[ThinLTO][TypeProfiling]Add annotated vtable GUID as referenced variables in per function summary, and update bitcode writer to create value-ids for these referenced vtables

llvm/include/llvm/ProfileData/InstrProf.h

@@ -283,14 +283,24 @@ void annotateValueSite(Module &M, Instruction &Inst,
 
 /// Extract the value profile data from \p Inst which is annotated with
 /// value profile meta data. Return false if there is no value data annotated,
-/// otherwise  return true.
+/// otherwise return true.
 bool getValueProfDataFromInst(const Instruction &Inst,
                               InstrProfValueKind ValueKind,
                               uint32_t MaxNumValueData,
                               InstrProfValueData ValueData[],
                               uint32_t &ActualNumValueData, uint64_t &TotalC,
                               bool GetNoICPValue = false);
 
+/// Extract the value profile data from \p Inst and returns them if \p Inst is
+/// annotated with value profile data. Returns nullptr otherwise. It's similar
+/// to `getValueProfDataFromInst` above except that an array is allocated only
+/// after a preliminary checking that the value profiles of kind `ValueKind`
+/// exist.
+std::unique_ptr<InstrProfValueData[]>
+getValueProfDataFromInst(const Instruction &Inst, InstrProfValueKind ValueKind,
+                         uint32_t MaxNumValueData, uint32_t &ActualNumValueData,
+                         uint64_t &TotalC, bool GetNoICPValue = false);
+
 inline StringRef getPGOFuncNameMetadataName() { return "PGOFuncName"; }
 
 /// Return the PGOFuncName meta data associated with a function.

llvm/lib/Analysis/ModuleSummaryAnalysis.cpp

@@ -124,6 +124,22 @@ static bool findRefEdges(ModuleSummaryIndex &Index, const User *CurUser,
         Worklist.push_back(Operand);
     }
   }
+
+  const Instruction *I = dyn_cast<Instruction>(CurUser);
+  if (I) {
+    uint32_t ActualNumValueData = 0;
+    uint64_t TotalCount = 0;
+    auto ValueDataArray = getValueProfDataFromInst(
+        *I, IPVK_VTableTarget, 24 /* MaxNumValueData */, ActualNumValueData,
+        TotalCount);
+
+    if (ValueDataArray.get()) {
+      for (uint32_t j = 0; j < ActualNumValueData; j++) {
+        RefEdges.insert(Index.getOrInsertValueInfo(
+            ValueDataArray[j].Value /* VTableGUID */));
+      }
+    }
+  }
   return HasBlockAddress;
 }
 

llvm/lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -199,14 +199,23 @@ class ModuleBitcodeWriterBase : public BitcodeWriterBase {
     for (const auto &GUIDSummaryLists : *Index)
       // Examine all summaries for this GUID.
       for (auto &Summary : GUIDSummaryLists.second.SummaryList)
-        if (auto FS = dyn_cast<FunctionSummary>(Summary.get()))
+        if (auto FS = dyn_cast<FunctionSummary>(Summary.get())) {
           // For each call in the function summary, see if the call
           // is to a GUID (which means it is for an indirect call,
           // otherwise we would have a Value for it). If so, synthesize
           // a value id.
           for (auto &CallEdge : FS->calls())
             if (!CallEdge.first.haveGVs() || !CallEdge.first.getValue())
               assignValueId(CallEdge.first.getGUID());
+
+          // For each referenced variables in the function summary, see if the
+          // variable is represented by a GUID (as opposed to a symbol to
+          // declarations or definitions in the module). If so, sythesize a
+          // value id.
+          for (auto &RefEdge : FS->refs())
+            if ((!RefEdge.haveGVs() || !RefEdge.getValue()))
+              assignValueId(RefEdge.getGUID());
+        }
   }
 
 protected:
@@ -4071,7 +4080,7 @@ void ModuleBitcodeWriterBase::writePerModuleFunctionSummaryRecord(
   NameVals.push_back(SpecialRefCnts.second); // worefcnt
 
   for (auto &RI : FS->refs())
-    NameVals.push_back(VE.getValueID(RI.getValue()));
+    NameVals.push_back(getValueId(RI));
 
   const bool UseRelBFRecord =
       WriteRelBFToSummary && !F.hasProfileData() &&

llvm/lib/ProfileData/InstrProf.cpp

@@ -1285,46 +1285,44 @@ void annotateValueSite(Module &M, Instruction &Inst,
   Inst.setMetadata(LLVMContext::MD_prof, MDNode::get(Ctx, Vals));
 }
 
