[Ffmpeg-devel-irc] ffmpeg-devel.log.20150915

[burek]

[00:22:44 CEST] <cone-326> ffmpeg 03Rodger Combs 07master:8e924629ebde: lavu/avstring: switch AV_ESCAPE_FLAGs to shift-based formatting
[01:51:21 CEST] <jcalve> Hello, I'm searching for a long lost friend, I believe he could have been here recently with "yumbox" as nick, if someone knows where I could contact him I'd appreciate a pm with any info, thanks
[03:55:16 CEST] <andrewrk> if I call avformat_open_input without setting AVFormatContext `pb` field, this file opens successfully; however if I set a custom avio, it says 'Invalid data found when processing input'
[03:55:39 CEST] <andrewrk> how can this be?
[03:56:04 CEST] <andrewrk> is the `whence` value of avio->seek different than the one from fseek?
[03:59:12 CEST] <BBB> no its the same
[03:59:21 CEST] <BBB> what do you set pb to?
[03:59:50 CEST] <BBB> its used in various places (it being setting pb) so it should work
[04:00:27 CEST] <andrewrk> read packet: return fread(buf, 1, buf_size, userdata->file)
[04:00:39 CEST] <andrewrk> write packet: return fwrite(buf, 1, buf_size, userdata->file)
[04:00:57 CEST] <andrewrk> seek: return fseek(userdata->file, offset, whence)
[04:01:03 CEST] <BBB> did you set flags?
[04:01:07 CEST] <BBB> did you open it already?
[04:01:18 CEST] <andrewrk> I'll put together a minimal test case
[04:01:24 CEST] <BBB> ok
[04:18:07 CEST] <andrewrk> BBB, https://gist.github.com/andrewrk/ad619e93727a63d676b3
[04:18:25 CEST] <andrewrk> I can supply a test file if you like
[04:18:49 CEST] <andrewrk> it works fine on some files, fails on others
[04:18:54 CEST] <andrewrk> note line 74
[04:28:17 CEST] <cone-269> ffmpeg 03Michael Niedermayer 07master:3e2ef00394b8: avformat/avidec: Workaround broken initial frame
[04:28:17 CEST] <cone-269> ffmpeg 03Michael Niedermayer 07master:59c1fec9fb94: avformat/avidec: Change (last)pos to 64bit
[04:29:29 CEST] <andrewrk> ok I think I figured it out. I should return ftell, not fseek return value
[04:29:39 CEST] <andrewrk> some docs on that would be really helpful
[04:29:48 CEST] <andrewrk> maybe I'll send a patch
[04:30:01 CEST] <genio> michaelni: thanks for the fix.  I'll test it out tomorrow
[04:53:51 CEST] <andrewrk> ffmpeg is passing 65536 for the 'whence' value
[04:53:57 CEST] <andrewrk> fseek says that is EINVAL
[04:58:37 CEST] <andrewrk> ok I just found the optional whence values
[06:34:31 CEST] <kinnera_> I would like to setup code base for contributing to ffmpeg.Can someone help me out?
[06:35:08 CEST] <kinnera> Even i need the same help
[08:36:31 CEST] <jasom> Sorry if this is the wrong place to ask, but I'm trying to debug ffmpeg getting a bogus PTS value, and was wondering if there is a way to get it to dump the PES header?
[08:37:02 CEST] <jasom> This is in a program stream, if it matters
[09:52:57 CEST] <cone-127> ffmpeg 03Paul B Mahol 07master:92bfc3cd0363: avcodec/dxv: remove code that is never reached
[09:59:42 CEST] <cone-127> ffmpeg 03Paul B Mahol 07master:1c853d5727e2: avfilter/avf_showcqt: draw text optionally
[10:11:05 CEST] <durandal_170> why we have mips code at all, its slowing down development
[10:15:08 CEST] <martijnb> still pretty relevant for stb's no?
