[Clang] Add __type_list_dedup builtin to deduplicate types in template arguments

ilya-biryukov · ilya-biryukov · commit a46885df62ff · 2024-09-04T14:57:26.000+02:00
This allows to deduplicate the type lists efficiently in C++. It is
possible to achieve the same effect via template metaprogramming, but
performance of the resulting code is much lower than in the compiler.

We have observed that this code is quite common in our internal codebase
and this builtin allows to have significant savings (up to 25% of
compile time on targets that already take 3 minutes to compile).
The same builtin is also used widely enough in the codebase that we
expect a savings from a long tail of uses, too, although it is hard to
put an estimate on this number in advance.

The implementation aims to be as simple as possible and relies on the
exsisting machinery for builtin templates.
diff --git a/clang-tools-extra/clangd/unittests/FindTargetTests.cpp b/clang-tools-extra/clangd/unittests/FindTargetTests.cpp
@@ -731,6 +731,12 @@ TEST_F(TargetDeclTest, BuiltinTemplates) {
     using type_pack_element = [[__type_pack_element]]<N, Pack...>;
   )cpp";
   EXPECT_DECLS("TemplateSpecializationTypeLoc", );
+
+  Code = R"cpp(
+    template <template <class...> Templ, class... Types>
+    using type_list_dedup = [[__type_list_dedup]]<Templ, Types...>;
+  )cpp";
+  EXPECT_DECLS("TemplateSpecializationTypeLoc", );
 }
 
 TEST_F(TargetDeclTest, MemberOfTemplate) {
diff --git a/clang/include/clang/AST/ASTContext.h b/clang/include/clang/AST/ASTContext.h
@@ -401,6 +401,9 @@ class ASTContext : public RefCountedBase<ASTContext> {
   /// The identifier '__type_pack_element'.
   mutable IdentifierInfo *TypePackElementName = nullptr;
 
+  /// The identifier '__type_list_dedup'.
+  mutable IdentifierInfo *TypeListDedupName = nullptr;
+
   QualType ObjCConstantStringType;
   mutable RecordDecl *CFConstantStringTagDecl = nullptr;
   mutable TypedefDecl *CFConstantStringTypeDecl = nullptr;
@@ -608,6 +611,7 @@ class ASTContext : public RefCountedBase<ASTContext> {
   mutable ExternCContextDecl *ExternCContext = nullptr;
   mutable BuiltinTemplateDecl *MakeIntegerSeqDecl = nullptr;
   mutable BuiltinTemplateDecl *TypePackElementDecl = nullptr;
+  mutable BuiltinTemplateDecl *TypeListDedupDecl = nullptr;
 
   /// The associated SourceManager object.
   SourceManager &SourceMgr;
@@ -1115,6 +1119,7 @@ class ASTContext : public RefCountedBase<ASTContext> {
   ExternCContextDecl *getExternCContextDecl() const;
   BuiltinTemplateDecl *getMakeIntegerSeqDecl() const;
   BuiltinTemplateDecl *getTypePackElementDecl() const;
+  BuiltinTemplateDecl *getTypeListDedupDecl() const;
 
   // Builtin Types.
   CanQualType VoidTy;
@@ -2006,6 +2011,12 @@ class ASTContext : public RefCountedBase<ASTContext> {
     return TypePackElementName;
   }
 
+  IdentifierInfo *getTypeListDedupName() const {
+    if (!TypeListDedupName)
+      TypeListDedupName = &Idents.get("__type_list_dedup");
+    return TypeListDedupName;
+  }
+
   /// Retrieve the Objective-C "instancetype" type, if already known;
   /// otherwise, returns a NULL type;
   QualType getObjCInstanceType() {
diff --git a/clang/include/clang/AST/DeclID.h b/clang/include/clang/AST/DeclID.h
@@ -83,6 +83,9 @@ enum PredefinedDeclIDs {
   /// The internal '__type_pack_element' template.
   PREDEF_DECL_TYPE_PACK_ELEMENT_ID,
 
+  /// The internal '__type_list_dedup' template.
+  PREDEF_DECL_TYPE_LIST_DEDUP_ID,
+
   /// The number of declaration IDs that are predefined.
   NUM_PREDEF_DECL_IDS
 };
diff --git a/clang/include/clang/Basic/Builtins.h b/clang/include/clang/Basic/Builtins.h
@@ -309,7 +309,10 @@ enum BuiltinTemplateKind : int {
   BTK__make_integer_seq,
 
