[Libva] API changes

[Ratin]

On Sun, Jan 13, 2013 at 10:59 PM, Xiang, Haihao <haihao.xiang at intel.com>wrote:

> On Sun, 2013-01-13 at 20:38 -0800, Ratin wrote:
> > Hi, I have noticed that since september / octorber of 2012, VAAPI
> > structs, signatures changed  so much that I had to throw out all my
> > decode/rendering code that I previously wrote and basically start
> > writing from scratch again.
>
> The official release has stable APIs. You are using staging branch,
> right ? The staging branch is created for new features, so some of the
> APIs for new features aren't stable enough.
>
>
Hi Xiang, Thanks for your reply, Yes, I am using the staging branch, but
irrespective of which branch I am working on, I think if the the basic
signatures should not change (for example vaCreateSurfaces signature
completely re-ordered) Things like VAEncPictureParameterBufferH264 does not
exist anymore, so the whole section of code which is based on this stucture
variable kind of falls apart.. I am sure it will not be changing anymore
soon - just curious to know if  a drastic change like that will happen soon
, should I wait until that moment etc.


BTW, I have a question regarding pushing the decode / render cycles to a
limit.  I have got success bulding the code base on ubuntu 13.04, kernel
version  3.7.0-7-generic, staging branches of inetl driver and VAAPI. I am
decoding h.264 video from a file, decode and then render (put surface). I
am pushing 9 1080P decode/render on a little Intel Nano box, with HD4000
with core i3. Initially the latency is nothing , load average is only .35,
cpu percentage is only 48 % . As time progresses, load average slowly gets
higher and higher, and at t+2 hr, load average is at 7.9, cpu usage is 312
%. What would cause something like this? I forced the cpu to goto max
performance via sysctl tunables so this should have been consistent. I am
also monitoring the cpu temperature and its gets hot but not alarming hot,
besides the temperature based cpu stepping thresholds kind disabled, as I
picked max cpu freq via kernel tunables.

The project I am doing requires the system to be up and running for days so
this is kind of important for me to keep some kind of consistency for days.
Do you guys do similar long term tests with high load and if yes, can you
elaborate what you have seen?

Thanks for your input..

Ratin






> > For example, the changes in test program avcenc is drastic. I was just
> > curious how often do you make these kind of core API changes? Is it
> > possible to follow some basic programming standards like, don't change
> > the signature drastically, use structs for signatures so its only
> > elements within strucs that changes, and also call version 2 of the
> > function from within function 1 internally , dont change version 1's
> > signature ..  etc?
> >
> >
> > Thanks
> >
> >
> > Ratin
> >
> > _______________________________________________
> > Libva mailing list
> > Libva at lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/libva
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/libva/attachments/20130114/964bb66d/attachment.html>