[10:16:55 CEST] <martijnb> I'm using ffmpeg predominantly on arm, but it wouldn't shock me if next month they ask me to do something on mips
[10:18:13 CEST] <martijnb> I get it can be a hassle, with x86 probably being by far the biggest target audience
[10:19:21 CEST] <martijnb> but x86 is never going to cut it for stb's or integrated systems
[10:19:35 CEST] <martijnb> sucks too much power, generates too much heat
[10:20:25 CEST] <nevcairiel> andrewrk: another thing missing in your paste is handling the filesize call to the seek callback, ie. if whence == AVSEEK_SIZE, then return the filesize
[10:53:13 CEST] <martijnb> clear
[10:53:19 CEST] <martijnb> err.. oops
[11:32:58 CEST] <cone-127> ffmpeg 03Paul B Mahol 07master:f2b1df735ed0: avfilter/avf_showfreqs: add lanczos and gauss windowing functions
[12:09:11 CEST] <cone-127> ffmpeg 03Paul B Mahol 07master:b0f985661678: avfilter: add stereowiden filter
[12:11:35 CEST] Action: durandal_170 going to add yet another stereo enhancer
[12:47:07 CEST] <BBB> andrewrk: I dont know if someone else helped yet, but check (with the one provided by libavformat, as well as your own) avio->seekable vs. ic->pb->seekable, they may have different values causing it to not work
[13:02:11 CEST] <cone-127> ffmpeg 03Ganesh Ajjanagadde 07master:e76be9cd450e: avfilter/vf_xbr: use the name 's' for the pointer to the private context
[13:45:59 CEST] <cone-127> ffmpeg 03Eloi BAIL 07master:bc3ea394154b: rtpdec: inform jitter buffer size
[13:46:00 CEST] <cone-127> ffmpeg 03Eloi BAIL 07master:0edf6c8b2a97: rtpdec: add a trace when jitter buffer is full
[14:13:16 CEST] <nevcairiel> kurosu: so the openhevc guys actually replied, one of them OKed the patch and the other one wanted to test a bit more
[14:24:10 CEST] <RiCON> hey, d3d11va is back
[15:06:32 CEST] <ubitux> durandal_170: f2b1df735 missing doc update?
[15:06:57 CEST] <durandal_170> oops
[15:31:10 CEST] <cone-127> ffmpeg 03Ronald S. Bultje 07master:a0d8a81075c0: vp9: switch min_tile_cols location so it shifts up instead of down.
[16:23:04 CEST] <BBB> is ganesh on irc?
[17:13:03 CEST] <cone-127> ffmpeg 03Ganesh Ajjanagadde 07master:b26a45c2c33b: avfilter/vf_vidstabdetect: use the name 's' for the pointer to the private context
[17:23:57 CEST] <cone-127> ffmpeg 03Paul B Mahol 07master:28e023377cfe: doc/filters: mention newly added windowing functions
[18:17:44 CEST] <cone-127> ffmpeg 03Michael Niedermayer 07master:7404f3bdb90e: lavc: Switch bitrate to 64bit unless compatibility with avconv was requested.
[18:17:45 CEST] <cone-127> ffmpeg 03Carl Eugen Hoyos 07master:c311713ca99c: lavf: Switch bitrate to 64bit unless compatibility with avconv was requested.
[18:17:46 CEST] <cone-127> ffmpeg 03Carl Eugen Hoyos 07master:b290972b8ff8: lavf: Switch probesize and max_analyze_duration to 64bit.
[18:23:25 CEST] <nevcairiel> way to push the same commit twice =P
[18:23:41 CEST] <nevcairiel> now APIchanges has dupes :)
[18:30:10 CEST] <durandal_170> when abi mess will finally be decided?
[18:31:08 CEST] <BBB> Gramner: ping
[18:32:37 CEST] <BBB> Gramner: why would http://pastebin.com/eFd8nhdL give essentially random output and fails? (the simd is fine, rly[tm] :) ) I think I misunderstand checkasm
[18:35:33 CEST] <nevcairiel> you give it random input, what do you expect
[18:37:07 CEST] <BBB> ?
