[jit][edge] Migrate to TypeFactory for jit types on mobile (#71516)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/71516

Mobile should be able to contruct dynamic types by default.
ghstack-source-id: 147498365

Test Plan:
CI.

**-48KB** binary size reduction for igios BSB.
UMBEX link: https://www.internalfb.com/intern/unigraph/explorer/?jsgq_traversal_spec=%7B%22builds%22%3A[%22bsb%3A422553426218394%5Cu0040base%22%2C%22bsb%3A422553426218394%5Cu0040diff%22]%7D&unigraph_project=UnigraphProjectMbex&is_mbex_redirected

Reviewed By: iseeyuan

Differential Revision: D33673958

fbshipit-source-id: 8600c04ae929283681971aae264d3774188df9cd
(cherry picked from commit 64ebcec09e)

aten/src/ATen/ParallelNative.cpp

@@ -308,7 +308,7 @@ c10::intrusive_ptr<c10::ivalue::Future> intraop_launch_future(
 #else
   // TODO: caffe2::PThreadPool only provides a data-parallel API.
   // Task parallelism is not currently supported.
-  auto future = c10::make_intrusive<c10::ivalue::Future>(NoneType::get());
+  auto future = c10::make_intrusive<c10::ivalue::Future>(c10::dynT<NoneType>());
   func();
   future->markCompleted();
   return future;

aten/src/ATen/core/custom_class.cpp

@@ -3,6 +3,7 @@
 #include <ATen/core/jit_type.h>
 #include <ATen/core/function_schema.h>
 #include <ATen/core/functional.h>
+#include <ATen/core/type_factory.h>
 
 #include <atomic>
 #include <unordered_map>
@@ -102,7 +103,7 @@ class_base::class_base(
 {
     detail::checkValidIdent(namespaceName, "Namespace name");
     detail::checkValidIdent(className, "Class name");
-    classTypePtr->addAttribute("capsule", at::CapsuleType::get());
+    classTypePtr->addAttribute("capsule", c10::TypeFactory::get<c10::CapsuleType>());
     c10::getCustomClassTypeMap().insert(
         {std::type_index(intrusivePtrClassTypeid), classTypePtr});
     c10::getCustomClassTypeMap().insert(

aten/src/ATen/core/dynamic_type.cpp

@@ -2,6 +2,7 @@
 
 #include <string>
 
+#include <ATen/core/class_type.h>
 #include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type.h>
 #include <ATen/core/type_factory.h>
@@ -198,6 +199,11 @@ TypePtr DynamicType::containedType(size_t i) const {
   return arguments_.elems.at(i).ty;
 }
 
+size_t DynamicType::containedTypeSize() const {
+  TORCH_INTERNAL_ASSERT(tag_ != Tag::Class);
+  return arguments_.elems.size();
+}
+
 TypeKind DynamicType::dynamicKind() const {
   switch (tag_) {
 #define CASE_TYPE(T, _, __) \
@@ -271,6 +277,16 @@ TypePtr DynamicType::fallback() const {
       return VarType::create(*name_);
     case Tag::AnyClass:
       return AnyClassType::get();
+    case Tag::QScheme:
+      return QSchemeType::get();
+    case Tag::Quantizer:
+      return QuantizerType::get();
+    case Tag::AnyEnum:
+      return AnyEnumType::get();
+    case Tag::RRef:
+      return RRefType::create(arguments_.elems[0].ty->fallback());
+    case Tag::Future:
+      return FutureType::create(arguments_.elems[0].ty->fallback());
     case Tag::Any:
       return AnyType::get();
   }

aten/src/ATen/core/dynamic_type.h

@@ -3,8 +3,6 @@
 #include <memory>
 #include <type_traits>
 
-#include <ATen/core/class_type.h>
-#include <ATen/core/ivalue.h>
 #include <ATen/core/jit_type_base.h>
 #include <c10/util/Optional.h>
 
@@ -53,8 +51,17 @@ constexpr DynamicTypeBits kDynamicClassTypeBit = DYNAMIC_TYPE_BIT(10);
   _(Storage, DYNAMIC_TYPE_BIT(16), 1)                                        \
   _(Var, DYNAMIC_TYPE_BIT(17), 0)                                            \
   _(AnyClass, (kDynamicClassTypeBit | kDynamicAnyTypeBit), 1)                \
+  _(QScheme, DYNAMIC_TYPE_BIT(18), 1)                                        \
+  _(Quantizer, DYNAMIC_TYPE_BIT(19), 1)                                      \
+  _(AnyEnum, DYNAMIC_TYPE_BIT(20), 1)                                        \
+  _(RRef, DYNAMIC_TYPE_BIT(21), 0)                                           \
+  _(Future, DYNAMIC_TYPE_BIT(22), 0)                                         \
   _(Any, 0xffffffff, 1)
 
