James Zern <jzern <at> google.com> writes: > On Fri, Feb 19, 2016 at 12:00 AM, Carl Eugen Hoyos wrote: > > James Zern <jzern <at> google.com> writes: > > > >> On Thu, Feb 18, 2016 at 11:40 PM, Carl Eugen Hoyos wrote: > >> > On Friday 19 February 2016 07:59:37 am James Zern wrote: > >> > > >> >> The decoder won't change when built on a big-endian > >> >> platform, so I don't think this is correct. > >> > > >> > New patch attached. > >> > >> Sorry, I meant all of it, high bitdepth is no different. > > > > I don't understand: Do you mean that the patch is incomplete > > and that further broken pix_fmts exist? Which ones? > > I meant I didn't think LE was incorrect to begin with, > am I missing something?
Difficult to say. Short answer: How can I reproduce this? (The fact that it works fine on BE with unpatched sources.) Long answer: I was honestly very angry with myself when I saw this code yesterday: I consider it an obvious bug that I (!) should have seen whenever this patch was sent to the mailing list. I spent some time yesterday to find BE hardware and to compile there (and git.videolan was extremely slow at the time and git pull tool ages) so I only tested very cursory but I did have the impression that my suspicion (that libvpx does not change the endianess of the input on BE which would not be a good idea performance-wise) is correct (for the one example I tested). So if you say now that it works fine for you, I probably made a mistake (and libvpx does something on be that makes it even slower). How I reproduce your findings? Thank you, Carl Eugen _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