   /// This names the __type_pack_element BuiltinTemplateDecl.
-  BTK__type_pack_element
+  BTK__type_pack_element,
+
+  /// This names the __type_list_dedup BuiltinTemplateDecl.
+  BTK__type_list_dedup,
 };
 
 } // end namespace clang
diff --git a/clang/lib/AST/ASTContext.cpp b/clang/lib/AST/ASTContext.cpp
@@ -1170,6 +1170,14 @@ ASTContext::getTypePackElementDecl() const {
   return TypePackElementDecl;
 }
 
+BuiltinTemplateDecl *
+ASTContext::getTypeListDedupDecl() const {
+  if (!TypeListDedupDecl)
+    TypeListDedupDecl =
+        buildBuiltinTemplateDecl(BTK__type_list_dedup, getTypeListDedupName());
+  return TypeListDedupDecl;
+}
+
 RecordDecl *ASTContext::buildImplicitRecord(StringRef Name,
                                             RecordDecl::TagKind TK) const {
   SourceLocation Loc;
diff --git a/clang/lib/AST/ASTImporter.cpp b/clang/lib/AST/ASTImporter.cpp
@@ -5476,6 +5476,9 @@ ExpectedDecl ASTNodeImporter::VisitBuiltinTemplateDecl(BuiltinTemplateDecl *D) {
   case BuiltinTemplateKind::BTK__type_pack_element:
     ToD = Importer.getToContext().getTypePackElementDecl();
     break;
+  case BuiltinTemplateKind::BTK__type_list_dedup:
+    ToD = Importer.getToContext().getTypeListDedupDecl();
+    break;
   }
   assert(ToD && "BuiltinTemplateDecl of unsupported kind!");
   Importer.MapImported(D, ToD);
diff --git a/clang/lib/AST/DeclTemplate.cpp b/clang/lib/AST/DeclTemplate.cpp
@@ -1608,13 +1608,44 @@ createTypePackElementParameterList(const ASTContext &C, DeclContext *DC) {
                                        nullptr);
 }
 
+static TemplateParameterList *
+createTypeListDedupParameterList(const ASTContext &C, DeclContext *DC) {
+  // template <typename ...> typename Templ
+  auto *InnerTs = TemplateTypeParmDecl::Create(
+      C, DC, SourceLocation(), SourceLocation(), /*Depth=*/1, /*Position=*/0,
+      /*Id=*/nullptr, /*Typename=*/true, /*ParameterPack=*/true,
+      /*HasTypeConstraint=*/false);
+  InnerTs->setImplicit(true);
+  auto *TemplateParamList = TemplateParameterList::Create(
+      C, SourceLocation(), SourceLocation(), {InnerTs}, SourceLocation(),
+      /*RequiresClause=*/nullptr);
+  auto *Template = TemplateTemplateParmDecl::Create(
+      C, DC, SourceLocation(), /*Depth=*/0,
+      /*Position=*/0, /*ParameterPack=*/false, /*Id=*/nullptr,
+      /*Typename=*/true, TemplateParamList);
+  Template->setImplicit(true);
+
+  // typename ...Ts
+  auto *Ts = TemplateTypeParmDecl::Create(
+      C, DC, SourceLocation(), SourceLocation(), /*Depth=*/0, /*Position=*/1,
+      /*Id=*/nullptr, /*Typename=*/true, /*ParameterPack=*/true,
+      /*HasTypeConstraint=*/false);
+  Ts->setImplicit(true);
+  return TemplateParameterList::Create(
+      C, SourceLocation(), SourceLocation(),
+      llvm::ArrayRef<NamedDecl *>({Template, Ts}), SourceLocation(),
+      /*RequiresClause=*/nullptr);
+}
+
 static TemplateParameterList *createBuiltinTemplateParameterList(
     const ASTContext &C, DeclContext *DC, BuiltinTemplateKind BTK) {
   switch (BTK) {
   case BTK__make_integer_seq:
     return createMakeIntegerSeqParameterList(C, DC);
   case BTK__type_pack_element:
     return createTypePackElementParameterList(C, DC);
+  case BTK__type_list_dedup:
+    return createTypeListDedupParameterList(C, DC);
   }
 
