Stowage/ladybird
2
0
Fork 0
mirror of https://github.com/LadybirdBrowser/ladybird.git synced 2025-10-19 15:43:20 +00:00
Code Issues Projects Releases Packages Wiki Activity
ladybird/Libraries/LibJS/Bytecode/Op.h

2952 lines
89 KiB
C
Raw Normal View History

LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
/*
Meta: Update my e-mail address everywhere
2024-10-04 13:19:50 +02:00
* Copyright (c) 2021-2023, Andreas Kling <andreas@ladybird.org>
LibJS: Add bytecode generation for BinaryOp::In
2021-06-07 21:13:37 +01:00
* Copyright (c) 2021, Linus Groh <linusg@serenityos.org>
LibJS: Generate bytecode for throw statements
2021-06-09 18:18:56 +02:00
* Copyright (c) 2021, Gunnar Beutner <gbeutner@serenityos.org>
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#pragma once
LibJS: Instantiate primitive array expressions using a single operation This will not meaningfully affect short array literals, but it does give us a bit of extra perf when evaluating huge array expressions like in Kraken/imaging-darkroom.js
2023-11-17 22:07:23 +02:00
#include <AK/FixedArray.h>
LibJS: Align Instructions as void* and roundup variably sized ones sizes Both is indeed needed, the standard alignment would have been 4, but some Instructions, like Jumps need an alignment of 8 Fixes #12127.
2022-09-09 16:47:42 +02:00
#include <AK/StdLibExtras.h>
LibJS: Implement bytecode generation for BigInts
2021-06-08 07:59:25 +02:00
#include <LibCrypto/BigInt/SignedBigInteger.h>
LibJS: Introduce Builtins Builtins are functions that can be detected during bytecode generation and enable fast-paths in the JIT.
2023-11-17 11:48:30 +01:00
#include <LibJS/Bytecode/Builtins.h>
LibJS: Add a separate "identifier table" to bytecode executables This is a specialized string table for storing identifiers only. Identifiers are always FlyStrings, which makes many common operations faster by allowing O(1) comparison.
2021-10-24 15:34:30 +02:00
#include <LibJS/Bytecode/IdentifierTable.h>
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
#include <LibJS/Bytecode/Instruction.h>
LibJS: Add basic support for while loops in the bytecode engine This introduces two new instructions: Jump and JumpIfFalse. Jumps are made to a Bytecode::Label, which is a simple object that represents a location in the bytecode stream. Note that you may not always know the target of a jump when adding the jump instruction itself, but we can just update the instruction later on during codegen once we know where the jump target is. The Bytecode::Interpreter now implements jumping via a jump slot that gets checked after each instruction to see if a jump is pending. If not, we just increment the PC as usual.
2021-06-04 12:07:38 +02:00
#include <LibJS/Bytecode/Label.h>
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
#include <LibJS/Bytecode/Operand.h>
LibJS/Bytecode: Don't reparse regular expressions on instantiation The RegExpLiteral AST node already has the parsed regex::Parser::Result so let's plumb that over to the bytecode executable instead of reparsing the regex every time NewRegExp is executed. ~12% speed-up on language/literals/regexp/S7.8.5_A2.1_T2.js in test262.
2023-07-13 10:49:07 +02:00
#include <LibJS/Bytecode/RegexTable.h>
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
#include <LibJS/Bytecode/Register.h>
LibJS/Bytecode: Reuse bytecode registers This patch adds a register freelist to Bytecode::Generator and switches all operands inside the generator to a new ScopedOperand type that is ref-counted and automatically frees the register when nothing uses it. This dramatically reduces the size of bytecode executable register windows, which were often in the several thousands of registers for large functions. Most functions now use less than 100 registers.
2024-05-07 21:36:56 +02:00
#include <LibJS/Bytecode/ScopedOperand.h>
LibJS: Store strings in a string table Instead of using Strings in the bytecode ops this adds a global string table to the Executable struct which individual operations can refer to using indices. This brings bytecode ops one step closer to being pointer free.
2021-06-09 10:02:01 +02:00
#include <LibJS/Bytecode/StringTable.h>
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
#include <LibJS/Heap/Cell.h>
LibJS: Drop "Record" suffix from all the *Environment record classes "Records" in the spec are basically C++ classes, so let's drop this mouthful of a suffix.
2021-07-01 12:24:46 +02:00
#include <LibJS/Runtime/Environment.h>
LibJS+LibWeb: Move IteratorOperations.h AOs to Iterator.h Rather than splitting the Iterator type and its AOs into two files, let's combine them into one file to match every other JS runtime object that we have.
2023-07-19 06:54:48 -04:00
#include <LibJS/Runtime/Iterator.h>
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
#include <LibJS/Runtime/Value.h>
LibJS: Avoid unnecessary ToObject conversion when resolving references When performing GetValue on a primitive type we do not need to perform the ToObject conversion as it will resolve to a property on the prototype object. To avoid this we skip the initial ToObject conversion on the base value as it only serves to get the primitive's boxed prototype. We further specialize on PrimitiveString in order to get efficient behaviour behaviour for the direct properties. Depending on the tests anywhere from 20 to 60%, with significant loop overhead.
2022-02-12 00:48:23 -08:00
#include <LibJS/Runtime/ValueTraits.h>
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
LibJS/Bytecode: Add codegen for "named evaluation if anonymous function" This gives anonymous functions the name from the LHS they are being assigned to. 171 new passes on test262. :^)
2023-06-23 14:27:42 +02:00
namespace JS {
Meta: Enforce newlines around namespaces This has come up several times during code review, so let's just enforce it using a new clang-format 20 option.
2025-05-13 07:06:33 -04:00
LibJS/Bytecode: Add codegen for "named evaluation if anonymous function" This gives anonymous functions the name from the LHS they are being assigned to. 171 new passes on test262. :^)
2023-06-23 14:27:42 +02:00
class FunctionExpression;
Meta: Enforce newlines around namespaces This has come up several times during code review, so let's just enforce it using a new clang-format 20 option.
2025-05-13 07:06:33 -04:00
LibJS/Bytecode: Add codegen for "named evaluation if anonymous function" This gives anonymous functions the name from the LHS they are being assigned to. 171 new passes on test262. :^)
2023-06-23 14:27:42 +02:00
}
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
namespace JS::Bytecode::Op {
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CreateRestParams final : public Instruction {
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
public:
CreateRestParams(Operand dst, u32 rest_index)
: Instruction(Type::CreateRestParams)
, m_dst(dst)
, m_rest_index(rest_index)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
private:
Operand m_dst;
u32 m_rest_index;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CreateArguments final : public Instruction {
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
public:
enum class Kind {
Mapped,
Unmapped,
};
LibJS: Use a local variable for arguments object when possible This allows us to skip allocating a function environment in cases where it was previously impossible because the arguments object needed a binding. This change does not bring visible improvement in Kraken or Octane benchmarks but seems useful to have anyway.
2024-05-21 09:32:51 +01:00
CreateArguments(Optional<Operand> dst, Kind kind, bool is_immutable)
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
: Instruction(Type::CreateArguments)
LibJS: Use a local variable for arguments object when possible This allows us to skip allocating a function environment in cases where it was previously impossible because the arguments object needed a binding. This change does not bring visible improvement in Kraken or Octane benchmarks but seems useful to have anyway.
2024-05-21 09:32:51 +01:00
, m_dst(dst)
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
, m_kind(kind)
, m_is_immutable(is_immutable)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Use a local variable for arguments object when possible This allows us to skip allocating a function environment in cases where it was previously impossible because the arguments object needed a binding. This change does not bring visible improvement in Kraken or Octane benchmarks but seems useful to have anyway.
2024-05-21 09:32:51 +01:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
if (m_dst.has_value())
visitor(m_dst.value());
}
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
private:
LibJS: Use a local variable for arguments object when possible This allows us to skip allocating a function environment in cases where it was previously impossible because the arguments object needed a binding. This change does not bring visible improvement in Kraken or Octane benchmarks but seems useful to have anyway.
2024-05-21 09:32:51 +01:00
Optional<Operand> m_dst;
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
Kind m_kind;
bool m_is_immutable { false };
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Mov final : public Instruction {
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Mov(Operand dst, Operand src)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Mov)
LibJS: Make sure that if expressions yield the correct value When evaluated as an expression "if (true) { 3 } else { 5 }" should yield 3. This updates the bytecode interpreter to make it so.
2021-06-07 22:05:09 +02:00
, m_dst(dst)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS: Make sure that if expressions yield the correct value When evaluated as an expression "if (true) { 3 } else { 5 }" should yield 3. This updates the bytecode interpreter to make it so.
2021-06-07 22:05:09 +02:00
{
}
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_src);
}
LibJS: Make sure that if expressions yield the correct value When evaluated as an expression "if (true) { 3 } else { 5 }" should yield 3. This updates the bytecode interpreter to make it so.
2021-06-07 22:05:09 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand src() const { return m_src; }
LibJS: Expose some internals of Instructions These will be needed in the future to allow optimization passes to check against these
2022-10-22 20:21:08 +02:00
LibJS: Make sure that if expressions yield the correct value When evaluated as an expression "if (true) { 3 } else { 5 }" should yield 3. This updates the bytecode interpreter to make it so.
2021-06-07 22:05:09 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_src;
LibJS: Make sure that if expressions yield the correct value When evaluated as an expression "if (true) { 3 } else { 5 }" should yield 3. This updates the bytecode interpreter to make it so.
2021-06-07 22:05:09 +02:00
};
LibJS/Bytecode: Add fast paths for many binary expression instructions By handling common cases like Int32 arithmetic directly in the instruction handler, we can avoid the cost of calling the generic helper functions in Value.cpp.
2024-02-20 11:59:46 +01:00
#define JS_ENUMERATE_COMMON_BINARY_OPS_WITH_FAST_PATH(O) \
O(Add, add) \
O(BitwiseAnd, bitwise_and) \
O(BitwiseOr, bitwise_or) \
O(BitwiseXor, bitwise_xor) \
O(GreaterThan, greater_than) \
O(GreaterThanEquals, greater_than_equals) \
LibJS/Bytecode: Add fast path for LeftShift with Int32 operands
2024-03-04 10:01:40 +01:00
O(LeftShift, left_shift) \
LibJS/Bytecode: Add fast paths for many binary expression instructions By handling common cases like Int32 arithmetic directly in the instruction handler, we can avoid the cost of calling the generic helper functions in Value.cpp.
2024-02-20 11:59:46 +01:00
O(LessThan, less_than) \
O(LessThanEquals, less_than_equals) \
O(Mul, mul) \
O(RightShift, right_shift) \
O(Sub, sub) \
LibJS: Use macros to generate the common unary/binary bytecode ops
2021-06-07 23:08:35 +02:00
O(UnsignedRightShift, unsigned_right_shift)
LibJS/Bytecode: Add fast paths for many binary expression instructions By handling common cases like Int32 arithmetic directly in the instruction handler, we can avoid the cost of calling the generic helper functions in Value.cpp.
2024-02-20 11:59:46 +01:00
#define JS_ENUMERATE_COMMON_BINARY_OPS_WITHOUT_FAST_PATH(O) \
O(Div, div) \
O(Exp, exp) \
O(Mod, mod) \
O(In, in) \
O(InstanceOf, instance_of) \
O(LooselyInequals, loosely_inequals) \
O(LooselyEquals, loosely_equals) \
O(StrictlyInequals, strict_inequals) \
LibJS/Bytecode: Add fast path for LeftShift with Int32 operands
2024-03-04 10:01:40 +01:00
O(StrictlyEquals, strict_equals)
LibJS/Bytecode: Add fast paths for many binary expression instructions By handling common cases like Int32 arithmetic directly in the instruction handler, we can avoid the cost of calling the generic helper functions in Value.cpp.
2024-02-20 11:59:46 +01:00
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
#define JS_DECLARE_COMMON_BINARY_OP(OpTitleCase, op_snake_case) \
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API OpTitleCase final : public Instruction { \
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
public: \
explicit OpTitleCase(Operand dst, Operand lhs, Operand rhs) \
: Instruction(Type::OpTitleCase) \
, m_dst(dst) \
, m_lhs(lhs) \
, m_rhs(rhs) \
{ \
} \
\
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const; \
ByteString to_byte_string_impl(Bytecode::Executable const&) const; \
void visit_operands_impl(Function<void(Operand&)> visitor) \
{ \
visitor(m_dst); \
visitor(m_lhs); \
visitor(m_rhs); \
} \
\
Operand dst() const { return m_dst; } \
Operand lhs() const { return m_lhs; } \
Operand rhs() const { return m_rhs; } \
\
private: \
Operand m_dst; \
Operand m_lhs; \
Operand m_rhs; \
LibJS: Use macros to generate the common unary/binary bytecode ops
2021-06-07 23:08:35 +02:00
};
LibJS/Bytecode: Add fast paths for many binary expression instructions By handling common cases like Int32 arithmetic directly in the instruction handler, we can avoid the cost of calling the generic helper functions in Value.cpp.
2024-02-20 11:59:46 +01:00
JS_ENUMERATE_COMMON_BINARY_OPS_WITHOUT_FAST_PATH(JS_DECLARE_COMMON_BINARY_OP)
JS_ENUMERATE_COMMON_BINARY_OPS_WITH_FAST_PATH(JS_DECLARE_COMMON_BINARY_OP)
LibJS: Use macros to generate the common unary/binary bytecode ops
2021-06-07 23:08:35 +02:00
#undef JS_DECLARE_COMMON_BINARY_OP
#define JS_ENUMERATE_COMMON_UNARY_OPS(O) \
O(BitwiseNot, bitwise_not) \
O(Not, not_) \
O(UnaryPlus, unary_plus) \
O(UnaryMinus, unary_minus) \
O(Typeof, typeof_)
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
#define JS_DECLARE_COMMON_UNARY_OP(OpTitleCase, op_snake_case) \
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API OpTitleCase final : public Instruction { \
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
public: \
OpTitleCase(Operand dst, Operand src) \
: Instruction(Type::OpTitleCase) \
, m_dst(dst) \
, m_src(src) \
{ \
} \
\
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const; \
ByteString to_byte_string_impl(Bytecode::Executable const&) const; \
void visit_operands_impl(Function<void(Operand&)> visitor) \
{ \
visitor(m_dst); \
visitor(m_src); \
} \
\
Operand dst() const { return m_dst; } \
Operand src() const { return m_src; } \
\
private: \
Operand m_dst; \
Operand m_src; \
LibJS: Use macros to generate the common unary/binary bytecode ops
2021-06-07 23:08:35 +02:00
};
JS_ENUMERATE_COMMON_UNARY_OPS(JS_DECLARE_COMMON_UNARY_OP)
#undef JS_DECLARE_COMMON_UNARY_OP
LibJS: Add bytecode instructions for a bunch of unary operators ~, !, +, -, typeof, and void.
2021-06-07 19:53:47 +01:00
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API NewObject final : public Instruction {
LibJS: Add a NewObject bytecode instruction for ObjectExpression :^)
2021-06-04 20:30:23 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit NewObject(Operand dst)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::NewObject)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Add a NewObject bytecode instruction for ObjectExpression :^)
2021-06-04 20:30:23 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
private:
Operand m_dst;
LibJS: Add a NewObject bytecode instruction for ObjectExpression :^)
2021-06-04 20:30:23 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API NewRegExp final : public Instruction {
LibJS: Add bytecode support for regexp literals
2021-06-19 17:17:40 -07:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
NewRegExp(Operand dst, StringTableIndex source_index, StringTableIndex flags_index, RegexTableIndex regex_index)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::NewRegExp)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Add bytecode support for regexp literals
2021-06-19 17:17:40 -07:00
, m_source_index(source_index)
, m_flags_index(flags_index)
LibJS/Bytecode: Don't reparse regular expressions on instantiation The RegExpLiteral AST node already has the parsed regex::Parser::Result so let's plumb that over to the bytecode executable instead of reparsing the regex every time NewRegExp is executed. ~12% speed-up on language/literals/regexp/S7.8.5_A2.1_T2.js in test262.
2023-07-13 10:49:07 +02:00
, m_regex_index(regex_index)
LibJS: Add bytecode support for regexp literals
2021-06-19 17:17:40 -07:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS: Add bytecode support for regexp literals
2021-06-19 17:17:40 -07:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS/JIT: Compile the NewRegExp bytecode instruction
2023-10-27 17:07:30 +02:00
StringTableIndex source_index() const { return m_source_index; }
StringTableIndex flags_index() const { return m_flags_index; }
RegexTableIndex regex_index() const { return m_regex_index; }
LibJS: Add bytecode support for regexp literals
2021-06-19 17:17:40 -07:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
LibJS: Add bytecode support for regexp literals
2021-06-19 17:17:40 -07:00
StringTableIndex m_source_index;
StringTableIndex m_flags_index;
LibJS/Bytecode: Don't reparse regular expressions on instantiation The RegExpLiteral AST node already has the parsed regex::Parser::Result so let's plumb that over to the bytecode executable instead of reparsing the regex every time NewRegExp is executed. ~12% speed-up on language/literals/regexp/S7.8.5_A2.1_T2.js in test262.
2023-07-13 10:49:07 +02:00
RegexTableIndex m_regex_index;
LibJS: Add bytecode support for regexp literals
2021-06-19 17:17:40 -07:00
};
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
#define JS_ENUMERATE_NEW_BUILTIN_ERROR_OPS(O) \
O(TypeError)
LibJS/Bytecode: Make execute_impl() return void for non-throwing ops This way we can't accidentally ignore exceptions.
2024-05-10 17:32:19 +02:00
#define JS_DECLARE_NEW_BUILTIN_ERROR_OP(ErrorName) \
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API New##ErrorName final : public Instruction { \
LibJS/Bytecode: Make execute_impl() return void for non-throwing ops This way we can't accidentally ignore exceptions.
2024-05-10 17:32:19 +02:00
public: \
New##ErrorName(Operand dst, StringTableIndex error_string) \
: Instruction(Type::New##ErrorName) \
, m_dst(dst) \
, m_error_string(error_string) \
{ \
} \
\
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const; \
LibJS/Bytecode: Make execute_impl() return void for non-throwing ops This way we can't accidentally ignore exceptions.
2024-05-10 17:32:19 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const; \
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor) \
{ \
visitor(m_dst); \
} \
LibJS/Bytecode: Make execute_impl() return void for non-throwing ops This way we can't accidentally ignore exceptions.
2024-05-10 17:32:19 +02:00
\
Operand dst() const { return m_dst; } \
StringTableIndex error_string() const { return m_error_string; } \
\
private: \
Operand m_dst; \
StringTableIndex m_error_string; \
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
};
JS_ENUMERATE_NEW_BUILTIN_ERROR_OPS(JS_DECLARE_NEW_BUILTIN_ERROR_OP)
#undef JS_DECLARE_NEW_BUILTIN_ERROR_OP
LibJS: Support object rest elements in the bytecode interpreter
2021-06-13 15:30:32 -07:00
// NOTE: This instruction is variable-width depending on the number of excluded names
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CopyObjectExcludingProperties final : public Instruction {
LibJS: Support object rest elements in the bytecode interpreter
2021-06-13 15:30:32 -07:00
public:
LibJS/Bytecode: Unroll the bytecode interpreter This commit adds a HANDLE_INSTRUCTION macro that expands to everything needed to handle a single instruction (invoking the handler function, checking for exceptions, and advancing the program counter). This gives a ~15% speed-up on a for loop microbenchmark, and makes basically everything faster.
2024-05-06 10:42:52 +02:00
static constexpr bool IsVariableLength = true;
LibJS/Bytecode: Reuse bytecode registers This patch adds a register freelist to Bytecode::Generator and switches all operands inside the generator to a new ScopedOperand type that is ref-counted and automatically frees the register when nothing uses it. This dramatically reduces the size of bytecode executable register windows, which were often in the several thousands of registers for large functions. Most functions now use less than 100 registers.
