burek021 at gmail.com
Sat May 24 02:05:02 CEST 2014
[00:00] <jamrial> haha, i see
[01:06] <jamrial> kurosu: i can send a 3dnow version of vector_clipf to make sure there are no yawns since everybody will be busy laughing :P
[01:07] <iive> I won't.
[01:40] <cone-833> ffmpeg.git 03James Almer 07master:1d36defe94c7: x86/dsputil: port ff_vector_clipf_sse to yasm
[01:41] <Daemon404> michaelni, ping
[01:46] <michaelni> Daemon404, pong
[01:47] <Daemon404> michaelni, http://git.videolan.org/?p=ffmpeg.git;a=commit;h=20fa3fb93d0f3d3eab2b1f63a03168f492fae047
[01:47] <Daemon404> this is causing sigabort on some files
[01:47] <Daemon404> apparently
[01:48] <michaelni> where can i find these files ?
[01:48] <Daemon404> let me see...
[01:51] <Daemon404> PM.
[01:58] <michaelni> and how do i reproduce it with that file ? (the sig would be from the muxer id assume)
[01:59] <Daemon404> its something to do with copying chapters
[01:59] Action: Daemon404 checks
[01:59] <Daemon404> (-map_chapters -1 'fixes' it for me)
[02:02] Action: Daemon404 greps logs
[02:06] <Daemon404> michaelni,
[02:06] <Daemon404> ffmpeg -i a.mov -vn -strict -2 a.mp4
[02:06] <Daemon404> ^
[02:06] <Daemon404> doesnt need to be aac obviously
[04:15] <BBB> kurosu: jamrial: can I please say how happy I am with your yams stuff? it's awesome, thank you
[04:16] <BBB> in general, any line of inline non-inlined asm converted to yasm brings a smile on my face, but when you add sse2/ssse3 and make it 2x as fast, I'm just happy
[04:17] <jamrial> you're welcome :P
[05:10] <cone-402> ffmpeg.git 03Michael Niedermayer 07master:744df0a0319c: avformat/movenc: Check that packet duration is valid in ff_mov_write_packet()
[05:10] <cone-402> ffmpeg.git 03Michael Niedermayer 07master:220df143b962: avformat/utils: Check end time in avpriv_new_chapter()
[05:10] <cone-402> ffmpeg.git 03Michael Niedermayer 07master:64bf8455442b: avformat/mov: Dont blindly trust the stream duration in seting chapter times
[05:11] <michaelni> Daemon404, should be fixed
[06:51] <jamrial> nevcairiel: can you test this patch http://pastie.org/9201161 and see if it fixes the fate failures on msvc 2013 x86_64?
[08:25] <kurosu> jamrial, but can you test that 3dnow patch ? ;)
[08:25] <kurosu> bye
[08:44] <ubitux> why is Don Moir arguying on such a meaningless issue? :D
[08:44] <ubitux> (re ffplay rotated video)
[15:43] <Daemon404> michaelni, cool
[15:45] <cone-556> ffmpeg.git 03James Almer 07master:6a4832caaede: x86/diracdsp: mark all functions as yasm
[17:55] <nevcairiel> jamrial: your patch didnt work
[17:55] <jamrial> wonder why it's failing only with msvc and not mingw64 then...
[17:59] <nevcairiel> how big is len usually? it seemed to have a rather astronomically large value
[18:00] <nevcairiel> i can debug in a bit
[18:01] <jamrial> ok, thanks
[18:14] <nevcairiel> jamrial: the "len" it reads into a register is 0x0000000900000040 ... somehow I think this should just be 0x40 without the 9 in the middle? needs register expansion from int to 64-bit?
[18:16] <nevcairiel> (also, only with your patch applied, otherwise it assigns the wrong arg to len in general)
[18:20] <jamrial> a disass of the old inline version with mingw64 shows a cltq, so i guess it's what you said
[18:21] <nevcairiel> or load it manually from the stack only reading 32-bit, instead of letting the macros generate it
[18:30] <nevcairiel> adding a movsxd lenq, lend fixes the issue in any case
[18:39] <nevcairiel> is that a reliable rule that using any operation on the 32-bit part of a register clears out its upper half?
[18:43] <nevcairiel> (practically, if i change the "shl" in that function to use lend instead of lenq, is it guaranteed to empty the upper half?)
[18:55] <jamrial> no idea
[18:55] <jamrial> does using "shl lend, 2" without the movsxd fix the failure there?