   llvm_unreachable("unhandled BuiltinTemplateKind!");
diff --git a/clang/lib/Lex/PPMacroExpansion.cpp b/clang/lib/Lex/PPMacroExpansion.cpp
@@ -1836,6 +1836,7 @@ void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
               // Report builtin templates as being builtins.
               .Case("__make_integer_seq", getLangOpts().CPlusPlus)
               .Case("__type_pack_element", getLangOpts().CPlusPlus)
+              .Case("__type_list_dedup", getLangOpts().CPlusPlus)
               // Likewise for some builtin preprocessor macros.
               // FIXME: This is inconsistent; we usually suggest detecting
               // builtin macros via #ifdef. Don't add more cases here.
diff --git a/clang/lib/Sema/SemaLookup.cpp b/clang/lib/Sema/SemaLookup.cpp
@@ -928,10 +928,15 @@ bool Sema::LookupBuiltin(LookupResult &R) {
         if (II == getASTContext().getMakeIntegerSeqName()) {
           R.addDecl(getASTContext().getMakeIntegerSeqDecl());
           return true;
-        } else if (II == getASTContext().getTypePackElementName()) {
+        }
+        if (II == getASTContext().getTypePackElementName()) {
           R.addDecl(getASTContext().getTypePackElementDecl());
           return true;
         }
+        if (II == getASTContext().getTypeListDedupName()) {
+          R.addDecl(getASTContext().getTypeListDedupDecl());
+          return true;
+        }
       }
 
       // Check if this is an OpenCL Builtin, and if so, insert its overloads.
diff --git a/clang/lib/Sema/SemaTemplate.cpp b/clang/lib/Sema/SemaTemplate.cpp
@@ -17,7 +17,9 @@
 #include "clang/AST/Expr.h"
 #include "clang/AST/ExprCXX.h"
 #include "clang/AST/RecursiveASTVisitor.h"
+#include "clang/AST/TemplateBase.h"
 #include "clang/AST/TemplateName.h"
+#include "clang/AST/TypeOrdering.h"
 #include "clang/AST/TypeVisitor.h"
 #include "clang/Basic/Builtins.h"
 #include "clang/Basic/DiagnosticSema.h"
@@ -38,9 +40,11 @@
 #include "clang/Sema/Template.h"
 #include "clang/Sema/TemplateDeduction.h"
 #include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/DenseSet.h"
 #include "llvm/ADT/SmallBitVector.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/StringExtras.h"
+#include "llvm/Support/raw_ostream.h"
 
 #include <iterator>
 #include <optional>
@@ -3132,12 +3136,12 @@ checkBuiltinTemplateIdType(Sema &SemaRef, BuiltinTemplateDecl *BTD,
                                        TemplateLoc, SyntheticTemplateArgs);
   }
 
-  case BTK__type_pack_element:
+  case BTK__type_pack_element: {
     // Specializations of
     //    __type_pack_element<Index, T_1, ..., T_N>
     // are treated like T_Index.
     assert(Converted.size() == 2 &&
-      "__type_pack_element should be given an index and a parameter pack");
+           "__type_pack_element should be given an index and a parameter pack");
 
     TemplateArgument IndexArg = Converted[0], Ts = Converted[1];
     if (IndexArg.isDependent() || Ts.isDependent())
@@ -3158,6 +3162,34 @@ checkBuiltinTemplateIdType(Sema &SemaRef, BuiltinTemplateDecl *BTD,
     int64_t N = Index.getExtValue();
     return Ts.getPackAsArray()[N].getAsType();
   }
+  case BTK__type_list_dedup: {
+    assert(Converted.size() == 2 &&
+           "__type_list_dedup should be given a template and a parameter pack");
+    TemplateArgument Template = Converted[0];
+    TemplateArgument Ts = Converted[1];
+    if (Template.isDependent() || Ts.isDependent())
+      return Context.getCanonicalTemplateSpecializationType(TemplateName(BTD),
+                                                            Converted);
+
+    assert(Template.getKind() == clang::TemplateArgument::Template);
+    assert(Ts.getKind() == clang::TemplateArgument::Pack);
+    TemplateArgumentListInfo SyntheticTemplateArgs;
+    llvm::DenseSet<QualType> Seen;
+    // Synthesize a new template argument list, removing duplicates.
+    for (auto T : Ts.getPackAsArray()) {
+      assert(T.getKind() == clang::TemplateArgument::Type);
+      if (!Seen.insert(T.getAsType().getCanonicalType()).second)
+        continue;
+      SyntheticTemplateArgs.addArgument(TemplateArgumentLoc(
+          TemplateArgument(T),
+          SemaRef.Context.getTrivialTypeSourceInfo(
+              T.getAsType(),
+              /*FIXME: add location*/ SourceLocation())));
+    }
+    return SemaRef.CheckTemplateIdType(Template.getAsTemplate(), TemplateLoc,
+                                       SyntheticTemplateArgs);
+  }
+  }
   llvm_unreachable("unexpected BuiltinTemplateDecl!");
 }
 
