[FFmpeg-devel] [PATCH] h264 parallelized,

[Andreas Öman]

Hi

Michael Niedermayer wrote:
> Hi
> 
> ok if md5 doesnt change for type 2 content and theres no meassureable
> speed loss for normal non type 2 content

I've tested with all h264-samples (i found) @ mphq + all the
type-2 content i have. MD5 does match. (I can make the
results available upon req.)

There is no measurable slow down for type 0/1 content.

However, when doing this I realized that there is no significant
speedup for type-2 content either (contrary to what i previously
claimed).
I have only done detailed speed tests on the entire parallelization
patch as a whole (found in a previous mail; 4666868E.10401 at olebyn.nu).
Probably, the addition of 'int simple' to xchg_mb_border() did
all of that (small) speedup.

Still, it might be worth to keep it separate to aid finding
regressions when doing bisect search (if you are worried
that it might possibly break something)