smoohta wrote:Well... in the video you posted you can see dmandelin's directory: http://people.mozilla.org/~dmandelin/Summit2010/ from which he opened the ants benchmark. If you snoop in it, you can see there's a browsers folder with moo1.zip (supposedly the win32 JM moo build he was using in the presentation) and a "regular" firefox-3.7a5pre.en-US.win32.zip... There's also an ants folder containing the cool ants benchmark
I've tried the moo build with the ants benchmark and it seems to work just like in the presentation, though I couldn't get JSNES to load due to some javascript errors...
I get a script timeout on the iGoogle homepage with that build.
Would it not seem sensible given real world doesn't match benchmarking to have a 'profiling enabled' build that acts like the heat mapping systems you see on websites, namely that the most highly stressed areas of the browser are highlighted for improvement and rarely used sections of code aren't prioritised. It stands to reason that while the two js engines system will be quick, a combined hybrid system would save a layer of processing namely tm > jm > native.
My suggestion would go beyond just js though and basicly time and account for all elements of the experience reporting back so developers can see where the most time could be shaved. The other side of this is that you could utilise the seperate layers of jm/tm as 'high speed'/'compatibility' modes.
Looks like Jaegermonkey has finally caught up with Tracemonkey on SunSpider. Tracemonkey has also just passed the <700ms line. http://www.arewefastyet.com
So we finally we have something insight. If we get a 15 - 20% Performance Win on SS with TM, Added the benefits of JM we should be close to V8 and Nitro....
iwod wrote:So we finally we have something insight. If we get a 15 - 20% Performance Win on SS with TM, Added the benefits of JM we should be close to V8 and Nitro....
I think time is the answer at that ,check the JaegerSpeed bug category to see the planned improvment in term of speed.
How significant are the Sunspider results though? Sunspider did not catch the fundamental problem that TraceMonkey simply fails in many real-life situations when there are no/not many long-running loops to optimize. So despite JägerMonkey being as fast as TraceMonkey on synthetic tests right now the former might actually perform already perform better because it covers more cases.
The previous signature has been removed again. Enjoy your month off, Erunno.
Erunno wrote:How significant are the Sunspider results though? Sunspider did not catch the fundamental problem that TraceMonkey simply fails in many real-life situations when there are no/not many long-running loops to optimize. So despite JägerMonkey being as fast as TraceMonkey on synthetic tests right now the former might actually perform already perform better because it covers more cases.
Yes you have right, but many people continue to compare browsers in term of speed using this syntetic tests.
Interesting JSNES part playing 60fps. Ants part arent impresive, my score its better 100-110fps without spider, 430-450 JS speed. With Spider and food, fps dont fall from 27fps with 32 JS speed.
Im waiting for JM, im sure will be very interesting part. Anyway, im agreement with you, and JavaScript its only a performance part of a whole.
Would the problem of lack of 'real world' benchmarks simply be a case of taking the javascript from say 10-20 popular sites, gathering it together and adjusting the output format, so that all parts have to pass and be of a certain speed to complete the output. Sounds tricky but shouldn't be that hard plus you could do a number of things to automate the process, you could for example setup a page that inserts the .js files from sites directly then uses the functions in the same way as the sites do, thus resulting in an 'always updated' benchmark.
Jsnes, sunspider etc are fine at what they do but the real test would be to throw massive amounts of 'common' content at it and judge which areas need improving.
mozillaZine is an independent Mozilla community and advocacy site. We're not affiliated or endorsed by the Mozilla Corporation but we love them just the same.