[TypeProf][PGO]Support skipping vtable comparisons for a class and its derived ones

llvm/lib/Transforms/Instrumentation/IndirectCallPromotion.cpp

@@ -132,6 +132,14 @@ static cl::opt<int> ICPMaxNumVTableLastCandidate(
     "icp-max-num-vtable-last-candidate", cl::init(1), cl::Hidden,
     cl::desc("The maximum number of vtable for the last candidate."));
 
+static cl::list<std::string> ICPIgnoredBaseTypes(
+    "icp-ignored-base-types", cl::Hidden,
+    cl::desc(
+        "A list of mangled vtable names. Classes specified by the vtables "
+        "and their derived ones will not be vtable-ICP'ed. Useful when the "
+        "profiled types and actual types in the optimized binary could be "
+        "different due to profiling limitations."));
+
 namespace {
 
 // The key is a vtable global variable, and the value is a map.
@@ -316,6 +324,8 @@ class IndirectCallPromoter {
 
   OptimizationRemarkEmitter &ORE;
 
+  const DenseSet<StringRef> &IgnoredBaseTypes;
+
   // A struct that records the direct target and it's call count.
   struct PromotionCandidate {
     Function *const TargetFunction;
@@ -391,10 +401,12 @@ class IndirectCallPromoter {
       Function &Func, Module &M, InstrProfSymtab *Symtab, bool SamplePGO,
       const VirtualCallSiteTypeInfoMap &VirtualCSInfo,
       VTableAddressPointOffsetValMap &VTableAddressPointOffsetVal,
+      const DenseSet<StringRef> &IgnoredBaseTypes,
       OptimizationRemarkEmitter &ORE)
       : F(Func), M(M), Symtab(Symtab), SamplePGO(SamplePGO),
         VirtualCSInfo(VirtualCSInfo),
-        VTableAddressPointOffsetVal(VTableAddressPointOffsetVal), ORE(ORE) {}
+        VTableAddressPointOffsetVal(VTableAddressPointOffsetVal), ORE(ORE),
+        IgnoredBaseTypes(IgnoredBaseTypes) {}
   IndirectCallPromoter(const IndirectCallPromoter &) = delete;
   IndirectCallPromoter &operator=(const IndirectCallPromoter &) = delete;
 
@@ -851,8 +863,25 @@ bool IndirectCallPromoter::isProfitableToCompareVTables(
     LLVM_DEBUG(dbgs() << "\n");
 
     uint64_t CandidateVTableCount = 0;
-    for (auto &[GUID, Count] : VTableGUIDAndCounts)
+
+    for (auto &[GUID, Count] : VTableGUIDAndCounts) {
       CandidateVTableCount += Count;
+      auto *VTableVar = Symtab->getGlobalVariable(GUID);
+
+      assert(VTableVar &&
+             "VTableVar must exist for GUID in VTableGUIDAndCounts");
+
+      SmallVector<MDNode *, 2> Types;
+      VTableVar->getMetadata(LLVMContext::MD_type, Types);
+
+      for (auto *Type : Types)
+        if (auto *TypeId = dyn_cast<MDString>(Type->getOperand(1).get()))
+          if (IgnoredBaseTypes.contains(TypeId->getString())) {
+            LLVM_DEBUG(dbgs() << "    vtable profiles should be ignored. Bail "
+                                 "out vtable comparison.");
+            return false;
+          }
+    }
 
     if (CandidateVTableCount < Candidate.Count * ICPVTablePercentageThreshold) {
       LLVM_DEBUG(
@@ -956,9 +985,15 @@ static bool promoteIndirectCalls(Module &M, ProfileSummaryInfo *PSI, bool InLTO,
   bool Changed = false;
   VirtualCallSiteTypeInfoMap VirtualCSInfo;
 
-  if (EnableVTableProfileUse)
+  DenseSet<StringRef> IgnoredBaseTypes;
+
+  if (EnableVTableProfileUse) {
     computeVirtualCallSiteTypeInfoMap(M, MAM, VirtualCSInfo);
 
+    for (StringRef Str : ICPIgnoredBaseTypes)
+      IgnoredBaseTypes.insert(Str);
+  }
+
   // VTableAddressPointOffsetVal stores the vtable address points. The vtable
   // address point of a given <vtable, address point offset> is static (doesn't
   // change after being computed once).
@@ -977,7 +1012,8 @@ static bool promoteIndirectCalls(Module &M, ProfileSummaryInfo *PSI, bool InLTO,
     auto &ORE = FAM.getResult<OptimizationRemarkEmitterAnalysis>(F);
 
     IndirectCallPromoter CallPromoter(F, M, &Symtab, SamplePGO, VirtualCSInfo,
-                                      VTableAddressPointOffsetVal, ORE);
+                                      VTableAddressPointOffsetVal,
+                                      IgnoredBaseTypes, ORE);
     bool FuncChanged = CallPromoter.processFunction(PSI);
     if (ICPDUMPAFTER && FuncChanged) {
       LLVM_DEBUG(dbgs() << "\n== IR Dump After =="; F.print(dbgs()));

llvm/test/Transforms/PGOProfile/icp_vtable_cmp.ll

@@ -1,7 +1,11 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
 
+; Tests that ICP compares vtables by checking IR.
 ; RUN: opt < %s -passes='pgo-icall-prom' -pass-remarks=pgo-icall-prom -enable-vtable-profile-use -icp-max-num-vtable-last-candidate=2 -S 2>&1 | FileCheck %s --check-prefixes=VTABLE-COMMON,VTABLE-CMP
+; Require exactly one vtable candidate for each function candidate. Tests that ICP compares function by checking IR.
 ; RUN: opt < %s -passes='pgo-icall-prom' -pass-remarks=pgo-icall-prom -enable-vtable-profile-use -icp-max-num-vtable-last-candidate=1 -S 2>&1 | FileCheck %s --check-prefixes=VTABLE-COMMON,FUNC-CMP
+; On top of line 4, ignore 'Base1' and its derived types for vtable-based comparison. Tests that ICP compares functions.
+; RUN: opt < %s -passes='pgo-icall-prom' -pass-remarks=pgo-icall-prom -enable-vtable-profile-use -icp-max-num-vtable-last-candidate=2 -icp-ignored-base-types='Base1' -S 2>&1 | FileCheck %s --check-prefixes=VTABLE-COMMON,FUNC-CMP
 
 target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"