[18:37:37 CEST] <BBB> the ref and new call get the same src pointer
[18:44:13 CEST] <wm4> durandal_170: the ABI mess should never have been a question in a sane world
[18:44:51 CEST] <Daemon404> it's almost as if an api thar requires accessing struct members isnt stable
[18:44:54 CEST] <Daemon404> that*
[18:45:44 CEST] <durandal_170> ubitux: will you add support for gimp curves?
[18:47:52 CEST] <cone-127> ffmpeg 03Paul B Mahol 07master:aaf4e703f6d9: avfilter/af_stereowiden: fix 2nd argument for av_calloc
[18:47:54 CEST] <wm4> Daemon404: the issue at hand is being ABI compatible with a hostile fork that has diverged for over 4 years
[18:48:02 CEST] <wm4> if you can't see the sheer insanity, they've got you
[18:48:03 CEST] <Daemon404> oh that
[18:48:17 CEST] <Daemon404> nobody seriously expects ABI compatability, and i would be shocked if even with that hack enabled, it were true
[18:49:06 CEST] <wm4> see flames on ML
[18:49:21 CEST] <Daemon404> i'd rather not
[18:49:25 CEST] <Daemon404> it's not going to be constructive
[18:49:30 CEST] <Daemon404> autists gonna autist
[18:49:36 CEST] <Daemon404> (are those even words?)
[18:59:57 CEST] <Gramner> BBB: did you printf both dst arrays to see what exactly differs?
[19:11:59 CEST] <BBB> Gramner: yes, theyre completely different
[19:12:01 CEST] <BBB> not a single similarity
[19:13:40 CEST] <BBB> Gramner: Im thinking that Im somehow associating the wrong ref with the wrong new func, if that makes any sense
[19:16:37 CEST] <Gramner> BBB: somewhat busy right now, but I can take a look later
[19:22:41 CEST] <cone-127> ffmpeg 03Ganesh Ajjanagadde 07master:f7b53c952778: avfilter/f_sendcmd: use the name 's' for the pointer to the private context
[19:26:36 CEST] <BBB> hm so now people dont like rename to s
[19:26:37 CEST] <BBB> ...
[19:27:49 CEST] <nevcairiel> i have no idea why someone would bother to rename shit in existing code anyway
[19:27:53 CEST] <nevcairiel> its like K&R just weirder
[19:28:54 CEST] <BBB> I dont care if other people do it
[19:28:57 CEST] <BBB> as long as I dont have to do it
[19:29:05 CEST] <BBB> Im fine with k&r, honestly
[19:29:29 CEST] <nevcairiel> I just format my code right the first time around :D
[19:29:31 CEST] <BBB> just dont ask me to obey some obscure detail randomly invented by some ffmpeg dev and then scrubbed under the k&r blanket
[19:29:49 CEST] <nevcairiel> I also hate these "dont include indent in your patch, it makes it hard to read" shit, just get a diff viewer which ignores whitespace
[19:30:12 CEST] <BBB> I agree
[19:30:20 CEST] <BBB> I think we stopped requiring that already for most patches
[19:30:33 CEST] <nevcairiel> carl still recommends it every chance he gets =p
[19:31:08 CEST] <nevcairiel> most annoyingly, people usually forget to send the re-indent patch after
[19:31:12 CEST] <nevcairiel> and you end up with a mess
[19:31:12 CEST] <nevcairiel> :D
[19:32:52 CEST] <iive> i've actually though of implementing 4'th patch line mode, one that indents whitespace. aka "+- " and "W"
[19:33:18 CEST] <iive> or "*"
[19:38:18 CEST] <kurosu> nevcairiel, about the openhevc guys, was any of the 2 a guy called Wassim? Did mraulet reply?
[19:39:02 CEST] <kierank> I met mraulet
[19:39:03 CEST] <nevcairiel> mraulet replied, he was the one that OKed it, the other one was erwan raffin
[19:39:04 CEST] <BBB> iive: diff -uw?
[19:39:32 CEST] <ubitux> durandal_170: no plan for gimp curves, patch welcome
[19:39:45 CEST] <iive> BBB: the resulting patch ignores the whitespice entirely.
