[FFmpeg-devel-irc] IRC log for 2010-09-12

[irc at mansr.com]

[13:04:59] <CIA-63> libswscale: ramiro * r32190 /trunk/libswscale/ (rgb2rgb.c rgb2rgb.h rgb2rgb_template.c): rgb2rgb: replace shuffle_bytes_2103() by optimized rgb32tobgr32()
[14:03:31] <mru> hmm, valgrind errors in lavfi-pixdesc_be
[14:03:42] <mru> might explain the crashes on sparc/openbsd
[14:54:25] <mru> found the bug
[16:10:03] <mru> and fixed it
[16:10:19] <mru> now if someone would review the fix...
[17:07:05] <J_Darnley> Will somebody please commit the rest of libavfilter!
[17:21:19] <mru> the rest?
[18:01:10] <J_Darnley> rotate and whatever else is still in the SoC repository
[18:01:28] <J_Darnley> mru ^
[18:02:32] <spaam> J_Darnley: maybe you need to send it to ffmpeg-devel :)
[18:04:07] <J_Darnley> Why don't I also put name in the license header?
[18:04:12] <J_Darnley> *my name
[18:15:39] <BBB> J_Darnley: we need people to make each filter go through review
[18:15:47] <BBB> you can't just commit that stuff, it'll be slow and awful
[18:15:58] <mru> I'm a bit concerned about libavfilter
[18:16:06] <mru> it has issues
[18:16:12] <Dark_Shikari> slow is fine IMO
[18:16:15] <mru> and the more filter are added, the harder they will be to fix
[18:16:51] <BBB> mru: sounds like swscale
[18:16:58] <mru> it's not that bad
[18:16:59] <mru> it's fixable
[18:17:00] <Dark_Shikari> lol
[18:17:26] <mru> it's mainly things like filter functions having no way of reporting an error back
[18:17:35] <spaam> mru: what issue does avfilter have?
[18:17:51] <mru> and I keep finding memory leaks and buffer overruns
[18:18:06] <wbs> and it seems to assume quite a bit on the behaviour of codecs, that not all of them fulfill (like supporting direct rendering with negative strides and such)
[18:18:21] <mru> imo all codecs should support that
[18:18:31] <mru> anything that doesn't needs to be fixed
[18:18:56] <_av500_> mru: the ti codecs dont ;)
[18:19:14] <mru> irrelevant and you know it
[18:19:45] <astrange> are they CODEC_CAP_DR1?
[18:19:47] <Dark_Shikari> I think that's a rather unreasonable requirement
[18:19:50] <Dark_Shikari> I mean, for one
[18:19:52] <mru> Dark_Shikari: why?
[18:19:59] <Dark_Shikari> at least only if they support direct rendering
[18:20:03] <mru> because it needs one instruction more on x86_64?
[18:20:25] <Dark_Shikari> mru: all of ffmpeg makes the assumption that strides fit in 32 bit
[18:20:38] <mru> if stride is specified as a signed type, one should reasonably assume it can be negative
[18:20:45] <mru> yes, signed 32-bit
[18:21:04] <mru> saste: ping
[18:21:07] <Dark_Shikari> then explain to me why it crashes in a large variety of C functions
[18:21:19] <mru> I haven't looked at that
[18:21:31] <mru> but ptr += stride certainly should work for negative stride
[18:21:35] <astrange> 14:01 < J_Darnley> rotate and whatever else is still in the SoC repository <- rotate wasn't from the soc repository
[18:21:47] <astrange> the author just never replied after i asked for better interpolation
[18:22:08] <Dark_Shikari> mru: I think at this point trying to fix every decoder to work with that is work better psent elsewhere
[18:22:21] <mru> now if someone is doing end=start+height*stride; while p<end, then I can see how it might crash...
[18:22:40] <astrange> although i can't remember how to do affine transforms with interpolation myself
[18:23:50] <Mehran> hello everyboyd
[18:24:44] <Dark_Shikari> mru: we could change all strides to intptr_t
[18:25:01] <mru> how would that help crashing C code?
[18:25:10] <Dark_Shikari> it'd eliminate the extra asm instruction
[18:25:22] <mru> and waste space
[18:25:29] <Dark_Shikari> space?  it still takes one register
[18:25:34] <Dark_Shikari> and 8 bytes on the stack
[18:25:46] <mru> and 64-bit prefixes on the instructions for x86
[18:26:05] <Mehran> I have a question about delete or pick a frequency from wave or mp3 file. everyone can help me?
[18:29:18] <_av500_> Mehran: wrong channel
[18:29:50] <Mehran> where is suiatable channel plz?
[18:30:37] <Mehran> _av500_: Where is Suitable channel? Thanks
[18:30:40] <Dark_Shikari> we're not the right channel to ask that question
[18:31:38] <BBB> Mehran: try #ffmpeg maybe
[18:31:49] <BBB> Dark_Shikari: in yasm, can I call functions that do not have a ff_ prefix?
[18:31:49] <Dark_Shikari> that doesn't sound like #ffmpeg either
[18:31:54] <Dark_Shikari> BBB: No.
[18:32:03] <Dark_Shikari> why would you want to?
[18:32:11] <BBB> call an internal (non-cglobal) function
[18:32:16] <Mehran> BBB: Thanks.
[18:32:19] <Dark_Shikari> It won't kill you if you add an ff_.
[18:32:20] <mru> BBB: I don't see why not
[18:32:36] <Dark_Shikari> mru: the current yasm call macro auto-prefixes functions
[18:32:46] <Dark_Shikari> same with function declarations
[18:32:48] <BBB> Dark_Shikari: so how do I declare the function then?
