[llvm] Revert "[BOLT][NFC] Register profiled functions once (#150622)" (PR #152597)

[Rafael Auler via llvm-commits]

https://github.com/rafaelauler created https://github.com/llvm/llvm-project/pull/152597

We are observing random crashes where addresses provided by the hardware (from LBR) randomly clashes with what LLVM uses in DenseSet as tombstones records (-2 << 12), which causes DenseSet to assert when ingesting this data while running the recently added registerProfiledFunctions().

>From 1fdf5abae9703302a6bd2e0d38dcaf0c951249be Mon Sep 17 00:00:00 2001
From: Rafael Auler <rafaelauler at meta.com>
Date: Thu, 7 Aug 2025 13:58:39 -0700
Subject: [PATCH] Revert "[BOLT][NFC] Register profiled functions once
 (#150622)"

Summary:
We are observing random crashes where addresses provided by the
hardware (from LBR) randomly clashes with what LLVM uses in DenseSet
as tombstones records (-2 << 12), which causes DenseSet to assert when
ingesting this data while running the recently added
registerProfiledFunctions().
---
 bolt/include/bolt/Profile/DataAggregator.h |  3 --
 bolt/lib/Profile/DataAggregator.cpp        | 39 ++++++++++------------
 2 files changed, 18 insertions(+), 24 deletions(-)

diff --git a/bolt/include/bolt/Profile/DataAggregator.h b/bolt/include/bolt/Profile/DataAggregator.h
index cb1b87f8d0d65..db0f6903185b7 100644
--- a/bolt/include/bolt/Profile/DataAggregator.h
+++ b/bolt/include/bolt/Profile/DataAggregator.h
@@ -502,9 +502,6 @@ class DataAggregator : public DataReader {
   /// entries).
   void imputeFallThroughs();
 
-  /// Register profiled functions for lite mode.
-  void registerProfiledFunctions();
-
   /// Debugging dump methods
   void dump() const;
   void dump(const PerfBranchSample &Sample) const;
diff --git a/bolt/lib/Profile/DataAggregator.cpp b/bolt/lib/Profile/DataAggregator.cpp
index c13fa6dbe582b..3604fdd3a94b4 100644
--- a/bolt/lib/Profile/DataAggregator.cpp
+++ b/bolt/lib/Profile/DataAggregator.cpp
@@ -581,26 +581,6 @@ void DataAggregator::imputeFallThroughs() {
     outs() << "BOLT-INFO: imputed " << InferredTraces << " traces\n";
 }
 
-void DataAggregator::registerProfiledFunctions() {
-  DenseSet<uint64_t> Addrs;
-  for (const auto &Trace : llvm::make_first_range(Traces)) {
-    if (Trace.Branch != Trace::FT_ONLY &&
-        Trace.Branch != Trace::FT_EXTERNAL_ORIGIN)
-      Addrs.insert(Trace.Branch);
-    Addrs.insert(Trace.From);
-  }
-
-  for (const auto [PC, _] : BasicSamples)
-    Addrs.insert(PC);
-
-  for (const PerfMemSample &MemSample : MemSamples)
-    Addrs.insert(MemSample.PC);
-
-  for (const uint64_t Addr : Addrs)
-    if (BinaryFunction *Func = getBinaryFunctionContainingAddress(Addr))
-      Func->setHasProfileAvailable();
-}
-
 Error DataAggregator::preprocessProfile(BinaryContext &BC) {
   this->BC = &BC;
 
@@ -623,7 +603,6 @@ Error DataAggregator::preprocessProfile(BinaryContext &BC) {
       exit(0);
   }
 
-  registerProfiledFunctions();
   return Error::success();
 }
 
@@ -1368,6 +1347,10 @@ std::error_code DataAggregator::parseAggregatedLBREntry() {
   }
 
   const uint64_t FromOffset = Addr[0]->Offset;
+  BinaryFunction *FromFunc = getBinaryFunctionContainingAddress(FromOffset);
+  if (FromFunc)
+    FromFunc->setHasProfileAvailable();
+
   int64_t Count = Counters[0];
   int64_t Mispreds = Counters[1];
 
@@ -1378,6 +1361,11 @@ std::error_code DataAggregator::parseAggregatedLBREntry() {
     return std::error_code();
   }
 
+  const uint64_t ToOffset = Addr[1]->Offset;
+  BinaryFunction *ToFunc = getBinaryFunctionContainingAddress(ToOffset);
+  if (ToFunc)
+    ToFunc->setHasProfileAvailable();
+
   /// For fall-through types, adjust locations to match Trace container.
   if (Type == FT || Type == FT_EXTERNAL_ORIGIN || Type == FT_EXTERNAL_RETURN) {
     Addr[2] = Location(Addr[1]->Offset); // Trace To
@@ -1625,6 +1613,9 @@ std::error_code DataAggregator::parseBranchEvents() {
   Traces.reserve(TraceMap.size());
   for (const auto &[Trace, Info] : TraceMap) {
     Traces.emplace_back(Trace, Info);
+    for (const uint64_t Addr : {Trace.Branch, Trace.From})
+      if (BinaryFunction *BF = getBinaryFunctionContainingAddress(Addr))
+        BF->setHasProfileAvailable();
   }
   clear(TraceMap);
 
@@ -1685,6 +1676,9 @@ std::error_code DataAggregator::parseBasicEvents() {
       continue;
     ++NumTotalSamples;
 
+    if (BinaryFunction *BF = getBinaryFunctionContainingAddress(Sample->PC))
+      BF->setHasProfileAvailable();
+
     ++BasicSamples[Sample->PC];
     EventNames.insert(Sample->EventName);
   }
@@ -1722,6 +1716,9 @@ std::error_code DataAggregator::parseMemEvents() {
     if (std::error_code EC = Sample.getError())
       return EC;
 
+    if (BinaryFunction *BF = getBinaryFunctionContainingAddress(Sample->PC))
+      BF->setHasProfileAvailable();
+
     MemSamples.emplace_back(std::move(Sample.get()));
   }