[mythtv] Experimental pvr 350 output patch

David Engel dlengel at attbi.com
Wed Dec 17 21:32:53 EST 2003


On Wed, Dec 17, 2003 at 07:57:11PM -0500, Isaac Richards wrote:
> My M10k is still compiling everything, but have you tried turning prebuffering 
> (V4L2_CID_IVTV_DEC_PREBUFFER = 1) back on or bumping the decode buffers back 
> up to 9 (V4L2_CID_IVTV_DEC_NUM_BUFFERS = 1)?  Might not need the second 
> buffer in ivtvdecoder that way.

I tried V4L2_CID_IVTV_DEC_PREBUFFER = 1, it didn't seem to help.  I
forgot to try V4L2_CID_IVTV_DEC_NUM_BUFFERS = 1.  I'll do that soon.

One thing I believe is contributing to the glitches is the way I'm
getting the current frame count.  I wanted the single stepping for
exact seeks to be as fast as possible so I'm calling the firmware
directly since the existing ioctl only returns the frame count as of
the last vsync.  This introduces a minimum 10ms, and sometimes 20 or
30ms, delay everytime through GetFrame.  

During normal playback, I can probably revert to the existing icotl to
eliminate the delay most of the time.  Even so, this can cause a
>=10ms delay too, we suspect because of a known firmware bug and the
driver not handling it correctly.

How fast is the RingBuffer at seeking backwards ~32-96k?  Becuase I'm
using non-blocking I/O for the mpeg data, the entire buffer usually
isn't written.  If we don't buffer the data in ivtvdecoder, we either
have to seek backwards evertime or only write 32k, driver-friendly
chunks at a time.

On Wed, Dec 17, 2003 at 08:49:47PM -0500, Isaac Richards wrote:
> Seems to work pretty well.  I didn't see any decoding glitches, either.  

Good.  I'll start preparing a message for Kevin and the ivtv guys.  I
know he's getting ready to release a new version, so hopefully I can
get the required changes in, in a backwards compatible manner, before
the new release.  Then, we can worry about commiting the Myth changes.

> Edit mode's funky =)

Yeah, I know, but at least it's functional.  I've got some ideas for
future improvements, but I want to get the existing stuff accepted and
handle some other todos before doing anymore with it.

David
-- 
David Engel
dlengel at attbi.com


More information about the mythtv-dev mailing list