+#define FORWARD_DECL_TYPE(NAME, _, __) struct NAME ## Type;
+  FORALL_DYNAMIC_TYPES(FORWARD_DECL_TYPE)
+#undef FORWARD_DECL_TYPE
+
 class DynamicType;
 using DynamicTypePtr = std::shared_ptr<DynamicType>;
 
@@ -142,6 +149,7 @@ class DynamicType : public SharedType {
   explicit DynamicType(Tag, c10::string_view, Arguments);
 
   TypePtr containedType(size_t) const override;
+  size_t containedTypeSize() const override;
   Tag tag() const {
     return tag_;
   }
@@ -154,6 +162,9 @@ class DynamicType : public SharedType {
   TypeKind dynamicKind() const;
 
   // Should be used only on the server side to restore static type information.
+#ifndef C10_MOBILE
+  TORCH_API
+#endif
   TypePtr fallback() const;
 
  private:
@@ -188,7 +199,7 @@ class DynamicType : public SharedType {
 
 template <typename T>
 struct DynamicTypeTrait {
-  static auto tagValue() {
+  C10_NOINLINE static auto tagValue() {
     TORCH_CHECK(false);
     return DynamicType::Tag::Any;
   }
@@ -201,7 +212,7 @@ C10_NOINLINE DynamicTypePtr makeBaseType(DynamicType::Tag tag);
 #define DYNAMIC_TYPE_TAG_VALUE(NAME, _, IS_BASE_TYPE)      \
   template <>                                              \
   struct TORCH_API DynamicTypeTrait<NAME##Type> {          \
-    static auto tagValue() {                               \
+    C10_ERASE static auto tagValue() {                     \
       return DynamicType::Tag::NAME;                       \
     }                                                      \
     static constexpr bool isBaseType = IS_BASE_TYPE;       \
@@ -214,19 +225,4 @@ C10_NOINLINE DynamicTypePtr makeBaseType(DynamicType::Tag tag);
 FORALL_DYNAMIC_TYPES(DYNAMIC_TYPE_TAG_VALUE)
 #undef DYNAMIC_TYPE_TAG_VALUE
 
-template <>
-struct IValue::TagType<c10::DynamicType> {
-  static DynamicType::Ptr get(const c10::IValue& v);
-};
-
-namespace ivalue {
-
-template <>
-struct TORCH_API TupleTypeFactory<c10::DynamicType> {
-  static DynamicTypePtr create(std::vector<TypePtr> elemTypes);
-  static DynamicTypePtr fallback(const Type&);
-};
-
-} // namespace ivalue
-
 } // namespace c10

aten/src/ATen/core/function_schema.h

@@ -390,7 +390,7 @@ struct FunctionSchema {
 
   // Check that inputs have the correct types and appends any missing default
   // values.
-  template <typename T = c10::Type>
+  template <typename T = c10::PlatformType>
   void checkAndNormalizeInputs(
       std::vector<IValue>& inputs,
       const std::unordered_map<std::string, IValue>& kwargs =

aten/src/ATen/core/function_schema_inl.h

@@ -293,7 +293,7 @@ inline void FunctionSchema::checkArg(
     TORCH_CHECK(
         false,
         formatTypeMismatchMsg(
-            argument, value.type()->repr_str(), pos));
+            argument, value.type<T>()->repr_str(), pos));
   }
 }
 

aten/src/ATen/core/ivalue.cpp

@@ -6,6 +6,7 @@
 #include <ATen/core/function.h>
 #include <ATen/core/jit_type.h>
 #include <ATen/core/stack.h>
+#include <ATen/core/type_factory.h>
 #include <c10/util/irange.h>
 #include <c10/util/StringUtil.h>
 #include <c10/util/hash.h>
@@ -403,6 +404,39 @@ bool IValue::is(const IValue& rhs) const {
   return lhs == rhs;
 }
 
