[gst-devel] alignment issues

Benjamin Otte in7y118 at public.uni-hamburg.de
Fri Apr 9 07:21:06 CEST 2004


On Thu, 8 Apr 2004, Ronald S. Bultje wrote:

> Hi,
>
> On Thu, 2004-04-08 at 21:26, Daniel Gazard wrote:
> > Thanks, I had tested your patch and I made a few minor changes.
> [..]
>
> All make sense, they were typos. I'm fine with the current patches. Dave
> gave the comment that he dislikes macro-generating macros, but I'm
> unsure how to do this differently without a lot of duplicate macro-code.
> So I'd vote for applying this.
>
I'm fine with those patches, too, apart from one thing:
The patches default to requiring alignment. While this is a certainly the
safer solution, it makes GStreamer slow when someone didn't bother to
check this for his arch.
So either we should put an extensive number of archs in there (currently
it's just x86 and I know it works on PPC and Sparc, too) or we should
default to unaligned - which is what I'd do. People porting Gst to a
platform that requires alignment will certainly know that their arch
requires it and look for that when it breaks and then add their arch in
the header. That's certainly a lot better than just being assumed slow...

Oh, and we should have a API doc page about porting GStreamer ;)

Benjamin





More information about the gstreamer-devel mailing list