2024-05-07 21:36:56 +02:00
CopyObjectExcludingProperties(Operand dst, Operand from_object, Vector<ScopedOperand> const& excluded_names)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::CopyObjectExcludingProperties)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Support object rest elements in the bytecode interpreter
2021-06-13 15:30:32 -07:00
, m_from_object(from_object)
, m_excluded_names_count(excluded_names.size())
{
for (size_t i = 0; i < m_excluded_names_count; i++)
m_excluded_names[i] = excluded_names[i];
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_from_object);
for (size_t i = 0; i < m_excluded_names_count; i++)
visitor(m_excluded_names[i]);
}
LibJS: Support object rest elements in the bytecode interpreter
2021-06-13 15:30:32 -07:00
LibJS: Overload Instruction::length() on variable-length instructions This allows the DISPATCH_NEXT() macro to do the math to find the next instruction boundary point without performing a function call.
2025-04-28 23:16:26 +02:00
size_t length() const { return length_impl(); }
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
size_t length_impl() const
LibJS: Store bytecode instruction length in instruction itself Instead of running a big switch statement on the opcode when checking how long an instruction is, we now simply store that in a member variable at construction time for instant access. This yields a 10.2% speed-up on Kraken/ai-astar :^)
2023-09-13 21:32:51 +02:00
{
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
return round_up_to_power_of_two(alignof(void*), sizeof(*this) + sizeof(Operand) * m_excluded_names_count);
LibJS: Store bytecode instruction length in instruction itself Instead of running a big switch statement on the opcode when checking how long an instruction is, we now simply store that in a member variable at construction time for instant access. This yields a 10.2% speed-up on Kraken/ai-astar :^)
2023-09-13 21:32:51 +02:00
}
LibJS: Support object rest elements in the bytecode interpreter
2021-06-13 15:30:32 -07:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand from_object() const { return m_from_object; }
LibJS/JIT: Compile the CopyObjectExcludingProperties instruction
2023-10-30 02:00:53 +01:00
size_t excluded_names_count() const { return m_excluded_names_count; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand const* excluded_names() const { return m_excluded_names; }
LibJS/JIT: Compile the CopyObjectExcludingProperties instruction
2023-10-30 02:00:53 +01:00
LibJS: Support object rest elements in the bytecode interpreter
2021-06-13 15:30:32 -07:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_from_object;
LibJS: Use u32 instead of size_t for bytecode instruction array sizes Instructions that have an embedded tail array are not going to have more than 2^32 elements in any remotely sane situation.
2025-04-05 23:59:18 +02:00
u32 m_excluded_names_count { 0 };
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_excluded_names[];
LibJS: Support object rest elements in the bytecode interpreter
2021-06-13 15:30:32 -07:00
};
LibJS: Generate bytecode for array expressions
2021-06-08 23:06:52 +02:00
// NOTE: This instruction is variable-width depending on the number of elements!
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API NewArray final : public Instruction {
LibJS: Generate bytecode for array expressions
2021-06-08 23:06:52 +02:00
public:
LibJS/Bytecode: Unroll the bytecode interpreter This commit adds a HANDLE_INSTRUCTION macro that expands to everything needed to handle a single instruction (invoking the handler function, checking for exceptions, and advancing the program counter). This gives a ~15% speed-up on a for loop microbenchmark, and makes basically everything faster.
2024-05-06 10:42:52 +02:00
static constexpr bool IsVariableLength = true;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit NewArray(Operand dst)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::NewArray)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Support array rest elements in the bytecode interpreter
2021-06-13 14:06:26 -07:00
, m_element_count(0)
{
}
LibJS/Bytecode: Make NewArray a variable-length instruction This removes a layer of indirection in the bytecode where we had to make sure all the initializer elements were laid out in sequential registers. Array expressions no longer clobber registers permanently, and they can be reused immediately afterwards.
2024-05-08 12:43:08 +02:00
NewArray(Operand dst, ReadonlySpan<ScopedOperand> elements)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::NewArray)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Make NewArray a variable-length instruction This removes a layer of indirection in the bytecode where we had to make sure all the initializer elements were laid out in sequential registers. Array expressions no longer clobber registers permanently, and they can be reused immediately afterwards.
2024-05-08 12:43:08 +02:00
, m_element_count(elements.size())
LibJS: Generate bytecode for array expressions
2021-06-08 23:06:52 +02:00
{
LibJS/Bytecode: Make NewArray a variable-length instruction This removes a layer of indirection in the bytecode where we had to make sure all the initializer elements were laid out in sequential registers. Array expressions no longer clobber registers permanently, and they can be reused immediately afterwards.
2024-05-08 12:43:08 +02:00
for (size_t i = 0; i < m_element_count; ++i)
m_elements[i] = elements[i];
LibJS: Generate bytecode for array expressions
2021-06-08 23:06:52 +02:00
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
for (size_t i = 0; i < m_element_count; i++)
visitor(m_elements[i]);
}
LibJS: Generate bytecode for array expressions
2021-06-08 23:06:52 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS: Overload Instruction::length() on variable-length instructions This allows the DISPATCH_NEXT() macro to do the math to find the next instruction boundary point without performing a function call.
2025-04-28 23:16:26 +02:00
size_t length() const { return length_impl(); }
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
size_t length_impl() const
LibJS: Add a basic pass manager and add some basic passes This commit adds a bunch of passes, the most interesting of which is a pass that merges blocks together, and a pass that places blocks that flow into each other next to each other, and a very simply pass that removes duplicate basic blocks. Note that this does not remove the jump at the end of each block in that pass to avoid scope creep in the passes.
2021-06-13 20:40:20 +04:30
{
LibJS/Bytecode: Make NewArray a variable-length instruction This removes a layer of indirection in the bytecode where we had to make sure all the initializer elements were laid out in sequential registers. Array expressions no longer clobber registers permanently, and they can be reused immediately afterwards.
2024-05-08 12:43:08 +02:00
return round_up_to_power_of_two(alignof(void*), sizeof(*this) + sizeof(Operand) * m_element_count);
LibJS: Expose some internals of Instructions These will be needed in the future to allow optimization passes to check against these
2022-10-22 20:21:08 +02:00
}
size_t element_count() const { return m_element_count; }
LibJS: Generate bytecode for array expressions
2021-06-08 23:06:52 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
LibJS: Use u32 instead of size_t for bytecode instruction array sizes Instructions that have an embedded tail array are not going to have more than 2^32 elements in any remotely sane situation.
2025-04-05 23:59:18 +02:00
u32 m_element_count { 0 };
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_elements[];
LibJS: Generate bytecode for array expressions
2021-06-08 23:06:52 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API NewPrimitiveArray final : public Instruction {
LibJS: Instantiate primitive array expressions using a single operation This will not meaningfully affect short array literals, but it does give us a bit of extra perf when evaluating huge array expressions like in Kraken/imaging-darkroom.js
2023-11-17 22:07:23 +02:00
public:
LibJS/Bytecode: Unroll the bytecode interpreter This commit adds a HANDLE_INSTRUCTION macro that expands to everything needed to handle a single instruction (invoking the handler function, checking for exceptions, and advancing the program counter). This gives a ~15% speed-up on a for loop microbenchmark, and makes basically everything faster.
2024-05-06 10:42:52 +02:00
static constexpr bool IsVariableLength = true;
LibJS/Bytecode: Make NewPrimitiveArray a variable-length instruction Instead of having a FixedArray with a separate heap allocation, we can just bake the primitive values into the instruction itself.
2024-03-03 12:37:28 +01:00
NewPrimitiveArray(Operand dst, ReadonlySpan<Value> elements)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::NewPrimitiveArray)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Make NewPrimitiveArray a variable-length instruction Instead of having a FixedArray with a separate heap allocation, we can just bake the primitive values into the instruction itself.
2024-03-03 12:37:28 +01:00
, m_element_count(elements.size())
LibJS: Instantiate primitive array expressions using a single operation This will not meaningfully affect short array literals, but it does give us a bit of extra perf when evaluating huge array expressions like in Kraken/imaging-darkroom.js
2023-11-17 22:07:23 +02:00
{
LibJS/Bytecode: Make NewPrimitiveArray a variable-length instruction Instead of having a FixedArray with a separate heap allocation, we can just bake the primitive values into the instruction itself.
2024-03-03 12:37:28 +01:00
for (size_t i = 0; i < m_element_count; ++i)
m_elements[i] = elements[i];
}
LibJS: Overload Instruction::length() on variable-length instructions This allows the DISPATCH_NEXT() macro to do the math to find the next instruction boundary point without performing a function call.
2025-04-28 23:16:26 +02:00
size_t length() const { return length_impl(); }
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
size_t length_impl() const
LibJS/Bytecode: Make NewPrimitiveArray a variable-length instruction Instead of having a FixedArray with a separate heap allocation, we can just bake the primitive values into the instruction itself.
2024-03-03 12:37:28 +01:00
{
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
return round_up_to_power_of_two(alignof(void*), sizeof(*this) + sizeof(Value) * m_element_count);
LibJS: Instantiate primitive array expressions using a single operation This will not meaningfully affect short array literals, but it does give us a bit of extra perf when evaluating huge array expressions like in Kraken/imaging-darkroom.js
2023-11-17 22:07:23 +02:00
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS: Instantiate primitive array expressions using a single operation This will not meaningfully affect short array literals, but it does give us a bit of extra perf when evaluating huge array expressions like in Kraken/imaging-darkroom.js
2023-11-17 22:07:23 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS/Bytecode: Make NewPrimitiveArray a variable-length instruction Instead of having a FixedArray with a separate heap allocation, we can just bake the primitive values into the instruction itself.
2024-03-03 12:37:28 +01:00
ReadonlySpan<Value> elements() const { return { m_elements, m_element_count }; }
LibJS: Instantiate primitive array expressions using a single operation This will not meaningfully affect short array literals, but it does give us a bit of extra perf when evaluating huge array expressions like in Kraken/imaging-darkroom.js
2023-11-17 22:07:23 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
LibJS: Use u32 instead of size_t for bytecode instruction array sizes Instructions that have an embedded tail array are not going to have more than 2^32 elements in any remotely sane situation.
2025-04-05 23:59:18 +02:00
u32 m_element_count { 0 };
LibJS/Bytecode: Make NewPrimitiveArray a variable-length instruction Instead of having a FixedArray with a separate heap allocation, we can just bake the primitive values into the instruction itself.
2024-03-03 12:37:28 +01:00
Value m_elements[];
LibJS: Instantiate primitive array expressions using a single operation This will not meaningfully affect short array literals, but it does give us a bit of extra perf when evaluating huge array expressions like in Kraken/imaging-darkroom.js
2023-11-17 22:07:23 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API AddPrivateName final : public Instruction {
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
public:
explicit AddPrivateName(IdentifierTableIndex name)
: Instruction(Type::AddPrivateName)
, m_name(name)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
private:
IdentifierTableIndex m_name;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ArrayAppend final : public Instruction {
LibJS: Add support for SpreadExpressions in array literals for bytecode For this it adds another opcode `Append $lhs` which appends the accumulator to the Array in $lhs, optionally spreading it.
2022-09-09 15:23:02 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
ArrayAppend(Operand dst, Operand src, bool is_spread)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ArrayAppend)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_src(src)
LibJS: Add support for SpreadExpressions in array literals for bytecode For this it adds another opcode `Append $lhs` which appends the accumulator to the Array in $lhs, optionally spreading it.
2022-09-09 15:23:02 +02:00
, m_is_spread(is_spread)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_src);
}
LibJS: Add a way to replace references to registers in Bytecode
2022-10-22 20:19:16 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand src() const { return m_src; }
LibJS/JIT: Compile the Append instruction
2023-10-29 17:25:49 +01:00
bool is_spread() const { return m_is_spread; }
LibJS: Add support for SpreadExpressions in array literals for bytecode For this it adds another opcode `Append $lhs` which appends the accumulator to the Array in $lhs, optionally spreading it.
2022-09-09 15:23:02 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_src;
LibJS: Add support for SpreadExpressions in array literals for bytecode For this it adds another opcode `Append $lhs` which appends the accumulator to the Array in $lhs, optionally spreading it.
2022-09-09 15:23:02 +02:00
bool m_is_spread = false;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ImportCall final : public Instruction {
LibJS/Bytecode: Add codegen for ImportCall Also moved most of the AST ImportCall::execute() into a helper so we can share the code.
2023-06-24 15:22:16 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
ImportCall(Operand dst, Operand specifier, Operand options)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ImportCall)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Add codegen for ImportCall Also moved most of the AST ImportCall::execute() into a helper so we can share the code.
2023-06-24 15:22:16 +02:00
, m_specifier(specifier)
, m_options(options)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_specifier);
visitor(m_options);
}
LibJS/Bytecode: Add codegen for ImportCall Also moved most of the AST ImportCall::execute() into a helper so we can share the code.
2023-06-24 15:22:16 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand specifier() const { return m_specifier; }
Operand options() const { return m_options; }
LibJS/JIT: Compile the ImportCall instruction
2023-10-29 23:41:33 +01:00
LibJS/Bytecode: Add codegen for ImportCall Also moved most of the AST ImportCall::execute() into a helper so we can share the code.
2023-06-24 15:22:16 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_specifier;
Operand m_options;
LibJS/Bytecode: Add codegen for ImportCall Also moved most of the AST ImportCall::execute() into a helper so we can share the code.
2023-06-24 15:22:16 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API IteratorToArray final : public Instruction {
LibJS: Support array rest elements in the bytecode interpreter
2021-06-13 14:06:26 -07:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit IteratorToArray(Operand dst, Operand iterator)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::IteratorToArray)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_iterator(iterator)
LibJS: Support array rest elements in the bytecode interpreter
2021-06-13 14:06:26 -07:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand iterator() const { return m_iterator; }
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_iterator);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
private:
Operand m_dst;
Operand m_iterator;
LibJS: Support array rest elements in the bytecode interpreter
2021-06-13 14:06:26 -07:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ConcatString final : public Instruction {
LibJS: Introduce an accumulator register to Bytecode::Interpreter This commit introduces the concept of an accumulator register to LibJS's bytecode interpreter. The accumulator register is always register 0, and most simple instructions use it for reading and writing. Not only does this slim down the AST, but it also simplifies a lot of the code. For example, the generate_bytecode methods no longer need to return an Optional<Register>, as any opcode which has a "return" value will always put it into the accumulator. This also renames the old Op::Load to Op::LoadImmediate, and uses Op::Load to load from a register into the accumulator. There is also an Op::Store to put the value in the accumulator into another register.
2021-06-07 20:58:36 -07:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit ConcatString(Operand dst, Operand src)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ConcatString)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_src(src)
LibJS: Introduce an accumulator register to Bytecode::Interpreter This commit introduces the concept of an accumulator register to LibJS's bytecode interpreter. The accumulator register is always register 0, and most simple instructions use it for reading and writing. Not only does this slim down the AST, but it also simplifies a lot of the code. For example, the generate_bytecode methods no longer need to return an Optional<Register>, as any opcode which has a "return" value will always put it into the accumulator. This also renames the old Op::Load to Op::LoadImmediate, and uses Op::Load to load from a register into the accumulator. There is also an Op::Store to put the value in the accumulator into another register.
2021-06-07 20:58:36 -07:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Introduce an accumulator register to Bytecode::Interpreter This commit introduces the concept of an accumulator register to LibJS's bytecode interpreter. The accumulator register is always register 0, and most simple instructions use it for reading and writing. Not only does this slim down the AST, but it also simplifies a lot of the code. For example, the generate_bytecode methods no longer need to return an Optional<Register>, as any opcode which has a "return" value will always put it into the accumulator. This also renames the old Op::Load to Op::LoadImmediate, and uses Op::Load to load from a register into the accumulator. There is also an Op::Store to put the value in the accumulator into another register.
2021-06-07 20:58:36 -07:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand src() const { return m_src; }
LibJS/JIT: Compile the ConcatString bytecode instruction
2023-10-29 00:06:54 +02:00
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_src);
}
LibJS: Introduce an accumulator register to Bytecode::Interpreter This commit introduces the concept of an accumulator register to LibJS's bytecode interpreter. The accumulator register is always register 0, and most simple instructions use it for reading and writing. Not only does this slim down the AST, but it also simplifies a lot of the code. For example, the generate_bytecode methods no longer need to return an Optional<Register>, as any opcode which has a "return" value will always put it into the accumulator. This also renames the old Op::Load to Op::LoadImmediate, and uses Op::Load to load from a register into the accumulator. There is also an Op::Store to put the value in the accumulator into another register.
2021-06-07 20:58:36 -07:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_src;
LibJS: Introduce an accumulator register to Bytecode::Interpreter This commit introduces the concept of an accumulator register to LibJS's bytecode interpreter. The accumulator register is always register 0, and most simple instructions use it for reading and writing. Not only does this slim down the AST, but it also simplifies a lot of the code. For example, the generate_bytecode methods no longer need to return an Optional<Register>, as any opcode which has a "return" value will always put it into the accumulator. This also renames the old Op::Load to Op::LoadImmediate, and uses Op::Load to load from a register into the accumulator. There is also an Op::Store to put the value in the accumulator into another register.
2021-06-07 20:58:36 -07:00
};
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
enum class EnvironmentMode {
Lexical,
Var,
};
LibJS/Bytecode: Split SetVariable into four separate instructions Instead of SetVariable having 2x2 modes for variable/lexical and initialize/set, those 4 modes are now separate instructions, which makes each instruction much less branchy.
2024-05-14 11:30:30 +02:00
enum class BindingInitializationMode {
Initialize,
Set,
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CreateLexicalEnvironment final : public Instruction {
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
public:
LibJS: Ensure capacity for created lexical and variable environments If the minimal amount of required bindings is known in advance, it could be used to ensure capacity to avoid resizing the internal vector that holds bindings.
2024-05-09 17:10:20 +02:00
explicit CreateLexicalEnvironment(u32 capacity = 0)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::CreateLexicalEnvironment)
LibJS: Ensure capacity for created lexical and variable environments If the minimal amount of required bindings is known in advance, it could be used to ensure capacity to avoid resizing the internal vector that holds bindings.
2024-05-09 17:10:20 +02:00
, m_capacity(capacity)
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Ensure capacity for created lexical and variable environments If the minimal amount of required bindings is known in advance, it could be used to ensure capacity to avoid resizing the internal vector that holds bindings.
2024-05-09 17:10:20 +02:00
private:
u32 m_capacity { 0 };
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CreateVariableEnvironment final : public Instruction {
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
public:
LibJS: Ensure capacity for created lexical and variable environments If the minimal amount of required bindings is known in advance, it could be used to ensure capacity to avoid resizing the internal vector that holds bindings.
2024-05-09 17:10:20 +02:00
explicit CreateVariableEnvironment(u32 capacity = 0)
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
: Instruction(Type::CreateVariableEnvironment)
LibJS: Ensure capacity for created lexical and variable environments If the minimal amount of required bindings is known in advance, it could be used to ensure capacity to avoid resizing the internal vector that holds bindings.
2024-05-09 17:10:20 +02:00
, m_capacity(capacity)
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Ensure capacity for created lexical and variable environments If the minimal amount of required bindings is known in advance, it could be used to ensure capacity to avoid resizing the internal vector that holds bindings.
2024-05-09 17:10:20 +02:00
private:
u32 m_capacity { 0 };
LibJS: Emit bytecode for function declaration instantiation By doing that all instructions required for instantiation are emitted once in compilation and then reused for subsequent calls, instead of running generic instantiation process for each call.
