Trent Piepho (xyzzy@xxxxxxxxxxxxx): > > No, I don't think it can output 4:2:0 packed. So, you're saying > > we can just edit the DMA program to not SKIP every other line > > (which is pretty wrong if the image is composed of two fields)? > > That would be > > There are two DMA programs, one for the top and one for the bottom > field. You would have the top program send the first line to chroma > plane line 1, skip the second line of the field, send the third line > to the chomra plane line 3, skip, line 5, skip, line 7, etc. The > bottom program would send the first line to chroma line 2, skip the > next, then line 4, etc. So, is that the way it is now? If so, why am I only getting chroma from the top field? > > nice, if you can also have it perform some averaging of chroma. > > Nope, no averaging. You'll need to get 4:2:2 and downsample yourself > if you want that. The data from the video scaler portion of the chip > comes out one line at a time to the DMA engine. There isn't any kind > of line store and average circuity. Ah, brutal. Regardless, I'd like chroma from both fields interlaced if the host I'm running on doesn't have the CPU to waste on me downsampling the chroma myself. > > Ideally I think of it like this: Let o == chroma + luma sample, x > > == luma sample. So, for 4:2:0 encoding, we get this: > > 1: o x o x o x o x > > 2: o x o x o x o x > > 3: o x o x o x o x > > 4: o x o x o x o x > > That looks like 4:2:2 to me. Um, yes, that's 4:2:2. My diagrams went on to show how I go from 4:2:2 to 4:2:0 ideally and how I think the bttv driver is currently doing it. -- Billy Biggs bbiggs@xxxxxxxxxxxx http://www.billybiggs.com/ wbiggs@xxxxxxxxxxxx
Attachment:
pgpe1PBrpPR9C.pgp
Description: PGP signature