From:  nathanrogers@google.com
Date:  17 Jun 2017 10:38:21 Hong Kong Time
Newsgroup:  news.mozilla.org/mozilla.dev.tech.js-engine.internals
Subject:  

Re: Possible performance regressions across different versions of SpiderMonkey in interpreted mode

NNTP-Posting-Host:  2620:0:1006:1:9858:f6f3:2a2b:a19

On Tuesday, May 30, 2017 at 5:44:34 AM UTC-7, David Teller wrote:
> Hi Nathan,
> 
>  I'm sorry I can't answer your questions. I haven't heard of any
> regression, but I don't think we perform regular testing on RPi. Do you,
> by any chance, have profiles for both versions? I can't promise that
> someone will have time to look at them, but if someone does, it would be
> helpful.
> 
> Cheers,
>  David
> 
> On 5/16/17 3:18 AM, nathanrogers--- via dev-tech-js-engine-internals wrote:
> > Hello,
> > 
> > I work on a project called Cobalt (cobalt.foo), it is an application container designed to run web apps (namely youtube.com/tv) on embedded TV devices.  We use SpiderMonkey as our JavaScript engine, and are currently in the process of rebasing from SpiderMonkey 24 to SpiderMonkey 45.  On our application specific benchmarks, which measure input latency, which effectively measures total JavaScript execution time, we are noticing a performance regression of about 40% when running in interpreted mode on a Raspberry Pi 1, which is our reference low end platform.
> > 
> > While we plan on investigating as much as possible on our side as well, I would like to ask, is a performance regression of this magnitude to be expected?  Our configuration of SpiderMonkey, HTML application, and bindings code are all held about as constant as they can be.  If yes, then are there any high ROI configuration (or even code) changes that we can make in order to best mitigate this, and if no, then in what areas should we begin looking into first?
> > 
> > Thanks,
> > 
> > -Nathan
> > _______________________________________________
> > dev-tech-js-engine-internals mailing list
> > dev-tech-js-engine-internals@lists.mozilla.org
> > https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals
> >

Hello David,

Thanks for being able to look into this.  In the drive folder below is profiling data (out.perf) from running the perf tool (http://www.brendangregg.com/perf.html) for 10 seconds while vertically scrolling on youtube.com/tv, as well as a flame graph (http://www.brendangregg.com/flamegraphs.html) generated directly from out.perf (graph.svg), and a flame graph filtered to isolate JavaScript (jsonly.svg).

https://drive.google.com/open?id=0BxoS5eVaXrQJXzFhTXM0Q2lURWc

Thanks,

-Nathan