2024-05-05 22:06:55 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CreatePrivateEnvironment final : public Instruction {
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
public:
explicit CreatePrivateEnvironment()
: Instruction(Type::CreatePrivateEnvironment)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API EnterObjectEnvironment final : public Instruction {
LibJS: Implement bytecode generation for WithStatement
2022-03-13 16:01:18 +03:30
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit EnterObjectEnvironment(Operand object)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::EnterObjectEnvironment)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_object(object)
LibJS: Implement bytecode generation for WithStatement
2022-03-13 16:01:18 +03:30
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand object() const { return m_object; }
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_object);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
private:
Operand m_object;
LibJS/JIT: Update "unwind context" stack in JIT code Until now, the unwind context stack has not been maintained by jitted code, which meant we were unable to support the `with` statement. This is a first step towards supporting that by making jitted code call out to C++ to update the unwind context stack when entering/leaving unwind contexts. We also introduce a new "Catch" bytecode instruction that moves the current exception into the accumulator. It's always emitted at the start of a "catch" block.
2023-11-11 23:19:46 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Catch final : public Instruction {
LibJS/JIT: Update "unwind context" stack in JIT code Until now, the unwind context stack has not been maintained by jitted code, which meant we were unable to support the `with` statement. This is a first step towards supporting that by making jitted code call out to C++ to update the unwind context stack when entering/leaving unwind contexts. We also introduce a new "Catch" bytecode instruction that moves the current exception into the accumulator. It's always emitted at the start of a "catch" block.
2023-11-11 23:19:46 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit Catch(Operand dst)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Catch)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/JIT: Update "unwind context" stack in JIT code Until now, the unwind context stack has not been maintained by jitted code, which meant we were unable to support the `with` statement. This is a first step towards supporting that by making jitted code call out to C++ to update the unwind context stack when entering/leaving unwind contexts. We also introduce a new "Catch" bytecode instruction that moves the current exception into the accumulator. It's always emitted at the start of a "catch" block.
2023-11-11 23:19:46 +01:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
private:
Operand m_dst;
LibJS: Implement bytecode generation for WithStatement
2022-03-13 16:01:18 +03:30
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API LeaveFinally final : public Instruction {
LibJS: Cleanup unwind state when transferring control out of a finalizer This does two things: * Clear exceptions when transferring control out of a finalizer Otherwise they would resurface at the end of the next finalizer (see test the new test case), or at the end of a function * Pop one scheduled jump when transferring control out of a finalizer This removes one old FIXME
2024-04-11 11:58:18 +02:00
public:
explicit LeaveFinally()
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::LeaveFinally)
LibJS: Cleanup unwind state when transferring control out of a finalizer This does two things: * Clear exceptions when transferring control out of a finalizer Otherwise they would resurface at the end of the next finalizer (see test the new test case), or at the end of a function * Pop one scheduled jump when transferring control out of a finalizer This removes one old FIXME
2024-04-11 11:58:18 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Cleanup unwind state when transferring control out of a finalizer This does two things: * Clear exceptions when transferring control out of a finalizer Otherwise they would resurface at the end of the next finalizer (see test the new test case), or at the end of a function * Pop one scheduled jump when transferring control out of a finalizer This removes one old FIXME
2024-04-11 11:58:18 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API RestoreScheduledJump final : public Instruction {
LibJS: Restore scheduled jumps in catch blocks without finalizers
2024-04-11 11:07:35 +02:00
public:
explicit RestoreScheduledJump()
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::RestoreScheduledJump)
LibJS: Restore scheduled jumps in catch blocks without finalizers
2024-04-11 11:07:35 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Restore scheduled jumps in catch blocks without finalizers
2024-04-11 11:07:35 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CreateVariable final : public Instruction {
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
public:
LibJS: Create const variables in ForIn/OfBodyEvaluation in strict mode Our implementation of environment.CreateImmutableBinding(name, true) in this AO was not correctly initializing const variables in strict mode. This would mean that constant declarations in for loop bodies would not throw if they were modified. To fix this, add a new parameter to CreateVariable to set strict mode. Also remove the vm.is_strict mode check here, as it doesn't look like anywhere in the spec will change strict mode depending on whether the script itself is running in script mode or not. This fixes two of our test-js tests, no change to test262.
2023-09-17 21:53:48 +12:00
explicit CreateVariable(IdentifierTableIndex identifier, EnvironmentMode mode, bool is_immutable, bool is_global = false, bool is_strict = false)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::CreateVariable)
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
, m_identifier(identifier)
, m_mode(mode)
, m_is_immutable(is_immutable)
LibJS/Bytecode: Create global variables before setting them This allows them to be accessed before assignment, and also prevents throwing in strict mode as we are trying to set a non-existent variable.
2022-07-17 19:06:44 +01:00
, m_is_global(is_global)
LibJS: Create const variables in ForIn/OfBodyEvaluation in strict mode Our implementation of environment.CreateImmutableBinding(name, true) in this AO was not correctly initializing const variables in strict mode. This would mean that constant declarations in for loop bodies would not throw if they were modified. To fix this, add a new parameter to CreateVariable to set strict mode. Also remove the vm.is_strict mode check here, as it doesn't look like anywhere in the spec will change strict mode depending on whether the script itself is running in script mode or not. This fixes two of our test-js tests, no change to test262.
2023-09-17 21:53:48 +12:00
, m_is_strict(is_strict)
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
LibJS/JIT: Compile the CreateVariable bytecode instruction
2023-10-29 02:26:15 +02:00
IdentifierTableIndex identifier() const { return m_identifier; }
EnvironmentMode mode() const { return m_mode; }
bool is_immutable() const { return m_is_immutable; }
bool is_global() const { return m_is_global; }
bool is_strict() const { return m_is_strict; }
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
private:
IdentifierTableIndex m_identifier;
EnvironmentMode m_mode;
LibJS/Bytecode: Create global variables before setting them This allows them to be accessed before assignment, and also prevents throwing in strict mode as we are trying to set a non-existent variable.
2022-07-17 19:06:44 +01:00
bool m_is_immutable : 4 { false };
bool m_is_global : 4 { false };
LibJS: Create const variables in ForIn/OfBodyEvaluation in strict mode Our implementation of environment.CreateImmutableBinding(name, true) in this AO was not correctly initializing const variables in strict mode. This would mean that constant declarations in for loop bodies would not throw if they were modified. To fix this, add a new parameter to CreateVariable to set strict mode. Also remove the vm.is_strict mode check here, as it doesn't look like anywhere in the spec will change strict mode depending on whether the script itself is running in script mode or not. This fixes two of our test-js tests, no change to test262.
2023-09-17 21:53:48 +12:00
bool m_is_strict { false };
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API InitializeLexicalBinding final : public Instruction {
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
public:
LibJS/Bytecode: Split SetVariable into four separate instructions Instead of SetVariable having 2x2 modes for variable/lexical and initialize/set, those 4 modes are now separate instructions, which makes each instruction much less branchy.
2024-05-14 11:30:30 +02:00
explicit InitializeLexicalBinding(IdentifierTableIndex identifier, Operand src)
: Instruction(Type::InitializeLexicalBinding)
, m_identifier(identifier)
, m_src(src)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS/Bytecode: Split SetVariable into four separate instructions Instead of SetVariable having 2x2 modes for variable/lexical and initialize/set, those 4 modes are now separate instructions, which makes each instruction much less branchy.
2024-05-14 11:30:30 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
IdentifierTableIndex identifier() const { return m_identifier; }
Operand src() const { return m_src; }
private:
IdentifierTableIndex m_identifier;
Operand m_src;
mutable EnvironmentCoordinate m_cache;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API InitializeVariableBinding final : public Instruction {
LibJS/Bytecode: Split SetVariable into four separate instructions Instead of SetVariable having 2x2 modes for variable/lexical and initialize/set, those 4 modes are now separate instructions, which makes each instruction much less branchy.
2024-05-14 11:30:30 +02:00
public:
explicit InitializeVariableBinding(IdentifierTableIndex identifier, Operand src)
: Instruction(Type::InitializeVariableBinding)
, m_identifier(identifier)
, m_src(src)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS/Bytecode: Split SetVariable into four separate instructions Instead of SetVariable having 2x2 modes for variable/lexical and initialize/set, those 4 modes are now separate instructions, which makes each instruction much less branchy.
2024-05-14 11:30:30 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
IdentifierTableIndex identifier() const { return m_identifier; }
Operand src() const { return m_src; }
private:
IdentifierTableIndex m_identifier;
Operand m_src;
mutable EnvironmentCoordinate m_cache;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API SetLexicalBinding final : public Instruction {
LibJS/Bytecode: Split SetVariable into four separate instructions Instead of SetVariable having 2x2 modes for variable/lexical and initialize/set, those 4 modes are now separate instructions, which makes each instruction much less branchy.
2024-05-14 11:30:30 +02:00
public:
explicit SetLexicalBinding(IdentifierTableIndex identifier, Operand src)
: Instruction(Type::SetLexicalBinding)
, m_identifier(identifier)
, m_src(src)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS/Bytecode: Split SetVariable into four separate instructions Instead of SetVariable having 2x2 modes for variable/lexical and initialize/set, those 4 modes are now separate instructions, which makes each instruction much less branchy.
2024-05-14 11:30:30 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
IdentifierTableIndex identifier() const { return m_identifier; }
Operand src() const { return m_src; }
private:
IdentifierTableIndex m_identifier;
Operand m_src;
mutable EnvironmentCoordinate m_cache;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API SetVariableBinding final : public Instruction {
LibJS/Bytecode: Split SetVariable into four separate instructions Instead of SetVariable having 2x2 modes for variable/lexical and initialize/set, those 4 modes are now separate instructions, which makes each instruction much less branchy.
2024-05-14 11:30:30 +02:00
public:
explicit SetVariableBinding(IdentifierTableIndex identifier, Operand src)
: Instruction(Type::SetVariableBinding)
LibJS: Fix all clang-tidy warnings in Bytecode/Op.h - Add missing explicit to constructors - Use move() where appropriate
2021-06-12 11:22:46 +02:00
, m_identifier(identifier)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
LibJS: Expose some internals of Instructions These will be needed in the future to allow optimization passes to check against these
2022-10-22 20:21:08 +02:00
IdentifierTableIndex identifier() const { return m_identifier; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand src() const { return m_src; }
LibJS: Expose some internals of Instructions These will be needed in the future to allow optimization passes to check against these
2022-10-22 20:21:08 +02:00
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
private:
LibJS: Add a separate "identifier table" to bytecode executables This is a specialized string table for storing identifiers only. Identifiers are always FlyStrings, which makes many common operations faster by allowing O(1) comparison.
2021-10-24 15:34:30 +02:00
IdentifierTableIndex m_identifier;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_src;
LibJS/Bytecode: Move environment variable caches into instructions These were out-of-line because we had some ideas about marking instruction streams PROT_READ only, but that seems pretty arbitrary and there's a lot of performance to be gained by putting these inline.
2024-05-11 18:28:03 +02:00
mutable EnvironmentCoordinate m_cache;
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetCalleeAndThisFromEnvironment final : public Instruction {
LibJS: Use correct `this` value when callee is a `with` binding If we're inside of a `with` statement scope, we have to take care to extract the correct `this` value for use in calls when calling a method on the binding object via an Identifier instead of a MemberExpression. This makes Vue.js work way better in the bytecode VM. :^) Also, 1 new pass on test262.
2023-08-01 14:33:58 +02:00
public:
LibJS/Bytecode: Move environment variable caches into instructions These were out-of-line because we had some ideas about marking instruction streams PROT_READ only, but that seems pretty arbitrary and there's a lot of performance to be gained by putting these inline.
2024-05-11 18:28:03 +02:00
explicit GetCalleeAndThisFromEnvironment(Operand callee, Operand this_value, IdentifierTableIndex identifier)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetCalleeAndThisFromEnvironment)
LibJS: Use correct `this` value when callee is a `with` binding If we're inside of a `with` statement scope, we have to take care to extract the correct `this` value for use in calls when calling a method on the binding object via an Identifier instead of a MemberExpression. This makes Vue.js work way better in the bytecode VM. :^) Also, 1 new pass on test262.
2023-08-01 14:33:58 +02:00
, m_identifier(identifier)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_callee(callee)
, m_this_value(this_value)
LibJS: Use correct `this` value when callee is a `with` binding If we're inside of a `with` statement scope, we have to take care to extract the correct `this` value for use in calls when calling a method on the binding object via an Identifier instead of a MemberExpression. This makes Vue.js work way better in the bytecode VM. :^) Also, 1 new pass on test262.
2023-08-01 14:33:58 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_callee);
visitor(m_this_value);
}
LibJS: Use correct `this` value when callee is a `with` binding If we're inside of a `with` statement scope, we have to take care to extract the correct `this` value for use in calls when calling a method on the binding object via an Identifier instead of a MemberExpression. This makes Vue.js work way better in the bytecode VM. :^) Also, 1 new pass on test262.
2023-08-01 14:33:58 +02:00
IdentifierTableIndex identifier() const { return m_identifier; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand callee() const { return m_callee; }
Operand this_() const { return m_this_value; }
LibJS: Use correct `this` value when callee is a `with` binding If we're inside of a `with` statement scope, we have to take care to extract the correct `this` value for use in calls when calling a method on the binding object via an Identifier instead of a MemberExpression. This makes Vue.js work way better in the bytecode VM. :^) Also, 1 new pass on test262.
2023-08-01 14:33:58 +02:00
private:
IdentifierTableIndex m_identifier;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_callee;
Operand m_this_value;
LibJS/Bytecode: Move environment variable caches into instructions These were out-of-line because we had some ideas about marking instruction streams PROT_READ only, but that seems pretty arbitrary and there's a lot of performance to be gained by putting these inline.
2024-05-11 18:28:03 +02:00
mutable EnvironmentCoordinate m_cache;
LibJS: Use correct `this` value when callee is a `with` binding If we're inside of a `with` statement scope, we have to take care to extract the correct `this` value for use in calls when calling a method on the binding object via an Identifier instead of a MemberExpression. This makes Vue.js work way better in the bytecode VM. :^) Also, 1 new pass on test262.
2023-08-01 14:33:58 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetBinding final : public Instruction {
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
public:
LibJS/Bytecode: Rename GetVariable => GetBinding
2024-05-14 11:32:04 +02:00
explicit GetBinding(Operand dst, IdentifierTableIndex identifier)
: Instruction(Type::GetBinding)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Fix all clang-tidy warnings in Bytecode/Op.h - Add missing explicit to constructors - Use move() where appropriate
2021-06-12 11:22:46 +02:00
, m_identifier(identifier)
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS: Expose some internals of Instructions These will be needed in the future to allow optimization passes to check against these
2022-10-22 20:21:08 +02:00
IdentifierTableIndex identifier() const { return m_identifier; }
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
LibJS: Add a separate "identifier table" to bytecode executables This is a specialized string table for storing identifiers only. Identifiers are always FlyStrings, which makes many common operations faster by allowing O(1) comparison.
2021-10-24 15:34:30 +02:00
IdentifierTableIndex m_identifier;
LibJS/Bytecode: Move environment variable caches into instructions These were out-of-line because we had some ideas about marking instruction streams PROT_READ only, but that seems pretty arbitrary and there's a lot of performance to be gained by putting these inline.
2024-05-11 18:28:03 +02:00
mutable EnvironmentCoordinate m_cache;
LibJS: Some more opcodes for the bytecode VM - NewString (allocates a new PrimitiveString from the GC heap) - GetVariable (retrieves a variable in the current scope) - SetVariable (assigns a variable in the current scope)
2021-06-03 18:26:32 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetInitializedBinding final : public Instruction {
LibJS: Optimize reading known-to-be-initialized `var` bindings `var` bindings are never in the temporal dead zone (TDZ), and so we know accessing them will not throw. We now take advantage of this by having a specialized environment binding value getter that doesn't check for exceptional cases. 1.08x speedup on JetStream.
2025-05-04 01:41:49 +02:00
public:
explicit GetInitializedBinding(Operand dst, IdentifierTableIndex identifier)
: Instruction(Type::GetInitializedBinding)
, m_dst(dst)
, m_identifier(identifier)
{
}
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
Operand dst() const { return m_dst; }
IdentifierTableIndex identifier() const { return m_identifier; }
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
private:
Operand m_dst;
IdentifierTableIndex m_identifier;
mutable EnvironmentCoordinate m_cache;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetGlobal final : public Instruction {
LibJS: Add optimized GetGlobal instruction to access global variables Using a special instruction to access global variables allows skipping the environment chain traversal for them and going directly to the module/global environment. Currently, this instruction only caches the offset for bindings that belong to the global object environment. However, there is also an opportunity to cache the offset in the global declarative record. This change results in a 57% increase in speed for imaging-gaussian-blur.js in Kraken.
2023-07-12 04:06:59 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
GetGlobal(Operand dst, IdentifierTableIndex identifier, u32 cache_index)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetGlobal)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Add optimized GetGlobal instruction to access global variables Using a special instruction to access global variables allows skipping the environment chain traversal for them and going directly to the module/global environment. Currently, this instruction only caches the offset for bindings that belong to the global object environment. However, there is also an opportunity to cache the offset in the global declarative record. This change results in a 57% increase in speed for imaging-gaussian-blur.js in Kraken.
2023-07-12 04:06:59 +02:00
, m_identifier(identifier)
, m_cache_index(cache_index)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Add optimized GetGlobal instruction to access global variables Using a special instruction to access global variables allows skipping the environment chain traversal for them and going directly to the module/global environment. Currently, this instruction only caches the offset for bindings that belong to the global object environment. However, there is also an opportunity to cache the offset in the global declarative record. This change results in a 57% increase in speed for imaging-gaussian-blur.js in Kraken.
2023-07-12 04:06:59 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS/JIT: Compile the GetGlobal bytecode instruction
2023-10-20 12:56:12 +02:00
IdentifierTableIndex identifier() const { return m_identifier; }
u32 cache_index() const { return m_cache_index; }
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS: Add optimized GetGlobal instruction to access global variables Using a special instruction to access global variables allows skipping the environment chain traversal for them and going directly to the module/global environment. Currently, this instruction only caches the offset for bindings that belong to the global object environment. However, there is also an opportunity to cache the offset in the global declarative record. This change results in a 57% increase in speed for imaging-gaussian-blur.js in Kraken.
2023-07-12 04:06:59 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
LibJS: Add optimized GetGlobal instruction to access global variables Using a special instruction to access global variables allows skipping the environment chain traversal for them and going directly to the module/global environment. Currently, this instruction only caches the offset for bindings that belong to the global object environment. However, there is also an opportunity to cache the offset in the global declarative record. This change results in a 57% increase in speed for imaging-gaussian-blur.js in Kraken.
