openML use two buffers pipes for transcoder. this could become a problem when using a lot of software transcoder : we must allocate a lot of buffer. maybe we can improve this by allowing one pipe for transcoder using this behaviour :
user set an image format for the way app->trancoder
user set an image format for the way transcode -> app
user set the streams parameters for both up and down way (the rates MUST be the same in such behaviour)
user allocate buffer for the most expensive format and queue them.
transcoder will perform operation and return output data in the same buffer memory space...
with such mechanism as long as all transcoder share the same rates we can allocate memory only for the most expensive buffers and share them among all transcoder.....