ladybird/Libraries/LibJS/Runtime/GeneratorObject.cpp
Andreas Kling d065171791
Some checks are pending
CI / macOS, arm64, Sanitizer, Clang (push) Waiting to run
CI / Linux, x86_64, Fuzzers, Clang (push) Waiting to run
CI / Linux, x86_64, Sanitizer, GNU (push) Waiting to run
CI / Linux, x86_64, Sanitizer, Clang (push) Waiting to run
Package the js repl as a binary artifact / Linux, arm64 (push) Waiting to run
Package the js repl as a binary artifact / macOS, arm64 (push) Waiting to run
Package the js repl as a binary artifact / Linux, x86_64 (push) Waiting to run
Run test262 and test-wasm / run_and_update_results (push) Waiting to run
Lint Code / lint (push) Waiting to run
Label PRs with merge conflicts / auto-labeler (push) Waiting to run
Push notes / build (push) Waiting to run
LibJS: Use property lookup caches for some of our hot C++ gets
We can use caching in a million more places. This is just me running JS
benchmarks and looking at which get() call sites were hot and putting
caches there.

Lots of nice speedups all over the place, some examples:

1.19x speedup on Octane/raytrace.js
1.13x speedup on Octane/earley-boyer.js
1.12x speedup on Kraken/ai-astar.js
1.10x speedup on Octane/box2d.js
1.08x speedup on Octane/gbemu.js
1.05x speedup on Octane/regexp.js
2025-10-14 15:47:38 +02:00

