeatureSpecs[FeatureID].Spec;
     if (Spec.isElementType<float>()) {
       auto *RF = FeatureLists[FeatureID]
-                     ->add_feature()
+                     .add_feature()
                      ->mutable_float_list()
                      ->mutable_value();
       RF->Resize(Spec.getElementCount(), 0.0);
       return reinterpret_cast<char *>(RF->mutable_data());
     } else if (Spec.isElementType<int32_t>() || Spec.isElementType<int64_t>()) {
       auto *RF = FeatureLists[FeatureID]
-                     ->add_feature()
+                     .add_feature()
                      ->mutable_int64_list()
                      ->mutable_value();
       RF->Resize(Spec.getElementCount(), 0);
@@ -315,17 +343,18 @@ public:
   }
 
   template <typename T> void logReward(T Value) {
+    assert(IncludeReward);
     if (RewardSpec.isElementType<float>())
-      Reward->add_feature()->mutable_float_list()->add_value(Value);
+      Reward.add_feature()->mutable_float_list()->add_value(Value);
     else if (RewardSpec.isElementType<int32_t>() ||
              RewardSpec.isElementType<int64_t>())
-      Reward->add_feature()->mutable_int64_list()->add_value(Value);
+      Reward.add_feature()->mutable_int64_list()->add_value(Value);
     else
       llvm_unreachable("Unsupported tensor type.");
   }
 
   size_t getNrRecords() const {
-    return FeatureLists.empty() ? 0 : FeatureLists[0]->feature().size();
+    return FeatureLists.empty() ? 0 : FeatureLists[0].feature().size();
   }
 };
 } // namespace llvm
@@ -538,5 +567,5 @@ char *Logger::addEntryAndGetFloatOrInt64Buffer(size_t FeatureID) {
   return reinterpret_cast<char *>(LoggerData->addNewTensor(FeatureID));
 }
 
-void Logger::print(raw_ostream &OS) { LoggerData->print(OS); }
+void Logger::flush(raw_ostream &OS) { LoggerData->flush(OS); }
 #endif // defined(LLVM_HAVE_TF_API)

llvm/unittests/Analysis/TFUtilsTest.cpp

@@ -179,7 +179,7 @@ TEST(TFUtilsTest, Logger) {
   L.logFloatReward(-3.0);
   std::string Result;
   raw_string_ostream OS(Result);
-  L.print(OS);
+  L.flush(OS);
 
   tensorflow::SequenceExample Expected;
   EXPECT_TRUE(Expected.ParseFromString(Result));
@@ -215,7 +215,7 @@ TEST(TFUtilsTest, LoggerInt32FeaturesAndReward) {
   L.logInt32Reward(-3);
   std::string Result;
   raw_string_ostream OS(Result);
-  L.print(OS);
+  L.flush(OS);
 
   tensorflow::SequenceExample Expected;
   EXPECT_TRUE(Expected.ParseFromString(Result));
@@ -250,7 +250,7 @@ TEST(TFUtilsTest, LoggerNoReward) {
 
   std::string Result;
   raw_string_ostream OS(Result);
-  L.print(OS);
+  L.flush(OS);
   tensorflow::SequenceExample Expected;
   EXPECT_TRUE(Expected.ParseFromString(Result));
   PROTO_CHECKER("the_float", float_list, 0, F00);
@@ -274,7 +274,7 @@ TEST(TFUtilsTest, LoggerFinalReward) {
   L.logFloatFinalReward(3.14);
   std::string Result;
   raw_string_ostream OS(Result);
-  L.print(OS);
+  L.flush(OS);
   const float Zero[]{0.0};
   const float R[]{3.14};
   tensorflow::SequenceExample Expected;