2023-07-12 04:06:59 +02:00
IdentifierTableIndex m_identifier;
u32 m_cache_index { 0 };
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API SetGlobal final : public Instruction {
LibJS: Add SetGlobal bytecode instruction for cached global writes Before this change, setting a global would end up as SetLexicalBinding. That instruction always failed to cache the access if the global was a property of the global object. 1.14x speedup on Octane/earley-boyer.js 2.04x speedup on MicroBench/for-of.js Note that MicroBench/for-of.js was more of a "set global" benchmark before this. After this change, it's actually a for..of benchmark. :^)
2025-05-01 23:58:38 +02:00
public:
SetGlobal(IdentifierTableIndex identifier, Operand src, u32 cache_index)
: Instruction(Type::SetGlobal)
, m_src(src)
, m_identifier(identifier)
, m_cache_index(cache_index)
{
}
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
Operand src() const { return m_src; }
IdentifierTableIndex identifier() const { return m_identifier; }
u32 cache_index() const { return m_cache_index; }
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
private:
Operand m_src;
IdentifierTableIndex m_identifier;
u32 m_cache_index { 0 };
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API DeleteVariable final : public Instruction {
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit DeleteVariable(Operand dst, IdentifierTableIndex identifier)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::DeleteVariable)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
, m_identifier(identifier)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS: Expose some internals of Instructions These will be needed in the future to allow optimization passes to check against these
2022-10-22 20:21:08 +02:00
IdentifierTableIndex identifier() const { return m_identifier; }
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
IdentifierTableIndex m_identifier;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetById final : public Instruction {
LibJS: Add GetById bytecode instruction for object property retrieval Same as PutById but in the other direction. :^)
2021-06-04 21:03:53 +02:00
public:
LibJS: Include identifier information in nullish property read access When a GetById / GetByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property read accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 11:26:10 -04:00
GetById(Operand dst, Operand base, IdentifierTableIndex property, Optional<IdentifierTableIndex> base_identifier, u32 cache_index)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetById)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_base(base)
LibJS: Fix all clang-tidy warnings in Bytecode/Op.h - Add missing explicit to constructors - Use move() where appropriate
2021-06-12 11:22:46 +02:00
, m_property(property)
LibJS: Include identifier information in nullish property read access When a GetById / GetByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property read accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 11:26:10 -04:00
, m_base_identifier(move(base_identifier))
LibJS/Bytecode: Cache object own property accesses The instructions GetById and GetByIdWithThis now remember the last-seen Shape, and if we see the same object again, we reuse the property offset from last time without doing a new lookup. This allows us to use Object::get_direct(), bypassing the entire lookup machinery and saving lots of time. ~23% speed-up on Kraken/ai-astar.js :^)
2023-07-08 17:43:26 +02:00
, m_cache_index(cache_index)
LibJS: Add GetById bytecode instruction for object property retrieval Same as PutById but in the other direction. :^)
2021-06-04 21:03:53 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
}
LibJS: Add GetById bytecode instruction for object property retrieval Same as PutById but in the other direction. :^)
2021-06-04 21:03:53 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
LibJS/JIT: Support the GetById bytecode op We can now do basic property (get) access in jitted code! :^)
2023-10-18 13:27:56 +02:00
IdentifierTableIndex property() const { return m_property; }
u32 cache_index() const { return m_cache_index; }
LibJS: Add GetById bytecode instruction for object property retrieval Same as PutById but in the other direction. :^)
2021-06-04 21:03:53 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
LibJS: Add a separate "identifier table" to bytecode executables This is a specialized string table for storing identifiers only. Identifiers are always FlyStrings, which makes many common operations faster by allowing O(1) comparison.
2021-10-24 15:34:30 +02:00
IdentifierTableIndex m_property;
LibJS: Include identifier information in nullish property read access When a GetById / GetByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property read accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 11:26:10 -04:00
Optional<IdentifierTableIndex> m_base_identifier;
LibJS/Bytecode: Cache object own property accesses The instructions GetById and GetByIdWithThis now remember the last-seen Shape, and if we see the same object again, we reuse the property offset from last time without doing a new lookup. This allows us to use Object::get_direct(), bypassing the entire lookup machinery and saving lots of time. ~23% speed-up on Kraken/ai-astar.js :^)
2023-07-08 17:43:26 +02:00
u32 m_cache_index { 0 };
LibJS: Add GetById bytecode instruction for object property retrieval Same as PutById but in the other direction. :^)
2021-06-04 21:03:53 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetCompletionFields final : public Instruction {
LibJS: Make async functions & generators faster with helper types Instead of returning internal generator results as ordinary JS::Objects with properties, we now use GeneratorResult and CompletionCell which both inherit from Cell directly and allow efficient access to state. 1.59x speedup on JetStream3/lazy-collections.js :^)
2025-03-31 09:32:39 +01:00
public:
GetCompletionFields(Operand type_dst, Operand value_dst, Operand completion)
: Instruction(Type::GetCompletionFields)
, m_type_dst(type_dst)
, m_value_dst(value_dst)
, m_completion(completion)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Make async functions & generators faster with helper types Instead of returning internal generator results as ordinary JS::Objects with properties, we now use GeneratorResult and CompletionCell which both inherit from Cell directly and allow efficient access to state. 1.59x speedup on JetStream3/lazy-collections.js :^)
2025-03-31 09:32:39 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_type_dst);
visitor(m_value_dst);
visitor(m_completion);
}
Operand type_dst() const { return m_type_dst; }
Operand value_dst() const { return m_value_dst; }
Operand completion() const { return m_completion; }
private:
Operand m_type_dst;
Operand m_value_dst;
Operand m_completion;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API SetCompletionType final : public Instruction {
LibJS: Make async functions & generators faster with helper types Instead of returning internal generator results as ordinary JS::Objects with properties, we now use GeneratorResult and CompletionCell which both inherit from Cell directly and allow efficient access to state. 1.59x speedup on JetStream3/lazy-collections.js :^)
2025-03-31 09:32:39 +01:00
public:
SetCompletionType(Operand completion, Completion::Type type)
: Instruction(Type::SetCompletionType)
, m_completion(completion)
, m_type(type)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Make async functions & generators faster with helper types Instead of returning internal generator results as ordinary JS::Objects with properties, we now use GeneratorResult and CompletionCell which both inherit from Cell directly and allow efficient access to state. 1.59x speedup on JetStream3/lazy-collections.js :^)
2025-03-31 09:32:39 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_completion);
}
Operand completion() const { return m_completion; }
private:
Operand m_completion;
Completion::Type m_type;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetByIdWithThis final : public Instruction {
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
GetByIdWithThis(Operand dst, Operand base, IdentifierTableIndex property, Operand this_value, u32 cache_index)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetByIdWithThis)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_base(base)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
, m_property(property)
, m_this_value(this_value)
LibJS/Bytecode: Cache object own property accesses The instructions GetById and GetByIdWithThis now remember the last-seen Shape, and if we see the same object again, we reuse the property offset from last time without doing a new lookup. This allows us to use Object::get_direct(), bypassing the entire lookup machinery and saving lots of time. ~23% speed-up on Kraken/ai-astar.js :^)
2023-07-08 17:43:26 +02:00
, m_cache_index(cache_index)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
visitor(m_this_value);
}
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
LibJS/JIT: Compile the GetByIdWithThis instruction
2023-10-29 21:39:02 +01:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand this_value() const { return m_this_value; }
LibJS/JIT: Compile the GetByIdWithThis instruction
2023-10-29 21:39:02 +01:00
u32 cache_index() const { return m_cache_index; }
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
IdentifierTableIndex m_property;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_this_value;
LibJS/Bytecode: Cache object own property accesses The instructions GetById and GetByIdWithThis now remember the last-seen Shape, and if we see the same object again, we reuse the property offset from last time without doing a new lookup. This allows us to use Object::get_direct(), bypassing the entire lookup machinery and saving lots of time. ~23% speed-up on Kraken/ai-astar.js :^)
2023-07-08 17:43:26 +02:00
u32 m_cache_index { 0 };
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetLength final : public Instruction {
LibJS/Bytecode: Add dedicated instruction for getting `length` property By doing this, we can remove all the special-case checks for `length` from the generic GetById and GetByIdWithThis code paths, making every other property lookup a bit faster. Small progressions on most benchmarks, and some larger progressions: - 12% on Octane/crypto.js - 6% on Kraken/ai-astar.js
2024-05-20 11:53:28 +02:00
public:
GetLength(Operand dst, Operand base, Optional<IdentifierTableIndex> base_identifier, u32 cache_index)
: Instruction(Type::GetLength)
, m_dst(dst)
, m_base(base)
, m_base_identifier(move(base_identifier))
, m_cache_index(cache_index)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS/Bytecode: Add dedicated instruction for getting `length` property By doing this, we can remove all the special-case checks for `length` from the generic GetById and GetByIdWithThis code paths, making every other property lookup a bit faster. Small progressions on most benchmarks, and some larger progressions: - 12% on Octane/crypto.js - 6% on Kraken/ai-astar.js
2024-05-20 11:53:28 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
}
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
u32 cache_index() const { return m_cache_index; }
private:
Operand m_dst;
Operand m_base;
Optional<IdentifierTableIndex> m_base_identifier;
u32 m_cache_index { 0 };
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetLengthWithThis final : public Instruction {
LibJS/Bytecode: Add dedicated instruction for getting `length` property By doing this, we can remove all the special-case checks for `length` from the generic GetById and GetByIdWithThis code paths, making every other property lookup a bit faster. Small progressions on most benchmarks, and some larger progressions: - 12% on Octane/crypto.js - 6% on Kraken/ai-astar.js
2024-05-20 11:53:28 +02:00
public:
GetLengthWithThis(Operand dst, Operand base, Operand this_value, u32 cache_index)
: Instruction(Type::GetLengthWithThis)
, m_dst(dst)
, m_base(base)
, m_this_value(this_value)
, m_cache_index(cache_index)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS/Bytecode: Add dedicated instruction for getting `length` property By doing this, we can remove all the special-case checks for `length` from the generic GetById and GetByIdWithThis code paths, making every other property lookup a bit faster. Small progressions on most benchmarks, and some larger progressions: - 12% on Octane/crypto.js - 6% on Kraken/ai-astar.js
2024-05-20 11:53:28 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
visitor(m_this_value);
}
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
Operand this_value() const { return m_this_value; }
u32 cache_index() const { return m_cache_index; }
private:
Operand m_dst;
Operand m_base;
Operand m_this_value;
u32 m_cache_index { 0 };
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetPrivateById final : public Instruction {
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit GetPrivateById(Operand dst, Operand base, IdentifierTableIndex property)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetPrivateById)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_base(base)
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
, m_property(property)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
}
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
LibJS/JIT: Compile the GetPrivateById instruction
2023-10-29 21:24:50 +01:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
IdentifierTableIndex m_property;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API HasPrivateId final : public Instruction {
LibJS/Bytecode: Support `in` binary operator for private fields 11 new passes on test262. :^)
2023-07-05 12:42:50 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
HasPrivateId(Operand dst, Operand base, IdentifierTableIndex property)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::HasPrivateId)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_base(base)
LibJS/Bytecode: Support `in` binary operator for private fields 11 new passes on test262. :^)
2023-07-05 12:42:50 +02:00
, m_property(property)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
}
LibJS/Bytecode: Support `in` binary operator for private fields 11 new passes on test262. :^)
2023-07-05 12:42:50 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
LibJS/JIT: Compile the HasPrivateId instruction
2023-10-30 00:28:13 +01:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Support `in` binary operator for private fields 11 new passes on test262. :^)
2023-07-05 12:42:50 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
LibJS/Bytecode: Support `in` binary operator for private fields 11 new passes on test262. :^)
2023-07-05 12:42:50 +02:00
IdentifierTableIndex m_property;
};
LibJS: Implement bytecode generation for all ObjectExpression properties
2022-03-31 00:59:58 +04:30
enum class PropertyKind {
Getter,
Setter,
KeyValue,
LibJS/Bytecode: Always make own properties in object expressions When building an object from an object expression, we don't want to go through the full property setting machinery. This patch adds a new PropertyKind::DirectKeyValue for PutById which guarantees that the property becomes an own property. This fixes an issue where setting the "__proto__" property in object expressions wasn't working right. 12 new passes on test262. :^)
2023-07-10 08:44:28 +02:00
DirectKeyValue, // Used for Object expressions. Always sets an own property, never calls a setter.
LibJS: Implement bytecode generation for all ObjectExpression properties
2022-03-31 00:59:58 +04:30
ProtoSetter,
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PutBySpread final : public Instruction {
LibJS: Add `PutBySpread` instruction This object property kind had completely different behaviour. By adding an instruction for it, we can remove a bunch of special casing, and avoid creating dummy `PropertyKey` values.
2024-11-01 22:00:32 +01:00
public:
PutBySpread(Operand base, Operand src)
: Instruction(Type::PutBySpread)
, m_base(base)
, m_src(src)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS: Add `PutBySpread` instruction This object property kind had completely different behaviour. By adding an instruction for it, we can remove a bunch of special casing, and avoid creating dummy `PropertyKey` values.
2024-11-01 22:00:32 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_base);
visitor(m_src);
}
Operand base() const { return m_base; }
Operand src() const { return m_src; }
private:
Operand m_base;
Operand m_src;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PutById final : public Instruction {
LibJS: Add PutById bytecode instruction for object property assignment Note that this is only used for non-computed accesses. Computed access is not yet implemented. :^)
2021-06-04 20:47:07 +02:00
public:
LibJS: Include identifier information in nullish property write access When a PutById / PutByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar = 1 Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz = 1 Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property write accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 12:10:52 -04:00
explicit PutById(Operand base, IdentifierTableIndex property, Operand src, PropertyKind kind, u32 cache_index, Optional<IdentifierTableIndex> base_identifier = {})
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::PutById)
LibJS: Devirtualize and pack the bytecode stream :^) This patch changes the LibJS bytecode to be a stream of instructions packed one-after-the-other in contiguous memory, instead of a vector of OwnPtr<Instruction>. This should be a lot more cache-friendly. :^) Instructions are also devirtualized and instead have a type field using a new Instruction::Type enum. To iterate over a bytecode stream, one must now use Bytecode::InstructionStreamIterator.
2021-06-07 15:12:43 +02:00
, m_base(base)
LibJS: Fix all clang-tidy warnings in Bytecode/Op.h - Add missing explicit to constructors - Use move() where appropriate
2021-06-12 11:22:46 +02:00
, m_property(property)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS: Implement bytecode generation for all ObjectExpression properties
2022-03-31 00:59:58 +04:30
, m_kind(kind)
LibJS: Add basic monomorphic caching for PutById property access This patch makes it possible for JS::Object::internal_set() to populate a CacheablePropertyMetadata, and uses this to implement a basic monomorphic cache for the most common form of property write access.
2023-11-08 20:51:26 +01:00
, m_cache_index(cache_index)
LibJS: Include identifier information in nullish property write access When a PutById / PutByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar = 1 Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz = 1 Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property write accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 12:10:52 -04:00
, m_base_identifier(move(base_identifier))
LibJS: Add PutById bytecode instruction for object property assignment Note that this is only used for non-computed accesses. Computed access is not yet implemented. :^)
2021-06-04 20:47:07 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_base);
visitor(m_src);
}
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand base() const { return m_base; }
LibJS/JIT: Compile the PutById bytecode instruction
2023-10-20 13:09:35 +02:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand src() const { return m_src; }
LibJS/JIT: Compile the PutById bytecode instruction
2023-10-20 13:09:35 +02:00
PropertyKind kind() const { return m_kind; }
LibJS: Add basic monomorphic caching for PutById property access This patch makes it possible for JS::Object::internal_set() to populate a CacheablePropertyMetadata, and uses this to implement a basic monomorphic cache for the most common form of property write access.
2023-11-08 20:51:26 +01:00
u32 cache_index() const { return m_cache_index; }
LibJS/JIT: Compile the PutById bytecode instruction
2023-10-20 13:09:35 +02:00
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_base;
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
IdentifierTableIndex m_property;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_src;
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
PropertyKind m_kind;
LibJS: Add basic monomorphic caching for PutById property access This patch makes it possible for JS::Object::internal_set() to populate a CacheablePropertyMetadata, and uses this to implement a basic monomorphic cache for the most common form of property write access.
2023-11-08 20:51:26 +01:00
u32 m_cache_index { 0 };
LibJS: Include identifier information in nullish property write access When a PutById / PutByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar = 1 Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz = 1 Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property write accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 12:10:52 -04:00
Optional<IdentifierTableIndex> m_base_identifier {};
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PutByIdWithThis final : public Instruction {
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
PutByIdWithThis(Operand base, Operand this_value, IdentifierTableIndex property, Operand src, PropertyKind kind, u32 cache_index)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::PutByIdWithThis)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
, m_base(base)
, m_this_value(this_value)
, m_property(property)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
, m_kind(kind)
LibJS: Add basic monomorphic caching for PutById property access This patch makes it possible for JS::Object::internal_set() to populate a CacheablePropertyMetadata, and uses this to implement a basic monomorphic cache for the most common form of property write access.
2023-11-08 20:51:26 +01:00
, m_cache_index(cache_index)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_base);
visitor(m_this_value);
visitor(m_src);
}
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand base() const { return m_base; }
Operand this_value() const { return m_this_value; }
LibJS/JIT: Compile the PutByIdWithThis instruction
2023-10-29 22:14:39 +01:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand src() const { return m_src; }
LibJS/JIT: Compile the PutByIdWithThis instruction
2023-10-29 22:14:39 +01:00
PropertyKind kind() const { return m_kind; }
LibJS: Add basic monomorphic caching for PutById property access This patch makes it possible for JS::Object::internal_set() to populate a CacheablePropertyMetadata, and uses this to implement a basic monomorphic cache for the most common form of property write access.
2023-11-08 20:51:26 +01:00
u32 cache_index() const { return m_cache_index; }
LibJS/JIT: Compile the PutByIdWithThis instruction
2023-10-29 22:14:39 +01:00
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_base;
Operand m_this_value;
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
IdentifierTableIndex m_property;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_src;
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
PropertyKind m_kind;
LibJS: Add basic monomorphic caching for PutById property access This patch makes it possible for JS::Object::internal_set() to populate a CacheablePropertyMetadata, and uses this to implement a basic monomorphic cache for the most common form of property write access.
2023-11-08 20:51:26 +01:00
u32 m_cache_index { 0 };
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PutPrivateById final : public Instruction {
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit PutPrivateById(Operand base, IdentifierTableIndex property, Operand src, PropertyKind kind = PropertyKind::KeyValue)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::PutPrivateById)
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
, m_base(base)
, m_property(property)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS/Bytecode: Support private class fields This is accomplished with two new instructions: - GetPrivateById - PutPrivateById Looks like 1616 new passes on test262. :^)
2023-06-23 08:16:17 +02:00
, m_kind(kind)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_base);
visitor(m_src);
}
LibJS: Add PutById bytecode instruction for object property assignment Note that this is only used for non-computed accesses. Computed access is not yet implemented. :^)
2021-06-04 20:47:07 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand base() const { return m_base; }
LibJS/JIT: Compile the PutPrivateById instruction
2023-10-29 22:20:47 +01:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand src() const { return m_src; }
LibJS/JIT: Compile the PutPrivateById instruction
2023-10-29 22:20:47 +01:00
LibJS: Add PutById bytecode instruction for object property assignment Note that this is only used for non-computed accesses. Computed access is not yet implemented. :^)
2021-06-04 20:47:07 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_base;
LibJS: Add a separate "identifier table" to bytecode executables This is a specialized string table for storing identifiers only. Identifiers are always FlyStrings, which makes many common operations faster by allowing O(1) comparison.
