[Ffmpeg-devel-irc] ffmpeg-devel.log.20131102

burek burek021 at gmail.com
Sun Nov 3 02:05:02 CET 2013


[00:41] <BBB> ubitux: not that fast ;)
[00:41] <ubitux> BBB: why did you start with 32 instead of 16 btw?
[00:41] <BBB> ubitux: taking some weird approaches so it's not exactly going fast
[00:42] <BBB> I guess I like suffering
[00:42] <BBB> ?
[00:42] <ubitux> :)
[00:42] <ubitux> it's really cool to have that much registers with xmm hehe
[00:43] <BBB> it's not enough for me ;)
[00:43] <ubitux> and it's a lot simpler now that i made 4x4 actually :p
[00:43] <ubitux> yes haha
[00:43] <ubitux> i guess so :)
[00:43] <ubitux> when idct_idct_{4x4,8x8} are done, i guess i'll have a look to the iadst variants
[00:44] <BBB> sure, or loopfilter
[00:44] <BBB> loopfilter is actually fun, I learned a lot from doing that in vp8
[00:44] <BBB> but do it in whatever order you prefer
[00:44] <BBB> most importantly, it should be fun
[00:45] <ubitux> ah?
[00:45] <ubitux> ok
[00:58] <cone-35> ffmpeg.git 03Mickaël Raulet 07master:38612379bffe: hevc: fix temporal_id(cherry picked from commit 5eac0df49d33bef7f3ffc5efde19839dddb554e8)
[00:58] <cone-35> ffmpeg.git 03Michael Niedermayer 07master:f578e5d9376a: avcodec/hevc: Adjust white-spaces to reduce difference to 064698d381e1e7790f21b0199a8930ea04e2e942
[01:03] <kurosu_> deblocking is most often a major pain
[01:03] <kurosu_> not enjoying it
[01:09] <ubitux> just like idct, i'll do the simpler part and let BBB do the major pain part
[01:09] Action: ubitux runs away
[01:09] <BBB> deblock is a friggin' pain, yes
[01:09] <BBB> but it's always a good feeling when you finish something like that
[01:09] <ubitux> :)
[01:09] <BBB> so I'm happy to do it
[01:09] <BBB> just will take a while
[01:11] <kurosu_> mc is cool: simple, very rewarding on the runtime :-)
[01:12] <ubitux> this one is already done afaict
[01:12] <kurosu_> with 32x32/16x16 transforms, I'm wondering if implementing transforms assuming upper coefficients are null will start to be beneficial
[01:13] <kurosu_> i was talking in a generic way, not speaking of vp9 in particular
[01:38] <cone-35> ffmpeg.git 03Michael Niedermayer 07master:647adc429050: h264: factor "if(h->sei_recovery_frame_cnt >= 0)" out
[09:32] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:8183e33225af: h264_refs: use named constant for setting frame_recovered
[09:38] <cone-155> ffmpeg.git 03Luca Barbato 07master:4a64e67988dd: flashsv: Refactor a little
[09:38] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:77c33424c175: Merge commit '4a64e67988dd01005efb1ae831bff14c1656b573'
[09:45] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:57070b1468ed: flashsv: Check diff_start diff_height values
[09:45] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:525289ed261f: Merge commit '57070b1468edc6ac8cb3696c817f3c943975d4c1'
[09:56] <cone-155> ffmpeg.git 03Anton Khirnov 07master:d1c229cdbc07: lavf: do not use int to store an int64
[09:56] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:02223f4dc4da: Merge commit 'd1c229cdbc07ef8eb703d68d1c5b509baee85245'
[10:01] <cone-155> ffmpeg.git 03Diego Biurrun 07master:23157d72b565: configure: Split test_cflags function off from check_cflags
[10:01] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:b4f95c96a577: Merge commit '23157d72b565e0228fec97f1eb059d4f8021a260'
[10:03] <ubitux> when i'm doing a mov reg, [pw_xxxx], can i assume pw_xxxx is aligned? (and thus use mova instead of movu)
[10:05] <michaelni> all global/static asm constants should be aligned
[10:05] <ubitux> ok, thx
[10:10] <cone-155> ffmpeg.git 03Diego Biurrun 07master:056fd4fe65e7: configure: Properly check for availability of -Wno-foo warning flags
[10:10] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:0db04ca01dfa: Merge commit '056fd4fe65e70b4dbca97bcf7faf8b7ce3df993f'
[10:18] <cone-155> ffmpeg.git 03Diego Biurrun 07master:1700b4e678ed: x86: vp8dsp: Split loopfilter code into a separate file
[10:18] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:7146eacfc5f2: Merge commit '1700b4e678ed329611a16b20d11e64b7abda4839'
[10:29] <cone-155> ffmpeg.git 03Luca Barbato 07master:037fea388e0d: flashsv: K&R formatting cosmetics
[10:29] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:0b82fdcc469e: Merge commit '037fea388e0df2a22441bc1ed86794152161baf1'
[10:36] <cone-155> ffmpeg.git 03Vittorio Giovara 07master:dc6ea00cd2b9: mpeg4video: K&R formatting cosmetics
[10:36] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:1344c0453615: Merge commit 'dc6ea00cd2b91b591e6726e5bf1d5e03a4a9bdd0'
[12:25] <ubitux> lol \r line breaks
[12:43] <ubitux> any suggestion to make a pshufw reg,reg,0 in xmm?