+template <typename T>
+inline bool IValue::isListOf() const {
+  // note: avoids calling type() to avoid extra referencing counting for the returned type.
+  if (!isList()) {
+    return false;
+  }
+  const auto& ty = static_cast<detail::ListImpl*>(payload.u.as_intrusive_ptr)->elementType;
+  if (ty->kind() == T::Kind) {
+    return true;
+  }
+  return *ty == *TypeFactory::get<T>();
+}
+
+bool IValue::isDoubleList() const {
+  return isListOf<c10::FloatType>();
+}
+
+bool IValue::isComplexDoubleList() const {
+  return isListOf<c10::ComplexType>();
+}
+
+bool IValue::isTensorList() const {
+  return isListOf<c10::TensorType>();
+}
+
+bool IValue::isIntList() const {
+  return isListOf<c10::IntType>();
+}
+
+bool IValue::isBoolList() const {
+  return isListOf<c10::BoolType>();
+}
+
 namespace {
 
 using IValueFormatter = std::function<void(std::ostream&, const IValue&)>;
@@ -430,7 +464,7 @@ std::ostream& printMaybeAnnotatedList(
     std::ostream& out,
     const IValue& the_list,
     IValueFormatter formatter) {
-  auto list_elem_type = the_list.type()->expectRef<ListType>().getElementType();
+  auto list_elem_type = the_list.type()->containedType(0);
   if (the_list.toListRef().size() == 0 ||
       !elementTypeCanBeInferredFromMembers(list_elem_type)) {
     out << "annotate(" << the_list.type()->annotation_str() << ", ";
@@ -925,7 +959,7 @@ c10::intrusive_ptr<ivalue::Object> ivalue::Object::deepcopy(IValue::HashAliasedI
   auto cu = type_.cu_;
   auto object = ivalue::Object::create(WeakOrStrongTypePtr(type_.cu_, type_.type_), type()->numAttributes());
   for (const auto i : c10::irange(slots_.size())) {
-    if (slots_[i].type() == c10::CapsuleType::get()) {
+    if (*slots_[i].type() == *c10::TypeFactory::get<CapsuleType>()) {
       // If we've gotten here, it means that we have *not* copied this
       // class via __getstate__ and __setstate__. That fact and the
       // fact that we have a Capsule attribute mean that this is a

aten/src/ATen/core/ivalue.h

@@ -6,6 +6,7 @@
 #include <ATen/core/custom_class.h>
 #include <ATen/core/ivalue_to.h>
 #include <ATen/core/jit_type_base.h>
+#include <ATen/core/type_factory.h>
 #include <c10/util/C++17.h>
 #include <c10/util/MaybeOwned.h>
 #include <c10/util/intrusive_ptr.h>
@@ -895,8 +896,8 @@ public:
     }
   }
 
-  template <typename T = c10::Type>
+  template <typename T = c10::PlatformType>
-  typename T::Ptr type() const;
+  TypePtr type() const;
 
   // Detect aliased tensors.
   struct HashAliasedIValue {

aten/src/ATen/core/ivalue_inl.h

@@ -586,6 +586,12 @@ struct TORCH_API TupleTypeFactory<TupleType> {
   static TupleTypePtr fallback(const Type& type);
 };
 
+template <>
+struct TORCH_API TupleTypeFactory<c10::DynamicType> {
+  static DynamicTypePtr create(std::vector<TypePtr> elemTypes);
+  static DynamicTypePtr fallback(const Type&);
+};
+
 struct TORCH_API Tuple : c10::intrusive_ptr_target {
  private:
   TupleElements elements_;
@@ -1915,39 +1921,6 @@ inline ivalue::Tuple& IValue::toTupleRef() const {
       payload.u.as_intrusive_ptr);
 }
 
-template <typename T>
-inline bool IValue::isListOf() const {
-  // note: avoids calling type() to avoid extra referencing counting for the returned type.
-  if (!isList()) {
-    return false;
-  }
-  const auto& ty = static_cast<detail::ListImpl*>(payload.u.as_intrusive_ptr)->elementType;
-  if (ty->kind() == T::Kind) {
-    return true;
-  }
-  return *ty == *T::get();
-}
-
-inline bool IValue::isDoubleList() const {
-  return isListOf<c10::FloatType>();
-}
-
-inline bool IValue::isComplexDoubleList() const {
-  return isListOf<c10::ComplexType>();
-}
-
-inline bool IValue::isTensorList() const {
-  return isListOf<c10::TensorType>();
-}
-
-inline bool IValue::isIntList() const {
-  return isListOf<c10::IntType>();
-}
-
-inline bool IValue::isBoolList() const {
-  return isListOf<c10::BoolType>();
-}
-
 inline IValue::IValue(c10::intrusive_ptr<ivalue::Tuple> v)
     : tag(Tag::Tuple), is_intrusive_ptr(true) {
   payload.u.as_intrusive_ptr = null_to_undefined_tensor(v.release());
@@ -2285,8 +2258,13 @@ struct IValue::TagType<c10::Type> {
   static TORCH_API c10::TypePtr get(const IValue&);
 };
 