[19:40:07 CEST] <BBB> isnt that what you want?
[19:40:15 CEST] <iive> BBB: no
[19:40:20 CEST] <kurosu> nevcairiel, the second name rings a bell, but I never met him
[19:40:33 CEST] <kurosu> kierank, at ibc I guess, given his new employer
[19:40:38 CEST] <kierank> yes
[19:41:13 CEST] <iive> BBB: with -w you ignore whitespace and the resulting patch marks the lines with " " aka, not changed.
[19:41:56 CEST] <iive> my proposal would not use "+-" for cosmetics, it would issue one line starting with new marker.
[19:42:49 CEST] <iive> for that line patch would work as if you have -l option.
[20:14:29 CEST] <cone-127> ffmpeg 03Lou Logan 07master:a62d4a01e87e: README.md: avoid Github pull requests
[20:15:21 CEST] <llogan> to be read by 0% of github pull requesters
[20:15:28 CEST] <philipl> good times.
[20:24:12 CEST] <BBB> Gramner: oh n/m I think I found it
[20:27:14 CEST] <BBB> yes it works now
[20:28:36 CEST] <Gramner> BBB: what was the problem?
[20:29:08 CEST] <BBB> we have fullpel and subpel functions, and you specify the mx/my for subpel functions
[20:29:23 CEST] <BBB> the c functions understand when subpel is called with m<dir>=0
[20:29:25 CEST] <BBB> the simd doesnt
[20:29:51 CEST] <BBB> (this wouldnt ever happen in the real decoder, but I did m<dir>=rnd()&15 instead of m<dir>=(rnd()%14)+1)
[20:30:06 CEST] <BBB> so it was calling a subpel function with m<dir>=0
[20:30:46 CEST] <BBB> (tldr; checkasm was fine, I was buggy)
[20:31:55 CEST] <BBB> so is the goal to write checkasm coverage for everything?
[20:32:11 CEST] <BBB> I mean, right now the coverage is & kinda & low :)
[20:32:25 CEST] <Gramner> well, gotta start somewhere
[20:32:52 CEST] <Gramner> ideally every asm function would have a test, yes
[20:32:56 CEST] <BBB> ok cool
[20:33:07 CEST] <BBB> Im thinking Ill do itxfm_add or ipred next
[20:33:13 CEST] <BBB> ipred is easy
[20:33:16 CEST] <BBB> itxfm_add will suck ass
[20:33:33 CEST] <Gramner> could also be a good target for gsoc and similar stuff as well I guess
[20:33:43 CEST] <BBB> depends on the type of function
[20:33:55 CEST] <BBB> I mean, writing a good test for itxfm_add in any codec is somewhat hard
[20:34:13 CEST] <BBB> because we typically have dc-only optimizations, and you can easily have overflows that dont happen on real input
[20:34:21 CEST] <BBB> so the input cannot be entirely random
[20:34:30 CEST] <BBB> it requires some understanding of a transform
[20:34:39 CEST] <Gramner> yeah, you need to know the limits of the input. and that's not always obvious
[20:34:48 CEST] <BBB> loopfilter is another one, random isnt great because you want to test different filter states
[20:35:03 CEST] <Gramner> also explicitly testing corner cases is useful in some cases
[20:35:10 CEST] <BBB> right, exactly
[20:35:28 CEST] <Gramner> since you can have failure that only happens once in a million in random inputs
[20:36:28 CEST] <cone-127> ffmpeg 03Ganesh Ajjanagadde 07master:0f711126b83e: avfilter/vsrc_cellauto: use the name 's' for the pointer to the private context
[20:39:33 CEST] <BBB> well at least I can do ipred I guess
[20:42:30 CEST] <BBB> Gramner: is the intention to have one file per module, or one file per function, or ...?
[20:42:44 CEST] <BBB> Gramner: like, would I start a new file for vp9 intra pred? or put it in the same file as mc?
