At 12:45 AM 6/6/2000, Marek Claussen wrote:
>If the CPU is not capable of guaranteeing real-time decoding for a SA
>bitstream on a full profile decoder, can it simply switch to profile 1
>or 2 decoding ?
>Or will this be non-normative ?
>Which role will the complexity vector, introduced by Giorgio, play in
>this context ?
Eric Scheirer wrote:
>The right way for a content author to be guaranteed of
>interruption-free performance is to use the profiling
>tool on his SAOL code, and only allow the content to
>be played on terminals that provided the required
>Level of performance. There is no way to have normative
>graceful degradation using the present system.
The priority bit has also been introduced to help the
implementation of CGD techniques. It can be associated
to score events to signal which of them are "more important"
others. Another bit is used for late-arriving events while
streaming. These things are described in the "event priority"
subclause (and following one) at the end of the "decoding
process" section. I personally believe in a strong way to the
NOTE in that subclause. Levels of a Profile are intended
to guarantee the author.
>Will it be possible to transmit additional information of the SA
>bitstream to make suggestions of the CPU load before starting the
>decoding process ?
IMO the only information that can be transimtted are the Profile and
Level of the bitstream. The Level gives a suggestion of the maximum,
more or less, you should have in the bitstream. More accurate
suggestions are not possible if you don't have a strictly normative
description of every part of the decoding process. Its complexity in
SA is often "decoder-dependent.".
>However, I agree that normative GD would be a useful
>thing to have, and I hope that as more implementations
>and content authors start to work with the cpuload
>variable, we will start to see what (if any) aspects of
>it might be made normative in a hypothetical future
>version (not corrigendum) of SAOL.
A "normative" CGD sounds to me like a "contradictio in
termini" (a contradiction of the words themselves). CGD is
an approximation of the correct result, I don't see why it
should be more normative than a lopass filter or an interpolation.
But since this is still hypothetical, I am not too worried :-))
Best regards to all,
Giorgio
__________________________________________________________________
Giorgio ZOIA
Integrated Systems Laboratory - DE/LSI - EPFL
CH-1015 Lausanne - SWITZERLAND
Phone: + 41 21 693 69 79 E-mail: Giorgio.Zoia@epfl.ch
Fax: +41 21 693 46 63
__________________________________________________________________
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 12:03:57 EST