mirror of
https://github.com/archlinuxarm/PKGBUILDs.git
synced 2024-11-28 22:57:37 +00:00
174 lines
8 KiB
Diff
174 lines
8 KiB
Diff
From ce71348a09f6689dd01a68db64b172191d0182d8 Mon Sep 17 00:00:00 2001
|
|
From: Andrey Kosyakov <caseq@chromium.org>
|
|
Date: Thu, 21 Dec 2023 18:38:38 +0000
|
|
Subject: [PATCH] [bindings] Use v8::Array::Iterate for converting script
|
|
wrappables
|
|
|
|
This changes CreateIDLSequenceFromV8Array to use the new
|
|
v8::Array::Iterate() operation.
|
|
This speeds up the "execBundles" part of the microbenchmark
|
|
at crbug.com/dawn/1858 by around 3x.
|
|
This depends on crrev.com/c/4846594 landing (and rolling) first.
|
|
|
|
This is a slight re-work of https://crrev.com/c/4847447/3,
|
|
originally by jkummerow@chromium.org
|
|
|
|
Bug: v8:14218, dawn:1858, 1511239
|
|
Change-Id: Ia266556d05b4d53e6942e12609d1c08882b4ff0f
|
|
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/5132129
|
|
Commit-Queue: Andrey Kosyakov <caseq@chromium.org>
|
|
Reviewed-by: Yuki Shiino <yukishiino@chromium.org>
|
|
Cr-Commit-Position: refs/heads/main@{#1240236}
|
|
---
|
|
.../bindings/core/v8/native_value_traits.h | 6 ++
|
|
.../core/v8/native_value_traits_impl.h | 91 ++++++++++++++++++-
|
|
2 files changed, 95 insertions(+), 2 deletions(-)
|
|
|
|
diff --git a/third_party/blink/renderer/bindings/core/v8/native_value_traits.h b/third_party/blink/renderer/bindings/core/v8/native_value_traits.h
|
|
index 1e5a0790df6d..a5c28b37e945 100644
|
|
--- a/third_party/blink/renderer/bindings/core/v8/native_value_traits.h
|
|
+++ b/third_party/blink/renderer/bindings/core/v8/native_value_traits.h
|
|
@@ -84,6 +84,12 @@ struct NativeValueTraitsBase {
|
|
std::is_pointer_v<ImplType> ||
|
|
requires(ImplType value) { value.IsNull(); };
|
|
|
|
+ // This should only be true for certain subclasses of ScriptWrappable
|
|
+ // that satisfy the assumptions of CreateIDLSequenceFromV8ArraySlow() with
|
|
+ // regards to how NativeValue() is implemented for the underlying type.
|
|
+ static constexpr bool supports_scriptwrappable_specific_fast_array_iteration =
|
|
+ false;
|
|
+
|
|
template <typename... ExtraArgs>
|
|
static decltype(auto) ArgumentValue(v8::Isolate* isolate,
|
|
int argument_index,
|
|
diff --git a/third_party/blink/renderer/bindings/core/v8/native_value_traits_impl.h b/third_party/blink/renderer/bindings/core/v8/native_value_traits_impl.h
|
|
index 5011503dcf1c..f085b6e90516 100644
|
|
--- a/third_party/blink/renderer/bindings/core/v8/native_value_traits_impl.h
|
|
+++ b/third_party/blink/renderer/bindings/core/v8/native_value_traits_impl.h
|
|
@@ -1037,10 +1037,86 @@ CreateIDLSequenceFromV8ArraySlow(v8::Isolate* isolate,
|
|
return {};
|
|
}
|
|
|
|
- typename NativeValueTraits<IDLSequence<T>>::ImplType result;
|
|
+ using ResultType = typename NativeValueTraits<IDLSequence<T>>::ImplType;
|
|
+ ResultType result;
|
|
result.ReserveInitialCapacity(length);
|
|
v8::Local<v8::Context> current_context = isolate->GetCurrentContext();
|
|
v8::TryCatch try_block(isolate);
|
|
+
|
|
+ // Fast path -- we're creating a sequence of script wrappables, which can be
|
|
+ // done by directly getting underlying object as long as array types are
|
|
+ // homogeneous. With ScriptWrappables, we don't expect to enter JS during
|
|
+ // iteration, so we can rely on v8::Array::Iterate() which is much faster than
|
|
+ // iterating an array on the client side of the v8. Additionally, for most
|
|
+ // subsptyes of ScriptWrappables, we can speed up type checks (see more on
|
|
+ // that below next to supports_scriptwrappable_specific_fast_array_iteration
|
|
+ // check.