[20:43:06 CEST] <nevcairiel> use common sense
[20:43:07 CEST] <nevcairiel> :)
[20:43:21 CEST] <Gramner> ^ basically, yes
[20:43:21 CEST] <nevcairiel> h264 split qpel and pred
[20:44:01 CEST] <BBB> Ill split it for now
[20:44:05 CEST] <BBB> if it gets crowded we can merge
[20:44:07 CEST] <Gramner> maybe use whatever naming standard the files in libavcodec/ uses?
[20:44:21 CEST] <BBB> hm.
[20:44:22 CEST] <BBB> ..
[20:47:42 CEST] <cone-127> ffmpeg 03Michael Niedermayer 07master:72db5e96fc56: avcodec/ffv1enc: Fix error message when the requested version does not support the requested features
[20:49:09 CEST] <BBB> I guess this is better then
[20:49:12 CEST] <jamrial> Gramner: did you see my email about the v210 checkasm problem with gcc asan?
[20:49:21 CEST] <BBB> oh I saw that too
[20:49:30 CEST] <Gramner> jamrial: yes, but I haven't had time to look into it
[20:49:40 CEST] <jamrial> ah ok
[20:52:06 CEST] <BBB> jamrial: you asked for vp9 checkasm coverage btw, so please review :-p
[20:56:29 CEST] <jamrial> BBB: i was compiling it just now :p
[20:59:11 CEST] <cone-127> ffmpeg 03Carl Eugen Hoyos 07master:040ddd27bfca: lavc/utils: Correctly show bit_rate >INT_MAX.
[21:06:35 CEST] <cone-127> ffmpeg 03Paul B Mahol 07master:eeaf19cee981: avfilter/vf_stereo3d: simplify blank calculations and check height
[21:09:02 CEST] <BBB> more tests! \o/
[21:09:13 CEST] <BBB> (admittedly, they were easy)
[21:10:24 CEST] <jamrial> you based this patch on the old version of the first
[21:10:39 CEST] <jamrial> vp9.c :p
[21:11:37 CEST] <BBB> new
[21:11:38 CEST] <BBB> not old
[21:11:49 CEST] <BBB> we decided to go with names identical to names in libavcodec/
[21:11:50 CEST] <BBB> so vp9.c
[21:12:00 CEST] <BBB> and then multiple tests inside that one file
[21:12:12 CEST] <BBB> <Gramner> maybe use whatever naming standard the files in libavcodec/ uses?
[21:12:31 CEST] <jamrial> oh right, mixed them up
[21:15:29 CEST] <BBB> Gramner: I fixed the checkasm prefix and const
[21:15:46 CEST] <BBB> Gramner: the op=1-only is something asan/valgrind would pick up, so its kind of nice maybe
[21:15:51 CEST] <BBB> not that I care a lot obviously
[21:16:29 CEST] <BBB> (and checkasm prefix/const also fixed in second patch, plus the buffer was too big - 64x64 instead of 32x32)
[21:28:54 CEST] <BBB> (so current versions https://github.com/rbultje/ffmpeg/commits/master, not re-sending to ML yet)
[21:37:34 CEST] <Gramner> BBB: no need for the overly complex type for the ipred variable, just use void*. Or skip the variable altogether and simply pass the rvalue of the assignment directly to check_func() (the latter works for the first patch as well)
[21:38:06 CEST] <BBB> hm true
[21:40:12 CEST] <BBB> new version pushed
[21:40:12 CEST] <jamrial> BBB: is it worth benching all the different 8tap functions? afaik the only difference between them are the filter values, so maybe you could try to run bench_new() only for the first, 8tap_smooth i think
[21:40:29 CEST] <BBB> smooth/sharp/regular are the same
[21:40:36 CEST] <BBB> as are all fullpel (dx=0&&dy==0) ones
[21:40:48 CEST] <BBB> but bilin is different
[21:41:02 CEST] <BBB> (we just dont have simd for it since I dont think there is any encoder that can create files for it)
[21:41:39 CEST] <BBB> Daemon404: tsk tsk tsk
[21:41:44 CEST] <Daemon404> ?
