[Ffmpeg-devel-irc] ffmpeg-devel.log.20151114

burek burek021 at gmail.com
Sun Nov 15 02:05:02 CET 2015


[01:12:55 CET] <cone-748> ffmpeg 03Michael Niedermayer 07master:44a7f17d0b20: avcodec/jpeg2000dec: Check for duplicate SIZ marker
[02:40:59 CET] <Timothy_Gu> how is textrel used in asm?
[02:43:36 CET] <nevcairiel> its secret magic sauce used for position-independent code on x64, if you want the nitty gritty details Gramner could probably enlighten you
[02:43:40 CET] <nevcairiel> er
[02:43:42 CET] <nevcairiel> on x86
[02:43:44 CET] <nevcairiel> its late :(
[02:44:48 CET] <Gramner> Timothy_Gu: x86-32 doesn't have any concept of ip-relative addressing so writing position-independant code on x86-32 is just horrible. and slow.
[02:45:11 CET] <Gramner> x86-64 doesn't have that problem though, so it's easy to avoid relocs there
[02:45:28 CET] <Timothy_Gu> ah that's why I never found any textrels in the 64-bit libs built locally
[02:45:46 CET] <Timothy_Gu> and i was wondering what the heck they are talking about
[03:04:08 CET] <cone-748> ffmpeg 03Michael Niedermayer 07master:4e16ad2868a1: avcodec/utils: Better check for channels in av_get_audio_frame_duration()
[03:04:09 CET] <cone-748> ffmpeg 03Michael Niedermayer 07master:df91aa034b82: avcodec/ivi: Check image dimensions
[03:14:34 CET] <BBB> Gramner: I think its fine to tell google to fix it for us, after all theyre causing it
[03:23:46 CET] <Gramner> also "fixing" the asm would likely make the code quite ugly and less maintainable
[03:26:01 CET] <J_Darnley> Is there a webpage somewhere that shows what the difference actually looks like in code?
[03:26:32 CET] <Gramner> compile some c with and without -fPIC i guess
[03:35:45 CET] <J_Darnley> Well I can't do that on Windows
[03:35:53 CET] <J_Darnley> I wonder if I can on ARM
[03:36:18 CET] <Gramner> you kinda need to do it on 32-bit x86 for it to make sense
[03:36:24 CET] <Gramner> ARM has ip-relative loads
[03:37:22 CET] <J_Darnley> Oh yes I forgot what I was looking for
[03:37:50 CET] Action: J_Darnley is as thick as four--yes four--short planks
[03:39:12 CET] <J_Darnley> Oh and my arch box isn't multilib
[03:39:20 CET] <rcombs> absolute addressing vs RIP-relative addressing
[03:39:22 CET] <J_Darnley> Well I don't care
[03:39:34 CET] <rcombs> or, well EIP on 32-bit
[03:39:51 CET] <rcombs> (and on 32-bit you need to do it explicitly)
[03:40:30 CET] <rcombs> but hey stop using 32-bit x86 in 2015
[03:48:50 CET] <cone-748> ffmpeg 03Ganesh Ajjanagadde 07master:6f520ce1a6c6: avutil/common: add av_rint64_clip
[04:35:08 CET] <BBB> rcombs: eip doesnt exist, you need to get the current instruction pointer  on function entry and then use that register in address instructions
[04:35:19 CET] <rcombs> oh hurr durr
[04:35:20 CET] <BBB> rcombs: (in case nobody else pointed that out)
[04:35:31 CET] <rcombs> forgot about that
[04:35:38 CET] <BBB> rcombs: rip is 64bit only, 32bit means 7 instead of 15 registers, and then this would make it 6 for some functions
[04:35:46 CET] <BBB> depending on how you implement it
[04:57:40 CET] <jamrial> some functions already are limited to 6 gprs with some compilers because of non aligned stack
[04:57:45 CET] <jamrial> and it's a pain to deal with
[13:18:00 CET] <cone-550> ffmpeg 03Michael Niedermayer 07master:ccba8aaff2ef: avcodec/avrndec: Use the AVFrame format instead of the context
[13:48:21 CET] <cone-550> ffmpeg 03Michael Niedermayer 07master:e04126072e98: avcodec/ffv1dec: Clear quant_table_count if its invalid
[13:48:22 CET] <cone-550> ffmpeg 03Michael Niedermayer 07master:17705f5d4f57: avcodec/flashsv: Check size before updating it
[14:39:05 CET] <cone-550> ffmpeg 03Hendrik Leppkes 07master:4c8d86eb18e2: avformat: implement SChannel SSP TLS protocol
[16:53:36 CET] <cone-550> ffmpeg 03Michael Niedermayer 07master:c8aaae8e0f15: avcodec/dpx: Move need_align to act per line
[17:06:49 CET] <cone-550> ffmpeg 03Ganesh Ajjanagadde 07master:f1726ad1b0d0: avcodec/ratecontrol: use predefined M_E
[17:06:50 CET] <cone-550> ffmpeg 03Ganesh Ajjanagadde 07master:c5fa42c69aec: avcodec/mpegvideo: use predefined M_PI
[17:06:51 CET] <cone-550> ffmpeg 03Ganesh Ajjanagadde 07master:11f0acf829e2: avcodec/mpegaudio: use predefined M_SQRT2
[17:06:52 CET] <cone-550> ffmpeg 03Ganesh Ajjanagadde 07master:618b3ae7d4a9: avcodec/simple_idct: use predefined M_SQRT2
[18:38:05 CET] <cone-550> ffmpeg 03Michael Niedermayer 07master:ac6ab77741f5: avcodec/rscc: Check input size for raw mode
[18:38:06 CET] <cone-550> ffmpeg 03Michael Niedermayer 07master:ebf5264cd6bb: avcodec/pgssubdec: Check dimensions for 0
[20:28:57 CET] <kierank> durandal_1707: seems afl-persistent with clang-fast is good
[20:29:01 CET] <kierank> fuzzing opus silk now
[20:29:05 CET] <kierank> speeds things up a bit
[20:30:15 CET] <durandal_1707> what's afl-persistent?
[20:31:09 CET] <kierank> https://github.com/mirrorer/afl/blob/master/experimental/persistent_demo/persistent_demo.c
[20:33:36 CET] <kierank> a lot of spurious crashes though
[21:23:31 CET] <jamrial> 6f520ce1 broke msvc 2012. llrint is not available and libavutil/libm.h is not included
[21:26:52 CET] <nevcairiel> and libm.h is a private header, while the one thats broken is public
[21:26:56 CET] <nevcairiel> good luck with that =p
[21:28:48 CET] <jamrial> av_rint64_clip is not even used anywhere
[21:29:29 CET] <J_Darnley> Just revert it?
[21:29:42 CET] <Daemon404> i dont understand the commit message
[21:29:51 CET] <Daemon404> the rational provided doesnt actually include rationale for adding the function
[21:29:57 CET] <Daemon404> only what it does
[21:35:16 CET] <jamrial> how about replacing it with (int64_t)rint()?
[21:35:55 CET] <nevcairiel> msdn only lists it for 2013
[21:35:58 CET] <jamrial> what the libm.h fallback does, basically
[21:36:15 CET] <nevcairiel> but maybe its wrong about rint
[21:36:52 CET] <nevcairiel> libm.h also has a fallback for rint, so maybe thats used then
[21:38:44 CET] <Daemon404> jamrial, before that: ask why it even exists
[21:38:53 CET] <Daemon404> adding more avutil functions for no reason? 
[21:38:59 CET] <Daemon404> to create libavcrt?
[21:39:00 CET] <nevcairiel> because it fixes some theoretical problem
[21:39:07 CET] <Daemon404> *nothing uses it*
[21:40:10 CET] <nevcairiel> thats why the problem is theoretical :D
[21:40:35 CET] <jamrial> i already wondered about that. and mentioned it's not even used
[21:40:44 CET] <Daemon404> [FFmpeg-devel] [PATCH] all: use M_SQRT1_2, M_SQRT2, M_PI <-- someone needs to check that this works
[21:40:48 CET] <jamrial> it was probably intended for some future patchset from ganesh
[21:40:50 CET] <Daemon404> those are hidden behind _USE_MATH_DEFINES in mavc
[21:40:54 CET] <Daemon404> or, they were.
[21:41:58 CET] <jamrial> Daemon404: msvc2013 and 2015 seem to work
[21:42:20 CET] <jamrial> nevcairiel's fate clients at least
[21:42:56 CET] <Daemon404> i mean pre-2013
[21:43:34 CET] <jamrial> we'll not know until the other issue is solved then :p
[21:44:36 CET] <jamrial> is it ok if a commit revert also reverts the lavu version bump?
[21:45:07 CET] <Daemon404> ask michaelni__ 
[21:45:09 CET] <Daemon404> im not sure
[21:45:31 CET] <J_Darnley> Yes?  Because the API is being reverted to its previous state?
[21:45:51 CET] <jamrial> alright
[21:46:07 CET] <J_Darnley> You should ask him first
[21:46:35 CET] <J_Darnley> Another argument is: it must always increase with API changes
[21:46:40 CET] <jamrial> michaelni__: ^
[21:46:47 CET] <J_Darnley> This is a change so it must be increased again.
[21:47:14 CET] <jamrial> the commit is 17 hours old. bumping it again is imo stupid
[21:47:33 CET] <nevcairiel> i would j ust leave the current version and remove it from apichanges
[21:47:52 CET] <jamrial> it's not even mentioned in apichanges
[21:48:08 CET] <nevcairiel> well then
[21:55:25 CET] <nevcairiel> hm, i think my compiler broke
[21:55:26 CET] <nevcairiel> LINK : fatal error LNK1104: cannot open file 'kernel32.lib'
[21:55:29 CET] <nevcairiel> that doesnt sound good =p
[22:00:23 CET] <Daemon404> how do you even manage that
[22:00:28 CET] <Daemon404> nukeing %LIB%?
[22:00:29 CET] <nevcairiel> no idea
[22:00:44 CET] <nevcairiel> i upgraded my win10 to 1511 today, maybe it fuzzed up something
[22:01:09 CET] <Daemon404> that sounds seriously sketchy if it caused that
[22:01:59 CET] <nevcairiel> it seems WindowsSdkDir is not being set properly
[22:02:15 CET] <nevcairiel> maybe its confused by the change in version
[22:02:37 CET] <Daemon404> version name in a directory? what could go wrong?
[22:05:06 CET] <nevcairiel> SO article with similar symptoms s ays it started after installing the WDK, which I did last week, and I may not have build ffmpeg with vs2015 since then, not on this box
[22:05:26 CET] <nevcairiel> https://connect.microsoft.com/VisualStudio/Feedback/Details/1656623
[22:05:40 CET] <nevcairiel> or https://connect.microsoft.com/VisualStudio/feedback/details/1610302/universalcrt-detection-breaks-when-windows-driver-kit-is-installed rather
[22:06:03 CET] <Daemon404> the wdk stuff is REALLY broken for me
[22:06:05 CET] <Daemon404> always has been
[22:06:12 CET] <Daemon404> uninstall actually says it cant find the uninstaller
[22:06:14 CET] <Daemon404> that kind of stuff
[22:06:21 CET] <Daemon404> and this is in a fresh vm
[22:06:28 CET] <nevcairiel> apparently marking the wdk folder as hidden fixes it
[22:06:29 CET] <nevcairiel> i shall try
[22:06:37 CET] <nevcairiel> i need the wdk to build something for work
[22:06:38 CET] <nevcairiel> so meh
[22:06:53 CET] <Daemon404> ... it relies on folder properties
[22:07:04 CET] <Daemon404> what kind of black magic do the msvc bat files use
[22:07:28 CET] <nevcairiel> well, it enumerates the installed versions based on the presence of their folder apparently
[22:07:36 CET] <nevcairiel> and if you hide it, its not enumarted =p
[22:08:04 CET] <J_Darnley> Good job MS!
[22:08:07 CET] <Daemon404> [21:02] <@Daemon404> version name in a directory? what could go wrong?
[22:10:54 CET] <nevcairiel> maybe they fix it in update 1, which should be out next week i think
[22:12:20 CET] <cone-506> ffmpeg 03Andreas Cadhalpun 07master:603ebab8d7b7: dds: disable palette flag for compressed dds
[22:45:07 CET] <durandal_1707> this ivr format is really strange
[23:06:56 CET] <cone-506> ffmpeg 03Michael Niedermayer 07master:a105f52855d0: avcodec/error_resilience: avoid accessing previous or next frames tables beyond height
[00:00:00 CET] --- Sun Nov 15 2015


More information about the Ffmpeg-devel-irc mailing list