Benchmarks & tests results

Tue Sep 27 16:32:27 UTC 2011

1. Hm OK IC ... 
Do we have a bug-tracker or something where we can store that stuff?
I would love to be the assignee for this bug ... but I fear I'll have forgotten about it when I find some time in 1-2weeks

2. sounds a bit like a HW-problem?

________________________________________
From: ext Lauri Laanmets [lauri.laanmets at proekspert.ee]
Sent: Monday, September 26, 2011 9:16 PM
To: Senyk Thomas (Nokia-DXM/Munich)
Cc: bog dan ro; necessitas-devel at kde.org
Subject: Re: Benchmarks & tests results

Hi,

1. About the Car3D wheel and door animation: I tracked town the cause in QML. "Animation.Infinite" returns correctly -1 in QML element scope but returns 4294967294 in script scope :) I have "no eye deer" why is that so. So just replacing the constant expression to -1 in two places does the trick.

2. About the opengl performance, I think I have to take my words back. I have downloaded a bunch of 3D apps from Android Market and they all have the same behavior. Maybe there is something in the Android or hardware but seems that pixels have longer response time in opengl. It is specifically visible in case of white font on blank screen but also picture edges and all other stuff. Therefore depending on the flick speed the overall font is more bright or less bright since edges in direction of moving go grey. The fastest speed, the bigger grey area. That is why it seems from distance like it's binking a little. Somehow Android menus doesn't behave like that.

Regards
Lauri

----- Original Message -----
From: "thomas senyk" <thomas.senyk at nokia.com>
To: "lauri laanmets" <lauri.laanmets at proekspert.ee>
Cc: "bog dan ro" <bog_dan_ro at yahoo.com>, necessitas-devel at kde.org
Sent: Monday, September 26, 2011 3:03:27 PM
Subject: RE: Benchmarks & tests results

That was a very old one ... and it does not apply.
If you do FullViewportUpdate (what you do if you're using the main.cpp from Car3D)
(... which you should to anyway on android (and any other eglfs platform))
you should be good.

So from my perspective it's a new/unknown bug/issue.

From: ext Lauri Laanmets [lauri.laanmets at proekspert.ee]
Sent: Monday, September 26, 2011 1:50 PM
To: Senyk Thomas (Nokia-DXM/Munich)
Cc: bog_dan_ro at yahoo.com; necessitas-devel at kde.org
Subject: Re: Benchmarks & tests results

Hi

Indeed, I forgot to mention about scaling wheels and doors. Although I was playing around with the source code (you know, round corners :) - in my mind I didn't disable that. I have to investigate why it doesn't work, when I boot up my Linux again. I'll let you know.

The flicking is not visible on the video since camera light sensors smooth it. It's not big but it is there. The effect is similar when having a bad deinterlace in videos - I hope know what I'm talking about. I remember a discussion about some double-buffer issue but cannot find it anymore in the history. What was it about?

Regards
Lauri

On 26.09.2011 14:19, thomas.senyk at nokia.com wrote:

Have you disable all the tranforms? (scaling wheel, rotating doors)

>From a performance perspective it looks ok (not great though)
I can't see the flicking/black-frames. Maybe a double-buffer problem?

From: necessitas-devel-bounces at kde.org [ necessitas-devel-bounces at kde.org ] on behalf of ext Lauri Laanmets [ lauri.laanmets at proekspert.ee ]
Sent: Monday, September 26, 2011 11:35 AM
To: BogDan
Cc: necessitas-devel at kde.org
Subject: Re: Benchmarks & tests results

Morning

Is there a short description somewhere how to run the tests and what are the prerequisites? The information coming to me at the moment is just so overwhelmingly big that I cannot cope with that :)

And also one question, the OpenGL performance looks very fast and stable but it's kind of a flicking. Seems like FPS is low or blank screens between frames (not sure but movement looks strange), am I doing something wrong?

Regards,
Lauri
PS: FYI: a short video of Car3D on Motorola Xoom: http://www.youtube.com/watch?v=lv_f96fqciQ

On 25.09.2011 20:26, BogDan wrote:

Hello folks,

I manage to run almost all tests on android, some of them are crashing but most of them are running and surprisingly a lot are passing all tests.
The tests where made using dfb51925e136a96ed0fcdcfed140f0ac6158771a compiled in debug mode.

Next step is to try to modify Nokia's bm[1] tools to fit our needs :
 - add device parameter or should we use host parameter instead? This parameter is very important to group benchmarks by device model ("ro.product.name" parameter of getprop command).
 - add authentication support: I think adding ssl support will be the easiest and the safest, this way we can automatically accept benchmark results from everyone who has a certificate signed with our server certificate.

Cheers,
BogDan.

[1] https://gitorious.org/~taipan/qt-labs/android-bm

_______________________________________________
Necessitas-devel mailing list Necessitas-devel at kde.org https://mail.kde.org/mailman/listinfo/necessitas-devel