[12:43] <ubitux> (copying first word all over the register)
[12:47] <ubitux> mmh maybe the SPLATW macro
[13:16] <cone-155> ffmpeg.git 03Martin Storsjö 07master:435214a757fc: Add an HDS live fragmenting muxer
[13:16] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:641358addcff: Merge remote-tracking branch 'qatar/master'
[13:16] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:ce7f1c76bd8f: avcodec/hevc: more whitespaces to reduce difference to 064698d381e1e7790f21b0199a8930ea04e2e942
[13:16] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:1a6948fa708d: libavcodec/hevc: reduce bracket differences to 064698d381e1e7790f21b0199a8930ea04e2e942
[13:22] <BBB> ubitux: oh btw, if you're interested, there's also vp9-mt that could be done, it's not super-hard (because vp9 is relatively straightforward) and should have quite big gains
[13:22] <ubitux> i hate working with threading
[13:31] <BBB> oh
[13:31] <BBB> ohwell
[14:15] <Compn> wbs : missing doc on hds muxer? 
[14:51] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:b23692b3a8e0: libavcodec/hevc: reduce whitespace differences to 064698d381e1e7790f21b0199a8930ea04e2e942
[14:51] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:6a84d17bd79d: libavcodec/hevcdsp_template: cosmetics to reduce diff to 064698d381e1e7790f21b0199a8930ea04e2e942
[14:51] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:69b3668b83eb: libavcodec/hevc: indention related cosmetics to reduce diff to 064698d381e1e7790f21b0199a8930ea04e2e942
[14:51] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:42dbe2d9b1d6: libavcodec/hevcdsp_template: random cosmetics to reduce diff to 064698d381e1e7790f21b0199a8930ea04e2e942
[14:53] <iive> does this mean libav have merged havc too?
[14:53] <iive> or these are toward the github repo?
[14:56] <kierank> merged hevc
[15:33] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:e877455ff98a: libavcodec/hevcdsp_template: whitespaces to reduce diff to 064698d381e1e7790f21b0199a8930ea04e2e942
[15:33] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:2c4f5736962a: libavcodec/hevc: random cosmetics to reduce diff to 064698d381e1e7790f21b0199a8930ea04e2e942
[18:31] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:35594c48caf0: libavcodec/hevc: random cosmetics to reduce diff to 064698d381e1e7790f21b0199a8930ea04e2e942
[18:31] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:a805e2e65301: avcodec/hevc: add "apply_defdispwin" alias for compatibilty with 064698d381e1e7790f21b0199a8930ea04e2e942
[18:37] <cone-155> ffmpeg.git 03Vittorio Giovara 07master:c68d4c230ad8: mpeg4video_parser: K&R formatting cosmetics
[18:38] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:1654e64bba80: Merge commit 'c68d4c230ad8ca85af3999a6af8e582c43620d58'
[18:44] <cone-155> ffmpeg.git 03Diego Biurrun 07master:5928b29f5399: os_support: Add #endif comments for better readability
[18:44] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:67d1d06225e9: Merge remote-tracking branch 'qatar/master'
[18:53] <MX070> hi !!!
