[mythtv] [patch] ViaSlice Interlaced Playback

Kenneth Aafløy ke-aa at frisurf.no
Wed Jun 23 15:22:18 EDT 2004


On Wednesday 23 June 2004 19:55, Steven Cruysberghs wrote:
> Hi Kenneth,
>
> Great work on the Viaslice stuff. I just got my epia M10k a few days ago
> and managed to get the hardware decoding to work (kernel 2.4.26-epia1
> that proved v4l_via.drv, CVS unichrome driver for via_drv.o and the
> opensource libddmpeg, PVR250 capture and X-resolution = 720x576). Today
> I tried adding this Interlaced playback patch and the qmatrix one from
> june 20.
>
> It works but I'm afraid I the only difference I notice is a flickering
> OSD when using the patches. Quality doesn't seem to improve.
> The image on the TV still seems deinterlaced to me.
[snip]

The qMatrix patch should yeild a sharper image, because only half of the 
tables was copied into the hw decoder before.

The interlacing patch was tested with a recorded pan sequence, which I put up 
at http://kenneth.aafloy.net/test-pan.mpg If you try to edit the line in 
videoout_viaslice.cpp: VideoOutputVIA::Show that says

	if (cur->progressive_sequence)
to
	if (0 || cur->progressive_sequence)

and recompile it will quickly disable the new interlacing code. Play the file 
a couple of times without interlacing enabled, then edit, recompile and watch 
the difference. It might be a little sluggish, because it's a fast moving 
scene, but the image quality is much better (vertical lines is gone).

The OSD is overlaid on top of the mpeg suface, and thus this deinterlacing 
does not apply.

The patch could certainly be improved on by blending (not overlaying) the OSD 
onto the MPEG surface, which also would give us color OSD, but this will 
require a driver patch. It has the sideeffect that MPEG2 decoding will be a 
little slower, because we need to copy the current frame to the (always) 
displaying frame, then blend the osd, and display this frame. It must be done 
this way because the hw mpeg decoder uses reference all frames as reference 
frames, and if we mess with one we mess with prediction.

Hope this'll change your list a bit :)

Kenneth


More information about the mythtv-dev mailing list