diff --git a/clang/lib/Serialization/ASTReader.cpp b/clang/lib/Serialization/ASTReader.cpp
@@ -23,6 +23,7 @@
 #include "clang/AST/DeclCXX.h"
 #include "clang/AST/DeclFriend.h"
 #include "clang/AST/DeclGroup.h"
+#include "clang/AST/DeclID.h"
 #include "clang/AST/DeclObjC.h"
 #include "clang/AST/DeclTemplate.h"
 #include "clang/AST/DeclarationName.h"
@@ -7897,6 +7898,11 @@ Decl *ASTReader::getPredefinedDecl(PredefinedDeclIDs ID) {
       return Context.TypePackElementDecl;
     NewLoaded = Context.getTypePackElementDecl();
     break;
+  case PREDEF_DECL_TYPE_LIST_DEDUP_ID:
+    if (Context.TypeListDedupDecl)
+      return Context.TypeListDedupDecl;
+    NewLoaded = Context.getTypeListDedupDecl();
+    break;
   case NUM_PREDEF_DECL_IDS:
     llvm_unreachable("Invalid decl ID");
     break;
diff --git a/clang/lib/Serialization/ASTWriter.cpp b/clang/lib/Serialization/ASTWriter.cpp
@@ -22,6 +22,7 @@
 #include "clang/AST/DeclCXX.h"
 #include "clang/AST/DeclContextInternals.h"
 #include "clang/AST/DeclFriend.h"
+#include "clang/AST/DeclID.h"
 #include "clang/AST/DeclObjC.h"
 #include "clang/AST/DeclTemplate.h"
 #include "clang/AST/DeclarationName.h"
@@ -5051,6 +5052,8 @@ void ASTWriter::PrepareWritingSpecialDecls(Sema &SemaRef) {
                      PREDEF_DECL_CF_CONSTANT_STRING_TAG_ID);
   RegisterPredefDecl(Context.TypePackElementDecl,
                      PREDEF_DECL_TYPE_PACK_ELEMENT_ID);
+  RegisterPredefDecl(Context.TypeListDedupDecl,
+                     PREDEF_DECL_TYPE_LIST_DEDUP_ID);
 
   const TranslationUnitDecl *TU = Context.getTranslationUnitDecl();
 
diff --git a/clang/test/Import/builtin-template/Inputs/S.cpp b/clang/test/Import/builtin-template/Inputs/S.cpp
@@ -14,3 +14,10 @@ using TypePackElement = __type_pack_element<i, T...>;
 
 template <int i>
 struct X;
+
+
+template <template <class...> class Templ, class...Types>
+using TypeListDedup = __type_list_dedup<Templ, Types...>;
+
+template <class ...Ts>
+struct TypeList {};
diff --git a/clang/test/Import/builtin-template/test.cpp b/clang/test/Import/builtin-template/test.cpp
@@ -1,9 +1,11 @@
 // RUN: clang-import-test -dump-ast -import %S/Inputs/S.cpp -expression %s -Xcc -DSEQ | FileCheck --check-prefix=CHECK-SEQ %s
 // RUN: clang-import-test -dump-ast -import %S/Inputs/S.cpp -expression %s -Xcc -DPACK | FileCheck --check-prefix=CHECK-PACK %s
-// RUN: clang-import-test -dump-ast -import %S/Inputs/S.cpp -expression %s -Xcc -DPACK -Xcc -DSEQ | FileCheck --check-prefixes=CHECK-SEQ,CHECK-PACK %s
+// RUN: clang-import-test -dump-ast -import %S/Inputs/S.cpp -expression %s -Xcc -DDEDUP | FileCheck --check-prefix=CHECK-DEDUP %s
+// RUN: clang-import-test -dump-ast -import %S/Inputs/S.cpp -expression %s -Xcc -DPACK -Xcc -DSEQ -Xcc -DDEDUP | FileCheck --check-prefixes=CHECK-SEQ,CHECK-PACK,CHECK-DEDUP %s
 