226 lines
10 KiB
C++

/*
* Copyright (c) 2021, Ali Mohammad Pur <mpfard@serenityos.org>
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#include <AK/TemporaryChange.h>
#include <LibJS/Bytecode/Generator.h>
#include <LibJS/Bytecode/Interpreter.h>
#include <LibJS/Runtime/CompletionCell.h>
#include <LibJS/Runtime/GeneratorObject.h>
#include <LibJS/Runtime/GeneratorPrototype.h>
#include <LibJS/Runtime/GeneratorResult.h>
#include <LibJS/Runtime/GlobalObject.h>
#include <LibJS/Runtime/Iterator.h>
namespace JS {
GC_DEFINE_ALLOCATOR(GeneratorObject);
ThrowCompletionOr<GC::Ref<GeneratorObject>> GeneratorObject::create(Realm& realm, Value initial_value, ECMAScriptFunctionObject* generating_function, NonnullOwnPtr<ExecutionContext> execution_context)
{
auto& vm = realm.vm();
// This is "g1.prototype" in figure-2 (https://tc39.es/ecma262/img/figure-2.png)
Value generating_function_prototype;
if (generating_function->kind() == FunctionKind::Async) {
// We implement async functions by transforming them to generator function in the bytecode
// interpreter. However an async function does not have a prototype and should not be
// changed thus we hardcode the prototype.
generating_function_prototype = realm.intrinsics().generator_prototype();
} else {
static Bytecode::PropertyLookupCache cache;
generating_function_prototype = TRY(generating_function->get(vm.names.prototype, cache));
}
auto generating_function_prototype_object = TRY(generating_function_prototype.to_object(vm));
auto object = realm.create<GeneratorObject>(realm, generating_function_prototype_object, move(execution_context));
object->m_generating_function = generating_function;
object->m_previous_value = initial_value;
return object;
}
GeneratorObject::GeneratorObject(Realm&, Object& prototype, NonnullOwnPtr<ExecutionContext> context, Optional<StringView> generator_brand)
: Object(ConstructWithPrototypeTag::Tag, prototype)
, m_execution_context(move(context))
, m_generator_brand(move(generator_brand))
{
}
void GeneratorObject::visit_edges(Cell::Visitor& visitor)
{
Base::visit_edges(visitor);
visitor.visit(m_generating_function);
visitor.visit(m_previous_value);
m_execution_context->visit_edges(visitor);
}
// 27.5.3.2 GeneratorValidate ( generator, generatorBrand ), https://tc39.es/ecma262/#sec-generatorvalidate
ThrowCompletionOr<GeneratorObject::GeneratorState> GeneratorObject::validate(VM& vm, Optional<StringView> const& generator_brand)
{
// 1. Perform ? RequireInternalSlot(generator, [[GeneratorState]]).
// 2. Perform ? RequireInternalSlot(generator, [[GeneratorBrand]]).
// NOTE: Already done by the caller of resume or resume_abrupt, as they wouldn't have a GeneratorObject otherwise.
// 3. If generator.[[GeneratorBrand]] is not the same value as generatorBrand, throw a TypeError exception.
if (m_generator_brand != generator_brand)
return vm.throw_completion<TypeError>(ErrorType::GeneratorBrandMismatch, m_generator_brand.value_or("<empty>"sv), generator_brand.value_or("<empty>"sv));
// 4. Assert: generator also has a [[GeneratorContext]] internal slot.
// NOTE: Done by already being a GeneratorObject.
// 5. Let state be generator.[[GeneratorState]].
auto state = m_generator_state;
// 6. If state is executing, throw a TypeError exception.
if (state == GeneratorState::Executing)
return vm.throw_completion<TypeError>(ErrorType::GeneratorAlreadyExecuting);
// 7. Return state.
return state;
}
ThrowCompletionOr<GeneratorObject::IterationResult> GeneratorObject::execute(VM& vm, Completion const& completion)
{
// Loosely based on step 4 of https://tc39.es/ecma262/#sec-generatorstart mixed with https://tc39.es/ecma262/#sec-generatoryield at the end.
auto generated_value = [](Value value) -> Value {
if (value.is_cell())
return static_cast<GeneratorResult const&>(value.as_cell()).result();
return value.is_special_empty_value() ? js_undefined() : value;
};
auto generated_continuation = [&](Value value) -> Optional<size_t> {
if (value.is_cell()) {
auto number_value = static_cast<GeneratorResult const&>(value.as_cell()).continuation();
if (number_value.is_null())
return {};
return static_cast<u64>(number_value.as_double());
}
return {};
};
auto compleion_cell = heap().allocate<CompletionCell>(completion);
auto& bytecode_interpreter = vm.bytecode_interpreter();
auto const next_block = generated_continuation(m_previous_value);
// We should never enter `execute` again after the generator is complete.
VERIFY(next_block.has_value());
auto next_result = bytecode_interpreter.run_executable(*m_generating_function->bytecode_executable(), next_block, compleion_cell);
vm.pop_execution_context();
auto result_value = move(next_result.value);
if (result_value.is_throw_completion()) {
// Uncaught exceptions disable the generator.
m_generator_state = GeneratorState::Completed;
return result_value.throw_completion();
}
m_previous_value = result_value.release_value();
bool done = !generated_continuation(m_previous_value).has_value();
m_generator_state = done ? GeneratorState::Completed : GeneratorState::SuspendedYield;
return IterationResult(generated_value(m_previous_value), done);
}
// 27.5.3.3 GeneratorResume ( generator, value, generatorBrand ), https://tc39.es/ecma262/#sec-generatorresume
ThrowCompletionOr<GeneratorObject::IterationResult> GeneratorObject::resume(VM& vm, Value value, Optional<StringView> const& generator_brand)
{
// 1. Let state be ? GeneratorValidate(generator, generatorBrand).
auto state = TRY(validate(vm, generator_brand));
// 2. If state is completed, return CreateIterResultObject(undefined, true).
if (state == GeneratorState::Completed) {
return IterationResult(js_undefined(), true);
}
// 3. Assert: state is either suspendedStart or suspendedYield.
VERIFY(state == GeneratorState::SuspendedStart || state == GeneratorState::SuspendedYield);
// 4. Let genContext be generator.[[GeneratorContext]].
auto& generator_context = m_execution_context;
// 5. Let methodContext be the running execution context.
auto const& method_context = vm.running_execution_context();
// 6. Suspend methodContext.
// 8. Push genContext onto the execution context stack; genContext is now the running execution context.
// NOTE: This is done out of order as to not permanently disable the generator if push_execution_context throws,
// as `resume` will immediately throw when [[GeneratorState]] is "executing", never allowing the state to change.
TRY(vm.push_execution_context(*generator_context, {}));
// 7. Set generator.[[GeneratorState]] to executing.
m_generator_state = GeneratorState::Executing;
// 9. Resume the suspended evaluation of genContext using NormalCompletion(value) as the result of the operation that suspended it. Let result be the value returned by the resumed computation.
auto result = execute(vm, normal_completion(value));
// 10. Assert: When we return here, genContext has already been removed from the execution context stack and methodContext is the currently running execution context.
VERIFY(&vm.running_execution_context() == &method_context);
// 11. Return ? result.
return result;
}
// 27.5.3.4 GeneratorResumeAbrupt ( generator, abruptCompletion, generatorBrand ), https://tc39.es/ecma262/#sec-generatorresumeabrupt
ThrowCompletionOr<GeneratorObject::IterationResult> GeneratorObject::resume_abrupt(JS::VM& vm, JS::Completion abrupt_completion, Optional<StringView> const& generator_brand)
{
// 1. Let state be ? GeneratorValidate(generator, generatorBrand).
auto state = TRY(validate(vm, generator_brand));
// 2. If state is suspendedStart, then
if (state == GeneratorState::SuspendedStart) {
// a. Set generator.[[GeneratorState]] to completed.
m_generator_state = GeneratorState::Completed;
// b. Once a generator enters the completed state it never leaves it and its associated execution context is never resumed. Any execution state associated with generator can be discarded at this point.
// We don't currently discard anything.
// c. Set state to completed.
state = GeneratorState::Completed;
}
// 3. If state is completed, then
if (state == GeneratorState::Completed) {
// a. If abruptCompletion.[[Type]] is return, then
if (abrupt_completion.type() == Completion::Type::Return) {
// i. Return CreateIterResultObject(abruptCompletion.[[Value]], true).
return IterationResult(abrupt_completion.value(), true);
}
// b. Return ? abruptCompletion.
return abrupt_completion;
}
// 4. Assert: state is suspendedYield.
VERIFY(state == GeneratorState::SuspendedYield);
// 5. Let genContext be generator.[[GeneratorContext]].
auto& generator_context = m_execution_context;
// 6. Let methodContext be the running execution context.
auto const& method_context = vm.running_execution_context();
// 7. Suspend methodContext.
// 9. Push genContext onto the execution context stack; genContext is now the running execution context.
// NOTE: This is done out of order as to not permanently disable the generator if push_execution_context throws,
// as `resume_abrupt` will immediately throw when [[GeneratorState]] is "executing", never allowing the state to change.
TRY(vm.push_execution_context(*generator_context, {}));
// 8. Set generator.[[GeneratorState]] to executing.
m_generator_state = GeneratorState::Executing;
// 10. Resume the suspended evaluation of genContext using abruptCompletion as the result of the operation that suspended it. Let result be the Completion Record returned by the resumed computation.
auto result = execute(vm, abrupt_completion);
// 11. Assert: When we return here, genContext has already been removed from the execution context stack and methodContext is the currently running execution context.
VERIFY(&vm.running_execution_context() == &method_context);
// 12. Return ? result.
return result;
}
}