-bool getValueProfDataFromInst(const Instruction &Inst,
-                              InstrProfValueKind ValueKind,
-                              uint32_t MaxNumValueData,
-                              InstrProfValueData ValueData[],
-                              uint32_t &ActualNumValueData, uint64_t &TotalC,
-                              bool GetNoICPValue) {
+MDNode *mayHaveValueProfileOfKind(const Instruction &Inst,
+                                  InstrProfValueKind ValueKind) {
   MDNode *MD = Inst.getMetadata(LLVMContext::MD_prof);
   if (!MD)
-    return false;
+    return nullptr;
 
-  unsigned NOps = MD->getNumOperands();
+  if (MD->getNumOperands() < 5)
+    return nullptr;
 
-  if (NOps < 5)
-    return false;
-
-  // Operand 0 is a string tag "VP":
   MDString *Tag = cast<MDString>(MD->getOperand(0));
-  if (!Tag)
-    return false;
-
-  if (!Tag->getString().equals("VP"))
-    return false;
+  if (!Tag || !Tag->getString().equals("VP"))
+    return nullptr;
 
   // Now check kind:
   ConstantInt *KindInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(1));
   if (!KindInt)
-    return false;
+    return nullptr;
   if (KindInt->getZExtValue() != ValueKind)
-    return false;
+    return nullptr;
+
+  return MD;
+}
 
+static bool getValueProfDataFromInst(const MDNode *const MD,
+                                     const uint32_t MaxNumDataWant,
+                                     InstrProfValueData ValueData[],
+                                     uint32_t &ActualNumValueData,
+                                     uint64_t &TotalC, bool GetNoICPValue) {
+  const unsigned NOps = MD->getNumOperands();
   // Get total count
   ConstantInt *TotalCInt = mdconst::dyn_extract<ConstantInt>(MD->getOperand(2));
   if (!TotalCInt)
     return false;
   TotalC = TotalCInt->getZExtValue();
-
   ActualNumValueData = 0;
 
   for (unsigned I = 3; I < NOps; I += 2) {
-    if (ActualNumValueData >= MaxNumValueData)
+    if (ActualNumValueData >= MaxNumDataWant)
       break;
     ConstantInt *Value = mdconst::dyn_extract<ConstantInt>(MD->getOperand(I));
     ConstantInt *Count =
@@ -1341,6 +1339,33 @@ bool getValueProfDataFromInst(const Instruction &Inst,
   return true;
 }
 
+std::unique_ptr<InstrProfValueData[]>
+getValueProfDataFromInst(const Instruction &Inst, InstrProfValueKind ValueKind,
+                         uint32_t MaxNumValueData, uint32_t &ActualNumValueData,
+                         uint64_t &TotalC, bool GetNoICPValue) {
+  MDNode *MD = mayHaveValueProfileOfKind(Inst, ValueKind);
+  if (!MD)
+    return nullptr;
+  auto ValueDataArray = std::make_unique<InstrProfValueData[]>(MaxNumValueData);
+  if (!getValueProfDataFromInst(MD, MaxNumValueData, ValueDataArray.get(),
+                                ActualNumValueData, TotalC, GetNoICPValue))
+    return nullptr;
+  return ValueDataArray;
+}
+
+bool getValueProfDataFromInst(const Instruction &Inst,
+                              InstrProfValueKind ValueKind,
+                              uint32_t MaxNumValueData,
+                              InstrProfValueData ValueData[],
+                              uint32_t &ActualNumValueData, uint64_t &TotalC,
+                              bool GetNoICPValue) {
+  MDNode *MD = mayHaveValueProfileOfKind(Inst, ValueKind);
+  if (!MD)
+    return false;
+  return getValueProfDataFromInst(MD, MaxNumValueData, ValueData,
+                                  ActualNumValueData, TotalC, GetNoICPValue);
+}
+
 MDNode *getPGOFuncNameMetadata(const Function &F) {
   return F.getMetadata(getPGOFuncNameMetadataName());
 }

llvm/test/Bitcode/thinlto-function-summary-vtableref-pgo.ll

@@ -1,41 +1,40 @@
 ; RUN: opt -module-summary %s -o %t.o
 
-; RUN: llvm-bcanalyzer -dump %t.o | FileCheck %s
+; RUN llvm-bcanalyzer -dump %t.o | FileCheck %s
 
 ; RUN: llvm-dis -o - %t.o | FileCheck %s --check-prefix=DIS
-
+; Round trip it through llvm-as
+; RUN: llvm-dis -o - %t.o | llvm-as -o - | llvm-dis -o - | FileCheck %s --check-prefix=DIS
 