|
|
+ if constexpr (std::is_base_of_v<ScriptWrappable, T>) {
|
|
+ struct CallbackData {
|
|
+ STACK_ALLOCATED();
|
|
+
|
|
+ public:
|
|
+ v8::Isolate* isolate;
|
|
+ v8::TypecheckWitness witness;
|
|
+ ResultType& result;
|
|
+ ExceptionState& exception_state;
|
|
+ CallbackData(v8::Isolate* isolate,
|
|
+ ResultType& result,
|
|
+ ExceptionState& exception_state)
|
|
+ : isolate(isolate),
|
|
+ witness(isolate),
|
|
+ result(result),
|
|
+ exception_state(exception_state) {}
|
|
+ };
|
|
+
|
|
+ CallbackData callback_data(isolate, result, exception_state);
|
|
+ v8::Array::IterationCallback callback = [](uint32_t index,
|
|
+ v8::Local<v8::Value> v8_element,
|
|
+ void* data) {
|
|
+ CallbackData* callback_data = reinterpret_cast<CallbackData*>(data);
|
|
+ // 3.4. Initialize Si to the result of converting nextItem to an IDL value
|
|
+ // of type T.
|
|
+ v8::TypecheckWitness& witness = callback_data->witness;
|
|
+ // We can speed up type check by taking advantage of V8's type witness,
|
|
+ // provided traits' NativeValue implementation doesn't have additional
|
|
+ // logic beyond checking the type and calling ToScriptWrappable().
|
|
+ if constexpr (
|
|
+ NativeValueTraits<
|
|
+ T>::supports_scriptwrappable_specific_fast_array_iteration) {
|
|
+ if (witness.Matches(v8_element)) {
|
|
+ auto&& value = ToScriptWrappable(v8_element.As<v8::Object>())
|
|
+ ->template ToImpl<T>();
|
|
+ callback_data->result.push_back(std::move(value));
|
|
+ return v8::Array::CallbackResult::kContinue;
|
|
+ }
|
|
+ }
|
|
+ auto&& element = NativeValueTraits<T>::NativeValue(
|
|
+ callback_data->isolate, v8_element, callback_data->exception_state);
|
|
+ if (callback_data->exception_state.HadException()) {
|
|
+ // It doesn't matter whether we return `kException` or `kBreak` here,
|
|
+ // as that only affects the return value of `v8_array->Iterate()`,
|
|
+ // which we are ignoring.
|
|
+ return v8::Array::CallbackResult::kException;
|
|
+ }
|
|
+ if constexpr (
|
|
+ NativeValueTraits<
|
|
+ T>::supports_scriptwrappable_specific_fast_array_iteration) {
|
|
+ witness.Update(v8_element);
|
|
+ }
|
|
+ callback_data->result.push_back(std::move(element));
|
|
+ return v8::Array::CallbackResult::kContinue;
|
|
+ };
|
|
+ if (!v8_array->Iterate(current_context, callback, &callback_data)
|
|
+ .IsJust()) {
|
|
+ if (try_block.HasCaught()) {
|
|
+ exception_state.RethrowV8Exception(try_block.Exception());
|
|
+ }
|
|
+ DCHECK(exception_state.HadException());
|
|
+ return {};
|
|
+ }
|
|
+ return result;
|
|
+ }
|
|
+
|
|
// Array length may change if array is mutated during iteration.
|
|
for (uint32_t i = 0; i < v8_array->Length(); ++i) {
|
|
v8::Local<v8::Value> v8_element;
|
|
@@ -1056,6 +1132,7 @@ CreateIDLSequenceFromV8ArraySlow(v8::Isolate* isolate,
|
|
return {};
|
|
result.push_back(std::move(element));
|
|
}
|
|
+
|
|
// 3.2. If next is false, then return an IDL sequence value of type
|
|
// sequence<T> of length i, where the value of the element at index j is Sj.
|
|
return result;
|
|
@@ -1398,6 +1475,7 @@ struct NativeValueTraits<T> : public NativeValueTraitsBase<T*> {
|
|
}
|
|
};
|
|
|
|
+// Interface types
|
|
template <typename T>
|
|
requires std::derived_from<T, CallbackInterfaceBase>
|
|
struct NativeValueTraits<IDLNullable<T>>
|
|
@@ -1470,12 +1548,21 @@ struct NativeValueTraits<T> : public NativeValueTraitsBase<T> {
|
|
template <typename T>
|
|
requires std::derived_from<T, ScriptWrappable>
|
|
struct NativeValueTraits<T> : public NativeValueTraitsBase<T*> {
|
|
+ // This signifies that CreateIDLSequenceFromV8ArraySlow() may apply
|
|
+ // certain optimization based on assumptions about `NativeValue()`
|
|
+ // implementation below. For subclasses of ScriptWrappable that have
|
|
+ // different implementation of NativeValue(), this should remain false.
|
|
+ static constexpr bool supports_scriptwrappable_specific_fast_array_iteration =
|
|
+ true;
|
|
+
|
|
static inline T* NativeValue(v8::Isolate* isolate,
|
|
v8::Local<v8::Value> value,
|
|
ExceptionState& exception_state) {
|
|
const WrapperTypeInfo* wrapper_type_info = T::GetStaticWrapperTypeInfo();
|
|
- if (V8PerIsolateData::From(isolate)->HasInstance(wrapper_type_info, value))
|
|
+ if (V8PerIsolateData::From(isolate)->HasInstance(wrapper_type_info,
|
|
+ value)) {
|
|
return ToScriptWrappable(value.As<v8::Object>())->template ToImpl<T>();
|
|
+ }
|
|
|
|
bindings::NativeValueTraitsInterfaceNotOfType(wrapper_type_info,
|
|
exception_state);
|