[gst-devel] Re: [gst-cvs] thomasvs gst-plugins: gst-plugins/ gst-plugins/sys/v4l/

[David Schleef]

On Tue, Apr 27, 2004 at 09:09:24PM +0200, Ronald Bultje wrote:
> And for Dave: I know we're not realtime, but we can aim to make it as
> realtime as possible. In this case, the amount of g_malloc()s can be
> greatly reduced without complicating code. That should be worth doing,
> even if we don't aim to be 100% realtime. I'm really not trying to make
> all buffer allocations realtime, I'm not that far off the world yet.

Sorry, I didn't need to go off on you like that.

Point is, that eliminating malloc()s in all gst/chain/loop code is
only 1% of the work needed to be done to make gstreamer and its
plugins real-time compatible.  I'm not entirely convinced that
moving g_malloc() calls out of the get function actually makes
the situation better -- it appears to just make the problems more
subtle.  One of our must-have features for correct real-time
operation is a real-time allocator, so in that case, it is
irrelevant whether the allocation is done in the get function or
not.

Real-time optimization is a lot like speed optimization -- there's
no point in doing it unless you have the ability to test that what
you are doing makes things better.  And just like speed optimization,
there's a lot of random information around that isn't rooted very
deeply in understanding.



dave...