On Montag 14 August 2006 18:35, Martin Wache wrote: > Stefan Lucke schrieb: > > There might be some other mystic ;-) . > > YUV (aka YV12) valid values are _not_ in the range 0 .. 255 ! > > Y and U/V values should be clamped: > > luma values to the range 16 .. 235 and > > chroma values to the range 15 .. 240 . > > > Of course you are right that YUV values are not in the range 0-255, but > I doubt that this causes the problems. > In this bugreport it was stated that the same osd switches from > correctly displayed to wrong colors: > > > Sometimes, if I leave some part of OSD > > on screen without touching it for a while, it will flick from being > > correct to being pink, and back again, as adverts come and go, at > > different resolutions. > > I can hardy imagine how a wrong range could cause that. Also the stripes > look like some scaling issue...
You are right the symptoms look like a scaling issue, but in that case it should be possible to reproduce that error on other cards too. So I try to look at things we did not consider yet . Another hint could be that most Xorg / XFree driver do a YV12 -> YUV2 conversion, but I failed to reprocude the error with vidix, OSD set to software and various horizontal resolutions ;-( . The only time I saw a out of range problem was here: http://sourceforge.net/mailarchive/forum.php?thread_id=272658&forum_id=5458 But there was no OSD involved and so failing values looked not regular. > > > Perhaps that could be tested in the C part of alpha mixing code -> > > AlphaBlend(). > > > That is one possibility, but using the correct range in ARGB_to_AYUV() > is probably the better solution. By the way, are you sure that the > calculation in ARGB_to_AYUV() is wrong? I must admit that I never > checked it, but at least the Y component seems to be correct. My suggestion to use AlphaBlend() is based on the fact that it is the last step in the mixing pipline and alpha 0 or 255 could lead to calulated luma 0. That's just a quick thought. -- Stefan Lucke _______________________________________________ Softdevice-devel mailing list [email protected] http://bat.berlios.de/mailman/listinfo/softdevice-devel
