Welcome back Andy.

On Sat, Nov 1, 2008 at 8:54 AM, Andy Walls <[EMAIL PROTECTED]> wrote:

> On Fri, 2008-10-31 at 15:46 -0400, Jeff Campbell wrote:
> > Hi Andy,
> >
> > My apologies for not getting back with the pci details, I've been away
> > from the hardware that has the problem for a few weeks (and using my
> > other box).  I'm now getting back at it.
> >
> > I wanted to followup on your suggestion that I use mplayer with a 16MB
> > buffer to play the stream if reading directly off /dev/video0.
> >
> > While that works, I was curious if there are any plans, or if it is
> > even possible, to do some buffering on the driver side so that the
> > resulting native TS is more stable.
>
> First what do you mean by native TS?  Analog captures default to an MPEG
> PS, but supposedly the chip can also be set to do a TS for analog
> capture (I haven't tested it yet).  Digital captures are just a
> pass-through of the TS from another chip.
>

Let me give an update here.  By setting the stream_type=1, you can get
native TS from the card/driver.  This is something Hans added support for a
few months ago.

This does in fact work and I have successfully multicasted it and played it
back.  It was generally watchable for test purposes but did suffer from the
jitter problem.  Ultimately I ended up experiencing significant audio/video
sync issues (over time) with the audio falling multiple seconds behind the
video. The actual video frame rate seemed fine (my input source was a DVD
player) and the audio, as I recall, was good but suffered occasional
dropouts.  I am looking in to a way to pre-buffer this before transmitting
it to try and smooth it out.

I have had some success adjusting buffer sizes on the client side that is
joining the multicast, but nothing conclusive yet due to various other
factors.

I am doing more testing tonight and will report any material findings.  So
far I've had no audio and audio sync issues, but there are a lot of
variables in my test setup so I don't want to say its the card or driver
just yet.

I am going to build the latest version of the driver as well as a
2.6.27.4kernel later tonight to test against.



>
> Well, I have some plans for buffers changes, but buffering up 8 or 16 MB
> in kernel space wasn't one of them.  The problem is one of non-uniform
> rate of delivery of audio and video buffers, I think.
>

Your description of non-uniform rate of deliver sounds correct and seems to
match the errors that I see in the messages window in the vlc client while
playing back a stream.


>
> I don't know how to get the chip to transfer 1 video buffer per frame,
> so the solution is to use smaller transfer buffers to get the chip to
> hand them over sooner.  The problem is the driver is simply written to
> stay with in the buffer count limit of 63 buffers usable by the firmware
> at any one time.  Making the buffers smaller to smooth out transfers
> while staying under 63 buffers makes for a very small overall buffer.
>
> I plan to change the driver to manage more than 63 buffers per stream
> and let the user control buffer sizes to meet their latency needs.  But
> right now I investigating some interrupt and mailbox problems to try to
> resolve problems with simultaneous analog and digital capture.
> Hopefully that will make things better too.
>

We all appreciate your efforts.  I'll try and do some more structured
testing and report back on what I find.

Great news about the pending improvements on simultaneous analog and digital
capture.


>
>
>
> >   VLC really chokes on it if you try and play it directly, due to late
> > picture and late sound.
>
> OK.  I never tested VLC.


I've been using 0.8.6 under linux but also install 0.9.4 under Vista today,
so I'll report anything that seems material.

-Jeff
_______________________________________________
ivtv-devel mailing list
[email protected]
http://ivtvdriver.org/mailman/listinfo/ivtv-devel

Reply via email to