[Ffmpeg-devel-irc] ffmpeg-devel.log.20140518

burek burek021 at gmail.com
Mon May 19 02:05:02 CEST 2014

[00:12] <kurosu> one reason could be the differing calling conventions, as I'm assuming the 5th arg is passed through the stack
[00:13] <kurosu> jamrial, might not be the case for x86_64, so an extra reg is needed there
[00:13] <kurosu> worth than that
[00:13] <kurosu> *worse
[00:14] <kurosu> "System V AMD64" passes the 6 first args in regs
[00:14] <kurosu> so I'm effectively overwriting them, and you can't save that many args in that case
[00:15] <kurosu> on the other hand, one does not have to handle win32/x86_32 so...
[00:46] <BBB> on win64, first 4 args are regs
[00:47] <BBB> on unix64 (including mac), it's 6
[00:47] <BBB> on 32bit, everything is stack
[00:48] <jamrial> kurosu: well, you effectively brought the function down to less than 8 xmm/gp regs, so it can be used on x86 builds as well now
[04:41] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:0323a55cd898: avformat/rtpenc: switch to AVFMT_FLAG_BITEXACT
[04:41] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:d524d424ec98: avformat/nutenc: switch to AVFMT_FLAG_BITEXACT
[04:41] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:fab50573bb92: avformat/mp3enc: switch to AVFMT_FLAG_BITEXACT
[04:41] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:0421409c5216: avformat/matroskaenc: switch to AVFMT_FLAG_BITEXACT
[04:41] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:9646ea63df78: avformat/mmf: switch to AVFMT_FLAG_BITEXACT
[04:41] <cone-266> ffmpeg.git 03Michael Niedermayer 07master:17c8a069167e: avformat/framehash: switch to AVFMT_FLAG_BITEXACT
[04:42] Action: michaelni hugs cone-266 
[04:58] <Zeranoe> Is this a known bug? http://ffmpeg.zeranoe.com/forum/viewtopic.php?f=7&t=1981
[09:20] <mraulet> kurosu, jamrial, https://ffmpeg.org/pipermail/ffmpeg-devel/2014-May/157734.html works for me
[09:30] <kurosu> jamrial, yes, but you'd have to disentangle (?) the assignment I think
[09:30] <kurosu> not really worth it because this function is seldom used (fades when the proper tool is on)
[09:31] <kurosu> mraulet, ok, couldn't verify, so glad to know
[09:35] <kurosu> anyway, to make it 32bits, the test must be changes to %if WIN64 || ARCH_X86_32
[09:35] <kurosu> don't know if there's a shorthand nowadays
[09:59] <kurosu>     pmulhw            m6, m0, m2
[09:59] <kurosu>     pmullw            m0, m2
[09:59] <kurosu>     punpckhwd         m1, m0, m6
[09:59] <kurosu>     punpcklwd         m0, m6
[09:59] <kurosu>     paddd             m0, m5
[09:59] <kurosu>     paddd             m1, m5
[10:00] <kurosu> isn't that just pmaddwd but with samples not interleaved ?
[10:00] <kurosu> (ignore the paddd actually)
[10:04] <Skyler_> is that trying to work around a lack of pmulld or something?
[10:09] <kurosu> Skyler_, ah right
[10:09] <kurosu> probably
[10:09] <kurosu> the funny thing is that the function is sse4
[10:10] <kurosu> this code really feels like it has been made for simplest reusability over thoroughness
[10:11] <kurosu> eg loading coeffs into specified regs, and too bad if some of the previous regs aren't used
[10:13] <Skyler_> pmaddwd is probably what you want there though
[10:13] <Skyler_> since it's 16x16->32, not 32x32->32
[10:14] <kurosu> well, I'm firstly trying to reduce the reg usage, removing the aforementioned holes
[10:15] <kurosu> not sure I'll get all the way here, because there's quite some work to do this for every function
[10:15] <kurosu> *all the way there
[10:42] <kurosu> michaelni / ...: please hold on, I'm squashing the next weighted pred reg realloc patch into the current one
[11:46] <j-b> 'morning
[12:00] <kurosu> oh wow, my macro loads coeffs into m14,m15 - core loops uses m0-m4 and coeffs - let's declare using 16 regs!