 // CHECK-SEQ:  BuiltinTemplateDecl {{.+}} <<invalid sloc>> <invalid sloc> implicit __make_integer_seq{{$}}
 // CHECK-PACK: BuiltinTemplateDecl {{.+}} <<invalid sloc>> <invalid sloc> implicit __type_pack_element{{$}}
+// CHECK-DEDUP: BuiltinTemplateDecl {{.+}} <<invalid sloc>> <invalid sloc> implicit __type_list_dedup{{$}}
 
 void expr() {
 #ifdef SEQ
@@ -20,4 +22,10 @@ void expr() {
   static_assert(__is_same(TypePackElement<0, X<0>, X<1>>, X<0>), "");
   static_assert(__is_same(TypePackElement<1, X<0>, X<1>>, X<1>), "");
 #endif
+
+#ifdef DEDUP
+  static_assert(__is_same(TypeListDedup<TypeList>, TypeList<>), "");
+  static_assert(__is_same(TypeListDedup<TypeList, int, double, int>, TypeList<int, double>), "");
+  static_assert(__is_same(TypeListDedup<TypeList, X<0>, X<1>, X<1>, X<2>, X<0>>, TypeList<X<0>, X<1>, X<2>>), "");
+#endif
 }
diff --git a/clang/test/PCH/type_list_dedup.cpp b/clang/test/PCH/type_list_dedup.cpp
@@ -0,0 +1,20 @@
+// RUN: %clang_cc1 -std=c++14 -x c++-header %s -emit-pch -o %t.pch
+// RUN: %clang_cc1 -std=c++14 -x c++ /dev/null -include-pch %t.pch
+
+// RUN: %clang_cc1 -std=c++14 -x c++-header %s -emit-pch -fpch-instantiate-templates -o %t.pch
+// RUN: %clang_cc1 -std=c++14 -x c++ /dev/null -include-pch %t.pch
+
+template <template <class...> class Templ, class...Types>
+using TypeListDedup = __type_list_dedup<Templ, Types...>;
+
+template <class ...Ts>
+struct TypeList {};
+
+template <int i>
+struct X {};
+
+void fn1() {
+  TypeList<int, double> l1 = TypeListDedup<TypeList, int, double, int>{};
+  TypeList<> l2 = TypeListDedup<TypeList>{};
+  TypeList<X<0>, X<1>> x1 = TypeListDedup<TypeList, X<0>, X<1>, X<0>, X<1>>{};
+}
diff --git a/clang/test/SemaTemplate/temp-param-list-dedup.cpp b/clang/test/SemaTemplate/temp-param-list-dedup.cpp
@@ -0,0 +1,33 @@
+// RUN: %clang_cc1 %s -verify
+// expected-no-diagnostics
+
+template <typename...> struct TypeList;
+
+static_assert(__is_same(
+  __type_list_dedup<TypeList, int, int*, int, double, float>,
+  TypeList<int, int*, double, float>));
+
+template <template<typename ...> typename Templ, typename ...Types>
+struct Dependent {
+  using empty_list = __type_list_dedup<Templ>;
+  using same = __type_list_dedup<Templ, Types...>;
+  using twice = __type_list_dedup<Templ, Types..., Types...>;
+  using dep_only_types = __type_list_dedup<TypeList, Types...>;
+  using dep_only_template = __type_list_dedup<Templ, int, double, int>;
+}; 
+
+static_assert(__is_same(Dependent<TypeList>::empty_list, TypeList<>));
+static_assert(__is_same(Dependent<TypeList>::same, TypeList<>));
+static_assert(__is_same(Dependent<TypeList>::twice, TypeList<>));
+static_assert(__is_same(Dependent<TypeList>::dep_only_types, TypeList<>));
+static_assert(__is_same(Dependent<TypeList>::dep_only_template, TypeList<int, double>));
+
+static_assert(__is_same(Dependent<TypeList, int*, double*, int*>::empty_list, TypeList<>));
+static_assert(__is_same(Dependent<TypeList, int*, double*, int*>::same, TypeList<int*, double*>));
+static_assert(__is_same(Dependent<TypeList, int*, double*, int*>::twice, TypeList<int*, double*>));
+static_assert(__is_same(Dependent<TypeList, int*, double*, int*>::dep_only_types, TypeList<int*, double*>));
+static_assert(__is_same(Dependent<TypeList, int*, double*, int*>::dep_only_template, TypeList<int, double>));
+
+// FIXME: tests for using template type alias as a template
+// FIXME: tests for errors
+// FIXME: tests for locations
