RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-07-25 19:37:35 +00:00
Code Issues Activity Actions

LibJS/JIT: Add fast path for GetById of Array.length

Browse source 
Array.length is magical (since it has to reflect the number of elements
in the object's property storage).

We now handle it specially in jitted code, giving us a massive speed-up
on Kraken/ai-astar.js (and probably many other things as well) :^)
This commit is contained in:
Andreas Kling 2023-11-12 15:04:23 +01:00
parent e41f0d9dec
commit b532dedc91
3 changed files with 65 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

60
Userland/Libraries/LibJS/JIT/Compiler.cpp
View file

@ -1239,6 +1239,66 @@ void Compiler::compile_get_by_id(Bytecode::Op::GetById const& op)
branch_if_object(ARG1, [&] {
extract_object_pointer(GPR0, ARG1);
// NOTE: Fast path for Array.length which magically reflects
// the "array-like size" of the array object's property storage.
if (m_bytecode_executable.get_identifier(op.property()) == "length"sv) {
Assembler::Label no_magical_length_property_case;
// if (!object.has_magical_length_property) goto no_magical_length_property_case;
m_assembler.mov8(
Assembler::Operand::Register(GPR1),
Assembler::Operand::Mem64BaseAndOffset(GPR0, Object::has_magical_length_property_offset()));
m_assembler.jump_if(
Assembler::Operand::Register(GPR1),
Assembler::Condition::EqualTo,
Assembler::Operand::Imm(0),
no_magical_length_property_case);
// NOTE: The base object has a magical "length" property, so now we just need
// to extract the "array-like size" from the object property storage.
// If we run into any issues, we'll jump to the slow case and figure things out in C++.
// GPR0 = object->indexed_properties().storage()
m_assembler.mov(
Assembler::Operand::Register(GPR0),
Assembler::Operand::Mem64BaseAndOffset(GPR0, Object::indexed_properties_offset() + IndexedProperties::storage_offset()));
// if (GPR0 == nullptr) goto slow_case;
m_assembler.jump_if(
Assembler::Operand::Register(GPR0),
Assembler::Condition::EqualTo,
Assembler::Operand::Imm(0),
slow_case);
// if (!GPR0->is_simple_storage()) goto slow_case;
m_assembler.mov8(
Assembler::Operand::Register(GPR1),
Assembler::Operand::Mem64BaseAndOffset(GPR0, IndexedPropertyStorage::is_simple_storage_offset()));
m_assembler.jump_if(
Assembler::Operand::Register(GPR1),
Assembler::Condition::EqualTo,
Assembler::Operand::Imm(0),
slow_case);
// accumulator = GPR0->array_like_size() | SHIFT_INT32_TAG
// return
m_assembler.mov(
Assembler::Operand::Register(GPR1),
Assembler::Operand::Mem64BaseAndOffset(GPR0, SimpleIndexedPropertyStorage::array_size_offset()));
m_assembler.mov(
Assembler::Operand::Register(GPR0),
Assembler::Operand::Imm(SHIFTED_INT32_TAG));
m_assembler.bitwise_or(
Assembler::Operand::Register(GPR1),
Assembler::Operand::Register(GPR0));
store_accumulator(GPR1);
m_assembler.jump(end);
no_magical_length_property_case.link(m_assembler);
}
// if (cache.shape != &object->shape()) goto slow_case;
m_assembler.mov(
Assembler::Operand::Register(GPR2),