2021-10-24 15:34:30 +02:00
IdentifierTableIndex m_property;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_src;
LibJS: Implement bytecode generation for all ObjectExpression properties
2022-03-31 00:59:58 +04:30
PropertyKind m_kind;
LibJS: Add PutById bytecode instruction for object property assignment Note that this is only used for non-computed accesses. Computed access is not yet implemented. :^)
2021-06-04 20:47:07 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API DeleteById final : public Instruction {
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit DeleteById(Operand dst, Operand base, IdentifierTableIndex property)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::DeleteById)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_base(base)
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
, m_property(property)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
}
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
LibJS/JIT: Compile the DeleteById instruction
2023-10-29 20:29:14 +01:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
IdentifierTableIndex m_property;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API DeleteByIdWithThis final : public Instruction {
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
DeleteByIdWithThis(Operand dst, Operand base, Operand this_value, IdentifierTableIndex property)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::DeleteByIdWithThis)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_base(base)
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
, m_this_value(this_value)
, m_property(property)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
visitor(m_this_value);
}
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
Operand this_value() const { return m_this_value; }
LibJS/JIT: Compile the DeleteByIdWithThis instruction
2023-10-29 22:05:09 +01:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
Operand m_this_value;
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
IdentifierTableIndex m_property;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetByValue final : public Instruction {
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
public:
LibJS: Include identifier information in nullish property read access When a GetById / GetByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property read accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 11:26:10 -04:00
GetByValue(Operand dst, Operand base, Operand property, Optional<IdentifierTableIndex> base_identifier = {})
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetByValue)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
, m_base(base)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_property(property)
LibJS: Include identifier information in nullish property read access When a GetById / GetByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property read accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 11:26:10 -04:00
, m_base_identifier(move(base_identifier))
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
visitor(m_property);
}
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
Operand property() const { return m_property; }
LibJS/Bytecode: Move GetByValue implementation to CommonImplementations
2023-10-20 12:38:43 +02:00
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
Operand m_property;
LibJS: Include identifier information in nullish property read access When a GetById / GetByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property read accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 11:26:10 -04:00
Optional<IdentifierTableIndex> m_base_identifier;
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetByValueWithThis final : public Instruction {
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
GetByValueWithThis(Operand dst, Operand base, Operand property, Operand this_value)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetByValueWithThis)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
, m_base(base)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_property(property)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
, m_this_value(this_value)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
visitor(m_property);
visitor(m_this_value);
}
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
Operand property() const { return m_property; }
Operand this_value() const { return m_this_value; }
LibJS/JIT: Compile the GetByValueWithThis instruction
2023-10-29 21:49:41 +01:00
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
Operand m_property;
Operand m_this_value;
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PutByValue final : public Instruction {
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
public:
LibJS: Include identifier information in nullish property write access When a PutById / PutByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar = 1 Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz = 1 Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property write accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 12:10:52 -04:00
PutByValue(Operand base, Operand property, Operand src, PropertyKind kind = PropertyKind::KeyValue, Optional<IdentifierTableIndex> base_identifier = {})
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::PutByValue)
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
, m_base(base)
, m_property(property)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS: Implement bytecode generation for all ObjectExpression properties
2022-03-31 00:59:58 +04:30
, m_kind(kind)
LibJS: Include identifier information in nullish property write access When a PutById / PutByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar = 1 Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz = 1 Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property write accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 12:10:52 -04:00
, m_base_identifier(move(base_identifier))
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_base);
visitor(m_property);
visitor(m_src);
}
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand base() const { return m_base; }
Operand property() const { return m_property; }
Operand src() const { return m_src; }
LibJS/JIT: Compile the PutByValue bytecode instruction
2023-10-21 15:55:15 +02:00
PropertyKind kind() const { return m_kind; }
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_base;
Operand m_property;
Operand m_src;
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
PropertyKind m_kind;
LibJS: Include identifier information in nullish property write access When a PutById / PutByValue bytecode operation results in accessing a nullish object, we now include the name of the property and the object being accessed in the exception message (if available). This should make it easier to debug live websites. For example, the following errors would all previously produce a generic error message of "ToObject on null or undefined": > foo = null > foo.bar = 1 Uncaught exception: [TypeError] Cannot access property "bar" on null object "foo" at <unknown> > foo = { bar: undefined } > foo.bar.baz = 1 Uncaught exception: [TypeError] Cannot access property "baz" on undefined object "foo.bar" at <unknown> Note we certainly don't capture all possible nullish property write accesses here. This just covers cases I've seen most on live websites; we can cover more cases as they arise.
2024-03-29 12:10:52 -04:00
Optional<IdentifierTableIndex> m_base_identifier;
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PutByValueWithThis final : public Instruction {
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
PutByValueWithThis(Operand base, Operand property, Operand this_value, Operand src, PropertyKind kind = PropertyKind::KeyValue)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::PutByValueWithThis)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
, m_base(base)
, m_property(property)
, m_this_value(this_value)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS/Bytecode: Use proper `this` for receiver in get/set for super expr Summary: Diff Tests: +14 ✅ -2 ❌ -12 📝
2023-07-02 19:26:31 +01:00
, m_kind(kind)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_base);
visitor(m_property);
visitor(m_this_value);
visitor(m_src);
}
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand base() const { return m_base; }
Operand property() const { return m_property; }
Operand this_value() const { return m_this_value; }
Operand src() const { return m_src; }
LibJS/JIT: Compile the PutByValueWithThis instruction
2023-10-30 01:32:20 +01:00
PropertyKind kind() const { return m_kind; }
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_base;
Operand m_property;
Operand m_this_value;
Operand m_src;
LibJS: Implement bytecode generation for all ObjectExpression properties
2022-03-31 00:59:58 +04:30
PropertyKind m_kind;
LibJS: Basic bytecode support for computed member expressions Expressions like foo[1 + 2] now work, and you can assign to them as well! :^)
2021-06-11 00:35:25 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API DeleteByValue final : public Instruction {
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
DeleteByValue(Operand dst, Operand base, Operand property)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::DeleteByValue)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
, m_base(base)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_property(property)
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
visitor(m_property);
}
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
Operand property() const { return m_property; }
LibJS/JIT: Compile the DeleteByValue instruction
2023-10-29 21:20:25 +01:00
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
Operand m_property;
LibJS/Bytecode: Implement the delete unary expression `delete` has to operate directly on Reference Records, so this introduces a new set of operations called DeleteByValue, DeleteVariable and DeleteById. They operate similarly to their Get counterparts, except they end in creating a (temporary) Reference and calling delete_ on it.
2022-03-27 19:50:09 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API DeleteByValueWithThis final : public Instruction {
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
DeleteByValueWithThis(Operand dst, Operand base, Operand this_value, Operand property)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::DeleteByValueWithThis)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
, m_base(base)
, m_this_value(this_value)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_property(property)
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
{
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand base() const { return m_base; }
Operand this_value() const { return m_this_value; }
Operand property() const { return m_property; }
LibJS/JIT: Compile the DeleteByValueWithThis instruction
2023-10-29 21:37:52 +01:00
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_base);
visitor(m_this_value);
visitor(m_property);
}
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_base;
Operand m_this_value;
Operand m_property;
LibJS/Bytecode: Generate bytecode for deleting super properties
2023-07-06 17:10:40 -04:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Jump final : public Instruction {
LibJS: Add basic support for while loops in the bytecode engine This introduces two new instructions: Jump and JumpIfFalse. Jumps are made to a Bytecode::Label, which is a simple object that represents a location in the bytecode stream. Note that you may not always know the target of a jump when adding the jump instruction itself, but we can just update the instruction later on during codegen once we know where the jump target is. The Bytecode::Interpreter now implements jumping via a jump slot that gets checked after each instruction to see if a jump is pending. If not, we just increment the PC as usual.
2021-06-04 12:07:38 +02:00
public:
LibJS: Generate bytecode in basic blocks instead of one big block This limits the size of each block (currently set to 1K), and gets us closer to a canonical, more easily analysable bytecode format. As a result of this, "Labels" are now simply entries to basic blocks. Since there is no more 'conditional' jump (as all jumps are always taken), JumpIf{True,False} are unified to JumpConditional, and JumpIfNullish is renamed to JumpNullish. Also fixes #7914 as a result of reimplementing the loop logic.
2021-06-09 06:49:58 +04:30
constexpr static bool IsTerminator = true;
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
explicit Jump(Label target)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Jump)
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
, m_target(target)
LibJS: Add basic support for while loops in the bytecode engine This introduces two new instructions: Jump and JumpIfFalse. Jumps are made to a Bytecode::Label, which is a simple object that represents a location in the bytecode stream. Note that you may not always know the target of a jump when adding the jump instruction itself, but we can just update the instruction later on during codegen once we know where the jump target is. The Bytecode::Interpreter now implements jumping via a jump slot that gets checked after each instruction to see if a jump is pending. If not, we just increment the PC as usual.
2021-06-04 12:07:38 +02:00
{
}
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_target);
}
LibJS: Add a basic pass manager and add some basic passes This commit adds a bunch of passes, the most interesting of which is a pass that merges blocks together, and a pass that places blocks that flow into each other next to each other, and a very simply pass that removes duplicate basic blocks. Note that this does not remove the jump at the end of each block in that pass to avoid scope creep in the passes.
2021-06-13 20:40:20 +04:30
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
auto& target() const { return m_target; }
LibJS: Generate bytecode in basic blocks instead of one big block This limits the size of each block (currently set to 1K), and gets us closer to a canonical, more easily analysable bytecode format. As a result of this, "Labels" are now simply entries to basic blocks. Since there is no more 'conditional' jump (as all jumps are always taken), JumpIf{True,False} are unified to JumpConditional, and JumpIfNullish is renamed to JumpNullish. Also fixes #7914 as a result of reimplementing the loop logic.
2021-06-09 06:49:58 +04:30
protected:
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
Label m_target;
LibJS: Add basic support for while loops in the bytecode engine This introduces two new instructions: Jump and JumpIfFalse. Jumps are made to a Bytecode::Label, which is a simple object that represents a location in the bytecode stream. Note that you may not always know the target of a jump when adding the jump instruction itself, but we can just update the instruction later on during codegen once we know where the jump target is. The Bytecode::Interpreter now implements jumping via a jump slot that gets checked after each instruction to see if a jump is pending. If not, we just increment the PC as usual.
2021-06-04 12:07:38 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API JumpIf final : public Instruction {
LibJS: Generate bytecode for do...while statements :^) This was quite straightforward using the same label/jump machinery that we added for while statements. The main addition here is a new JumpIfTrue bytecode instruction.
2021-06-04 12:20:44 +02:00
public:
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
constexpr static bool IsTerminator = true;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit JumpIf(Operand condition, Label true_target, Label false_target)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::JumpIf)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_condition(condition)
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
, m_true_target(true_target)
, m_false_target(false_target)
LibJS: Generate bytecode for do...while statements :^) This was quite straightforward using the same label/jump machinery that we added for while statements. The main addition here is a new JumpIfTrue bytecode instruction.
2021-06-04 12:20:44 +02:00
{
}
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_true_target);
visitor(m_false_target);
}
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_condition);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand condition() const { return m_condition; }
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
auto& true_target() const { return m_true_target; }
auto& false_target() const { return m_false_target; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
private:
Operand m_condition;
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
Label m_true_target;
Label m_false_target;
LibJS: Generate bytecode for do...while statements :^) This was quite straightforward using the same label/jump machinery that we added for while statements. The main addition here is a new JumpIfTrue bytecode instruction.
2021-06-04 12:20:44 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API JumpTrue final : public Instruction {
LibJS/Bytecode: Turn JumpIf condition,@a,@next into JumpTrue/JumpFalse If one of the jump targets is the very next block, we can convert the jump instruction into a smaller JumpTrue or JumpFalse.
2024-05-06 10:15:17 +02:00
public:
constexpr static bool IsTerminator = true;
explicit JumpTrue(Operand condition, Label target)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::JumpTrue)
LibJS/Bytecode: Turn JumpIf condition,@a,@next into JumpTrue/JumpFalse If one of the jump targets is the very next block, we can convert the jump instruction into a smaller JumpTrue or JumpFalse.
2024-05-06 10:15:17 +02:00
, m_condition(condition)
, m_target(target)
{
}
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_target);
}
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_condition);
}
LibJS/Bytecode: Turn JumpIf condition,@a,@next into JumpTrue/JumpFalse If one of the jump targets is the very next block, we can convert the jump instruction into a smaller JumpTrue or JumpFalse.
2024-05-06 10:15:17 +02:00
Operand condition() const { return m_condition; }
auto& target() const { return m_target; }
private:
Operand m_condition;
Label m_target;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API JumpFalse final : public Instruction {
LibJS/Bytecode: Turn JumpIf condition,@a,@next into JumpTrue/JumpFalse If one of the jump targets is the very next block, we can convert the jump instruction into a smaller JumpTrue or JumpFalse.
2024-05-06 10:15:17 +02:00
public:
constexpr static bool IsTerminator = true;
explicit JumpFalse(Operand condition, Label target)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::JumpFalse)
LibJS/Bytecode: Turn JumpIf condition,@a,@next into JumpTrue/JumpFalse If one of the jump targets is the very next block, we can convert the jump instruction into a smaller JumpTrue or JumpFalse.
2024-05-06 10:15:17 +02:00
, m_condition(condition)
, m_target(target)
{
}
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_target);
}
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_condition);
}
LibJS/Bytecode: Turn JumpIf condition,@a,@next into JumpTrue/JumpFalse If one of the jump targets is the very next block, we can convert the jump instruction into a smaller JumpTrue or JumpFalse.
2024-05-06 10:15:17 +02:00
Operand condition() const { return m_condition; }
auto& target() const { return m_target; }
private:
Operand m_condition;
Label m_target;
};
LibJS/Bytecode: Add fast paths for compare-and-jump with 2 numbers When comparing two numbers, we can avoid a lot of implicit type conversion nonsense and go straight to comparison, saving time in the most common case.
2024-05-13 09:23:53 +02:00
#define JS_ENUMERATE_COMPARISON_OPS(X) \
X(LessThan, less_than, <) \
X(LessThanEquals, less_than_equals, <=) \
X(GreaterThan, greater_than, >) \
X(GreaterThanEquals, greater_than_equals, >=) \
X(LooselyEquals, loosely_equals, ==) \
X(LooselyInequals, loosely_inequals, !=) \
X(StrictlyEquals, strict_equals, ==) \
X(StrictlyInequals, strict_inequals, !=)
#define DECLARE_COMPARISON_OP(op_TitleCase, op_snake_case, numeric_operator) \
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Jump##op_TitleCase final : public Instruction { \
LibJS/Bytecode: Do basic compare-and-jump peephole optimization We now fuse sequences like [LessThan, JumpIf] to JumpLessThan. This is only allowed for temporaries (i.e VM registers) with no other references to them.
2024-05-09 15:13:31 +02:00
public: \
constexpr static bool IsTerminator = true; \
\
explicit Jump##op_TitleCase(Operand lhs, Operand rhs, Label true_target, Label false_target) \
: Instruction(Type::Jump##op_TitleCase) \
, m_lhs(lhs) \
, m_rhs(rhs) \
, m_true_target(true_target) \
, m_false_target(false_target) \
{ \
} \
\
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const; \
LibJS/Bytecode: Do basic compare-and-jump peephole optimization We now fuse sequences like [LessThan, JumpIf] to JumpLessThan. This is only allowed for temporaries (i.e VM registers) with no other references to them.
2024-05-09 15:13:31 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const; \
void visit_labels_impl(Function<void(Label&)> visitor) \
{ \
visitor(m_true_target); \
visitor(m_false_target); \
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
} \
void visit_operands_impl(Function<void(Operand&)> visitor) \
{ \
visitor(m_lhs); \
visitor(m_rhs); \
LibJS/Bytecode: Do basic compare-and-jump peephole optimization We now fuse sequences like [LessThan, JumpIf] to JumpLessThan. This is only allowed for temporaries (i.e VM registers) with no other references to them.
2024-05-09 15:13:31 +02:00
} \
\
Operand lhs() const { return m_lhs; } \
Operand rhs() const { return m_rhs; } \
auto& true_target() const { return m_true_target; } \
auto& false_target() const { return m_false_target; } \
\
private: \
Operand m_lhs; \
Operand m_rhs; \
Label m_true_target; \
Label m_false_target; \
};
JS_ENUMERATE_COMPARISON_OPS(DECLARE_COMPARISON_OP)
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API JumpNullish final : public Instruction {
LibJS: Implement bytecode ops for logical expressions
2021-06-08 02:18:47 +02:00
public:
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
constexpr static bool IsTerminator = true;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit JumpNullish(Operand condition, Label true_target, Label false_target)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::JumpNullish)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_condition(condition)
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
, m_true_target(true_target)
, m_false_target(false_target)
LibJS: Implement bytecode ops for logical expressions
2021-06-08 02:18:47 +02:00
{
}
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_true_target);
visitor(m_false_target);
}
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_condition);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand condition() const { return m_condition; }
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
auto& true_target() const { return m_true_target; }
auto& false_target() const { return m_false_target; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
private:
Operand m_condition;
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
Label m_true_target;
Label m_false_target;
LibJS: Implement bytecode ops for logical expressions
2021-06-08 02:18:47 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API JumpUndefined final : public Instruction {
LibJS: Add JumpUndefined bytecode
2021-06-13 12:24:40 -07:00
public:
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
constexpr static bool IsTerminator = true;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit JumpUndefined(Operand condition, Label true_target, Label false_target)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::JumpUndefined)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_condition(condition)
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
, m_true_target(true_target)
, m_false_target(false_target)
LibJS: Add JumpUndefined bytecode
2021-06-13 12:24:40 -07:00
{
}
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_true_target);
visitor(m_false_target);
}
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_condition);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand condition() const { return m_condition; }
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
auto& true_target() const { return m_true_target; }
auto& false_target() const { return m_false_target; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
private:
Operand m_condition;
LibJS/Bytecode: Make each Jump instruction inherit Instruction directly Before this change, all JumpFoo instructions inherited from Jump, which forced the unconditional Jump to have an unusued "false target" member. Also, labels were unnecessarily wrapped in Optional<>. By defining each jump instruction separately, they all shrink in size, and all ambiguity is removed.
2024-05-05 11:52:14 +02:00
Label m_true_target;
Label m_false_target;
LibJS: Add JumpUndefined bytecode
2021-06-13 12:24:40 -07:00
};
LibJS/Bytecode: Reorder Call instruction members to make it smaller
2024-05-10 07:12:18 +02:00
enum class CallType : u8 {
LibJS/Bytecode: Rename Call and SuperCall to &WithArgumentArray Forcing every function call to allocate a new Array just to accommodate spread parameters is not very nice, so let's start moving towards making this a special case rather than the general (and only) case.
2023-07-02 15:59:54 +02:00
Call,
Construct,
DirectEval,
};
LibJS: Very basic support for "new" construction in bytecode VM This patch adds a CallType to the Bytecode::Op::Call instruction, which can be either Call or Construct. We then generate Construct calls for the NewExpression AST node. When executed, these get fed into VM::construct().
2021-06-10 23:01:49 +02:00
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Call final : public Instruction {
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
public:
LibJS/Bytecode: Unroll the bytecode interpreter This commit adds a HANDLE_INSTRUCTION macro that expands to everything needed to handle a single instruction (invoking the handler function, checking for exceptions, and advancing the program counter). This gives a ~15% speed-up on a for loop microbenchmark, and makes basically everything faster.
2024-05-06 10:42:52 +02:00
static constexpr bool IsVariableLength = true;
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
Call(Operand dst, Operand callee, Operand this_value, ReadonlySpan<ScopedOperand> arguments, Optional<StringTableIndex> expression_string = {})
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Call)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
, m_callee(callee)
, m_this_value(this_value)
LibJS/Bytecode: Put arguments directly in the Call instruction Instead of having Call refer to a range of VM registers, it now has a trailing list of argument operands as part of the instruction. This means we no longer have to shuffle every argument value into a register before making a call, making bytecode smaller & faster. :^)
2024-02-20 15:59:45 +01:00
, m_argument_count(arguments.size())
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
, m_expression_string(expression_string)
{
for (size_t i = 0; i < arguments.size(); ++i)
m_arguments[i] = arguments[i];
}
LibJS: Overload Instruction::length() on variable-length instructions This allows the DISPATCH_NEXT() macro to do the math to find the next instruction boundary point without performing a function call.
2025-04-28 23:16:26 +02:00
size_t length() const { return length_impl(); }
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
size_t length_impl() const
{
return round_up_to_power_of_two(alignof(void*), sizeof(*this) + sizeof(Operand) * m_argument_count);
}
Operand dst() const { return m_dst; }
Operand callee() const { return m_callee; }
Operand this_value() const { return m_this_value; }
Optional<StringTableIndex> const& expression_string() const { return m_expression_string; }
u32 argument_count() const { return m_argument_count; }
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_callee);
visitor(m_this_value);
for (size_t i = 0; i < m_argument_count; i++)
visitor(m_arguments[i]);
}
private:
Operand m_dst;
Operand m_callee;
Operand m_this_value;
u32 m_argument_count { 0 };
Optional<StringTableIndex> m_expression_string;
Operand m_arguments[];
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CallBuiltin final : public Instruction {
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
public:
static constexpr bool IsVariableLength = true;
CallBuiltin(Operand dst, Operand callee, Operand this_value, ReadonlySpan<ScopedOperand> arguments, Builtin builtin, Optional<StringTableIndex> expression_string = {})
: Instruction(Type::CallBuiltin)
, m_dst(dst)
, m_callee(callee)
, m_this_value(this_value)
, m_argument_count(arguments.size())
LibJS: Introduce Builtins Builtins are functions that can be detected during bytecode generation and enable fast-paths in the JIT.