[18:55] <nevcairiel> yea, thats why I ask
[18:55] <jamrial> it at least doesn't break anything here
[18:55] <nevcairiel> on top of your earlier patch, that is
[18:56] <nevcairiel> but yes, it works just fine with that change
[18:56] <jamrial> ok, guess then i'll go with that. better than adding an extra mov
[18:56] <nevcairiel> too bad the cglobal macro doesn't make manual loading of args easier
[18:56] <nevcairiel> then you could just load 32-bit
[18:58] <jamrial> yeah, improving PROLOGUE is marked as TODO in x86inc
[20:45] <cone-205> ffmpeg.git 03Niels Möller 07master:6b7b8585d735: dca: Convert dca_dmixtable to integers
[20:45] <cone-205> ffmpeg.git 03Michael Niedermayer 07master:c867be03f61e: Merge commit '6b7b8585d73548f08b98959df39d80b97677662f'
[20:45] <cone-205> ffmpeg.git 03Michael Niedermayer 07master:4e4dbb99831d: dcadec: fix xxch_dmix_coeff and xxch_dmix_sf after merge
[21:13] <cone-205> ffmpeg.git 03Vittorio Giovara 07master:21f68c2489cb: avcodec: bump version after rotation api
[21:13] <cone-205> ffmpeg.git 03Michael Niedermayer 07master:d062db788012: Merge commit '21f68c2489cba2a1a4a41d0c5c828266e6162800'
[21:21] <cone-205> ffmpeg.git 03Carl Eugen Hoyos 07master:27d460fe9821: Refuse to mux H.264 with fourcc H264 into avi without startcode.
[21:21] <cone-205> ffmpeg.git 03Michael Niedermayer 07master:9340ced47b5b: Merge remote-tracking branch 'cehoyos/master'
[22:09] <cone-205> ffmpeg.git 03Michael Niedermayer 07master:6da06ef6bbc0: avcodec/dcadec: make dca_dmix_code() 650% faster
[22:12] <nevcairiel> are compilers smart enough to replace pow(2, -15) with a constant? that seems kinda too smart for them!
[22:13] <thardin> check
[22:13] <nevcairiel> looks like gcc does it
[22:15] <wm4> why not use a fp constant
[22:16] <wm4> 0x1p-15 or so? (not sure if the - goes there)
[22:19] <nevcairiel> if it would be a power of ten, that might be easier, but powers of two dont have an easy syntax, so it would be something like 3.0517578125e-5
[22:20] <nevcairiel> (or in hex, if you prefer)
[22:20] <wm4> nevcairiel: that is a power of 2
[22:21] <nevcairiel> the hex syntax is power of 2, not 10 like the decimal syntax?
[22:22] <wm4> yes, it's power of 2
[22:22] <wm4> the advantage of hex floats are that they can represent floats exactly
[22:23] <nevcairiel> its a C99 feature, i wonder if msvc supports that :D
[22:24] <cone-205> ffmpeg.git 03Michael Niedermayer 07master:9ccb5455ca8c: avcodec/dcadec: use a constant instead of assuming every compiler can optimize pow(2,-15)
[22:24] <nevcairiel> guess we'll find out!
[22:25] <nevcairiel> at the very least i expect older msvc to break
[22:25] <michaelni> if something breaks ill replace it by a dezimal constant
[22:26] <michaelni> that is please ping me if it breaks somewhere
[22:27] <nevcairiel> libavcodec/dcadec.c(2110) : error C2059: syntax error : 'bad suffix on number'
[22:27] <nevcairiel> well, no luck!
[22:28] <wm4> it might actually be in C++11, but yeah, even if MSVC adds it to C, it probably will break with older versions
[22:31] <nevcairiel> msvc also doesn't optimize it out, the pow call remains if i put it back
[22:33] <nevcairiel> anyhow michaelni, obligatory ping, it does indeed break msvc, even the latest version
[22:34] <ubitux> michaelni: don't forget to add a com along the constant though
[22:45] <cone-205> ffmpeg.git 03Michael Niedermayer 07master:6b88f22e8953: avcodec/dcadec: dont use hex float, msvc doesnt support it
[22:52] <ubitux> so libav is going into the mountain... to reindent code as fast as possible?
[23:03] <ubitux> btw, should i apply the ffplay rotation patchset?
[23:06] <cone-205> ffmpeg.git 03James Almer 07master:518cbf9b4a0a: x86/dsputil: fix VECTOR_CLIP_INT32 macro
[23:07] <michaelni> ubitux, either that or send cuz a mail
[23:08] <michaelni> cuS I meant
[23:08] <ubitux> too bad he isn't on irc
[23:08] <ubitux> i guess i'll ping
[23:08] <michaelni> yes, he probably missed the patch or is busy
[23:08] <Plorkyeran_> +wm4> it might actually be in C++11 <-- it isn't
[23:09] <Plorkyeran_> c++11 only imported the c99 standard library
[23:09] <Plorkyeran_> and not the language changes
[23:11] <wm4> I didn't look, but g++ printed a wanring in c++98 mode, but not c++11 mode
[23:30] <cone-205> ffmpeg.git 03James Almer 07master:584327f22fbf: x86/dsputil: fix argument declaration in vector_clipf
[23:30] <cone-205> ffmpeg.git 03Clément BSsch 07master:fa972dcf1a36: avutil/display: use hypot().
[23:30] <cone-205> ffmpeg.git 03Clément BSsch 07master:e63a17bfdc99: avformat/mov: use av_display_rotation_get() for rotate metadata.
[23:30] <cone-205> ffmpeg.git 03Clément BSsch 07master:08c51e12b1c3: ffplay: support rotated video.
[23:34] <Timothy_Gu> michaelni: did you see Diego's patch to Libav about huffyuvdsp
[23:35] <Timothy_Gu> Is that the same thing as losslessdsp?
[23:46] <michaelni> i think its not exactly the same
[00:00] --- Sat May 24 2014
More information about the Ffmpeg-devel-irc