[ThinLTO] Shrink FunctionSummary by 8 bytes (#107706)

During the ThinLTO indexing step for one of our large applications, we create 4 million instances of FunctionSummary. Changing: std::vector<EdgeTy> CallGraphEdgeList; to: SmallVector<EdgeTy, 0> CallGraphEdgeList; in FunctionSummary reduces the size of each instance by 8 bytes. The rest of the patch makes the same change to other places so that the types stay compatible across function boundaries.
llvm · Sep 7, 2024 · 51d3829 · 51d3829
1 parent a1e06f7
commit 51d3829
Show file tree

Hide file tree

Showing 7 changed files with 30 additions and 24 deletions.
diff --git a/llvm/include/llvm/AsmParser/LLParser.h b/llvm/include/llvm/AsmParser/LLParser.h
@@ -394,7 +394,7 @@ namespace llvm {
     bool parseGVFlags(GlobalValueSummary::GVFlags &GVFlags);
     bool parseGVarFlags(GlobalVarSummary::GVarFlags &GVarFlags);
     bool parseOptionalFFlags(FunctionSummary::FFlags &FFlags);
-    bool parseOptionalCalls(std::vector<FunctionSummary::EdgeTy> &Calls);
+    bool parseOptionalCalls(SmallVectorImpl<FunctionSummary::EdgeTy> &Calls);
     bool parseHotness(CalleeInfo::HotnessType &Hotness);
     bool parseOptionalTypeIdInfo(FunctionSummary::TypeIdInfo &TypeIdInfo);
     bool parseTypeTests(std::vector<GlobalValue::GUID> &TypeTests);

diff --git a/llvm/include/llvm/IR/ModuleSummaryIndex.h b/llvm/include/llvm/IR/ModuleSummaryIndex.h
@@ -851,7 +851,7 @@ class FunctionSummary : public GlobalValueSummary {
   /// Create an empty FunctionSummary (with specified call edges).
   /// Used to represent external nodes and the dummy root node.
   static FunctionSummary
-  makeDummyFunctionSummary(std::vector<FunctionSummary::EdgeTy> Edges) {
+  makeDummyFunctionSummary(SmallVectorImpl<FunctionSummary::EdgeTy> &&Edges) {
     return FunctionSummary(
         FunctionSummary::GVFlags(
             GlobalValue::LinkageTypes::AvailableExternallyLinkage,
@@ -880,7 +880,9 @@ class FunctionSummary : public GlobalValueSummary {
   FFlags FunFlags;
 
   /// List of <CalleeValueInfo, CalleeInfo> call edge pairs from this function.
-  std::vector<EdgeTy> CallGraphEdgeList;
+  /// We use SmallVector<ValueInfo, 0> instead of std::vector<ValueInfo> for its
+  /// smaller memory footprint.
+  SmallVector<EdgeTy, 0> CallGraphEdgeList;
 
   std::unique_ptr<TypeIdInfo> TIdInfo;
 
@@ -910,7 +912,7 @@ class FunctionSummary : public GlobalValueSummary {
 public:
   FunctionSummary(GVFlags Flags, unsigned NumInsts, FFlags FunFlags,
                   SmallVectorImpl<ValueInfo> &&Refs,
-                  std::vector<EdgeTy> CGEdges,
+                  SmallVectorImpl<EdgeTy> &&CGEdges,
                   std::vector<GlobalValue::GUID> TypeTests,
                   std::vector<VFuncId> TypeTestAssumeVCalls,
                   std::vector<VFuncId> TypeCheckedLoadVCalls,
@@ -957,7 +959,7 @@ class FunctionSummary : public GlobalValueSummary {
   /// Return the list of <CalleeValueInfo, CalleeInfo> pairs.
   ArrayRef<EdgeTy> calls() const { return CallGraphEdgeList; }
 
-  std::vector<EdgeTy> &mutableCalls() { return CallGraphEdgeList; }
+  SmallVector<EdgeTy, 0> &mutableCalls() { return CallGraphEdgeList; }
 
   void addCall(EdgeTy E) { CallGraphEdgeList.push_back(E); }
 
@@ -1535,7 +1537,7 @@ class ModuleSummaryIndex {
       discoverNodes(ValueInfo(HaveGVs, &S), FunctionHasParent);
     }
 