+template <>
+struct IValue::TagType<c10::DynamicType> {
+  static TORCH_API c10::TypePtr get(const IValue&);
+};
+
 template <typename T>
-typename T::Ptr IValue::type() const {
+TypePtr IValue::type() const {
   return IValue::TagType<T>::get(*this);
 }
 

aten/src/ATen/core/jit_type.h

@@ -5,6 +5,7 @@
 #include <ATen/core/TensorBody.h>
 #include <ATen/core/functional.h>
 #include <ATen/core/symbol.h>
+#include <ATen/core/type_factory.h>
 #include <ATen/core/qualified_name.h>
 #include <c10/util/TypeList.h>
 #include <c10/util/Optional.h>
@@ -1730,7 +1731,8 @@ struct getTypePtr_<c10::QScheme> final {
 template <>
 struct getTypePtr_<at::Generator> final {
   static decltype(auto) call() {
-    return OptionalType::create(GeneratorType::get());
+    return TypeFactory::create<OptionalType>(
+        TypeFactory::get<GeneratorType>());
   }
 };
 template <>
@@ -1798,7 +1800,8 @@ struct getTypePtr_<c10::Dict<K, V>> final {
 template <class T>
 struct getTypePtr_<at::optional<T>> final {
   static const auto& call() {
-    static auto type = OptionalType::create(getTypePtr_<T>::call());
+    static auto type = TypeFactory::create<OptionalType>(
+        getTypePtr_<T>::call());
     return type;
   }
 };

aten/src/ATen/core/jit_type_base.h

@@ -558,6 +558,9 @@ struct TORCH_API Type {
   virtual TypePtr containedType(size_t i) const {
     return containedTypes().at(i);
   }
+  virtual size_t containedTypeSize() const {
+    return containedTypes().size();
+  }
   // create a new version of this type, replacing its contained types with
   // contained_types
   TypePtr withContained(std::vector<TypePtr> contained_types);

aten/src/ATen/core/type_factory.cpp

@@ -1,5 +1,7 @@
 #include <ATen/core/type_factory.h>
 
+#include <ATen/core/jit_type.h>
+
 namespace c10 {
 
 // Dtype constraints are not constrained in compilation. Therefore, we map
@@ -56,4 +58,11 @@ const std::unordered_map<std::string, c10::TypePtr>& DefaultTypeFactory::
   return map;
 }
 
+c10::TypePtr DefaultTypeFactory::createNamedTuple(
+    const std::string& name,
+    const std::vector<c10::string_view>& fields,
+    const std::vector<c10::TypePtr>& types) {
+  return c10::TupleType::createNamed(name, fields, types);
+}
+
 } // namespace c10

aten/src/ATen/core/type_factory.h

@@ -1,12 +1,19 @@
 #pragma once
 
-#include <ATen/core/dynamic_type.h>
-#include <ATen/core/jit_type.h>
 #include <type_traits>
+#include <unordered_map>
+
+#include <ATen/core/dynamic_type.h>
+#include <ATen/core/jit_type_base.h>
+#include <c10/macros/Macros.h>
 
 namespace c10 {
 
-struct TORCH_API DynamicTypeFactory {
+template <typename T>
+struct TORCH_API TypeFactoryBase {};
+
+template <>
+struct TORCH_API TypeFactoryBase<c10::DynamicType> {
   template <typename T, typename... Args>
   static c10::DynamicTypePtr create(TypePtr ty, Args&&... args) {
     return std::make_shared<c10::DynamicType>(
@@ -29,26 +36,40 @@ struct TORCH_API DynamicTypeFactory {
         name,
         c10::DynamicType::Arguments(fields, types));
   }
+  template <typename T>
+  C10_ERASE static c10::DynamicTypePtr createNamed(const std::string& name) {
+    return std::make_shared<c10::DynamicType>(
+        c10::DynamicTypeTrait<T>::tagValue(),
+        name,
+        c10::DynamicType::Arguments{});
+  }
+  template <typename T>
+  C10_ERASE static c10::DynamicTypePtr get() {
+    return DynamicTypeTrait<T>::getBaseType();
+  }
   static const std::unordered_map<std::string, c10::TypePtr>& basePythonTypes();
 };
 
