[FFmpeg-devel] [PATCH] VP8 EMU_EDGE support

Ronald S. Bultje rsbultje
Tue Dec 28 18:37:36 CET 2010


Hi,

On Mon, Dec 27, 2010 at 9:25 PM, Ronald S. Bultje <rsbultje at gmail.com> wrote:
> On Sun, Dec 26, 2010 at 8:07 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
>> On Sat, Dec 25, 2010 at 11:35:09AM -0500, Ronald S. Bultje wrote:
>>> On Sat, Dec 25, 2010 at 10:09 AM, Ronald S. Bultje <rsbultje at gmail.com> wrote:
>>> > patch adds support for that. Complications are that as far as I can
>>> > see, there's no restrictions on referencing off-screen pixels, in fact
>>> > these are defined (127/129), so I had to add some hacks to handle
>>> > that. Please review and let's discuss if these are acceptable or if we
>>> > want other ways to accomplish the same. I've added two new intra
>>> > prediction modes, DC_127 and DC_129, which are identical to DC_128 but
>>> > different value. For more comples VP8 edge intra pred cases where it
>>> > depends on off-frame pixels, I emulate the edge in a temp 5x8 buffer.
>>> > I can add more hacky intra pred cases for this but I don't think it's
>>> > important, after all the number of MBs that reference off-frame pixel
>>> > data is merely one row + one col...
>>> >
>>> > Make fate-vp8 fails with -flags emu_edge. I think this isn't because
>>> > displayed data output changes, because diff says PNGs are identical
>>> > before/after. Maybe framemd5 includes edge pixels in its pixel data?
>>> > Also, pixel data looks visually identical and photoshop difference
>>> > mask shows no differences.
>>>
>>> Correction: the no-pixel-diff was b/c I misplaced -flags emu_edge in
>>> the commandline while debugging this. Attached patch makes fate-vp8
>>> pass with -flags emu_edge added.
>>>
>>> Ronald
>>
>>> ?h264pred.c | ? 68 +++++++++++++++++++++++-
>>> ?h264pred.h | ? 12 +++-
>>
>> ok if needed
>
> Unfortunately needed.
>
> Jason asked me to improve the names of the pred modes a bit, so here's
> a new patch with changed names, some comments, and so on. I'll apply
> this tomorrow if there's no further comments.

And applied.

Ronald



More information about the ffmpeg-devel mailing list