JSStackFrame Evisceration: Difference between revisions

← Older edit

JSStackFrame Evisceration (view source)

Revision as of 17:47, 30 September 2010

812 bytes added , 30 September 2010

→‎Members to remove

Lw

Confirmed users

367

edits

@@ Line 1: / Line 1: @@
-== Removing JSStackFrame members / things on the call path<br> ==
+This leaves sizeof(JSStackFrame) == 10 words (11 on 32-bit), but we only need to initialize 4 of those words on the common call path; the rest can be lazily (or never) initialized.
-This would leave sizeof(JSStackFrame) == 6 words.<br>
+=== Members to remove ===
-=== Directly remove<br> ===
+Sorted in estimated order of benefit / difficulty:
-*[https://bugzilla.mozilla.org/show_bug.cgi?id=535912 blockChain]<br>
+{| width="95%" cellspacing="4" cellpadding="1" border="0"
-*[https://bugzilla.mozilla.org/show_bug.cgi?id=539144 argc/argv] (lw)<br>
+|-
-*[https://bugzilla.mozilla.org/show_bug.cgi?id=540675 callerVersion] (not just a predictable branch)<br>
+| '''Task'''
-*[https://bugzilla.mozilla.org/show_bug.cgi?id=550639 rval] (harder than originally expected, but its a uint64 on x86, so 2x points)<br>
+| '''Size (wks)'''
-*[https://bugzilla.mozilla.org/show_bug.cgi?id=557375 thisv] (uint64 on x86, 2x points)<br>
+| '''Assignee'''
-*annotation - remove UniversalXPConnect / ultimately, the calls to JS_{Get,Set}FrameAnnotation in nsScriptSecurityManager.cpp must go<br>
+| '''Note'''
-*displaySave - remove display optimization (cdleary)<br>
+|-
+| X[https://bugzilla.mozilla.org/show_bug.cgi?id=539144 argc/argv/fun/script/thisv]
-=== Remove from method-jit frame<br> ===
+| 3
+| lw
-*ncode (merge with savedPC)<br>
+| Requires StackSegment [https://bugzilla.mozilla.org/show_bug.cgi?id=579183 changes].
+|-
-=== Remove in non-interpreter stack frame<br> ===
+| X[https://bugzilla.mozilla.org/show_bug.cgi?id=595073 ncode]
+| 3
-*fun<br>
+|
-*script<br>
+| Merge it with savedPC in method-jit.  Requires building map HW PC --> bytecode (which we sortof already heave
-*hookData (?)<br>
+|-
-*imacpc (?)
+| [https://bugzilla.mozilla.org/show_bug.cgi?id=593882 scopeChain]
+| .5
-== Remove from call path ==
+|
+|-
-*PIC for fast natives calls
+| X[https://bugzilla.mozilla.org/show_bug.cgi?id=577708 displaySave]
-*PIC/fast path for interpreted inline call
+| 1
-*Remove empty script check (not just a predictable branch)
+| cdleary
+| Remove display optimization for great justice!
+|-
+| XhookData
+| .1
+| lw
+| Lazily initialize using JSStackFrame::flags.  Included in bug 539144.
+|-
+| X[https://bugzilla.mozilla.org/show_bug.cgi?id=546848 annotation]
+| 3
+| sayrer
+| Trivial if we can remove callers in nsScriptSecurityManager. Alas, that is not trivial, so this is going to take a while.  For now, we lazily initialize which gives practically all the benefit.
+|-
+| X[https://bugzilla.mozilla.org/show_bug.cgi?id=540675 callerVersion]
+| .5
+| cdleary
+| Mostly the challenge is just understanding the actual use case.
+|-
+| X[https://bugzilla.mozilla.org/show_bug.cgi?id=535912 blockChain]
+| 1
+| wmmcloskey
+| A bit tricky
+|-
+| [https://bugzilla.mozilla.org/show_bug.cgi?id=595073 rval]
+| 1
+|
+| Lazily initialize with a JSStackFrame::flags bit and avoid ever writing in the hot JM call/return paths.
+|-
+| X[https://bugzilla.mozilla.org/show_bug.cgi?id=586358 imacpc]
+| 1
+| cdleary
+| Leave uninitialized in call path and use JSStackFrame::flags to indicate whether there is or is not an imacpc.
+|}