2023-11-17 11:48:30 +01:00
, m_builtin(builtin)
LibJS/Bytecode: Reorder Call instruction members to make it smaller
2024-05-10 07:12:18 +02:00
, m_expression_string(expression_string)
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
{
LibJS/Bytecode: Put arguments directly in the Call instruction Instead of having Call refer to a range of VM registers, it now has a trailing list of argument operands as part of the instruction. This means we no longer have to shuffle every argument value into a register before making a call, making bytecode smaller & faster. :^)
2024-02-20 15:59:45 +01:00
for (size_t i = 0; i < arguments.size(); ++i)
m_arguments[i] = arguments[i];
}
LibJS: Overload Instruction::length() on variable-length instructions This allows the DISPATCH_NEXT() macro to do the math to find the next instruction boundary point without performing a function call.
2025-04-28 23:16:26 +02:00
size_t length() const { return length_impl(); }
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
size_t length_impl() const
LibJS/Bytecode: Put arguments directly in the Call instruction Instead of having Call refer to a range of VM registers, it now has a trailing list of argument operands as part of the instruction. This means we no longer have to shuffle every argument value into a register before making a call, making bytecode smaller & faster. :^)
2024-02-20 15:59:45 +01:00
{
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
return round_up_to_power_of_two(alignof(void*), sizeof(*this) + sizeof(Operand) * m_argument_count);
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand callee() const { return m_callee; }
Operand this_value() const { return m_this_value; }
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
Optional<StringTableIndex> const& expression_string() const { return m_expression_string; }
LibJS: Handle `Call`s argument range during `EliminateLoads` pass Otherwise we could forget to emit stores those registers, thinking that they are only directly accessed.
2023-07-03 15:15:24 +02:00
u32 argument_count() const { return m_argument_count; }
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
Builtin const& builtin() const { return m_builtin; }
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_callee);
visitor(m_this_value);
for (size_t i = 0; i < m_argument_count; i++)
visitor(m_arguments[i]);
}
private:
Operand m_dst;
Operand m_callee;
Operand m_this_value;
u32 m_argument_count { 0 };
Builtin m_builtin;
Optional<StringTableIndex> m_expression_string;
Operand m_arguments[];
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CallConstruct final : public Instruction {
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
public:
static constexpr bool IsVariableLength = true;
LibJS: Remove unused `this` value from CallConstruct instruction There's no `this` value prior in the caller context, and this was never actually used by CallConstruct.
2025-04-06 22:50:55 +02:00
CallConstruct(Operand dst, Operand callee, ReadonlySpan<ScopedOperand> arguments, Optional<StringTableIndex> expression_string = {})
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
: Instruction(Type::CallConstruct)
, m_dst(dst)
, m_callee(callee)
, m_argument_count(arguments.size())
, m_expression_string(expression_string)
{
for (size_t i = 0; i < arguments.size(); ++i)
m_arguments[i] = arguments[i];
}
LibJS: Overload Instruction::length() on variable-length instructions This allows the DISPATCH_NEXT() macro to do the math to find the next instruction boundary point without performing a function call.
2025-04-28 23:16:26 +02:00
size_t length() const { return length_impl(); }
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
size_t length_impl() const
{
return round_up_to_power_of_two(alignof(void*), sizeof(*this) + sizeof(Operand) * m_argument_count);
}
Operand dst() const { return m_dst; }
Operand callee() const { return m_callee; }
Optional<StringTableIndex> const& expression_string() const { return m_expression_string; }
u32 argument_count() const { return m_argument_count; }
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_callee);
for (size_t i = 0; i < m_argument_count; i++)
visitor(m_arguments[i]);
}
private:
Operand m_dst;
Operand m_callee;
u32 m_argument_count { 0 };
Optional<StringTableIndex> m_expression_string;
Operand m_arguments[];
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CallDirectEval final : public Instruction {
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
public:
static constexpr bool IsVariableLength = true;
CallDirectEval(Operand dst, Operand callee, Operand this_value, ReadonlySpan<ScopedOperand> arguments, Optional<StringTableIndex> expression_string = {})
: Instruction(Type::CallDirectEval)
, m_dst(dst)
, m_callee(callee)
, m_this_value(this_value)
, m_argument_count(arguments.size())
, m_expression_string(expression_string)
{
for (size_t i = 0; i < arguments.size(); ++i)
m_arguments[i] = arguments[i];
}
LibJS: Overload Instruction::length() on variable-length instructions This allows the DISPATCH_NEXT() macro to do the math to find the next instruction boundary point without performing a function call.
2025-04-28 23:16:26 +02:00
size_t length() const { return length_impl(); }
LibJS: Split `Call` Instruction by `CallType` Instead of branching on the `CallType` at runtime in a hot instruction like `Call`, lets emit separate instructions for each call type during codegen.
2024-10-31 22:47:30 +01:00
size_t length_impl() const
{
return round_up_to_power_of_two(alignof(void*), sizeof(*this) + sizeof(Operand) * m_argument_count);
}
Operand dst() const { return m_dst; }
Operand callee() const { return m_callee; }
Operand this_value() const { return m_this_value; }
Optional<StringTableIndex> const& expression_string() const { return m_expression_string; }
u32 argument_count() const { return m_argument_count; }
LibJS: Introduce Builtins Builtins are functions that can be detected during bytecode generation and enable fast-paths in the JIT.
2023-11-17 11:48:30 +01:00
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_callee);
visitor(m_this_value);
for (size_t i = 0; i < m_argument_count; i++)
visitor(m_arguments[i]);
}
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_callee;
Operand m_this_value;
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
u32 m_argument_count { 0 };
LibJS/Bytecode: Reorder Call instruction members to make it smaller
2024-05-10 07:12:18 +02:00
Optional<StringTableIndex> m_expression_string;
LibJS/Bytecode: Put arguments directly in the Call instruction Instead of having Call refer to a range of VM registers, it now has a trailing list of argument operands as part of the instruction. This means we no longer have to shuffle every argument value into a register before making a call, making bytecode smaller & faster. :^)
2024-02-20 15:59:45 +01:00
Operand m_arguments[];
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API CallWithArgumentArray final : public Instruction {
LibJS/Bytecode: Rename Call and SuperCall to &WithArgumentArray Forcing every function call to allocate a new Array just to accommodate spread parameters is not very nice, so let's start moving towards making this a special case rather than the general (and only) case.
2023-07-02 15:59:54 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
CallWithArgumentArray(CallType type, Operand dst, Operand callee, Operand this_value, Operand arguments, Optional<StringTableIndex> expression_string = {})
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::CallWithArgumentArray)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Support basic function calls in the bytecode world :^) This patch adds the Call bytecode instruction which is emitted for the CallExpression AST node. It's pretty barebones and doesn't handle 'this' values properly, etc. But it can perform basic function calls! :^) Note that the called function will *not* execute as bytecode, but will simply fall back into the old codepath and use the AST interpreter.
2021-06-05 15:15:30 +02:00
, m_callee(callee)
, m_this_value(this_value)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_arguments(arguments)
LibJS: Very basic support for "new" construction in bytecode VM This patch adds a CallType to the Bytecode::Op::Call instruction, which can be either Call or Construct. We then generate Construct calls for the NewExpression AST node. When executed, these get fed into VM::construct().
2021-06-10 23:01:49 +02:00
, m_type(type)
LibJS: Forward a string aproximation of the CallExpression to Call Ops This gives us better debug output when analysing calls to `undefined` and also fixes multiple test-js cases expecting an `(evaluated from $Expression)` in the error message. This also refactors out the generation of that string, to avoid code duplication with the AST interpreter.
2022-10-01 01:36:06 +02:00
, m_expression_string(expression_string)
LibJS: Support basic function calls in the bytecode world :^) This patch adds the Call bytecode instruction which is emitted for the CallExpression AST node. It's pretty barebones and doesn't handle 'this' values properly, etc. But it can perform basic function calls! :^) Note that the called function will *not* execute as bytecode, but will simply fall back into the old codepath and use the AST interpreter.
2021-06-05 15:15:30 +02:00
{
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
CallType call_type() const { return m_type; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand callee() const { return m_callee; }
Operand this_value() const { return m_this_value; }
Operand arguments() const { return m_arguments; }
LibJS/Bytecode: Add Call opcode for fixed-argument-count calls This avoids the overhead of allocating a new Array on every function call, saving a substantial amount of time and avoiding GC thrash. This patch only makes use of Op::Call in CallExpression. There are other places we should codegen this op. We should also do the same for super expression calls. ~5% speed-up on Kraken/stanford-crypto-ccm.js
2023-07-02 16:33:00 +02:00
Optional<StringTableIndex> const& expression_string() const { return m_expression_string; }
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_callee);
visitor(m_this_value);
visitor(m_arguments);
}
LibJS: Devirtualize and pack the bytecode stream :^) This patch changes the LibJS bytecode to be a stream of instructions packed one-after-the-other in contiguous memory, instead of a vector of OwnPtr<Instruction>. This should be a lot more cache-friendly. :^) Instructions are also devirtualized and instead have a type field using a new Instruction::Type enum. To iterate over a bytecode stream, one must now use Bytecode::InstructionStreamIterator.
2021-06-07 15:12:43 +02:00
LibJS: Support basic function calls in the bytecode world :^) This patch adds the Call bytecode instruction which is emitted for the CallExpression AST node. It's pretty barebones and doesn't handle 'this' values properly, etc. But it can perform basic function calls! :^) Note that the called function will *not* execute as bytecode, but will simply fall back into the old codepath and use the AST interpreter.
2021-06-05 15:15:30 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_callee;
Operand m_this_value;
Operand m_arguments;
LibJS: Very basic support for "new" construction in bytecode VM This patch adds a CallType to the Bytecode::Op::Call instruction, which can be either Call or Construct. We then generate Construct calls for the NewExpression AST node. When executed, these get fed into VM::construct().
2021-06-10 23:01:49 +02:00
CallType m_type;
LibJS: Forward a string aproximation of the CallExpression to Call Ops This gives us better debug output when analysing calls to `undefined` and also fixes multiple test-js cases expecting an `(evaluated from $Expression)` in the error message. This also refactors out the generation of that string, to avoid code duplication with the AST interpreter.
2022-10-01 01:36:06 +02:00
Optional<StringTableIndex> m_expression_string;
LibJS: Support basic function calls in the bytecode world :^) This patch adds the Call bytecode instruction which is emitted for the CallExpression AST node. It's pretty barebones and doesn't handle 'this' values properly, etc. But it can perform basic function calls! :^) Note that the called function will *not* execute as bytecode, but will simply fall back into the old codepath and use the AST interpreter.
2021-06-05 15:15:30 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API SuperCallWithArgumentArray : public Instruction {
LibJS: Implement SuperCall for the Bytecode-VM
2022-08-30 18:03:02 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit SuperCallWithArgumentArray(Operand dst, Operand arguments, bool is_synthetic)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::SuperCallWithArgumentArray)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_arguments(arguments)
LibJS: Implement SuperCall for the Bytecode-VM
2022-08-30 18:03:02 +02:00
, m_is_synthetic(is_synthetic)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_arguments);
}
LibJS: Implement SuperCall for the Bytecode-VM
2022-08-30 18:03:02 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand arguments() const { return m_arguments; }
LibJS/JIT: Compile the SuperCallWithArgumentArray instruction
2023-10-29 15:34:01 +01:00
bool is_synthetic() const { return m_is_synthetic; }
LibJS: Implement SuperCall for the Bytecode-VM
2022-08-30 18:03:02 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_arguments;
LibJS: Implement SuperCall for the Bytecode-VM
2022-08-30 18:03:02 +02:00
bool m_is_synthetic;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API NewClass final : public Instruction {
LibJS: NewClass bytecode instruction This adds a the NewClass bytecode instruction, enough of it is implemented for it to show it in the bytecode (js -d).
2021-06-30 15:42:13 -03:00
public:
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
static constexpr bool IsVariableLength = true;
explicit NewClass(Operand dst, Optional<Operand> super_class, ClassExpression const& class_expression, Optional<IdentifierTableIndex> lhs_name, ReadonlySpan<Optional<ScopedOperand>> elements_keys)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::NewClass)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_super_class(super_class)
LibJS: NewClass bytecode instruction This adds a the NewClass bytecode instruction, enough of it is implemented for it to show it in the bytecode (js -d).
2021-06-30 15:42:13 -03:00
, m_class_expression(class_expression)
LibJS: Use the IdentifierTable for NewFunction and NewClass lhs names This makes them trivially copyable, which is an assumption multiple optimizations use when rebuilding the instruction stream. This fixes most optimized crashes in the test262 suite.
2023-06-28 18:17:13 +02:00
, m_lhs_name(lhs_name)
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
, m_element_keys_count(elements_keys.size())
{
for (size_t i = 0; i < m_element_keys_count; i++) {
if (elements_keys[i].has_value())
m_element_keys[i] = elements_keys[i]->operand();
}
}
LibJS: Overload Instruction::length() on variable-length instructions This allows the DISPATCH_NEXT() macro to do the math to find the next instruction boundary point without performing a function call.
2025-04-28 23:16:26 +02:00
size_t length() const { return length_impl(); }
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
size_t length_impl() const
LibJS: NewClass bytecode instruction This adds a the NewClass bytecode instruction, enough of it is implemented for it to show it in the bytecode (js -d).
2021-06-30 15:42:13 -03:00
{
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
return round_up_to_power_of_two(alignof(void*), sizeof(*this) + sizeof(Optional<Operand>) * m_element_keys_count);
LibJS: NewClass bytecode instruction This adds a the NewClass bytecode instruction, enough of it is implemented for it to show it in the bytecode (js -d).
2021-06-30 15:42:13 -03:00
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
if (m_super_class.has_value())
visitor(m_super_class.value());
for (size_t i = 0; i < m_element_keys_count; i++) {
if (m_element_keys[i].has_value())
visitor(m_element_keys[i].value());
}
}
LibJS: NewClass bytecode instruction This adds a the NewClass bytecode instruction, enough of it is implemented for it to show it in the bytecode (js -d).
2021-06-30 15:42:13 -03:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Optional<Operand> const& super_class() const { return m_super_class; }
LibJS/JIT: Compile the NewClass bytecode instruction
2023-10-29 02:59:50 +01:00
ClassExpression const& class_expression() const { return m_class_expression; }
Optional<IdentifierTableIndex> const& lhs_name() const { return m_lhs_name; }
LibJS: NewClass bytecode instruction This adds a the NewClass bytecode instruction, enough of it is implemented for it to show it in the bytecode (js -d).
2021-06-30 15:42:13 -03:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Optional<Operand> m_super_class;
LibJS: NewClass bytecode instruction This adds a the NewClass bytecode instruction, enough of it is implemented for it to show it in the bytecode (js -d).
2021-06-30 15:42:13 -03:00
ClassExpression const& m_class_expression;
LibJS: Use the IdentifierTable for NewFunction and NewClass lhs names This makes them trivially copyable, which is an assumption multiple optimizations use when rebuilding the instruction stream. This fixes most optimized crashes in the test262 suite.
2023-06-28 18:17:13 +02:00
Optional<IdentifierTableIndex> m_lhs_name;
LibJS: Use u32 instead of size_t for bytecode instruction array sizes Instructions that have an embedded tail array are not going to have more than 2^32 elements in any remotely sane situation.
2025-04-05 23:59:18 +02:00
u32 m_element_keys_count { 0 };
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
Optional<Operand> m_element_keys[];
LibJS: NewClass bytecode instruction This adds a the NewClass bytecode instruction, enough of it is implemented for it to show it in the bytecode (js -d).
2021-06-30 15:42:13 -03:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API NewFunction final : public Instruction {
LibJS: Add a new EnterScope bytecode instruction This is intended to perform the same duties as enter_scope() does in the AST tree-walk interpreter: - Hoisted function declaration processing - Hoisted variable declaration processing - ... maybe more This first cut only implements the function declaration processing.
2021-06-05 15:14:09 +02:00
public:
LibJS: Change NewFunction instruction to accept FunctionNode Preparation for upcoming changes where NewFunction will have to be used with FunctionDeclaration node.
2024-05-09 05:18:23 +00:00
explicit NewFunction(Operand dst, FunctionNode const& function_node, Optional<IdentifierTableIndex> lhs_name, Optional<Operand> home_object = {})
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::NewFunction)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Perform function instantiation in bytecode This replaces Bytecode::Op::EnterScope with a new NewFunction op that instantiates a ScriptFunction from a given FunctionNode (AST). This is then used to instantiate the local functions directly from bytecode when entering a ScopeNode. :^)
2021-06-10 00:49:23 +02:00
, m_function_node(function_node)
LibJS: Use the IdentifierTable for NewFunction and NewClass lhs names This makes them trivially copyable, which is an assumption multiple optimizations use when rebuilding the instruction stream. This fixes most optimized crashes in the test262 suite.
2023-06-28 18:17:13 +02:00
, m_lhs_name(lhs_name)
LibJS/Bytecode: Set "home object" of functions within object expression We manage this by having a stack of home objects in Generator, and then adding an optional home object parameter to the NewFunction instruction.
2023-06-16 10:25:05 +02:00
, m_home_object(move(home_object))
LibJS: Add a new EnterScope bytecode instruction This is intended to perform the same duties as enter_scope() does in the AST tree-walk interpreter: - Hoisted function declaration processing - Hoisted variable declaration processing - ... maybe more This first cut only implements the function declaration processing.
2021-06-05 15:14:09 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
if (m_home_object.has_value())
visitor(m_home_object.value());
}
LibJS: Add a new EnterScope bytecode instruction This is intended to perform the same duties as enter_scope() does in the AST tree-walk interpreter: - Hoisted function declaration processing - Hoisted variable declaration processing - ... maybe more This first cut only implements the function declaration processing.
2021-06-05 15:14:09 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS: Change NewFunction instruction to accept FunctionNode Preparation for upcoming changes where NewFunction will have to be used with FunctionDeclaration node.
2024-05-09 05:18:23 +00:00
FunctionNode const& function_node() const { return m_function_node; }
LibJS/JIT: Compile the NewFunction bytecode instruction
2023-10-21 15:49:04 +02:00
Optional<IdentifierTableIndex> const& lhs_name() const { return m_lhs_name; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Optional<Operand> const& home_object() const { return m_home_object; }
LibJS/JIT: Compile the NewFunction bytecode instruction
2023-10-21 15:49:04 +02:00
LibJS: Add a new EnterScope bytecode instruction This is intended to perform the same duties as enter_scope() does in the AST tree-walk interpreter: - Hoisted function declaration processing - Hoisted variable declaration processing - ... maybe more This first cut only implements the function declaration processing.
2021-06-05 15:14:09 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
LibJS: Change NewFunction instruction to accept FunctionNode Preparation for upcoming changes where NewFunction will have to be used with FunctionDeclaration node.
2024-05-09 05:18:23 +00:00
FunctionNode const& m_function_node;
LibJS: Use the IdentifierTable for NewFunction and NewClass lhs names This makes them trivially copyable, which is an assumption multiple optimizations use when rebuilding the instruction stream. This fixes most optimized crashes in the test262 suite.
2023-06-28 18:17:13 +02:00
Optional<IdentifierTableIndex> m_lhs_name;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Optional<Operand> m_home_object;
LibJS: Add a new EnterScope bytecode instruction This is intended to perform the same duties as enter_scope() does in the AST tree-walk interpreter: - Hoisted function declaration processing - Hoisted variable declaration processing - ... maybe more This first cut only implements the function declaration processing.
