[Ffmpeg-devel-irc] ffmpeg-devel.log.20130208

burek burek021 at gmail.com
Sat Feb 9 02:05:03 CET 2013


[00:15] <saste> anybody going to review the opencl deshake patch?
[00:21] <cone-93> ffmpeg.git 03Michael Niedermayer 07master:0d194ee51ed4: vc1dec: Fix destination pointers in context for fields
[00:25] <michaelni> saste, iam lazy, you can review it
[00:25] <saste> michaelni, i'm busy
[00:25] <saste> i'll try to review it anyway, in the weekend
[00:26] <saste> also i know no s*#t about opencl
[00:27] <michaelni> anyone has any odd platforms/hw that are missing from fate.ffmpeg.org ? ... and wiling to setup fate clients ?
[00:28] <michaelni> we now also have a SH4 client since a few days (qemu based though)
[00:29] <saste> michaelni, i have a ZX spectrum, might that be helpful?
[00:30] <michaelni> can it run ffmpeg / fate ?
[00:42] <michaelni> saste, you could optimize and port ffmpeg to ZX spectrum ;)
[00:43] <michaelni> maybe a striped down version that does support just one container and codec or so
[00:43] <saste> uhm in 128KiB it's going to be hard
[00:44] <michaelni> uhm why ? 
[00:44] <michaelni> there was a 3d first person shooter in less than 64k IIRC
[00:44] <michaelni> not on  ZX spectrum though id guess ;)
[00:51] <gnafu> If anyone could do it, it would be Fabrice Bellard.  And so I would think a project he started could be transformed to do it.
[00:51] <gnafu> ;D
[01:13] <saste> well at least we could have a player on zx
[01:13] <saste> after all we have an encoder for c64
[01:48] <ubitux> hey
[01:49] <ubitux> michaelni: a00c4b4db  seriously? oO
[01:49] <michaelni> yes seriously
[01:49] <michaelni> no joke
[01:50] <ubitux> why does this happen?
[01:50] <michaelni> no idea, gcc i guess
[01:51] <michaelni> but + is nicer for gcc
[01:51] <michaelni> it allows o2*=a2; o2+= ...
[01:51] <michaelni> with - that style would not be possible for gcc to assemble to
[01:51] <michaelni> but i didnt check if gcc actually does this
[01:52] <ubitux> fun
[01:53] <ubitux> kyowa, tokyo desu!~
[01:53] Action: ubitux &
[02:12] <Compn> michaelni : wei was asking in the dxva2 mail "Also I want to ask a question that is the chroma_format always stored in the ext sequence header?"
[02:12] <Compn> he has some other questions in that mail too
[02:27] <michaelni> Compn, what mail (subj/date) ?
[02:54] <michaelni> Compn, found the mail & replied
[03:00] <Compn> oop, ok good :)
[03:00] <Compn> thanks for looking at it 
[11:45] <cone-778> ffmpeg.git 03Paul B Mahol 07master:810cd0dd9c03: riff: chunks must be at 2 byte boundary
[12:14] <durandal_1707> michaelni: the fil filters is buggy, it interleaves incorrectly
[12:51] <michaelni> durandal_1707, libmpcodecs vf_fil ? 
[12:51] <michaelni> does the isssue happen just with ffmpeg or also mplayer ?
[12:51] <michaelni> its a mplayer filter and not maintained in ffmpeg
[12:52] <durandal_1707> michaelni: i dont have mplayer installed
[12:52] <durandal_1707> i just did fil=i,fil and expect to see same output
[12:55] <michaelni> fil was designed to be fast (0 copy) IIRC it can be wrong sometimes
[12:55] <durandal_1707> i want to port it
[12:56] <durandal_1707> so i want to know what is expected result
[13:00] <michaelni> i think these artifacts are expected for stride != width
[13:06] <durandal_1707> michaelni: so its not useful at all?
[13:07] <michaelni> might be usefull if you have a slow box and need it for realtime playback ...
[13:08] <michaelni> but the copy would not be that slow so its probably not that usefull
[13:08] <durandal_1707> michaelni: so what about making port that would do copy and be morae usefull
[13:09] <michaelni> yes if it cant do it without copy & without artofacts it should copy
[13:10] <michaelni> maybe with a user override switch that allows forcing non copy
[13:10] <michaelni> but that would have very few users
[13:10] <michaelni> also see vf_il
[13:24] <durandal_1707> there is no filter that blends to images?
[13:24] <durandal_1707> *two
[13:36] <Compn> durandal_1707 : blend how ?
[13:37] <Compn> durandal_1707 : the stereo3d puts images side by side 
[13:37] <Compn> or pad or that other filter
[13:40] <iive> or interlace, or one above the other, or codes them red/blue .. ;)
[14:16] <cone-778> ffmpeg.git 03Michael Niedermayer 07master:47335126c049: ff_end_tag: assert that the start is aligned.
[14:16] <cone-778> ffmpeg.git 03James Almer 07master:89946c840375: dxva2: Prevent redefinition of _WIN32_WINNT
[16:14] <durandal_1707> ubitux: when you will become live again?
[16:14] <ubitux> around wednesday
[16:15] <durandal_1707> ok, don't drink too much blood
[16:20] <durandal_1707> michaelni: so fil is just "fast" variant of il?
[16:28] <michaelni> yes
[16:28] <durandal_1707> ok, why they exist at all?
[16:30] <michaelni> il/fil allow for example frame based filters to be easily applied to fields
[16:30] <michaelni> probably they have other uses too
[16:31] <durandal_1707> hmm, so i will just port il - could i pick less retarded name?
[16:32] <durandal_1707> and fil will be just special case of il
[16:43] <durandal_1707> michaelni: hmm mp=il=d gives obviously broken output here
[16:55] <durandal_1707> nvm, i deinterleave only luma plane ...
[16:55] <durandal_1707> michaelni: are you ok to relicense this trivial code to LGPL?
[18:14] <michaelni> durandal_1707, the vf_il.c (4289byte) code (the parts i wrote) can be used under LGPL, its just some memcpies anyway
[23:52] <cone-778> ffmpeg.git 03Michael Niedermayer 07master:b8820e7aa338: dsp: make put_no_rnd_pixels_tab 4x4 big
[00:00] --- Sat Feb  9 2013


More information about the Ffmpeg-devel-irc mailing list