JavaScriptCore:
2009-07-07 Gavin Barraclough <[email protected]>
Reviewed by Sam Weinig.
Stop loading constants into the register file.
Instead, use high register values (highest bit bar the sign bit set) to indicate
constants in the instruction stream, and when we encounter such a value load it
directly from the CodeBlock.
Since constants are no longer copied into the register file, this patch renders
the 'unexpected constant' mechanism redundant, and removes it.
2% improvement, thanks to Sam Weinig.
- bytecode/CodeBlock.cpp:
(JSC::CodeBlock::dump):
(JSC::CodeBlock::CodeBlock):
(JSC::CodeBlock::mark):
(JSC::CodeBlock::shrinkToFit):
- bytecode/CodeBlock.h:
(JSC::CodeBlock::isTemporaryRegisterIndex):
(JSC::CodeBlock::constantRegister):
(JSC::CodeBlock::isConstantRegisterIndex):
(JSC::CodeBlock::getConstant):
(JSC::ExecState::r):
- bytecode/Opcode.h:
- bytecompiler/BytecodeGenerator.cpp:
(JSC::BytecodeGenerator::preserveLastVar):
(JSC::BytecodeGenerator::BytecodeGenerator):
(JSC::BytecodeGenerator::addConstantValue):
(JSC::BytecodeGenerator::emitEqualityOp):
(JSC::BytecodeGenerator::emitLoad):
(JSC::BytecodeGenerator::emitResolveBase):
(JSC::BytecodeGenerator::emitResolveWithBase):
(JSC::BytecodeGenerator::emitNewError):
- bytecompiler/BytecodeGenerator.h:
(JSC::BytecodeGenerator::emitNode):
- interpreter/CallFrame.h:
(JSC::ExecState::noCaller):
(JSC::ExecState::hasHostCallFrameFlag):
(JSC::ExecState::addHostCallFrameFlag):
(JSC::ExecState::removeHostCallFrameFlag):
- interpreter/Interpreter.cpp:
(JSC::Interpreter::resolve):
(JSC::Interpreter::resolveSkip):
(JSC::Interpreter::resolveGlobal):
(JSC::Interpreter::resolveBase):
(JSC::Interpreter::resolveBaseAndProperty):
(JSC::Interpreter::resolveBaseAndFunc):
(JSC::Interpreter::dumpRegisters):
(JSC::Interpreter::throwException):
(JSC::Interpreter::createExceptionScope):
(JSC::Interpreter::privateExecute):
(JSC::Interpreter::retrieveArguments):
- jit/JIT.cpp:
(JSC::JIT::privateCompileMainPass):
- jit/JITInlineMethods.h:
(JSC::JIT::emitLoadDouble):
(JSC::JIT::emitLoadInt32ToDouble):
- jit/JITOpcodes.cpp:
(JSC::JIT::emit_op_new_error):
(JSC::JIT::emit_op_enter):
(JSC::JIT::emit_op_enter_with_activation):
- parser/Nodes.cpp:
(JSC::DeleteResolveNode::emitBytecode):
(JSC::DeleteValueNode::emitBytecode):
(JSC::PrefixResolveNode::emitBytecode):
- runtime/JSActivation.cpp:
(JSC::JSActivation::JSActivation):
- wtf/Platform.h:
LayoutTests:
2009-07-07 Gavin Barraclough <[email protected]>
Reviewed by Geoff Garen.
fast/js/global-recursion-on-full-stack.html is a little finicky.
The test recurses down the JS stack to find the bottom (catching this with an exception),
then tries to call a host function (document.write), switch writes new '<script>' code,
and expects this code to be run, then expects this code to try to call 'f();' again,
which it expects to fail, and it expects to catch that exception. However it is possible
that one of the earlier stages (the call to document.write, entering the interpreter to
run the new global code) will fail, depending on exactly how much stack space was free at
the point the last call to f() failed.
Tweak the test to make it work.
- fast/js/global-recursion-on-full-stack.html:
