[MPlayer-dev-eng] [PATCH] VFCTRL_PAUSE_UPDATE

[Jason Tackaberry]

On Sat, 2005-09-17 at 18:18 +0200, Reimar Döffinger wrote:
> Using vo_draw_alpha_yv12 is a bad idea though, unless I finally can
> convince everyone to switch to more sane values for the OSD alpha.

It would be interesting to update vo_draw_alpha_yv12 (and all the code
that uses it) down the road, and change vf_overlay to use that, if
that's a realistic goal.  But for now I think it's more realistic to
merge vf_overlay as it is, because the less external code it touches the
better chance it has in being merged.

> But this made me see something else: your blend function uses unsigned
> saturate, but the U and V values are actually signed, so for those a
> different function using signed saturate should be used.

I mentioned this on IRC but will repeat here for those following the
thread: it seems that chroma planes are supposed to be unsigned, as all
planes in mp_image_s are declared as unsigned, plus
DOCS/tech/colorspace.txt makes a reference to U and V values being
offset +128.

> Also I wonder why in the premultiply_alpha_byte function you aren't
> using multiply_alpha...

Yes, I should be.  I've updated it.

Also, you mentioned in another email (and again on IRC) that I should be
doing 255-alpha in the premultiplication stage, not during blend.  I've
updated the code for this as well.

> movl,cmpl->mov, cmp won't compile on 64bit systems otherwise.

Fixed.

> lockbyte should be volatile

Yes, good point.  Shouldn't bgra_imgbuf also be volatile since
bgra_imgbuf is simply lockbyte+16? 

> \n in front of the messages and using perror is not good IMHO.

Merged (with some modifications) your changes to use strerror().

> fix a few includes (esp. missing malloc.h causes crash on 64 bit
> systems), compile with -Wall from time to time.

Merged your fixes to the #includes block. 

> Fix a problem when the timer overflows and use GetTimeMS to reduce the
> likelyhood of overflows.

Merged this fix, as well as the others that were in your attachment.
(In blend_image, you used memcpy_pic when alpha==256, which is good, but
you copied from the source mpi, rather than the overlay.  Fixed that.)

I do compile with -Wall from time to time.  I should make sure I do so
before submitting a patch here.

I'll post an updated patch shortly.  Thanks for all your help so far,
Reimar.  Have a good holiday!

Cheers,
Jason.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 229 bytes
Desc: This is a digitally signed message part
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/attachments/20050917/08fce7ed/attachment.pgp>