-    std::vector<FunctionSummary::EdgeTy> Edges;
+    SmallVector<FunctionSummary::EdgeTy, 0> Edges;
     // create edges to all roots in the Index
     for (auto &P : FunctionHasParent) {
       if (P.second)
@@ -1544,9 +1546,11 @@ class ModuleSummaryIndex {
     }
     if (Edges.empty()) {
       // Failed to find root - return an empty node
-      return FunctionSummary::makeDummyFunctionSummary({});
+      return FunctionSummary::makeDummyFunctionSummary(
+          SmallVector<FunctionSummary::EdgeTy, 0>());
     }
-    auto CallGraphRoot = FunctionSummary::makeDummyFunctionSummary(Edges);
+    auto CallGraphRoot =
+        FunctionSummary::makeDummyFunctionSummary(std::move(Edges));
     return CallGraphRoot;
   }
 
@@ -1894,10 +1898,11 @@ template <> struct GraphTraits<ValueInfo> {
     return P.first;
   }
   using ChildIteratorType =
-      mapped_iterator<std::vector<FunctionSummary::EdgeTy>::iterator,
+      mapped_iterator<SmallVector<FunctionSummary::EdgeTy, 0>::iterator,
                       decltype(&valueInfoFromEdge)>;
 
-  using ChildEdgeIteratorType = std::vector<FunctionSummary::EdgeTy>::iterator;
+  using ChildEdgeIteratorType =
+      SmallVector<FunctionSummary::EdgeTy, 0>::iterator;
 
   static NodeRef getEntryNode(ValueInfo V) { return V; }
 

diff --git a/llvm/include/llvm/IR/ModuleSummaryIndexYAML.h b/llvm/include/llvm/IR/ModuleSummaryIndexYAML.h
@@ -230,7 +230,7 @@ template <> struct CustomMappingTraits<GlobalValueSummaryMapTy> {
               FSum.CanAutoHide,
               static_cast<GlobalValueSummary::ImportKind>(FSum.ImportType)),
           /*NumInsts=*/0, FunctionSummary::FFlags{}, std::move(Refs),
-          ArrayRef<FunctionSummary::EdgeTy>{}, std::move(FSum.TypeTests),
+          SmallVector<FunctionSummary::EdgeTy, 0>{}, std::move(FSum.TypeTests),
           std::move(FSum.TypeTestAssumeVCalls),
           std::move(FSum.TypeCheckedLoadVCalls),
           std::move(FSum.TypeTestAssumeConstVCalls),

diff --git a/llvm/lib/Analysis/ModuleSummaryAnalysis.cpp b/llvm/lib/Analysis/ModuleSummaryAnalysis.cpp
@@ -309,7 +309,7 @@ static void computeFunctionSummary(
   // Map from callee ValueId to profile count. Used to accumulate profile
   // counts for all static calls to a given callee.
   MapVector<ValueInfo, CalleeInfo, DenseMap<ValueInfo, unsigned>,
-            std::vector<std::pair<ValueInfo, CalleeInfo>>>
+            SmallVector<FunctionSummary::EdgeTy, 0>>
       CallGraphEdges;
   SetVector<ValueInfo, SmallVector<ValueInfo, 0>> RefEdges, LoadRefEdges,
       StoreRefEdges;
@@ -964,7 +964,7 @@ ModuleSummaryIndex llvm::buildModuleSummaryIndex(
                         /* HasUnknownCall */ true,
                         /* MustBeUnreachable */ false},
                     SmallVector<ValueInfo, 0>{},
-                    ArrayRef<FunctionSummary::EdgeTy>{},
+                    SmallVector<FunctionSummary::EdgeTy, 0>{},
                     ArrayRef<GlobalValue::GUID>{},
                     ArrayRef<FunctionSummary::VFuncId>{},
                     ArrayRef<FunctionSummary::VFuncId>{},

diff --git a/llvm/lib/AsmParser/LLParser.cpp b/llvm/lib/AsmParser/LLParser.cpp
@@ -9396,7 +9396,7 @@ bool LLParser::parseFunctionSummary(std::string Name, GlobalValue::GUID GUID,
       /*Live=*/false, /*IsLocal=*/false, /*CanAutoHide=*/false,
       GlobalValueSummary::Definition);
   unsigned InstCount;
-  std::vector<FunctionSummary::EdgeTy> Calls;
+  SmallVector<FunctionSummary::EdgeTy, 0> Calls;
   FunctionSummary::TypeIdInfo TypeIdInfo;
   std::vector<FunctionSummary::ParamAccess> ParamAccesses;
   SmallVector<ValueInfo, 0> Refs;
@@ -9685,7 +9685,8 @@ bool LLParser::parseOptionalFFlags(FunctionSummary::FFlags &FFlags) {
 /// Call ::= '(' 'callee' ':' GVReference
 ///            [( ',' 'hotness' ':' Hotness | ',' 'relbf' ':' UInt32 )]?
 ///            [ ',' 'tail' ]? ')'
-bool LLParser::parseOptionalCalls(std::vector<FunctionSummary::EdgeTy> &Calls) {
+bool LLParser::parseOptionalCalls(
+    SmallVectorImpl<FunctionSummary::EdgeTy> &Calls) {
   assert(Lex.getKind() == lltok::kw_calls);
   Lex.Lex();
 

diff --git a/llvm/lib/Bitcode/Reader/BitcodeReader.cpp b/llvm/lib/Bitcode/Reader/BitcodeReader.cpp
@@ -986,10 +986,9 @@ class ModuleSummaryIndexBitcodeReader : public BitcodeReaderBase {
       uint64_t Offset,
       DenseMap<unsigned, GlobalValue::LinkageTypes> &ValueIdToLinkageMap);
   SmallVector<ValueInfo, 0> makeRefList(ArrayRef<uint64_t> Record);
-  std::vector<FunctionSummary::EdgeTy> makeCallList(ArrayRef<uint64_t> Record,
-                                                    bool IsOldProfileFormat,
-                                                    bool HasProfile,
-                                                    bool HasRelBF);
+  SmallVector<FunctionSummary::EdgeTy, 0>
+  makeCallList(ArrayRef<uint64_t> Record, bool IsOldProfileFormat,
+               bool HasProfile, bool HasRelBF);
   Error parseEntireSummary(unsigned ID);
   Error parseModuleStringTable();
   void parseTypeIdCompatibleVtableSummaryRecord(ArrayRef<uint64_t> Record);
@@ -7378,11 +7377,11 @@ ModuleSummaryIndexBitcodeReader::makeRefList(ArrayRef<uint64_t> Record) {
   return Ret;
 }
 
-std::vector<FunctionSummary::EdgeTy>
+SmallVector<FunctionSummary::EdgeTy, 0>
 ModuleSummaryIndexBitcodeReader::makeCallList(ArrayRef<uint64_t> Record,
                                               bool IsOldProfileFormat,
                                               bool HasProfile, bool HasRelBF) {
-  std::vector<FunctionSummary::EdgeTy> Ret;
+  SmallVector<FunctionSummary::EdgeTy, 0> Ret;
   // In the case of new profile formats, there are two Record entries per
   // Edge. Otherwise, conservatively reserve up to Record.size.
   if (!IsOldProfileFormat && (HasProfile || HasRelBF))
@@ -7670,7 +7669,7 @@ Error ModuleSummaryIndexBitcodeReader::parseEntireSummary(unsigned ID) {
           ArrayRef<uint64_t>(Record).slice(RefListStartIndex, NumRefs));
       bool HasProfile = (BitCode == bitc::FS_PERMODULE_PROFILE);
       bool HasRelBF = (BitCode == bitc::FS_PERMODULE_RELBF);
-      std::vector<FunctionSummary::EdgeTy> Calls = makeCallList(
+      SmallVector<FunctionSummary::EdgeTy, 0> Calls = makeCallList(
           ArrayRef<uint64_t>(Record).slice(CallGraphEdgeStartIndex),
           IsOldProfileFormat, HasProfile, HasRelBF);
       setSpecialRefs(Refs, NumRORefs, NumWORefs);
@@ -7824,7 +7823,7 @@ Error ModuleSummaryIndexBitcodeReader::parseEntireSummary(unsigned ID) {
       SmallVector<ValueInfo, 0> Refs = makeRefList(
           ArrayRef<uint64_t>(Record).slice(RefListStartIndex, NumRefs));
       bool HasProfile = (BitCode == bitc::FS_COMBINED_PROFILE);
-      std::vector<FunctionSummary::EdgeTy> Edges = makeCallList(
+      SmallVector<FunctionSummary::EdgeTy, 0> Edges = makeCallList(
           ArrayRef<uint64_t>(Record).slice(CallGraphEdgeStartIndex),
           IsOldProfileFormat, HasProfile, false);
       ValueInfo VI = std::get<0>(getValueInfoFromValueId(ValueID));

diff --git a/llvm/lib/IR/ModuleSummaryIndex.cpp b/llvm/lib/IR/ModuleSummaryIndex.cpp
@@ -37,7 +37,8 @@ static cl::opt<bool> ImportConstantsWithRefs(
 constexpr uint32_t FunctionSummary::ParamAccess::RangeWidth;
 
 FunctionSummary FunctionSummary::ExternalNode =
-    FunctionSummary::makeDummyFunctionSummary({});
+    FunctionSummary::makeDummyFunctionSummary(
+        SmallVector<FunctionSummary::EdgeTy, 0>());
 
 GlobalValue::VisibilityTypes ValueInfo::getELFVisibility() const {
   bool HasProtected = false;