[jit] Reduce refcounting of Types (pytorch#65345)

Summary:
Pull Request resolved: pytorch#65345

FooType::get() can return a const reference. Inconveniently, converting shared_ptr<FooType> to shared_ptr<Type> requires a copy & refcount bump, so to properly take advantage of this in unshapedType() we need to take a const Type& in isSubtypeOf(), which is good practice anyway -- don't require a shared_ptr if you don't need to take ownership.
ghstack-source-id: 140044165

Test Plan:
CI

perf says c10::unshapedType time decreased from 2.8% to 2.2% during static runtime startup, though I expect this to be generally beneficial.

Reviewed By: hlu1

Differential Revision: D31027361

fbshipit-source-id: 676feb81db9f74ad7b8651d8774f4ecb4cfa6ab8

Author: swolchok

aten/src/ATen/BatchedFallback.cpp

@@ -36,7 +36,7 @@ static bool areAnyArgumentsTensorList(const FunctionSchema& schema) {
   return std::any_of(
       schema.arguments().begin(),
       schema.arguments().end(),
-      [] (const Argument& arg) { return arg.type()->isSubtypeOf(ListType::ofTensors()); });
+      [] (const Argument& arg) { return arg.type()->isSubtypeOf(*ListType::ofTensors()); });
 }
 
 // Returns if an operator is in-place. An operator is inplace if:

aten/src/ATen/core/List_inl.h

@@ -63,7 +63,7 @@ List<T> toTypedList(impl::GenericList list) {
   // as List<Tensor> before we changed that argument to be List<optional<Tensor>>. When deserializing, we
   // have list.use_count() == 1 and can deserialize the List<Tensor> directly as List<optional<Tensor>>.
   TORCH_CHECK(*list.impl_->elementType == *getTypePtr<T>()
-    || (list.use_count() == 1 && list.impl_->elementType->isSubtypeOf(getTypePtr<T>()))
+    || (list.use_count() == 1 && list.impl_->elementType->isSubtypeOf(*getTypePtr<T>()))
     , "Tried to cast a List<", toString(list.impl_->elementType), "> to a List<", toString(getTypePtr<T>()), ">. Types mismatch.");
   return List<T>(std::move(list.impl_));
 }

aten/src/ATen/core/dispatch/DispatchKeyExtractor.h

@@ -172,13 +172,13 @@ struct TORCH_API DispatchKeyExtractor final {
         " arguments but this PyTorch build only supports ", c10::utils::bitset::NUM_BITS());
     c10::utils::bitset dispatch_arg_indices_reverse;
     for (size_t index = 0; index < schema.arguments().size(); ++index) {
-      if (schema.arguments()[index].type()->isSubtypeOf(TensorType::get()) ||
+      if (schema.arguments()[index].type()->isSubtypeOf(*TensorType::get()) ||
           schema.arguments()[index].type()->isSubtypeOf(
-              ListType::ofTensors()) ||
+              *ListType::ofTensors()) ||
           schema.arguments()[index].type()->isSubtypeOf(
-              ListType::ofOptionalTensors()) ||
+              *ListType::ofOptionalTensors()) ||
           schema.arguments()[index].type()->isSubtypeOf(
-              OptionalType::ofTensor())) {
+              *OptionalType::ofTensor())) {
         dispatch_arg_indices_reverse.set(schema.arguments().size() - 1 - index);
       }
     }

aten/src/ATen/core/function_schema_inl.h

@@ -76,7 +76,7 @@ inline bool Argument::isBackwardCompatibleWith(
     if (lhs->kwarg_only() && !rhs->kwarg_only()) {
       return false;
     }
-    if (!rhs->type()->isSubtypeOfExt(lhs->type(), why_not)) {
+    if (!rhs->type()->isSubtypeOfExt(*lhs->type(), why_not)) {
       return false;
     }
     if (rhs->default_value().has_value() &&
@@ -179,7 +179,7 @@ inline void FunctionSchema::checkArg(
     // Fast-path for the common case
     return;
   }
-  if (!value.type()->isSubtypeOf(argument.type())) {
+  if (!value.type()->isSubtypeOf(*argument.type())) {
     TORCH_CHECK(
         false,
         formatTypeMismatchMsg(
@@ -304,7 +304,7 @@ inline bool isSubtypeOfList(
     if (c.name() != p.name()) {
       return false;
     }
-    if (!c.type()->isSubtypeOfExt(p.type(), why_not)) {
+    if (!c.type()->isSubtypeOfExt(*p.type(), why_not)) {
       return false;
     }
   }