Closed Bug 502145 Opened 14 years ago Closed 14 years ago

TM: shortcut lookup flag calculation when on trace

Categories

(Core :: JavaScript Engine, defect)

x86
macOS
defect
Not set
normal

Tracking

()

RESOLVED FIXED

People

(Reporter: gal, Assigned: gal)

References

Details

Attachments

(1 file, 2 obsolete files)

Instead of re-calculating lookup flags on trace (Detecting ...), calculate at recording time and just peek into the side exit structure. This is an 8ms perf win on SS on my machine.
Attached patch patch (obsolete) — Splinter Review
This of course also avoids falling of trace in InferFlags, which we do a lot right now.
Assignee: general → gal
Attachment #386662 - Flags: review?(igor)
Blocks: 460904
Attached patch static analysis fix (obsolete) — Splinter Review
Attachment #386662 - Attachment is obsolete: true
Attachment #386698 - Flags: review?(igor)
Attachment #386662 - Flags: review?(igor)
Attachment #386698 - Attachment is obsolete: true
Attachment #386699 - Flags: review?(igor)
Attachment #386698 - Flags: review?(igor)
Attachment #386699 - Flags: review?(igor) → review+
Pushed to the try server along with the js_AllocStack patch.
Could this have caused an across the board increase of about 150ms on SS?  I'm now seeing this consistently using build: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.2a1pre) Gecko/20090706 Minefield/3.6a1pre (.NET CLR 3.5.30729) ID:20090706032334.  The build prior to this landing averaged about 1600ms on SS, this build averages about 1750ms.

~B
(In reply to comment #6)
> Could this have caused an across the board increase of about 150ms on SS?  I'm
> now seeing this consistently using build: Mozilla/5.0 (Windows; U; Windows NT
> 5.1; en-US; rv:1.9.2a1pre) Gecko/20090706 Minefield/3.6a1pre (.NET CLR
> 3.5.30729) ID:20090706032334.  The build prior to this landing averaged about
> 1600ms on SS, this build averages about 1750ms.

Is this witha  mozilla-central build or a tracemonkey tree build?
Tracemonkey tree build from this AM: tinderbox-builds/tracemonkey-win32/1246875814/

~B
Testing again. I am not opposed to backing out until this is clarified.
TM tip gives me the best score we ever had on my box for SS on a macosx shell build (922ms). Might be different in the browser or on windows. Anyone in the mood to test this in an opt build?
I just tested TM tip (24ea6a78f889) and got 1186 ms in browser, which is better than anything I've gotten before: (SS results) http://tinyurl.com/nby9a8.
http://hg.mozilla.org/mozilla-central/rev/753d18be63fc
Status: NEW → RESOLVED
Closed: 14 years ago
Resolution: --- → FIXED
Depends on: 514790
You need to log in before you can comment on or make changes to this bug.