[21:41:57 CEST] <BBB> you have to learn to pick your fights :-p
[21:42:04 CEST] <Daemon404> context?
[21:42:08 CEST] <BBB> x264dev
[21:42:13 CEST] <Daemon404> o ok
[21:42:26 CEST] <Daemon404> no fights, only lulz
[21:43:21 CEST] <j-b> for the lulz
[21:47:22 CEST] <jamrial> BBB: yeah, which is why benching them all is not necessary. testing them is obviously ok, but you could disable running the benchmark for the latter two by checking filter_names[filter] == 0 or similar before calling bench_new()
[21:47:25 CEST] <jamrial> not sure how feasible is that
[21:47:48 CEST] <BBB> ok
[21:47:53 CEST] <jamrial> i say this because after the vp9mc patch running checkasm --bench is like three times slower than before
[21:50:29 CEST] <Gramner> BBB: the whole step thingy in the ipred randomize_buffers() seems kind of weird imo
[21:50:45 CEST] <BBB> Gramner: it was just so I wouldnt need to write long lines
[21:51:09 CEST] <BBB> else you get for (k = -1 * SIZEOF_PIXEL; k < .. * SIZEOF_PIXEL; k += 4)
[21:51:15 CEST] <BBB> it was getting long
[21:52:08 CEST] <BBB> jamrial: https://github.com/rbultje/ffmpeg/commit/c689368187d3bc40d9d50d77e5d82333f0d840c1 ?
[21:52:12 CEST] <BBB> (I can squash)
[21:54:09 CEST] <Gramner> what about the second half? it's basically 4 / SIZEOF_PIXEL * SIZEOF_PIXEL
[21:54:49 CEST] <Gramner> also the top half is unaligned due to the -1, AV_WN32A is aligned only i think
[21:56:22 CEST] <BBB> hm...
[21:56:23 CEST] <BBB> ok
[22:00:15 CEST] <andrewrk> nevcairiel, thank you. After finding out about the extra whence API, my issues are resolved.
[22:00:38 CEST] <andrewrk> the docs for the seek callback function should probably mention that
[22:01:21 CEST] <BBB> Gramner: fixed
[22:02:04 CEST] <jamrial> BBB: with that change the bilin 8bpp functions are not being benchmarked anymore
[22:02:40 CEST] <BBB> ?
[22:02:43 CEST] <BBB> they should
[22:02:57 CEST] <BBB> it only skips if filter == 12
[22:02:59 CEST] <BBB> er...
[22:03:04 CEST] <BBB> filter == 3 && bit_depth == 12
[22:03:34 CEST] <jamrial> hold on
[22:03:44 CEST] <BBB> it probably doesnt bench because theres no simd
[22:03:50 CEST] <BBB> (there was never any simd)
[22:06:27 CEST] <jamrial> without that change i get http://pastebin.com/cUuZDNZ6 but with it those are gone
[22:07:27 CEST] <BBB> they have no subpel suffix
[22:07:30 CEST] <BBB> so theyre fullpel only
[22:07:33 CEST] <BBB> I can rename that
[22:07:34 CEST] <BBB> sec
[22:11:46 CEST] <BBB> jamrial: is this better?
[22:11:53 CEST] <BBB> it now gives different names to subpel and fullpel functions
[22:12:48 CEST] <jamrial> did you push it to github?
[22:13:23 CEST] <BBB> https://github.com/rbultje/ffmpeg/commit/7eb376a0cfe0af0e02e79a6c05c5eb714b192861
[22:13:24 CEST] <BBB> yes
[22:17:35 CEST] <jamrial> yeah, looks good
[22:17:49 CEST] <BBB> ty
[22:17:54 CEST] <BBB> Gramner: also good?
[22:18:42 CEST] <Gramner> BBB: you could also start the loop at -4 instead of having the if bit_depth stuff in randomize_buffers(). that just writes a few unused bytes extra
[22:18:43 CEST] <jamrial> huh, i'm getting a segmentation fault with ipred now after this latest change
[22:18:56 CEST] <BBB> Gramner: ok
[22:19:29 CEST] <jamrial> Program received signal SIGSEGV, Segmentation fault.