+using DynamicTypeFactory = TypeFactoryBase<c10::DynamicType>;
+
 // Helper functions for constructing DynamicTypes inline.
 template <
     typename T,
     std::enable_if_t<DynamicTypeTrait<T>::isBaseType, int> = 0>
-DynamicTypePtr dynT() {
+C10_ERASE DynamicTypePtr dynT() {
-  return DynamicTypeTrait<T>::getBaseType();
+  return DynamicTypeFactory::get<T>();
 }
 
 template <
     typename T,
     typename... Args,
     std::enable_if_t<!DynamicTypeTrait<T>::isBaseType, int> = 0>
-DynamicTypePtr dynT(Args&&... args) {
+C10_ERASE DynamicTypePtr dynT(Args&&... args) {
   return DynamicTypeFactory::create<T>(std::forward<Args>(args)...);
 }
 
-struct TORCH_API DefaultTypeFactory {
+template <>
+struct TORCH_API TypeFactoryBase<c10::Type> {
   template <typename T, typename... Args>
   static c10::TypePtr create(TypePtr ty, Args&&... args) {
     return T::create(std::move(ty), std::forward<Args>(args)...);
@@ -60,18 +81,28 @@ struct TORCH_API DefaultTypeFactory {
   static c10::TypePtr createNamedTuple(
       const std::string& name,
       const std::vector<c10::string_view>& fields,
-      const std::vector<c10::TypePtr>& types) {
+      const std::vector<c10::TypePtr>& types);
-    return c10::TupleType::createNamed(name, fields, types);
+  template <typename T>
+  C10_ERASE static c10::TypePtr createNamed(const std::string& name) {
+    return T::create(name);
   }
   static const std::unordered_map<std::string, c10::TypePtr>& basePythonTypes();
+  template <typename T>
+  C10_ERASE static c10::TypePtr get() {
+    return T::get();
+  }
 };
 
-using TypeFactory =
+using DefaultTypeFactory = TypeFactoryBase<c10::Type>;
+
+using PlatformType =
 #ifdef C10_MOBILE
-    DynamicTypeFactory
+    c10::DynamicType
 #else
-    DefaultTypeFactory
+    c10::Type
 #endif
     ;
 
+using TypeFactory = TypeFactoryBase<PlatformType>;
+
 } // namespace c10

c10/macros/Macros.h

@@ -225,6 +225,16 @@ using namespace c10::hip;
 #define C10_ALWAYS_INLINE inline
 #endif
 
+#if defined(_MSC_VER)
+#define C10_ATTR_VISIBILITY_HIDDEN
+#elif defined(__GNUC__)
+#define C10_ATTR_VISIBILITY_HIDDEN __attribute__((__visibility__("hidden")))
+#else
+#define C10_ATTR_VISIBILITY_HIDDEN
+#endif
+
+#define C10_ERASE C10_ALWAYS_INLINE C10_ATTR_VISIBILITY_HIDDEN
+
 // C10_FALLTHROUGH - Annotate fallthrough to the next case in a switch.
 #if C10_HAS_CPP_ATTRIBUTE(fallthrough)
 #define C10_FALLTHROUGH [[fallthrough]]

torch/csrc/jit/frontend/schema_matching.cpp

@@ -16,6 +16,9 @@ namespace torch {
 namespace jit {
 
 static inline TypePtr unwrapOptional(TypePtr opt_type) {
+  if (auto dyn = opt_type->castRaw<c10::DynamicType>()) {
+    return unwrapOptional(dyn->fallback());
+  }
   if (auto unwrap_list_type = opt_type->cast<OptionalType>()) {
     return unwrap_list_type->getElementType();
   }
@@ -282,12 +285,17 @@ static bool varargsCanBeUsedAsList(
   bool is_last_argument = arg_index + 1 == schema.arguments().size() ||
       schema.arguments()[arg_index + 1].kwarg_only();
 
+  auto arg_type = arg.type();
+  if (auto dyn = arg_type->castRaw<c10::DynamicType>()) {
+    arg_type = dyn->fallback();
+  }
+
   // The formal must be a list
-  bool argument_is_list = arg.type()->kind() == TypeKind::ListType;
+  bool argument_is_list = arg_type->kind() == TypeKind::ListType;
 
