[jitter] Re: HD performance with jit.qt.movie in @window direct mode

But with this method, should you want to queue more than one movie  
back to back, as separate files, dont you get the  weird QT window  
pop over? Can you queue movies seamlessly back to back without  
leaving fullscreen at the end of one movie?

Curious , as I have not had the chance to try this in any capacity,  
if the above is possible, I wish I knew a few days ago! :)

======

Back to jitter land,

Jitter 1.6.3/Max 4.6.3

I have a patch (attached to this mail) I am using for a performance,  
an incredibly paired down/simple qt/gl patch, on a quad 2.5 Ghz G5  
with 8GB of ram and NVidia Quadro FX 4500. It cues with movies with  
19 jit.qt.movies with @preroll 1, @unique 1, @colormode uyvy, @adapt  
1 and @loop 0.

Only one movie plays at a time, and bangs are gated and switched  
'around' all others so only one is active. All movies are stopped  
untill needed. I have one GL object, a slab, @automatic 0. I use a  
uyvy2rgba shader and a simple 'fade to color' custom shader. I have  
playback issues if the shaders are not in the pipeline as well, just  
so you know.

There are no ui objects. not even an fpsgui. Main timing is triggered  
by a qmetro 2.

Playing DVCProHD (that is, 1280x720p at 60Hz) footage with this simple  
of a patch will continuously stutter. Very visibly.

Direct to window leads to even worse performance (oddly enough).  if  
I follow the direct to window noaccel route in fullscreen, I get  
consistent very low frame rates in full screen, regardless of menubar  
being visible or windows being overlayed. Playback seems somewhat  
better, but still not in any way shape or form 'right' in a windowed  
mode. Replacing 'fullscreen' with a rect message that still takes up  
the full screen still results in impressive stuttering for such a  
machine. I have this behavior on the quad and on my macbook pro.

For reference, I have benchmarked the drives (they are 90% empty) and  
there is more than sufficient disk througput.

Movies can be long, but nothing but playback at rate 1 occurs. this  
is simply sending 'start' to the movies.

I have confirmed that Quicktime and Final Cut can play back these  
clips to the same size output (second monitor) without any visible  
framerate drops, hiccups or glitches.

Converting to UYVU seemed to somewhat improve the situation, but  
stuttering still happens.

I have attached a version of the patch with cuing missing. I feel as  
though I havent gotten a straight answer from anyone about this when  
I send a patch, except the 'it works for me'. Maybe Im better at  
seeing stuttering?

I know this topic is a little sore, so apologies Cycling, I really  
hope this is not seen as a nagging/get your shit together/etc sort of  
a rant, I am attaching the patch in an attempt to isolate or identify  
any strategies I (and others) can use to, in the real world, on real  
systems in real patches, get top performance with Jitter for high  
quality playback.

I feel i should note that, myself withstanding, other people working  
with me have noted the stuttering, but deem it somewhat acceptable  
(so we are sticking with Jitter for its flexibility and ease to  
change cues/timing etc). So it isnt "just me".

Ive also played back these files with the movie and slab help files  
with the @adapt 1 added to jit.qt.movie, and see the same stuttering  
behavior.

I do not expect or want any immediate fixes for this either, im  
simply curious at this point what the response will be. Again,  
apologies for bringing /extending this sore topic. Thanks in advance.

p.s. Yes, I am aware that other codecs might free the CPU and  
scheduler to do more, but, at this point, I dont think that is much  
the issue?

-------------- next part --------------
A non-text attachment was scrubbed...
Name: TBOD_CUE
Type: application/octet-stream
Size: 8090 bytes
Desc: not available
Url : http://www.cycling74.com/pipermail/jitter/attachments/20070907/4bb72014/TBOD_CUE.obj
-------------- next part --------------


v a d e //

www.vade.info
abstrakt.vade.info