Changeset 37086 in webkit for trunk/JavaScriptCore/kjs/JSGlobalObject.cpp

Timestamp:

Sep 29, 2008, 5:46:25 PM (17 years ago)

Author:

[email protected]

Message:

JavaScriptCore:

2008-09-29 Geoffrey Garen <​[email protected]>

Reviewed by Cameron Zwarich.

Store the callee ScopeChain, not the caller ScopeChain, in the call frame
header. Nix the "scopeChain" local variable and ExecState::m_scopeChain, and
access the callee ScopeChain through the call frame header instead.

Profit: call + return are simpler, because they don't have to update the
"scopeChain" local variable, or ExecState::m_scopeChain.

Because CTI keeps "r" in a register, reading the callee ScopeChain relative
to "r" can be very fast, in any cases we care to optimize.

0% speedup on empty function call benchmark. (5.5% speedup in bytecode.)
0% speedup on SunSpider. (7.5% speedup on controlflow-recursive.)
2% speedup on SunSpider --v8.
2% speedup on v8 benchmark.

• VM/CTI.cpp: Changed scope chain access to read the scope chain from the call frame header. Sped up op_ret by changing it not to fuss with the "scopeChain" local variable or ExecState::m_scopeChain.

• VM/CTI.h: Updated CTI trampolines not to take a ScopeChainNode* argument, since that's stored in the call frame header now.

• VM/Machine.cpp: Access "scopeChain" and "codeBlock" through new helper functions that read from the call frame header. Updated functions operating on ExecState::m_callFrame to account for / take advantage of the fact that Exec:m_callFrame is now never NULL.

Fixed a bug in op_construct, where it would use the caller's default
object prototype, rather than the callee's, when constructing a new object.

• VM/Machine.h: Made some helper functions available. Removed ScopeChainNode* arguments to a lot of functions, since the ScopeChainNode* is now stored in the call frame header.

• VM/RegisterFile.h: Renamed "CallerScopeChain" to "ScopeChain", since that's what it is now.

• kjs/DebuggerCallFrame.cpp: Updated for change to ExecState signature.

• kjs/ExecState.cpp:
• kjs/ExecState.h: Nixed ExecState::m_callFrame, along with the unused isGlobalObject function.

• kjs/JSGlobalObject.cpp:
• kjs/JSGlobalObject.h: Gave the global object a fake call frame in which to store the global scope chain, since our code now assumes that it can always read the scope chain out of the ExecState's call frame.

JavaScriptGlue:

2008-09-29 Geoffrey Garen <​[email protected]>

Not reviewed.

Forwarding headers to fix the build.

• ForwardingHeaders/kjs/CTI.h: Copied from ForwardingHeaders/kjs/ExecState.h.
• ForwardingHeaders/kjs/ustring.h: Copied from ForwardingHeaders/kjs/ExecState.h.
• ForwardingHeaders/masm: Added.
• ForwardingHeaders/masm/X86Assembler.h: Added.
• ForwardingHeaders/profiler: Added.
• ForwardingHeaders/profiler/Profiler.h: Added.

LayoutTests:

2008-09-29 Geoffrey Garen <​[email protected]>

Reviewed by Cameron Zwarich.

Test case for which prototype is used when calling "new" across windows.

• fast/js/construct-global-object-expected.txt: Added.
• fast/js/construct-global-object.html: Added.

File:

• trunk/JavaScriptCore/kjs/JSGlobalObject.cpp (modified) (1 diff)

trunk/JavaScriptCore/kjs/JSGlobalObject.cpp

@@ -140 +140 @@
     d()->debugger = 0;
 
-    d()->globalExec.set(new ExecState(this, thisValue, d()->globalScopeChain.node()));
+    d()->globalExec.set(new ExecState(this, thisValue, d()->globalCallFrame + RegisterFile::CallFrameHeaderSize));
 
     d()->profileGroup = 0;