   // matching varargs of typevar list nyi
   bool typevar_list = argument_is_list &&
-      arg.type()->castRaw<ListType>()->getElementType()->cast<VarType>();
+      arg_type->castRaw<ListType>()->getElementType()->cast<VarType>();
 
   // it must not be a broadcasting list like int[3],
   // otherwise a single int is a valid input

torch/csrc/jit/frontend/schema_type_parser.cpp

@@ -41,32 +41,33 @@ namespace jit {
 
 TypePtr SchemaTypeParser::parseBaseType() {
   static std::unordered_map<std::string, TypePtr> type_map = {
-      {"Generator", GeneratorType::get()},
+      {"Generator", c10::TypeFactory::get<GeneratorType>()},
-      {"Dimname", StringType::get()},
+      {"Dimname", c10::TypeFactory::get<StringType>()},
-      {"ScalarType", IntType::get()},
+      {"ScalarType", c10::TypeFactory::get<IntType>()},
-      {"Layout", IntType::get()},
+      {"Layout", c10::TypeFactory::get<IntType>()},
-      {"MemoryFormat", IntType::get()},
+      {"MemoryFormat", c10::TypeFactory::get<IntType>()},
-      {"Storage", StorageType::get()},
+      {"Storage", c10::TypeFactory::get<StorageType>()},
-      {"QScheme", QSchemeType::get()},
+      {"QScheme", c10::TypeFactory::get<QSchemeType>()},
-      {"Quantizer", QuantizerType::get()},
+      {"Quantizer", c10::TypeFactory::get<QuantizerType>()},
       {"ConstQuantizerPtr",
-       IntType::get()}, // TODO This type should be removed from the schema
+       c10::TypeFactory::get<IntType>()}, // TODO This type should be removed
-                        // parser, it should use the custom class mechanism
+                                          // from the schema parser, it should
+                                          // use the custom class mechanism
                                           // instead. @jerryzh
-      {"Device", DeviceObjType::get()},
+      {"Device", c10::TypeFactory::get<DeviceObjType>()},
-      {"Stream", StreamObjType::get()},
+      {"Stream", c10::TypeFactory::get<StreamObjType>()},
-      {"Scalar", NumberType::get()},
+      {"Scalar", c10::TypeFactory::get<NumberType>()},
-      {"str", StringType::get()},
+      {"str", c10::TypeFactory::get<StringType>()},
-      {"float", FloatType::get()},
+      {"float", c10::TypeFactory::get<FloatType>()},
-      {"complex", ComplexType::get()},
+      {"complex", c10::TypeFactory::get<ComplexType>()},
-      {"int", IntType::get()},
+      {"int", c10::TypeFactory::get<IntType>()},
-      {"bool", BoolType::get()},
+      {"bool", c10::TypeFactory::get<BoolType>()},
-      {"None", NoneType::get()},
+      {"None", c10::TypeFactory::get<NoneType>()},
-      {"NoneType", NoneType::get()},
+      {"NoneType", c10::TypeFactory::get<NoneType>()},
-      {"Capsule", CapsuleType::get()},
+      {"Capsule", c10::TypeFactory::get<CapsuleType>()},
-      {"Any", at::AnyType::get()},
+      {"Any", c10::TypeFactory::get<c10::AnyType>()},
-      {"AnyClassType", at::AnyClassType::get()},
+      {"AnyClassType", c10::TypeFactory::get<c10::AnyClassType>()},
-      {"AnyEnumType", at::AnyEnumType::get()},
+      {"AnyEnumType", c10::TypeFactory::get<c10::AnyEnumType>()},
   };
   auto tok = L.cur();
   if (!L.nextIf(TK_NONE) && !L.nextIf(TK_NONE_TYPE)) {
@@ -79,7 +80,7 @@ TypePtr SchemaTypeParser::parseBaseType() {
     if (text.size() > 0 && islower(text[0])) {
       // lower case identifiers that are not otherwise valid types
       // are treated as type variables
-      return VarType::create(text);
+      return c10::TypeFactory::createNamed<VarType>(text);
     }
     throw ErrorReport(tok.range) << "unknown type specifier";
   }
@@ -313,7 +314,7 @@ std::pair<TypePtr, c10::optional<AliasInfo>> SchemaTypeParser::parseType() {
         alias_info->addContainedType(std::move(*r.second));
       }
     });
-    value = TupleType::create(std::move(types));
+    value = c10::TypeFactory::create<TupleType>(std::move(types));
   } else if (L.cur().kind == TK_IDENT && L.cur().text() == "Future") {
     L.next(); // Future
     L.expect('(');
@@ -321,7 +322,7 @@ std::pair<TypePtr, c10::optional<AliasInfo>> SchemaTypeParser::parseType() {
     auto subtype = std::move(p.first);
     auto subalias = std::move(p.second);
     L.expect(')');
-    value = FutureType::create(subtype);
+    value = c10::TypeFactory::create<FutureType>(subtype);
   } else if (L.cur().kind == TK_IDENT && L.cur().text() == "RRef") {
     L.next(); // RRef
     L.expect('(');
@@ -329,10 +330,10 @@ std::pair<TypePtr, c10::optional<AliasInfo>> SchemaTypeParser::parseType() {
     auto subtype = std::move(p.first);
     auto subalias = std::move(p.second);
     L.expect(')');
-    value = RRefType::create(subtype);
+    value = c10::TypeFactory::create<RRefType>(subtype);
   } else if (L.cur().kind == TK_IDENT && L.cur().text() == "Tensor") {
     L.next();
-    value = TensorType::get();
+    value = c10::TypeFactory::get<TensorType>();
     alias_info = parseAliasAnnotation();
   } else if (L.cur().kind == TK_IDENT && L.cur().text() == "Dict") {
     L.next();
@@ -342,7 +343,7 @@ std::pair<TypePtr, c10::optional<AliasInfo>> SchemaTypeParser::parseType() {
     auto value_type = parseType().first;
     L.expect(')');
     alias_info = parseAliasAnnotation();
-    value = DictType::create(key_type, value_type);
+    value = c10::TypeFactory::create<DictType>(key_type, value_type);
   } else if (L.cur().kind == TK_IDENT && L.cur().text() == "Union") {
     L.next();
     L.expect('(');
@@ -395,7 +396,7 @@ std::pair<TypePtr, c10::optional<AliasInfo>> SchemaTypeParser::parseType() {
     if (L.cur().kind == '[' && L.lookahead().kind == ']') {
       L.next(); // [
       L.next(); // ]
-      value = ListType::create(value);
+      value = c10::TypeFactory::create<ListType>(value);
       auto container = parseAliasAnnotation();
       if (container && alias_info) {
         container->addContainedType(std::move(*alias_info));

torch/csrc/jit/ir/ir.h

@@ -1485,6 +1485,9 @@ inline Value::Value(Node* node_, size_t offset_)
 
 inline Value* Value::setType(TypePtr type) {
   AT_ASSERT(type);
+  if (auto dyn = type->castRaw<c10::DynamicType>()) {
+    type = dyn->fallback();
+  }
   type_ = std::move(type);
   for (Use& use : uses_) {
     use.user->op_ = nullptr;

torch/csrc/jit/serialization/export_module.cpp

@@ -108,7 +108,11 @@ std::pair<IValue, IValue> getFunctionTuple(
   static const std::string torch_prefix("__torch__");
   static const std::string class_prefix("__torch__.torch.classes");
 
-  for (const TypePtr& t : mobile_code.types_) {
+  for (const TypePtr& ty : mobile_code.types_) {
+    auto t = ty;
+    if (auto dyn = t->castRaw<c10::DynamicType>()) {
+      t = dyn->fallback();
+    }
     std::string type_str = t->annotation_str();
     if (t->kind() == TypeKind::TupleType) {
       TORCH_CHECK(
@@ -216,9 +220,13 @@ std::pair<IValue, IValue> getFunctionTuple(
         arg.type()->annotation_str(type_printer) => mangled unique name of the
         module/submodule
       */
+      auto arg_type = arg.type();
+      if (auto dyn = arg_type->castRaw<c10::DynamicType>()) {
+        arg_type = dyn->fallback();
+      }
       argTables.emplace_back(Table({
           {"name", arg.name()},
-          {"type", arg.type()->annotation_str(type_printer)},
+          {"type", arg_type->annotation_str(type_printer)},
           {"default_value", arg.default_value()},
       }));
     }

torch/csrc/jit/serialization/pickler.cpp

@@ -575,7 +575,11 @@ void Pickler::endTypeTag(const IValue& ivalue) {
 
   // Push the dict type
   TORCH_INTERNAL_ASSERT(ivalue.type());
-  pushString(ivalue.type()->annotation_str());
+  auto type = ivalue.type();
+  if (auto dyn = type->castRaw<c10::DynamicType>()) {
+    type = dyn->fallback();
+  }
+  pushString(type->annotation_str());
 