[18:53] <MX070> ive not been in here before as ive tried to use google to help me with my issues but right now im a little stuck....
[18:53] <MX070> anyone here firstly??
[18:59] <cone-155> ffmpeg.git 03Timothy Gu 07master:d3211cfaedcc: avcodec/libmp3lame: add ABR support
[18:59] <iive> no. user and usage support is provided in #ffmpeg, this channel is primary for development of ffmpeg.
[19:00] <MX070> ah ok no probs
[19:00] <MX070> is there another chatroom where i might be able to get some help?
[19:01] <mark4o> MX070: /join #ffmpeg
[19:02] <MX070> thank you !
[19:19] <ubitux> SWAP is really cool and all, but really, to debug, that's a hell :))
[20:21] <ubitux> Timothy should really fix his mailer :(
[20:37] <BBB> ubitux: since when is assembly for debugging? :)
[20:37] <ubitux> i never use a debugger&
[20:37] <ubitux> except for asm :p
[20:38] <BBB> good
[20:38] <BBB> disass
[20:38] <BBB> info all-registers
[20:38] <ubitux> BBB: current method is copy pasting http://pastie.org/8450857 for each new run
[20:39] <ubitux> using "ni"
[20:39] <ubitux> and comparing with c av_log'ed
[20:39] <BBB> lol
[20:39] <ubitux> but after SWAP, i'm lost (or the process of debugging get slowed down significantly :p
[20:43] <BBB> disass, info all-registers
[20:43] <BBB> really
[20:43] <BBB> it's awesome
[20:43] <BBB> or, well
[20:43] <BBB> let me restate that
[20:44] <BBB> it is slightly less 100% horrific
[20:44] <ubitux> i'm not sure how that will help
[20:45] <ubitux> i'm interested in tracing the changing values of each xmm registers instr. after instr.
[20:45] <ubitux> but i would love to the see some colors with gdb after each ni
[20:45] <ubitux> to show which register just got updated
[20:46] <ubitux> ...and some fucking 0 padding for values ffs, align is completely fucked up :(
[20:47] <ubitux> also, info all-registers is way too cluttered up for xmm registers
[20:50] <ubitux> maybe i should try radare
[20:50] <ubitux> (re)try
[20:51] <kurosu> p/x $xmm0 ?
[20:52] <ubitux> kurosu: yep this sucks
[20:52] <BBB> stepi?
[20:52] <ubitux> mmh?
[20:52] <ubitux> what are you guys trying to answer?
[20:52] <ubitux> :))
[20:52] <cone-155> ffmpeg.git 03James Zern 07master:a8bcc956fd66: libvpxenc: split vp8/vp9 options array
[20:52] <BBB> if you want to see registers instruction by instruction, just use stepi, followed by info all-registers
[20:53] <BBB> it gives you all information you may ever want
[20:53] <BBB> and a lot of stuff you don't want
[20:53] <ubitux> 20:38:56 <@ubitux> BBB: current method is copy pasting http://pastie.org/8450857 for each new run
[20:53] <ubitux> 20:39:01 <@ubitux> using "ni"
[20:53] <ubitux> i find this better ^
[20:53] <ubitux> because it displays automatically after each next-instruction
[20:53] <ubitux> and i step in with <enter>
[20:56] <BBB> that's sweet
[20:57] <ubitux> yes but: http://pastie.org/8450898 random misalign, always a pain to crawl to see which reg changed
[21:00] <ubitux> i wrote a gdb function yesterday
[21:00] <ubitux> to make a personnal print
[21:00] <ubitux> but that's so slow it's unusable
[21:00] <ubitux> (like 4-5 sec to print all xmm reg)
[21:03] <kierank> what's wrong with info all registers
[21:03] <kierank> or whatever the command is
[21:03] <ubitux> kierank: http://ubitux.fr/pub/pics/_whats-wrong-with-info-all-reg.png
[21:03] <kierank> ah
[21:03] <kierank> :)
[21:04] <ubitux> i don't feel like reading this after each instruction
[21:04] <ubitux> not even once actually
[21:04] <Daemon404> ubitux, i usually have gdb dump it to a file
[21:04] <Daemon404> and review it elsewhere via some sort of tail plugin
[21:05] <ubitux> and you tail -f it?
[21:05] <Daemon404> (yes its shit)
[21:05] <ubitux> haha
[21:05] <Daemon404> usually in a proper text editor
[21:05] <ubitux> i'm pretty sure i can have a small script doing some monitoring and do the color thing i'm looking for
[21:05] <Daemon404> yes
[21:06] <Daemon404> but really it is gdb's fault
[21:06] <Daemon404> for havign such a shit ui
[21:06] <ubitux> yup
[21:06] <ubitux> i miss ollydbg from the old days
[21:06] <ubitux> (does it still exist?)
[21:06] <Daemon404> ... i still use ollydbg to patch stuff 
[21:06] <Daemon404> <_<
[21:06] <ubitux> i liked that ui
[21:06] <Daemon404> the ui layout is ok
[21:07] <Daemon404> the poop-brown color could use a change
[21:07] <ubitux> not sure it's appropriate for xmm dbg though
[21:07] <Daemon404> ollydbg's ui is similar to MSVC's
[21:07] <Daemon404> for debug
[21:07] <Daemon404> though i have no idea how it handles asm
[21:07] <Daemon404> i'd probably use IDA nowadays for that
[21:09] <ubitux> *expensive*
[21:09] <Daemon404> there's a free version
[21:09] <ubitux> usable?
[21:09] <Daemon404> older release, still useful
[21:09] <Daemon404> for asm
[21:09] <Daemon404> not so much for decompiling 
[21:09] <Daemon404> my RE is done in a 'liberated' version
[21:10] <ubitux> :/
[21:10] <Daemon404> https://www.hex-rays.com/products/ida/support/download_freeware.shtml
[21:10] <Daemon404> ida works very well in wine
[21:11] <Daemon404> newer vers have linux builds
[21:11] <Daemon404> not sure what version live debugging was added for remote linux procs
[21:13] <ubitux> :/
[21:14] <ubitux> sounds like a lot of pain
[21:14] <Daemon404> its a pain in the butt until you figure all teh stuff out
[21:14] <Daemon404> then it's very productive
[21:15] <ubitux> i doubt i'll be able to have a working workflow with a wine+oldida to debug simd in elf binaries
[21:15] <Daemon404> i had that working
[21:15] <Daemon404> on my version anyway
[21:15] <Daemon404> enjoy your gdb!
[21:15] <ubitux> i'm looking again at radare
[21:16] <Daemon404> ... they sure could use a better home page
[21:16] <Daemon404> 90% of their front page is tshirts and twitter
[21:16] <Daemon404> real crddible
[21:36] <ubitux> ida 450-2700¬
[21:36] <ubitux> nice
[21:37] <durandal_1707> whassup?
[21:40] <JEEB> ubitux, also you have to have the basic license for a year first
[21:40] <JEEB> and only then you can apply (yes, apply!) for the plugins like decompiler
[21:41] <JEEB> (oh, and buying an IDA license also means applying)
[21:41] <JEEB> (it's not a done deal that you will actually get the license)
[21:41] <ubitux> i think i'll just mess with my c/p gdb template for a while
[21:41] <ubitux> and continue swearing in front of my screen
[21:43] <durandal_1707> ubitux: why you need decompiler?
[21:44] <ubitux> debugger
[21:48] <ubitux> durandal_1707: debugging simd
[22:20] <wm4> does "name at address dot bla" really prevent spam?
[22:20] <wm4> you'd think the spammers figured that one out already...
[22:21] <ubitux> limit damage i guess
[22:24] <Daemon404> fizzbuzz for spammers
[23:00] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:8d88ed9a1fb4: avcodec/ffv1enc: pass through pts
[23:00] <cone-155> ffmpeg.git 03Michael Niedermayer 07master:1fb3b494fac7: ffv1enc: store 2pass statistics at the end
[23:00] <Daemon404> ffv1 has 2 pass?
[23:00] <Daemon404> lolwut?
[23:03] <JEEB> I never knew...
[23:10] <nevcairiel> isnt it lossless? i wonder what 2pass does for lossless
[23:11] <superdump> something to do with context i expect
[00:00] --- Sun Nov  3 2013


More information about the Ffmpeg-devel-irc mailing list