[12:03] <J_Darnley> You should use them if you have them!
[12:06] <kurosu> I fear that plepere develops under linux and never sees the dark side of such a code :(
[12:59] <BBB> don't you want pmaddubsw?
[13:00] <BBB> I mean, for 8bit at least
[13:00] <cone-397> ffmpeg.git 03Anton Khirnov 07master:e1b66778b6ee: lavc: remove the locking code in avcodec_close()
[13:00] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:ff17d8b56ec8: Merge commit 'e1b66778b6ee82a192b5895e23c4e135f7269326'
[13:09] <cone-397> ffmpeg.git 03Anton Khirnov 07master:fd056029f45a: lavc: add avcodec_free_context().
[13:09] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:68c05185e229: Merge commit 'fd056029f45a9f6d213d9fce8165632042511d4f'
[13:15] <cone-397> ffmpeg.git 03Anton Khirnov 07master:d256ed78ffe2: 4xm: allocate extradata properly.
[13:15] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:6a510909242a: Merge commit 'd256ed78ffe202a4dcc8d625becffc716bfa3977'
[13:25] <cone-397> ffmpeg.git 03Anton Khirnov 07master:b136564909ad: a64: remove unneeded struct qualifier
[13:25] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:c7c11058abb7: Merge commit 'b136564909adec2043be4c94ade08b67ad443124'
[13:31] <cone-397> ffmpeg.git 03Anton Khirnov 07master:3ef6c5264b25: a64: check that extradata exists before reading from it
[13:31] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:47a721348a5f: Merge commit '3ef6c5264b2590781b4ed556443ff49709dd45fb'
[13:42] <cone-397> ffmpeg.git 03Anton Khirnov 07master:e4dc1000d7bb: yuv4mpeg: split the demuxer and muxer into separate files
[13:42] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:56d3cd1455ea: Merge commit 'e4dc1000d7bbbcb5b45cf9849fc5315f19578e37'
[13:54] <J_Darnley> I wonder what benefit that ^^ is supposed to bring.  I didn't think y4m was so complicated that you needed to split
[14:08] <BBB> J_Darnley: smaller binary for demuxed-only builds, likely
[14:08] <cone-397> ffmpeg.git 03Anton Khirnov 07master:b513bf6f69e2: yuv4mpegdec: do not set coded_frame properties
[14:08] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:213e82b55c28: Merge commit 'b513bf6f69e26e724de6d5dca642c3582dcd0517'
[14:11] <cone-397> ffmpeg.git 03Anton Khirnov 07master:ab3fdaa7131e: yop: only set extradata size after it has been successfully allocated
[14:11] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:4fac4447d153: Merge commit 'ab3fdaa7131e2fd8e33006daea25a282724c01f0'
[14:19] <cone-397> ffmpeg.git 03Anton Khirnov 07master:c3311d472a75: avienc: sanitize variable naming in write_header()
[14:19] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:4f32efdff007: Merge commit 'c3311d472a7528c67f76d0d061704ae70a99b32e'
[14:28] <cone-397> ffmpeg.git 03Anton Khirnov 07master:7a5f4f685304: lavf: extend avg_frame_rate documentation.