[22:19:29 CEST] <jamrial> 0x0000000000583a87 in ff_vp9dsp_intrapred_init_10 (
[22:19:30 CEST] <jamrial>     dsp=dsp at entry=0x1ff026501b00163)
[22:19:30 CEST] <jamrial>     at D:/msys/ffmpeg/src/libavcodec/vp9dsp_template.c:1114
[22:19:30 CEST] <jamrial> 1114        init_intra_pred(TX_4X4,   4x4);
[22:19:48 CEST] <Gramner> and you could probably drop "vp9" from the static function names actually since it's kind of implied. "check_mc" is fine
[22:24:10 CEST] <BBB> Gramner: fixed
[22:25:18 CEST] <BBB> jamrial: I have no idea ...
[22:25:30 CEST] <jamrial> let me check this new version
[22:26:40 CEST] <BBB> try this new one
[22:26:45 CEST] <BBB> l/a were too small for 10/12 bpp
[22:26:52 CEST] <BBB> that may cause issues if these functions run
[22:29:33 CEST] <jamrial> yeah, works now
[22:29:44 CEST] <BBB> ok cool
[22:30:34 CEST] <BBB> should I re-send a patch to ML, or ok to push directly?
[22:33:21 CEST] <jamrial> BBB: if Gramner oks it here after you fixed his lattest comment then just push it
[22:33:27 CEST] <jamrial> or don't wait for him and push it now :p
[22:33:32 CEST] <BBB> Ill wait for him
[22:35:57 CEST] <jamrial> now to get my hands on some new ssse3-enabled machines to see if palignr is faster in the end or not
[22:36:03 CEST] <BBB> lol
[22:36:10 CEST] <BBB> try kieranks obe2
[22:37:24 CEST] <jamrial> you tested it on an intel cpu with ssse3 but no sse4, right?
[22:38:38 CEST] <BBB> I have sse4
[22:38:41 CEST] <BBB> but no avx
[22:39:04 CEST] <Gramner> BBB: hmm, I'm thinking we should call it vp9dsp instead of vp9. the other checkasm tests uses the same name as the header from libavcodec/ and the header is called vp9dsp.h
[22:39:23 CEST] <BBB> crap
[22:39:28 CEST] <BBB> does git handle renames?
[22:39:53 CEST] <nevcairiel> git mv vp9.c vp9dsp.c
[22:40:46 CEST] <Gramner> if you change that I'm ok with pushing
[22:41:59 CEST] <BBB> Gramner: changed
[22:42:14 CEST] <BBB> I also renamed the checkasm function to checkasm_check_vp9dsp
[22:42:24 CEST] <BBB> and the entry in checkasm.c is now called vp9dsp instead of vp9"
[22:42:37 CEST] <BBB> so it says  - vp9dsp.mc           [OK]
[22:42:40 CEST] <Gramner> yes, they should all be the same
[22:43:14 CEST] <BBB> ok let me push then \o/
[22:44:37 CEST] <cone-127> ffmpeg 03Ronald S. Bultje 07master:084451e1e4ff: checkasm: add vp9 MC tests.
[22:44:38 CEST] <cone-127> ffmpeg 03Ronald S. Bultje 07master:bbd44e124a59: checkasm: add vp9 intra pred tests.
[22:45:19 CEST] <BBB> let me write some 10/12bpp assembly now
[23:27:56 CEST] <Gramner> jamrial: http://pastebin.com/4f3dVbJx should fix the gcc asan failure
[23:44:15 CEST] <cone-127> ffmpeg 03Thierry Foucu 07master:164cd9b42af7: libavformat/riff.c: Add support for RV40 codec in AVI
[23:59:16 CEST] <cone-127> ffmpeg 03Ganesh Ajjanagadde 07master:6d1801b5aaf9: configure: make assignment of variable to '=' explicit
[00:00:00 CEST] --- Wed Sep 16 2015