MozillaZine


Javascript Performance Thread

Discussion about official Mozilla Firefox builds
WonderCsabo
 
Posts: 2230
Joined: June 25th, 2010, 9:29 am
Location: Budapest, Hungary

Post Posted August 31st, 2011, 2:24 am

We wont see a performance boost, because the AWFY page uses Firefox builds witch already contain TI.

dao-g
 
Posts: 670
Joined: January 31st, 2005, 5:37 am
Location: Berlin

Post Posted August 31st, 2011, 3:02 am

WonderCsabo wrote:We wont see a performance boost, because the AWFY page uses Firefox builds witch already contain TI.

http://arewefastyet.com/?a=b&view=regress does, http://arewefastyet.com/ doesn't.

greg86
 
Posts: 50
Joined: January 28th, 2007, 11:33 am
Location: Germany

Post Posted August 31st, 2011, 4:49 am

where did u see a boost on http://www.arewefastyet.com/?a=b&view=regress&runs=300 ???
i only see regressions since middle of august

dbcooper.dk
 
Posts: 895
Joined: March 14th, 2010, 3:44 am

Post Posted August 31st, 2011, 5:03 am

iwod wrote:Great, next would be to further optimize for V8 benchmarks to wipe the smug face off Chrome.


That will come with the implementation of incremental and generational GC.

Ver Greeneyes
 
Posts: 1030
Joined: June 28th, 2008, 4:57 am

Post Posted August 31st, 2011, 5:09 am

The main page will display the TI-enabled results once bug 683410 lands.

Blechnum
 
Posts: 26
Joined: October 28th, 2010, 1:33 am

Post Posted August 31st, 2011, 6:48 am

Finally something happening!
Here is what I'm seeing (compared to two days ago): V8 improved from 4109 to 4607, Kraken improved from 5351.2 ms to 4573.1 ms and Sunspider also went down from 258 ms to 246.3 ms. Nice. But Peacekeaper took a turn for the worse I'm afraid: the score went down from 5241 to 4671.
There are also regressions in some of the subtests in Kraken and V8 but I don't got the numbers.
Oh well, at least things are moving again. :D

WonderCsabo
 
Posts: 2230
Joined: June 25th, 2010, 9:29 am
Location: Budapest, Hungary

Post Posted August 31st, 2011, 8:06 am

I think the about:config pref for TI should contain the word "jit". In that case, you can search for "jit", and all JS engine prefs come up.
I remember some dev suggested the same for profiling. Guys, do you think it worths opening a bug?

Mark342
 
Posts: 277
Joined: September 14th, 2010, 4:15 am

Post Posted August 31st, 2011, 8:38 am

Make sure you guys are testing the latest version of V8 (version 6) when comparing inference on/off
javascript.options.typeinference set to true:
V8: 5472
Highest scores:
Crypto: 13707
DeltaBlue: 10546

other scores:
Richards: 7692

Sunspider: 232.4ms +/- 0.9%

Kraken: 3961.6ms +/- 0.6%

---------------------------------------------------
javscript.options.typeinference set to false:
V8: 4193
Highest scores:
Richards: 6563
Crypto: 5816

Other scores:
DeltaBlue: 4708

Sunspider:
260.1ms +/- 2.0%

Kraken:
5322.7ms +/- 0.4%
--------------------------------------------

Note: the browser must be restarted for javascript.options.typeinference to take affect

UrbenLegend
 
Posts: 307
Joined: August 18th, 2007, 11:12 am

Post Posted August 31st, 2011, 9:43 am

Seems like there's a performance regression for http://bellard.org/jslinux/

It used to boot up in 6 seconds. Now with TI it boots up in 13 seconds.

EDIT: Found bug https://bugzilla.mozilla.org/show_bug.cgi?id=670885

nadark
 
Posts: 491
Joined: November 12th, 2010, 9:12 am

Post Posted August 31st, 2011, 10:50 am

Mark342 wrote:Make sure you guys are testing the latest version of V8 (version 6) when comparing inference on/off
javascript.options.typeinference set to true:
V8: 5472
Highest scores:
Crypto: 13707
DeltaBlue: 10546

other scores:
Richards: 7692

Sunspider: 232.4ms +/- 0.9%

Kraken: 3961.6ms +/- 0.6%

---------------------------------------------------
javscript.options.typeinference set to false:
V8: 4193
Highest scores:
Richards: 6563
Crypto: 5816

Other scores:
DeltaBlue: 4708

Sunspider:
260.1ms +/- 2.0%

Kraken:
5322.7ms +/- 0.4%
--------------------------------------------

Note: the browser must be restarted for javascript.options.typeinference to take affect



I think I read somewhere that there's some overhead for including TI in the browser and turning off the pref only doesn't get rid of it due to the complexity of the code needing a lot of changes in the javascript engine to work , since most people don't run with TI disabled. If you could, could you test with the build just before TI landed on the trunk.

dutchguy
 
Posts: 199
Joined: October 14th, 2003, 12:39 am

Post Posted September 2nd, 2011, 12:09 am

They removed the 'old' engines from the regression page :)

Zlip792

User avatar
 
Posts: 1340
Joined: May 7th, 2011, 1:47 pm
Location: Pakistan

Post Posted September 2nd, 2011, 12:13 am

Is TI graph is in actual (JM+TI) graph? or JM+TI will be added later?

iwod
 
Posts: 1033
Joined: July 18th, 2003, 10:09 pm

Post Posted September 2nd, 2011, 3:59 am

Anyone know when is Incremental GC landing? It seems to be read and finished

https://bugzilla.mozilla.org/show_bug.cgi?id=641027

But dont know what is stopping it to land.

Astrophizz
 
Posts: 235
Joined: March 13th, 2011, 1:15 pm

Post Posted September 2nd, 2011, 7:05 am

They also removed the 'old' engines from the regular page and didn't replace it with TI...

phuzi0n
 
Posts: 517
Joined: June 23rd, 2010, 5:48 pm

Post Posted September 2nd, 2011, 9:31 am

Is Firebug killing javascript performance after TI landed for anyone else? I tried updating to the latest stable Firebug 1.8.2 and then to 1.9.0a1 but it still slows down v8 benchmark by ~4x and sunspider by ~2x. I made sure that Firebug's javascript profiling is disabled (with Firebug profiling enabled v8 is ~10x slower).

Return to Firefox Builds


Who is online

Users browsing this forum: No registered users and 2 guests