[14:28] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:8636b11bdc6e: Merge commit '7a5f4f6853047eea74a6443d73f4d74d52b2799f'
[14:51] <cone-397> ffmpeg.git 03Anton Khirnov 07master:93afb6c98df8: avconv: set output avg_frame_rate when known
[14:51] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:8334850a9985: Merge commit '93afb6c98df876b15e3d911a9450ad55f92080ce'
[14:58] <cone-397> ffmpeg.git 03Anton Khirnov 07master:79f2c426fde6: dv: do not set codec timebase
[14:58] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:b06741d26794: Merge commit '79f2c426fde6e71c40b29504112d0528b85be623'
[16:05] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:bf718bdf12be: avcodec/crystalhd: allocate padding for extradata
[16:05] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:947ed8f292f5: avcodec/ffv1enc: allocate padding for extradata
[16:05] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:5441eb460cb4: avcodec/movtextenc: allocate padding for extradata
[16:05] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:6e8fe448154e: avcodec/wmv2enc: allocate padding for extradata and check malloc failure
[16:15] <cone-397> ffmpeg.git 03Olivier Langlois 07master:0a150670476b: lavd: Use av_gettime_relative()
[16:30] <cone-397> ffmpeg.git 03Christophe Gisquet 07master:f1793fe9cdf1: x86: hevc_mc: specify coefficients registers
[16:55] <kurosu> BBB: some intermediate computations are specified to be on 32 bits
[17:37] <BBB> kurosu: but do they use the upper 16 bits?
[17:38] <kurosu> well, the input is on 16bits, but I haven't checked what the specs says on that
[17:38] <kurosu> plus I don't really intend to
[17:39] <kurosu> plus, this hevc thing is a hobby, it's not like my regular job is to implement hevc ;)
[17:40] <BBB> are you sure about that :-p
[17:40] <BBB> oh, right, implement
[17:40] <BBB> hm...
[17:40] <BBB> ok
[18:25] <ubitux> Daemon404: so, no comment on writing_filters.txt, i thought you were complaining about that a while ago
[18:25] <ubitux> ?
[19:08] <Daemon404> ubitux, i have not read it
[19:08] <Daemon404> just got home from the beach
[19:08] <Daemon404> making dinner, so later.
[20:14] <kierank> vps_max_num_reorder_pics out of range: 7
[20:14] <kierank> for hevc
[20:25] <nevcairiel> the check in the code appears correct as the spec describes it
[20:25] <cone-397> ffmpeg.git 03Anshul 07master:73a60633143b: ffprobe: fix crash happening because of new streams occuring
[21:03] <cone-397> ffmpeg.git 03Anton Khirnov 07master:33082af7560a: doc/APIchanges: fill in missing hashes and dates
[21:03] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:6eefc578a5c4: Merge commit '33082af7560a18179b52734665ef66f812902d33'
[21:05] <cone-397> ffmpeg.git 03Anton Khirnov 07master:968a62a9d9e3: doc/APIchanges: mark the release 10 branch point
[21:05] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:a6ba5184a5e2: Merge commit '968a62a9d9e3a785ce1f79b80facfe09c6c14544'
[21:11] <cone-397> ffmpeg.git 03Anton Khirnov 07master:d246231e4714: wavenc: use codec descriptors to get the codec name
[21:11] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:c01bcce4959a: Merge commit 'd246231e4714119faac6c7acd881d3b687bb8b11'
[22:21] <Daemon404> ^ because lags for ffmpeg had to include an fp emulator to be portable
[22:22] <kierank> EBROKENBYDESIGN
[22:23] <cone-397> ffmpeg.git 03Anton Khirnov 07master:6656370b8583: avconv: set the "encoder" tag when transcoding
[22:23] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:1e49439f04f4: Merge commit '6656370b858329ca07a60a2de954d5e90daa0206'
[22:23] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:bbee02ed9c18: ffmpeg: Skip writing the version when -flags bitexact is used
[23:22] <cone-397> ffmpeg.git 03Anton Khirnov 07master:0ba5299a805e: movenc: use the "encoder" metadata tag to write stsd Compressorname
[23:22] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:5771efde87c1: Merge commit '0ba5299a805e9ccaef1a757381fc2ada4d54b8a1'
[23:22] <cone-397> ffmpeg.git 03Michael Niedermayer 07master:0c152fe916e3: ffmpeg: prefix encoder with "Lavc " in bitexact mode
[23:48] <cone-397> ffmpeg.git 03James Almer 07master:057ebf122276: x86/hevc_deblock: remove some duplicated instructions
[00:00] --- Mon May 19 2014

More information about the Ffmpeg-devel-irc mailing list