[Gstreamer-openmax] Proposal of new implementation

Felipe Contreras felipe.contreras at gmail.com
Wed Jan 26 05:41:46 PST 2011


2011/1/25 Benjamin Gaignard <benjamin.gaignard at linaro.org>:
> you are right, the operation flow are very similar but we create an
> overloaded gst buffer (gstomxbuffer) to allow zero copy in gstreamer graph

Note that gst-openmax already offers zero-copy support through the
"share hacks".

> With gstomxbuffer the sequence is no more:
> emptybufferdone -> memcpy in gst buffer -> fillthisbuffer -> pad push
> gstbuffer
> it becomes:
> emptybufferdone -> pad push gstomxbuffer -> gstomxbuffer unref ->
> fillthisbuffer

I did implement this GstOmxBuffer at some point (maybe I already have
the branch somewhere), however, I did not notice any advantage, it
just made the code more complicated.

I have been planning to reorganize the buffer handling code, as right
now it's buggy, and I would like to isolate the "share hack"
implementation so it's ugliness doesn't affect the rest of the code.
Similarly, this GstOmxBuffer strategy could be isolated the same way.

If I truly had something against GstOmxBuffer, I don't remember
anymore what that was.

Cheers.

-- 
Felipe Contreras




More information about the Gstreamer-openmax mailing list