 ; CHECK: <GLOBALVAL_SUMMARY_BLOCK
 ; CHECK-NEXT:   <VERSION op0=9/>
 ; CHECK-NEXT:   <FLAGS op0=0/>
-; The `VALUE_GUID` below represents the "_ZN4Base4funcEv" referenced by the
-; indirect call instruction.
-; CHECK-NEXT:   <VALUE_GUID op0=8 op1=5459407273543877811/>
-; <PERMODULE_PROFILE> has the format [valueid, flags, instcount, funcflags,
-;                                     numrefs, rorefcnt, worefcnt,
-;                                     n x (valueid, hotness+tailcall)]
-; CHECK-NEXT:   <PERMODULE_PROFILE abbrevid=4 op0=0 op1=0 op2=4 op3=256 op4=0 op5=0 op6=0 op7=8 op8=2/>
+; "_ZTV4Base" referenced by the instruction that loads vtable from object.
+; CHECK-NEXT:   <VALUE_GUID op0=8 op1=1960855528937986108/>
+; "_ZN4Base4funcEv" referenced by the indirect call instruction.
+; CHECK-NEXT:   <VALUE_GUID op0=7 op1=5459407273543877811/>
+; CHECK-NEXT:   <PERMODULE abbrevid=5 op0=0 op1=0 op2=4 op3=256 op4=1 op5=1 op6=0 op7=8 op8=7/>
 ; CHECK-NEXT:  </GLOBALVAL_SUMMARY_BLOCK>
 
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
-define i32 @_Z4testP4Base(ptr %0) !prof !0 {
-  %2 = load ptr, ptr %0, !prof !1
+define i32 @_Z4testP4Base(ptr %0) {
+  %2 = load ptr, ptr %0, !prof !0
   %3 = load ptr, ptr %2
-  %4 = tail call i32 %3(ptr %0), !prof !2
+  %4 = tail call i32 %3(ptr %0), !prof !1
   ret i32 %4
 }
 
-!0 = !{!"function_entry_count", i32 150}
 ; 1960855528937986108 is the MD5 hash of _ZTV4Base
-!1 = !{!"VP", i32 2, i64 1600, i64 1960855528937986108, i64 1600}
+!0 = !{!"VP", i32 2, i64 1600, i64 1960855528937986108, i64 1600}
 ; 5459407273543877811 is the MD5 hash of _ZN4Base4funcEv
-!2 = !{!"VP", i32 0, i64 1600, i64 5459407273543877811, i64 1600}
+!1 = !{!"VP", i32 0, i64 1600, i64 5459407273543877811, i64 1600}
 
 ; ModuleSummaryIndex stores <guid, global-value summary> map in std::map; so
 ; global value summares are printed out in the order that gv's guid increases.
-; DIS: ^0 = module: (path: "/usr/local/google/home/mingmingl/llvm-fork/llvm-project/build/test/Bitcode/Output/thinlto-function-summary-vtableref-pgo.ll.tmp.o", hash: (0, 0, 0, 0, 0))
-; DIS: ^1 = gv: (guid: 5459407273543877811)
-; DIS: ^2 = gv: (name: "_Z4testP4Base", summaries: (function: (module: ^0, flags: (linkage: external, visibility: default, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 4, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 0, returnDoesNotAlias: 0, noInline: 0, alwaysInline: 0, noUnwind: 0, mayThrow: 0, hasUnknownCall: 1, mustBeUnreachable: 0), calls: ((callee: ^1, hotness: none))))) ; guid = 15857150948103218965
-; DIS: ^3 = blockcount: 0
+; DIS: ^0 = module: (path: "{{.*}}", hash: (0, 0, 0, 0, 0))
+; DIS-NEXT: ^1 = gv: (guid: 1960855528937986108)
+; DIS-NEXT: ^2 = gv: (guid: 5459407273543877811)
+; DIS-NEXT: ^3 = gv: (name: "_Z4testP4Base", summaries: (function: (module: ^0, flags: (linkage: external, visibility: default, notEligibleToImport: 0, live: 0, dsoLocal: 0, canAutoHide: 0), insts: 4, funcFlags: (readNone: 0, readOnly: 0, noRecurse: 0, returnDoesNotAlias: 0, noInline: 0, alwaysInline: 0, noUnwind: 0, mayThrow: 0, hasUnknownCall: 1, mustBeUnreachable: 0), calls: ((callee: ^2, hotness: none)), refs: (readonly ^1)))) ; guid = 15857150948103218965
+; DIS-NEXT: ^4 = blockcount: 0