2021-06-05 15:14:09 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API BlockDeclarationInstantiation final : public Instruction {
LibJS/Bytecode: Leave BlockDeclarationInstantiation in C++ Instead of implementing this AO in bytecode, we now have an instruction for it that simply invokes the C++ implementation. This allows us to simplify Bytecode::Generator quite a bit by removing all the variable scope tracking.
2023-06-16 16:34:47 +02:00
public:
explicit BlockDeclarationInstantiation(ScopeNode const& scope_node)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::BlockDeclarationInstantiation)
LibJS/Bytecode: Leave BlockDeclarationInstantiation in C++ Instead of implementing this AO in bytecode, we now have an instruction for it that simply invokes the C++ implementation. This allows us to simplify Bytecode::Generator quite a bit by removing all the variable scope tracking.
2023-06-16 16:34:47 +02:00
, m_scope_node(scope_node)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Leave BlockDeclarationInstantiation in C++ Instead of implementing this AO in bytecode, we now have an instruction for it that simply invokes the C++ implementation. This allows us to simplify Bytecode::Generator quite a bit by removing all the variable scope tracking.
2023-06-16 16:34:47 +02:00
LibJS/JIT: Compile the BlockDeclarationInstantiation instruction
2023-10-29 15:13:28 +01:00
ScopeNode const& scope_node() const { return m_scope_node; }
LibJS/Bytecode: Leave BlockDeclarationInstantiation in C++ Instead of implementing this AO in bytecode, we now have an instruction for it that simply invokes the C++ implementation. This allows us to simplify Bytecode::Generator quite a bit by removing all the variable scope tracking.
2023-06-16 16:34:47 +02:00
private:
ScopeNode const& m_scope_node;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Return final : public Instruction {
LibJS: Compile ScriptFunctions into bytecode and run them that way :^) If there's a current Bytecode::Interpreter in action, ScriptFunction will now compile itself into bytecode and execute in that context. This patch also adds the Return bytecode instruction so that we can actually return values from called functions. :^) Return values are propagated from callee to caller via the caller's $0 register. Bytecode::Interpreter now keeps a stack of register "windows". These are not very efficient, but it should be pretty straightforward to convert them to e.g a sliding register window architecture later on. This is pretty dang cool! :^)
2021-06-05 15:53:36 +02:00
public:
LibJS: Generate bytecode in basic blocks instead of one big block This limits the size of each block (currently set to 1K), and gets us closer to a canonical, more easily analysable bytecode format. As a result of this, "Labels" are now simply entries to basic blocks. Since there is no more 'conditional' jump (as all jumps are always taken), JumpIf{True,False} are unified to JumpConditional, and JumpIfNullish is renamed to JumpNullish. Also fixes #7914 as a result of reimplementing the loop logic.
2021-06-09 06:49:58 +04:30
constexpr static bool IsTerminator = true;
LibJS: Make Op::Return value required It turns out we do not have any scenario where this is not provided.
2025-05-16 12:28:51 +12:00
explicit Return(Operand value)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Return)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_value(value)
LibJS: Compile ScriptFunctions into bytecode and run them that way :^) If there's a current Bytecode::Interpreter in action, ScriptFunction will now compile itself into bytecode and execute in that context. This patch also adds the Return bytecode instruction so that we can actually return values from called functions. :^) Return values are propagated from callee to caller via the caller's $0 register. Bytecode::Interpreter now keeps a stack of register "windows". These are not very efficient, but it should be pretty straightforward to convert them to e.g a sliding register window architecture later on. This is pretty dang cool! :^)
2021-06-05 15:53:36 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
LibJS: Make Op::Return value required It turns out we do not have any scenario where this is not provided.
2025-05-16 12:28:51 +12:00
visitor(m_value);
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
LibJS: Make Op::Return value required It turns out we do not have any scenario where this is not provided.
2025-05-16 12:28:51 +12:00
Operand const& value() const { return m_value; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
private:
LibJS: Make Op::Return value required It turns out we do not have any scenario where this is not provided.
2025-05-16 12:28:51 +12:00
Operand m_value;
LibJS: Compile ScriptFunctions into bytecode and run them that way :^) If there's a current Bytecode::Interpreter in action, ScriptFunction will now compile itself into bytecode and execute in that context. This patch also adds the Return bytecode instruction so that we can actually return values from called functions. :^) Return values are propagated from callee to caller via the caller's $0 register. Bytecode::Interpreter now keeps a stack of register "windows". These are not very efficient, but it should be pretty straightforward to convert them to e.g a sliding register window architecture later on. This is pretty dang cool! :^)
2021-06-05 15:53:36 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Increment final : public Instruction {
LibJS: Implement bytecode generation for UpdateExpression :^) Added Increment and Decrement bytecode ops to support this. Postfix updates use a temporary register to preserve the original value. Note that this patch only implements Identifier updates. Member expression updates are a TODO.
2021-06-09 11:40:38 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit Increment(Operand dst)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Increment)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Implement bytecode generation for UpdateExpression :^) Added Increment and Decrement bytecode ops to support this. Postfix updates use a temporary register to preserve the original value. Note that this patch only implements Identifier updates. Member expression updates are a TODO.
2021-06-09 11:40:38 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
private:
Operand m_dst;
LibJS: Implement bytecode generation for UpdateExpression :^) Added Increment and Decrement bytecode ops to support this. Postfix updates use a temporary register to preserve the original value. Note that this patch only implements Identifier updates. Member expression updates are a TODO.
2021-06-09 11:40:38 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PostfixIncrement final : public Instruction {
LibJS/Bytecode: Dedicated instructions for postfix increment/decrement Instead of splitting the postfix variants into ToNumeric + Inc/Dec, we now have dedicated PostfixIncrement and PostfixDecrement instructions that handle both outputs in one go.
2024-02-20 11:45:01 +01:00
public:
explicit PostfixIncrement(Operand dst, Operand src)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::PostfixIncrement)
LibJS/Bytecode: Dedicated instructions for postfix increment/decrement Instead of splitting the postfix variants into ToNumeric + Inc/Dec, we now have dedicated PostfixIncrement and PostfixDecrement instructions that handle both outputs in one go.
2024-02-20 11:45:01 +01:00
, m_dst(dst)
, m_src(src)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS/Bytecode: Dedicated instructions for postfix increment/decrement Instead of splitting the postfix variants into ToNumeric + Inc/Dec, we now have dedicated PostfixIncrement and PostfixDecrement instructions that handle both outputs in one go.
2024-02-20 11:45:01 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_src);
}
LibJS/Bytecode: Dedicated instructions for postfix increment/decrement Instead of splitting the postfix variants into ToNumeric + Inc/Dec, we now have dedicated PostfixIncrement and PostfixDecrement instructions that handle both outputs in one go.
2024-02-20 11:45:01 +01:00
Operand dst() const { return m_dst; }
Operand src() const { return m_src; }
private:
Operand m_dst;
Operand m_src;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Decrement final : public Instruction {
LibJS: Implement bytecode generation for UpdateExpression :^) Added Increment and Decrement bytecode ops to support this. Postfix updates use a temporary register to preserve the original value. Note that this patch only implements Identifier updates. Member expression updates are a TODO.
2021-06-09 11:40:38 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit Decrement(Operand dst)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Decrement)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Implement bytecode generation for UpdateExpression :^) Added Increment and Decrement bytecode ops to support this. Postfix updates use a temporary register to preserve the original value. Note that this patch only implements Identifier updates. Member expression updates are a TODO.
2021-06-09 11:40:38 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
private:
Operand m_dst;
LibJS: Implement bytecode generation for UpdateExpression :^) Added Increment and Decrement bytecode ops to support this. Postfix updates use a temporary register to preserve the original value. Note that this patch only implements Identifier updates. Member expression updates are a TODO.
2021-06-09 11:40:38 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PostfixDecrement final : public Instruction {
LibJS/Bytecode: Perform ToNumeric on accumulator before postfix inc/dec This ensures we get the expected behavior of code like: let a = [] let b = a++ (Where b should be 0, not [], because JavaScript.)
2023-06-16 10:51:40 +02:00
public:
LibJS/Bytecode: Dedicated instructions for postfix increment/decrement Instead of splitting the postfix variants into ToNumeric + Inc/Dec, we now have dedicated PostfixIncrement and PostfixDecrement instructions that handle both outputs in one go.
2024-02-20 11:45:01 +01:00
explicit PostfixDecrement(Operand dst, Operand src)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::PostfixDecrement)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_src(src)
LibJS/Bytecode: Perform ToNumeric on accumulator before postfix inc/dec This ensures we get the expected behavior of code like: let a = [] let b = a++ (Where b should be 0, not [], because JavaScript.)
2023-06-16 10:51:40 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_src);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand src() const { return m_src; }
private:
Operand m_dst;
Operand m_src;
LibJS/Bytecode: Perform ToNumeric on accumulator before postfix inc/dec This ensures we get the expected behavior of code like: let a = [] let b = a++ (Where b should be 0, not [], because JavaScript.)
2023-06-16 10:51:40 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Throw final : public Instruction {
LibJS: Generate bytecode for throw statements
2021-06-09 18:18:56 +02:00
public:
constexpr static bool IsTerminator = true;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit Throw(Operand src)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Throw)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS: Generate bytecode for throw statements
2021-06-09 18:18:56 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand src() const { return m_src; }
private:
Operand m_src;
LibJS: Generate bytecode for throw statements
2021-06-09 18:18:56 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ThrowIfNotObject final : public Instruction {
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
ThrowIfNotObject(Operand src)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ThrowIfNotObject)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand src() const { return m_src; }
private:
Operand m_src;
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ThrowIfNullish final : public Instruction {
LibJS/Bytecode: Throw on destructuring object assignment to nullish LHS 24 new passes on test262. :^)
2023-06-25 08:50:07 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit ThrowIfNullish(Operand src)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ThrowIfNullish)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
LibJS/Bytecode: Throw on destructuring object assignment to nullish LHS 24 new passes on test262. :^)
2023-06-25 08:50:07 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand src() const { return m_src; }
private:
Operand m_src;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ThrowIfTDZ final : public Instruction {
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
public:
explicit ThrowIfTDZ(Operand src)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ThrowIfTDZ)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_src(src)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_src);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand src() const { return m_src; }
private:
Operand m_src;
LibJS/Bytecode: Throw on destructuring object assignment to nullish LHS 24 new passes on test262. :^)
2023-06-25 08:50:07 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API EnterUnwindContext final : public Instruction {
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
public:
LibJS: Make EnterUnwindContext a terminator op Otherwise a basic block could have multiple outgoing edges without having much reason to do so.
2021-06-13 20:39:40 +04:30
constexpr static bool IsTerminator = true;
Revert "LibJS/Bytecode: Bring back the bytecode optimization pipeline" This reverts commit 5b29974bfa9b14224c5a1dc4feb41e22eda1e0f2.
2024-03-06 08:36:19 +01:00
EnterUnwindContext(Label entry_point)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::EnterUnwindContext)
LibJS: Make EnterUnwindContext a terminator op Otherwise a basic block could have multiple outgoing edges without having much reason to do so.
2021-06-13 20:39:40 +04:30
, m_entry_point(move(entry_point))
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
{
}
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_entry_point);
}
LibJS: Add a basic pass manager and add some basic passes This commit adds a bunch of passes, the most interesting of which is a pass that merges blocks together, and a pass that places blocks that flow into each other next to each other, and a very simply pass that removes duplicate basic blocks. Note that this does not remove the jump at the end of each block in that pass to avoid scope creep in the passes.
2021-06-13 20:40:20 +04:30
auto& entry_point() const { return m_entry_point; }
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
private:
LibJS: Make EnterUnwindContext a terminator op Otherwise a basic block could have multiple outgoing edges without having much reason to do so.
2021-06-13 20:39:40 +04:30
Label m_entry_point;
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ScheduleJump final : public Instruction {
LibJS: Generate unwind chains for break in Bytecode This uses a newly added instruction `ScheduleJump` This instruction tells the finally proceeding it, that instead of jumping to it's next block it should jump to the designated block.
2022-11-25 16:15:34 +01:00
public:
// Note: We use this instruction to tell the next `finally` block to
// continue execution with a specific break/continue target;
constexpr static bool IsTerminator = true;
ScheduleJump(Label target)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ScheduleJump)
LibJS: Generate unwind chains for break in Bytecode This uses a newly added instruction `ScheduleJump` This instruction tells the finally proceeding it, that instead of jumping to it's next block it should jump to the designated block.
2022-11-25 16:15:34 +01:00
, m_target(target)
{
}
Label target() const { return m_target; }
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_target);
}
LibJS: Generate unwind chains for break in Bytecode This uses a newly added instruction `ScheduleJump` This instruction tells the finally proceeding it, that instead of jumping to it's next block it should jump to the designated block.
2022-11-25 16:15:34 +01:00
private:
Label m_target;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API LeaveLexicalEnvironment final : public Instruction {
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
public:
LibJS/Bytecode: Simplify creating/leaving lexical environment Since we no longer need to create or leave var environments directly in bytecode, we can streamline the two instructions by making them always operate on the lexical environment.
2023-06-16 16:43:24 +02:00
LeaveLexicalEnvironment()
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::LeaveLexicalEnvironment)
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: More properly implement scoping rules in bytecode codegen Now we emit CreateVariable and SetVariable with the appropriate initialization/environment modes, much closer to the spec. This makes a whole lot of things like let/const variables, function and variable hoisting and some other things work :^)
2022-02-12 19:48:45 +03:30
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API LeavePrivateEnvironment final : public Instruction {
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
public:
LeavePrivateEnvironment()
: Instruction(Type::LeavePrivateEnvironment)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Stop using execute_ast_node() for class property evaluation Instead, generate bytecode to execute their AST nodes and save the resulting operands inside the NewClass instruction. Moving property expression evaluation to happen before NewClass execution also moves along creation of new private environment and its population with private members (private members should be visible during property evaluation). Before: - NewClass After: - CreatePrivateEnvironment - AddPrivateName - ... - AddPrivateName - NewClass - LeavePrivateEnvironment
2024-05-11 22:54:41 +00:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API LeaveUnwindContext final : public Instruction {
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
public:
LeaveUnwindContext()
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::LeaveUnwindContext)
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ContinuePendingUnwind final : public Instruction {
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
public:
constexpr static bool IsTerminator = true;
LibJS: Fix all clang-tidy warnings in Bytecode/Op.h - Add missing explicit to constructors - Use move() where appropriate
2021-06-12 11:22:46 +02:00
explicit ContinuePendingUnwind(Label resume_target)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ContinuePendingUnwind)
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
, m_resume_target(resume_target)
{
}
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_resume_target);
}
LibJS: Add a basic pass manager and add some basic passes This commit adds a bunch of passes, the most interesting of which is a pass that merges blocks together, and a pass that places blocks that flow into each other next to each other, and a very simply pass that removes duplicate basic blocks. Note that this does not remove the jump at the end of each block in that pass to avoid scope creep in the passes.
2021-06-13 20:40:20 +04:30
auto& resume_target() const { return m_resume_target; }
LibJS: Implement bytecode generation for try..catch..finally EnterUnwindContext pushes an unwind context (exception handler and/or finalizer) onto a stack. LeaveUnwindContext pops the unwind context from that stack. Upon return to the interpreter loop we check whether the VM has an exception pending. If no unwind context is available we return from the loop. If an exception handler is available we clear the VM's exception, put the exception value into the accumulator register, clear the unwind context's handler and jump to the handler. If no handler is available but a finalizer is available we save the exception value + metadata (for later use by ContinuePendingUnwind), clear the VM's exception, pop the unwind context and jump to the finalizer. ContinuePendingUnwind checks whether a saved exception is available. If no saved exception is available it jumps to the resume label. Otherwise it stores the exception into the VM. The Jump after LeaveUnwindContext could be integrated into the LeaveUnwindContext instruction. I've kept them separate for now to make the bytecode more readable. > try { 1; throw "x" } catch (e) { 2 } finally { 3 }; 4 1: [ 0] EnterScope [ 10] EnterUnwindContext handler:@4 finalizer:@3 [ 38] EnterScope [ 48] LoadImmediate 1 [ 60] NewString 1 ("x") [ 70] Throw <for non-terminated blocks: insert LeaveUnwindContext + Jump @3 here> 2: [ 0] LoadImmediate 4 3: [ 0] EnterScope [ 10] LoadImmediate 3 [ 28] ContinuePendingUnwind resume:@2 4: [ 0] SetVariable 0 (e) [ 10] EnterScope [ 20] LoadImmediate 2 [ 38] LeaveUnwindContext [ 3c] Jump @3 String Table: 0: e 1: x
2021-06-10 15:04:38 +02:00
private:
Label m_resume_target;
};
LibJS: Generate bytecode for entering nested lexical environments This adds a new PushLexicalEnvironment instruction that creates a new LexicalEnvironment and pushes it on the VM's scope stack. There is no corresponding PopLexicalEnvironment instruction yet, so this will behave incorrectly with let/const scopes for example.
2021-06-10 22:12:21 +02:00
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Yield final : public Instruction {
LibJS: Implement generator functions (only in bytecode mode)
2021-06-11 01:38:30 +04:30
public:
constexpr static bool IsTerminator = true;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit Yield(Label continuation_label, Operand value)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Yield)
LibJS: Implement generator functions (only in bytecode mode)
2021-06-11 01:38:30 +04:30
, m_continuation_label(continuation_label)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_value(value)
LibJS: Implement generator functions (only in bytecode mode)
2021-06-11 01:38:30 +04:30
{
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit Yield(nullptr_t, Operand value)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Yield)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_value(value)
LibJS: Implement generator functions (only in bytecode mode)
2021-06-11 01:38:30 +04:30
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
if (m_continuation_label.has_value())
visitor(m_continuation_label.value());
}
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_value);
}
LibJS: Add a basic pass manager and add some basic passes This commit adds a bunch of passes, the most interesting of which is a pass that merges blocks together, and a pass that places blocks that flow into each other next to each other, and a very simply pass that removes duplicate basic blocks. Note that this does not remove the jump at the end of each block in that pass to avoid scope creep in the passes.
