Emit metadata if there is a profile hash mismatch

[tmsriram]

Summary

This patch emits "instr_prof_hash_mismatch" function annotation metadata if there is a hash mismatch while applying instrumented profiles.

During the PGO optimized build using instrumented profiles, if the CFG of the function has changed since generating the profile, a hash mismatch is encountered. This patch emits this information as annotation metadata. We plan to use this with Propeller which is done at the machine IR level. Propeller is usually applied on top of PGO and a hash mismatch during PGO could be used to detect source drift.

[davidxl]

llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp

1254

Do you need to check the string as well?

[tmsriram]

llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp

1254

Thanks. I was debating what to do here, I had it as checking the string first. However, I noticed function metadata with annotation is only added here so I was not sure. Let me change it to explicitly check the string.

[tmsriram]

Check if metadata already exists.

[tmsriram]

Update patch.

[snehasish]

llvm/test/Transforms/PGOProfile/hash_mismatch_metadata.ll

2

Would it be useful to add a test to check that existing metadata is preserved?

[davidxl]

llvm/test/Transforms/PGOProfile/hash_mismatch_metadata.ll

2

That (end-to-end) can be added as a profile runtime test.

[tmsriram]

llvm/test/Transforms/PGOProfile/hash_mismatch_metadata.ll

2

I wasnt sure how to do it but I found a way. I can introduce a synthetic annotation. Let me update.

[tmsriram]

Update test to check if the old metadata is retained and the new one is not duplicated.

[tmsriram]

Update patch.

[snehasish]

lgtm

[xur]

lgtm too

[davidxl]

lgtm

llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp

for (auto &E : Edges) {
 
     getBBInfo(E->SrcBB).UnknownCountOutEdge--;
     getBBInfo(E->DestBB).UnknownCountInEdge--;
     return;
   }
   llvm_unreachable("Cannot find the unknown count edge");
 }
 
+// Emit function metadata indicating PGO profile mismatch.
+static void annotateFunctionWithHashMismatch(Function &F,
+                                             LLVMContext &ctx) {
+  const char MetadataName[] = "instr_prof_hash_mismatch";
+  SmallVector<Metadata *, 2> Names;
+  // If this metadata already exists, ignore.
+  if (MDTuple *Tuple =
+      cast<MDTuple>(F.getMetadata(LLVMContext::MD_annotation))) {
+    for (auto &N : Tuple->operands()) {
+      if (cast<MDString>(N.get())->getString() == MetadataName)
+        return;
+      Names.push_back(N.get());
+    }
+  }
+
+  MDBuilder MDB(ctx);
+  Names.push_back(MDB.createString(MetadataName));
+  MDNode *MD = MDTuple::get(ctx, Names);
+  F.setMetadata(LLVMContext::MD_annotation, MD);
+}
+
 // Read the profile from ProfileFileName and assign the value to the
 // instrumented BB and the edges. This function also updates ProgramMaxCount.
 // Return true if the profile are successfully read, and false on errors.
 bool PGOUseFunc::readCounters(IndexedInstrProfReader *PGOReader, bool &AllZeros,
                               bool &AllMinusOnes) {
   auto &Ctx = M->getContext();
   Expected<InstrProfRecord> Result =
       PGOReader->getInstrProfRecord(FuncInfo.FuncName, FuncInfo.FunctionHash);
handleAllErrors(std::move(E), [&](const InstrProfError &IPE) {
                  Err == instrprof_error::malformed) {
         IsCS ? NumOfCSPGOMismatch++ : NumOfPGOMismatch++;
         SkipWarning =
             NoPGOWarnMismatch ||
             (NoPGOWarnMismatchComdat &&
              (F.hasComdat() ||
               F.getLinkage() == GlobalValue::AvailableExternallyLinkage));
         LLVM_DEBUG(dbgs() << "hash mismatch (skip=" << SkipWarning << ")");
+        // Emit function metadata indicating PGO profile mismatch.
+        annotateFunctionWithHashMismatch(F, M->getContext());
       }
 
       LLVM_DEBUG(dbgs() << " IsCS=" << IsCS << "\n");
       if (SkipWarning)
         return;
 
       std::string Msg = IPE.message() + std::string(" ") + F.getName().str() +
                         std::string(" Hash = ") +

llvm/test/Transforms/PGOProfile/Inputs/hash_mismatch_metadata.proftext

+:ir
+:entry_first
+foo
+# Func Hash:
+11111
+# Num Counters:
+1
+# Counter Values:
+100

llvm/test/Transforms/PGOProfile/hash_mismatch_metadata.ll

+; RUN: llvm-profdata merge %S/Inputs/hash_mismatch_metadata.proftext -o %t.profdata
+; RUN: opt < %s -mtriple=x86_64-linux-gnu -passes=pgo-instr-use -pgo-test-profile-file=%t.profdata -S | FileCheck %s
+
+define void @foo() {
+entry:
+   ret void
+}
+
+; CHECK: !{{[0-9]+}} = !{!"instr_prof_hash_mismatch"}