   // Pop the dict and type into a tuple
   push<PickleOpCode>(PickleOpCode::TUPLE2);

torch/csrc/jit/serialization/unpickler.cpp

@@ -34,7 +34,7 @@ static void restoreAccurateTypeTagsIfPossible(const IValue& root) {
 // of the contained objects and cannot restore the tags.
 void restoreAccurateTypeTags(const IValue& root, const TypePtr& type_tag) {
   struct Work {
-    TypePtr static_type;
+    TypePtr type;
     IValue value;
   };
   std::vector<Work> to_process = {{type_tag, root}};
@@ -53,7 +53,11 @@ void restoreAccurateTypeTags(const IValue& root, const TypePtr& type_tag) {
       }
       scanned.emplace_hint(it, key);
     }
-    switch (w.static_type->kind()) {
+    auto kind = w.type->kind();
+    if (auto dyn = w.type->castRaw<c10::DynamicType>()) {
+      kind = dyn->dynamicKind();
+    }
+    switch (kind) {
       case TensorType::Kind:
       case StorageType::Kind:
       case NumberType::Kind:
@@ -83,52 +87,37 @@ void restoreAccurateTypeTags(const IValue& root, const TypePtr& type_tag) {
         // no op, there is nothing to tag
         break;
       case DynamicType::Kind:
+      case UnionType::Kind:
       case EnumType::Kind:
         // TODO(gmagogsfm): Implement serialization/deserialization of Enum.
         TORCH_INTERNAL_ASSERT(false);
       case TupleType::Kind: {
         auto t = w.value.toTuple();
-        auto ttype = w.static_type->expect<TupleType>();
+        for (size_t i = 0; i < w.type->containedTypeSize(); ++i) {
-        for (size_t i = 0; i < ttype->containedTypes().size(); ++i) {
+          Work elem = {w.type->containedType(i), t->elements().at(i)};
-          Work elem = {ttype->containedTypes().at(i), t->elements().at(i)};
           to_process.emplace_back(std::move(elem));
         }
       } break;
       case FutureType::Kind: {
         auto f = w.value.toFuture();
-        auto t = w.static_type->expect<FutureType>();
         if (f->completed()) {
-          Work elem = {t->getElementType(), f->value()};
+          Work elem = {w.type->containedType(0), f->value()};
           to_process.emplace_back(std::move(elem));
         }
       } break;
       case OptionalType::Kind: {
         if (!w.value.isNone()) {
-          auto t = w.static_type->expect<OptionalType>();
+          Work elem = {w.type->containedType(0), w.value};
-          Work elem = {t->getElementType(), w.value};
           to_process.emplace_back(std::move(elem));
         }
       } break;
-      case UnionType::Kind: {
-        auto t = w.static_type->expect<UnionType>();
-        if (t->containedTypes().size() == 2 &&
-            t->canHoldType(*NoneType::get())) {
-          if (!w.value.isNone()) {
-            auto inner = t->containedTypes()[0] != NoneType::get()
-                ? t->containedTypes()[0]
-                : t->containedTypes()[1];
-            Work elem = {inner, w.value};
-            to_process.emplace_back(std::move(elem));
-          }
-        }
-      } break;
       case ListType::Kind: {
         // specialized lists do not need their type refined, so we can exit
         // early here
         if (!w.value.isList()) {
           break;
         }
-        auto elem_type = w.static_type->castRaw<ListType>()->getElementType();
+        auto elem_type = w.type->containedType(0);
         auto lst = w.value.toList();
         lst.unsafeSetElementType(elem_type);
         for (const IValue item : lst) {
@@ -137,13 +126,14 @@ void restoreAccurateTypeTags(const IValue& root, const TypePtr& type_tag) {
         }
       } break;
       case DictType::Kind: {
-        auto dt = w.static_type->cast<DictType>();
         auto d = w.value.toGenericDict();
-        d.unsafeSetKeyType(dt->getKeyType());
+        auto keyType = w.type->containedType(0);
-        d.unsafeSetValueType(dt->getValueType());
+        auto valType = w.type->containedType(1);
+        d.unsafeSetKeyType(keyType);
+        d.unsafeSetValueType(valType);
         for (const auto& item : d) {
-          Work kelem = {dt->getKeyType(), item.key()};
+          Work kelem = {keyType, item.key()};
-          Work velem = {dt->getValueType(), item.value()};
+          Work velem = {valType, item.value()};
           to_process.emplace_back(std::move(kelem));
           to_process.emplace_back(std::move(velem));
         }