2021-06-13 20:40:20 +04:30
auto& continuation() const { return m_continuation_label; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand value() const { return m_value; }
LibJS: Implement generator functions (only in bytecode mode)
2021-06-11 01:38:30 +04:30
private:
Optional<Label> m_continuation_label;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_value;
LibJS: Implement generator functions (only in bytecode mode)
2021-06-11 01:38:30 +04:30
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API PrepareYield final : public Instruction {
LibJS: Prepare yield object before re-routing it through finally
2024-05-18 17:25:43 +02:00
public:
explicit PrepareYield(Operand dest, Operand value)
: Instruction(Type::PrepareYield)
, m_dest(dest)
, m_value(value)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS: Prepare yield object before re-routing it through finally
2024-05-18 17:25:43 +02:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dest);
visitor(m_value);
}
Operand destination() const { return m_dest; }
Operand value() const { return m_value; }
private:
Operand m_dest;
Operand m_value;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Await final : public Instruction {
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
public:
constexpr static bool IsTerminator = true;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit Await(Label continuation_label, Operand argument)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Await)
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
, m_continuation_label(continuation_label)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_argument(argument)
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Flatten bytecode to a contiguous representation Instead of keeping bytecode as a set of disjoint basic blocks on the malloc heap, bytecode is now a contiguous sequence of bytes(!) The transformation happens at the end of Bytecode::Generator::generate() and the only really hairy part is rerouting jump labels. This required solving a few problems: - The interpreter execution loop had to change quite a bit, since we were storing BasicBlock pointers all over the place, and control transfer was done by redirecting the interpreter's current block. - Exception handlers & finalizers are now stored per-bytecode-range in a side table in Executable. - The interpreter now has a plain program counter instead of a stream iterator. This actually makes error stack generation a bit nicer since we just have to deal with a number instead of reaching into the iterator. This yields a 25% performance improvement on this microbenchmark: for (let i = 0; i < 1_000_000; ++i) { } But basically everything gets faster. :^)
2024-05-06 06:44:08 +02:00
void visit_labels_impl(Function<void(Label&)> visitor)
{
visitor(m_continuation_label);
}
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_argument);
}
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
auto& continuation() const { return m_continuation_label; }
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand argument() const { return m_argument; }
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
private:
Label m_continuation_label;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_argument;
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetIterator final : public Instruction {
LibJS: Implement array destructuring for the bytecode interpreter
2021-06-13 13:40:48 -07:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
GetIterator(Operand dst, Operand iterable, IteratorHint hint = IteratorHint::Sync)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetIterator)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_iterable(iterable)
LibJS/Bytecode: Implement for await of Summary: Diff Tests: +391 ✅ +15 ❌ +2 💥️ -408 📝
2023-06-27 19:24:34 +01:00
, m_hint(hint)
LibJS: Implement array destructuring for the bytecode interpreter
2021-06-13 13:40:48 -07:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_iterable);
}
LibJS/Bytecode: Implement for await of Summary: Diff Tests: +391 ✅ +15 ❌ +2 💥️ -408 📝
2023-06-27 19:24:34 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand iterable() const { return m_iterable; }
LibJS/JIT: Compile the GetIterator instruction
2023-10-29 15:53:36 +01:00
IteratorHint hint() const { return m_hint; }
LibJS/Bytecode: Implement for await of Summary: Diff Tests: +391 ✅ +15 ❌ +2 💥️ -408 📝
2023-06-27 19:24:34 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_iterable;
LibJS/Bytecode: Implement for await of Summary: Diff Tests: +391 ✅ +15 ❌ +2 💥️ -408 📝
2023-06-27 19:24:34 +01:00
IteratorHint m_hint { IteratorHint::Sync };
LibJS: Implement array destructuring for the bytecode interpreter
2021-06-13 13:40:48 -07:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetObjectFromIteratorRecord final : public Instruction {
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
GetObjectFromIteratorRecord(Operand object, Operand iterator_record)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetObjectFromIteratorRecord)
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
, m_object(object)
, m_iterator_record(iterator_record)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_object);
visitor(m_iterator_record);
}
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand object() const { return m_object; }
Operand iterator_record() const { return m_iterator_record; }
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_object;
Operand m_iterator_record;
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetNextMethodFromIteratorRecord final : public Instruction {
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
GetNextMethodFromIteratorRecord(Operand next_method, Operand iterator_record)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetNextMethodFromIteratorRecord)
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
, m_next_method(next_method)
, m_iterator_record(iterator_record)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_next_method);
visitor(m_iterator_record);
}
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand next_method() const { return m_next_method; }
Operand iterator_record() const { return m_iterator_record; }
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_next_method;
Operand m_iterator_record;
LibJS: Stop converting between Object <-> IteratorRecord all the time This patch makes IteratorRecord an Object. Although it's not exposed to author code, this does allow us to store it in a VM register. Now that we can store it in a VM register, we don't need to convert it back and forth between IteratorRecord and Object when accessing it from bytecode. The big win here is avoiding 3 [[Get]] accesses on every iteration step of for..of loops. There are also a bunch of smaller efficiencies gained. 20% speed-up on this microbenchmark: function go(a) { for (const p of a) { } } const a = []; a.length = 1_000_000; go(a);
2023-12-07 10:44:41 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetMethod final : public Instruction {
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
GetMethod(Operand dst, Operand object, IdentifierTableIndex property)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetMethod)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_object(object)
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
, m_property(property)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_object);
}
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand object() const { return m_object; }
LibJS/JIT: Compile the GetMethod instruction
2023-10-30 00:04:52 +01:00
IdentifierTableIndex property() const { return m_property; }
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
Operand m_object;
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
IdentifierTableIndex m_property;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetObjectPropertyIterator final : public Instruction {
LibJS: Implement bytecode generation for For-In/Of statements This also implements the rather interesting behaviour that #12772 relies on, so this fixes that bug in BC mode (the AST interp remains affected).
2022-03-18 20:18:19 +03:30
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
GetObjectPropertyIterator(Operand dst, Operand object)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetObjectPropertyIterator)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_object(object)
LibJS: Implement bytecode generation for For-In/Of statements This also implements the rather interesting behaviour that #12772 relies on, so this fixes that bug in BC mode (the AST interp remains affected).
2022-03-18 20:18:19 +03:30
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_object);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand object() const { return m_object; }
private:
Operand m_dst;
Operand m_object;
LibJS: Implement bytecode generation for For-In/Of statements This also implements the rather interesting behaviour that #12772 relies on, so this fixes that bug in BC mode (the AST interp remains affected).
2022-03-18 20:18:19 +03:30
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API IteratorClose final : public Instruction {
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
IteratorClose(Operand iterator_record, Completion::Type completion_type, Optional<Value> completion_value)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::IteratorClose)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_iterator_record(iterator_record)
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
, m_completion_type(completion_type)
, m_completion_value(completion_value)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_iterator_record);
}
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand iterator_record() const { return m_iterator_record; }
LibJS/JIT: Compile the IteratorClose instruction
2023-10-29 16:59:53 +01:00
Completion::Type completion_type() const { return m_completion_type; }
Optional<Value> const& completion_value() const { return m_completion_value; }
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_iterator_record;
LibJS/Bytecode: Implement yield*
2022-12-09 18:48:57 +00:00
Completion::Type m_completion_type { Completion::Type::Normal };
Optional<Value> m_completion_value;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API AsyncIteratorClose final : public Instruction {
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
AsyncIteratorClose(Operand iterator_record, Completion::Type completion_type, Optional<Value> completion_value)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::AsyncIteratorClose)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_iterator_record(iterator_record)
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
, m_completion_type(completion_type)
, m_completion_value(completion_value)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_iterator_record);
}
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand iterator_record() const { return m_iterator_record; }
LibJS/JIT: Compile the AsyncIteratorClose instruction
2023-10-30 02:17:24 +01:00
Completion::Type completion_type() const { return m_completion_type; }
Optional<Value> const& completion_value() const { return m_completion_value; }
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_iterator_record;
LibJS/Bytecode: Add Await and AsyncIteratorClose instructions
2023-07-14 21:42:43 +01:00
Completion::Type m_completion_type { Completion::Type::Normal };
Optional<Value> m_completion_value;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API IteratorNext final : public Instruction {
LibJS: Implement array destructuring for the bytecode interpreter
2021-06-13 13:40:48 -07:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
IteratorNext(Operand dst, Operand iterator_record)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::IteratorNext)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
, m_iterator_record(iterator_record)
LibJS: Implement array destructuring for the bytecode interpreter
2021-06-13 13:40:48 -07:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
visitor(m_iterator_record);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
Operand iterator_record() const { return m_iterator_record; }
private:
Operand m_dst;
Operand m_iterator_record;
LibJS: Implement array destructuring for the bytecode interpreter
2021-06-13 13:40:48 -07:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API IteratorNextUnpack final : public Instruction {
LibJS: Skip iterator result object allocation in for..of and for..in Introduce special instruction for `for..of` and `for..in` loop that skips `{ value, done }` result object allocation if iterator is builtin (array, map, set, string). This reduces GC pressure significantly and avoids extracting the `value` and `done` properties. This change makes this micro benchmark 48% faster on my computer: ```js const arr = new Array(10_000_000); let counter = 0; for (let _ of arr) { counter++; } ```
2025-04-30 16:31:26 +03:00
public:
LibJS: Optimize array destructuring assignment for builtin iterators ...by avoiding `{ value, done }` iterator result value allocation. This change applies the same otimization 81b6a11 added for `for..in` and `for..of`. Makes following micro benchmark go 22% faster on my computer: ```js function f() { const arr = []; for (let i = 0; i < 10_000_000; i++) { arr.push([i]); } let sum = 0; for (let [i] of arr) { sum += i; } } f(); ```
2025-05-01 16:05:24 +03:00
IteratorNextUnpack(Operand dst_value, Operand dst_done, Operand iterator_record)
: Instruction(Type::IteratorNextUnpack)
LibJS: Skip iterator result object allocation in for..of and for..in Introduce special instruction for `for..of` and `for..in` loop that skips `{ value, done }` result object allocation if iterator is builtin (array, map, set, string). This reduces GC pressure significantly and avoids extracting the `value` and `done` properties. This change makes this micro benchmark 48% faster on my computer: ```js const arr = new Array(10_000_000); let counter = 0; for (let _ of arr) { counter++; } ```
2025-04-30 16:31:26 +03:00
, m_dst_value(dst_value)
, m_dst_done(dst_done)
, m_iterator_record(iterator_record)
{
}
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst_value);
visitor(m_dst_done);
visitor(m_iterator_record);
}
Operand dst_value() const { return m_dst_value; }
Operand dst_done() const { return m_dst_done; }
Operand iterator_record() const { return m_iterator_record; }
private:
Operand m_dst_value;
Operand m_dst_done;
Operand m_iterator_record;
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ResolveThisBinding final : public Instruction {
LibJS: Implement 'this' in the bytecode VM ThisExpression now emits a "ResolveThisBinding" bytecode op, which simply loads the VM's current 'this' binding into the accumulator.
2021-10-24 14:43:00 +02:00
public:
LibJS/Bytecode: Always resolve `this` binding into dedicated register We already have a dedicated register slot for `this`, so instead of having ResolveThisBinding take a `dst` operand, just write the value directly into the `this` register every time.
2024-05-31 20:41:29 +02:00
ResolveThisBinding()
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ResolveThisBinding)
LibJS: Implement 'this' in the bytecode VM ThisExpression now emits a "ResolveThisBinding" bytecode op, which simply loads the VM's current 'this' binding into the accumulator.
2021-10-24 14:43:00 +02:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS/Bytecode: Always resolve `this` binding into dedicated register We already have a dedicated register slot for `this`, so instead of having ResolveThisBinding take a `dst` operand, just write the value directly into the `this` register every time.
2024-05-31 20:41:29 +02:00
void visit_operands_impl(Function<void(Operand&)>) { }
LibJS: Implement 'this' in the bytecode VM ThisExpression now emits a "ResolveThisBinding" bytecode op, which simply loads the VM's current 'this' binding into the accumulator.
2021-10-24 14:43:00 +02:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API ResolveSuperBase final : public Instruction {
LibJS/Bytecode: Implement initial support for super member expressions
2023-05-15 18:45:16 +01:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit ResolveSuperBase(Operand dst)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::ResolveSuperBase)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Implement initial support for super member expressions
2023-05-15 18:45:16 +01:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
private:
Operand m_dst;
LibJS/Bytecode: Implement initial support for super member expressions
2023-05-15 18:45:16 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetNewTarget final : public Instruction {
LibJS/Bytecode: Add support for new.target
2022-03-19 19:40:21 +00:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit GetNewTarget(Operand dst)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetNewTarget)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Add support for new.target
2022-03-19 19:40:21 +00:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
private:
Operand m_dst;
LibJS/Bytecode: Add support for new.target
2022-03-19 19:40:21 +00:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API GetImportMeta final : public Instruction {
LibJS: Implement import.meta for bytecode
2023-07-11 23:07:12 -04:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
explicit GetImportMeta(Operand dst)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::GetImportMeta)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS: Implement import.meta for bytecode
2023-07-11 23:07:12 -04:00
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
private:
Operand m_dst;
LibJS: Implement import.meta for bytecode
2023-07-11 23:07:12 -04:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API TypeofBinding final : public Instruction {
LibJS/Bytecode: Make typeof return "undefined" on unresolvable IDs Previously it would throw instead of returning "undefined" for `typeof Identifier` if Identifier does not exist.
2022-06-11 23:12:22 +01:00
public:
LibJS/Bytecode: Rename TypeofVariable => TypeofBinding
2024-06-14 09:37:26 +02:00
TypeofBinding(Operand dst, IdentifierTableIndex identifier)
: Instruction(Type::TypeofBinding)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_dst(dst)
LibJS/Bytecode: Make typeof return "undefined" on unresolvable IDs Previously it would throw instead of returning "undefined" for `typeof Identifier` if Identifier does not exist.
2022-06-11 23:12:22 +01:00
, m_identifier(identifier)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
ThrowCompletionOr<void> execute_impl(Bytecode::Interpreter&) const;
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_dst);
}
LibJS/Bytecode: Make typeof return "undefined" on unresolvable IDs Previously it would throw instead of returning "undefined" for `typeof Identifier` if Identifier does not exist.
2022-06-11 23:12:22 +01:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand dst() const { return m_dst; }
LibJS/JIT: Compile the TypeofVariable bytecode instruction
2023-10-21 15:26:03 +02:00
IdentifierTableIndex identifier() const { return m_identifier; }
LibJS/Bytecode: Make typeof return "undefined" on unresolvable IDs Previously it would throw instead of returning "undefined" for `typeof Identifier` if Identifier does not exist.
2022-06-11 23:12:22 +01:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_dst;
LibJS/Bytecode: Make typeof return "undefined" on unresolvable IDs Previously it would throw instead of returning "undefined" for `typeof Identifier` if Identifier does not exist.
2022-06-11 23:12:22 +01:00
IdentifierTableIndex m_identifier;
LibJS/Bytecode: Give TypeofBinding instructions a lookup cache These can use an EnvironmentCoordinate for caching, just like normal binding lookups. Saves a bunch of time for repeated typeof checks.
2024-06-14 13:57:51 +02:00
mutable EnvironmentCoordinate m_cache;
LibJS/Bytecode: Make typeof return "undefined" on unresolvable IDs Previously it would throw instead of returning "undefined" for `typeof Identifier` if Identifier does not exist.
2022-06-11 23:12:22 +01:00
};
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API End final : public Instruction {
LibJS: Update bytecode generator to use local variables - Update ECMAScriptFunctionObject::function_declaration_instantiation to initialize local variables - Introduce GetLocal, SetLocal, TypeofLocal that will be used to operate on local variables. - Update bytecode generator to emit instructions for local variables
2023-07-05 02:17:10 +02:00
public:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
constexpr static bool IsTerminator = true;
explicit End(Operand value)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::End)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_value(value)
LibJS: Update bytecode generator to use local variables - Update ECMAScriptFunctionObject::function_declaration_instantiation to initialize local variables - Introduce GetLocal, SetLocal, TypeofLocal that will be used to operate on local variables. - Update bytecode generator to emit instructions for local variables
2023-07-05 02:17:10 +02:00
{
}
Everywhere: Rename {Deprecated => Byte}String This commit un-deprecates DeprecatedString, and repurposes it as a byte string. As the null state has already been removed, there are no other particularly hairy blockers in repurposing this type as a byte string (what it _really_ is). This commit is auto-generated: $ xs=$(ack -l \bDeprecatedString\b\|deprecated_string AK Userland \ Meta Ports Ladybird Tests Kernel) $ perl -pie 's/\bDeprecatedString\b/ByteString/g; s/deprecated_string/byte_string/g' $xs $ clang-format --style=file -i \ $(git diff --name-only | grep \.cpp\|\.h) $ gn format $(git ls-files '*.gn' '*.gni')
2023-12-16 17:49:34 +03:30
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_value);
}
LibJS: Update bytecode generator to use local variables - Update ECMAScriptFunctionObject::function_declaration_instantiation to initialize local variables - Introduce GetLocal, SetLocal, TypeofLocal that will be used to operate on local variables. - Update bytecode generator to emit instructions for local variables
2023-07-05 02:17:10 +02:00
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand value() const { return m_value; }
LibJS/JIT: Compile the TypeofLocal bytecode instruction
2023-10-27 16:22:14 +02:00
LibJS: Update bytecode generator to use local variables - Update ECMAScriptFunctionObject::function_declaration_instantiation to initialize local variables - Introduce GetLocal, SetLocal, TypeofLocal that will be used to operate on local variables. - Update bytecode generator to emit instructions for local variables
2023-07-05 02:17:10 +02:00
private:
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
Operand m_value;
LibJS: Update bytecode generator to use local variables - Update ECMAScriptFunctionObject::function_declaration_instantiation to initialize local variables - Introduce GetLocal, SetLocal, TypeofLocal that will be used to operate on local variables. - Update bytecode generator to emit instructions for local variables
2023-07-05 02:17:10 +02:00
};
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
LibJS: Enable EXPLICIT_SYMBOL_EXPORT
2025-06-28 21:39:13 -07:00
class JS_API Dump final : public Instruction {
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
public:
explicit Dump(StringView text, Operand value)
LibJS/Bytecode: Remove Instruction::m_length field Now that the interpreter is unrolled, we can advance the program counter manually based on the current instruction type. This makes most instructions a bit smaller. :^)
2024-05-06 11:09:09 +02:00
: Instruction(Type::Dump)
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
, m_text(text)
, m_value(value)
{
}
Revert "LibJS: Add StackFrame to avoid indirection in VM register access" This reverts commit 36bb2824a67c21a481955a2699b47862976fe692. Although this was faster on my M3 MacBook Pro, other Apple machines disagree, including our benchmark runner. So let's revert it.
2025-04-29 16:08:42 +02:00
void execute_impl(Bytecode::Interpreter&) const;
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
ByteString to_byte_string_impl(Bytecode::Executable const&) const;
LibJS: Join locals, constants and registers into single vector Merging registers, constants and locals into single vector means: - Better data locality - No need to check type in Interpreter::get() and Interpreter::set() which are very hot functions Performance improvement is visible in almost all Octane and Kraken tests.
2024-05-12 18:49:03 +02:00
void visit_operands_impl(Function<void(Operand&)> visitor)
{
visitor(m_value);
}
LibJS/Bytecode: Move to a new bytecode format This patch moves us away from the accumulator-based bytecode format to one with explicit source and destination registers. The new format has multiple benefits: - ~25% faster on the Kraken and Octane benchmarks :^) - Fewer instructions to accomplish the same thing - Much easier for humans to read(!) Because this change requires a fundamental shift in how bytecode is generated, it is quite comprehensive. Main implementation mechanism: generate_bytecode() virtual function now takes an optional "preferred dst" operand, which allows callers to communicate when they have an operand that would be optimal for the result to go into. It also returns an optional "actual dst" operand, which is where the completion value (if any) of the AST node is stored after the node has "executed". One thing of note that's new: because instructions can now take locals as operands, this means we got rid of the GetLocal instruction. A side-effect of that is we have to think about the temporal deadzone (TDZ) a bit differently for locals (GetLocal would previously check for empty values and interpret that as a TDZ access and throw). We now insert special ThrowIfTDZ instructions in places where a local access may be in the TDZ, to maintain the correct behavior. There are a number of progressions and regressions from this test: A number of async generator tests have been accidentally fixed while converting the implementation to the new bytecode format. It didn't seem useful to preserve bugs in the original code when converting it. Some "does eval() return the correct completion value" tests have regressed, in particular ones related to propagating the appropriate completion after control flow statements like continue and break. These are all fairly obscure issues, and I believe we can continue working on them separately. The net test262 result is a progression though. :^)
2024-02-04 08:00:54 +01:00
private:
StringView m_text;
Operand m_value;
};
LibJS: Start fleshing out a bytecode for the JavaScript engine :^) This patch begins the work of implementing JavaScript execution in a bytecode VM instead of an AST tree-walk interpreter. It's probably quite naive, but we have to start somewhere. The basic idea is that you call Bytecode::Generator::generate() on an AST node and it hands you back a Bytecode::Block filled with instructions that can then be interpreted by a Bytecode::Interpreter. This first version only implements two instructions: Load and Add. :^) Each bytecode block has infinity registers, and the interpreter resizes its register file to fit the block being executed. Two new `js` options are added in this patch as well: `-d` will dump the generated bytecode `-b` will execute the generated bytecode Note that unless `-d` and/or `-b` are specified, none of the bytecode related stuff in LibJS runs at all. This is implemented in parallel with the existing AST interpreter. :^)
2021-06-03 10:46:30 +02:00
}
Reference in a new issue Copy permalink