[18:32:50] <mru> oh, I didn't know call was a macro
[18:32:59] <BBB> mangle(ff_blabla): ?
[18:33:02] <Dark_Shikari> there is nothing wrong with making everything ff_
[18:33:06] <Dark_Shikari> it won't eat you
[18:33:07] <mru> of course not
[18:33:09] <Dark_Shikari> it doesn't take extra code space
[18:33:13] <Dark_Shikari> because it's done automatically.
[18:33:25] <Dark_Shikari> so just let your local symbols be ff_ too.
[18:34:58] <BBB> Undefined symbols:
[18:34:58] <BBB>   "_ff_x264_add8x4_idct_sse2", referenced from:
[18:35:06] <BBB> that's the error I get (compilation itself succeeds fine)
[18:35:27] <BBB> I declared the function simply as x264_add8x4_idct_sse2:\n<actual code>
[18:35:53] <mru> the call macro probably adds the underscore on underscore-encumbered systems
[18:36:14] <BBB> 00000000000026e0 t x264_add8x4_idct_sse2
[18:36:17] <BBB> the ff is missing also
[18:36:24] <mru> of course
[18:36:34] <mru> what would have added it there?
[18:36:36] <BBB> but the problem is that if I add those like mangle(ff_x264_..):\n<code>
[18:36:39] <BBB> then compilation fails
[18:37:03] <BBB> because it can't find the declaration for the symbol x264_...[etc]
[18:37:31] <BBB> so compilation expects me to remove the ff_, but linking expects me to add them, because call adds them
[18:37:36] <BBB> it's a little weird
[18:37:44] <mru> the linker doesn't do anything
[18:38:28] <mru> Dark_Shikari: I don't see the call macro adding any prefix?
[18:38:36] <thresh> gah, crappy chinese HW, replacing the power supply worked to debrick wl500gp
[18:39:48] <mru> ah, cglobal redefines the name
[18:39:49] <mru> evil
[18:40:04] <BBB> ooooo wait now I see
[18:40:05] <BBB> nevermind
[18:40:11] <BBB> the error is in h264dsp_mmx.c
[18:40:15] <BBB> which I didn't update yet
[18:40:17] * BBB kicks himself
[18:41:44] <mru> let me help
[18:41:47] <mru> /kick BBB
[18:42:45] <Dark_Shikari> mru: yes, quite evil =p
[18:43:10] <Dark_Shikari> btw if we want we can remove the x264_ prefix, it might work better that way
[18:43:13] <BBB> Dark_Shikari: is r0m something else than the memory location of the first argument?
[18:43:16] <Dark_Shikari> because right now we effectively have two prefixes
[18:43:22] <Dark_Shikari> BBB: no, that's it
[18:43:23] <BBB> no, x264 is fine, I fucked up
[18:43:32] <Dark_Shikari> No, I mean we might want to remove that
[18:43:36] <Dark_Shikari> since prefixes are auto-added
[18:43:39] <BBB> for "add r0m, gprsize"
[18:43:41] <Dark_Shikari> so x264_ is no longer in the x264 asm either
[18:43:52] <Dark_Shikari> BBB: if you want yasm to keep track of your changes to the stack pointer
[18:43:52] <Dark_Shikari> use ADD
[18:44:09] <BBB> I want to increase the value of the address
[18:44:23] <Dark_Shikari> oh, I see
[18:44:26] <Dark_Shikari> you're not changing the stack pointer
[18:44:30] <Dark_Shikari> you're changing r0m while it's in memory
[18:44:31] <BBB> i.e. add dword [esp+bla], gprsize
[18:44:33] <Dark_Shikari> WARNING
[18:44:33] <BBB> yes
[18:44:34] <Dark_Shikari> x86_64.
[18:44:39] <BBB> this is x86-32 only
[18:44:41] <Dark_Shikari> ok, good.
[18:45:01] <BBB> for x86-64 I use r10
[18:45:28] <wbs> mru: btw, how far did you debug the compiler bug causing the gsm test failures on gcc 4.2 on ppc/sparc?
[18:45:35] <BBB> or can I not add to memory straight away?
[18:45:42] <BBB> i.e. should I read it and then re-write it?
[18:45:46] <mru> wbs: not at all
[18:45:55] <mru> feel free to give it a shot
[18:46:45] <wbs> mru: ok.. I had a quick look, and it seems that it inlines/reorders things incorrectly regarding the filter_value function - adding av_noinline to that function makes it work
[18:46:59] <mru> hmm
[18:47:43] <mru> hehe, that's the kind of construct compilers often fail on
[18:47:48] <BBB> Dark_Shikari: what is r0mp then?
[18:48:12] <BBB> changing it to mov r0, r0m; add r0, gprsize; mov r0m, r0 fixes it, but so does changing it to add r0mp, gprsize
[18:48:15] <BBB> I wonder what r0mp is
[18:48:39] <Dark_Shikari> it relates to win64
[18:48:45] <Dark_Shikari> and the way it handles pointers on the stack
[18:48:47] <Dark_Shikari> or something like that
[18:48:48] <wbs> mru: which construct?
[18:49:11] <mru> I'm not sure
[18:49:25] <mru> it just has a familiar look
[18:50:23] <wbs> ok
[18:50:41] <BBB> ugh shit now I have to test this crap on win64 also
[18:50:48] <BBB> maybe I'll just commit it and see if win64 breaks
[18:50:52] <BBB> if it does I can always blame ramiro
[18:52:32] <BBB> the weird thing is that r0mp and r0m evaluate to the same disassembly in gdb
[18:52:37] <BBB> ohwell
[18:52:42] <BBB> make fate works as well
[18:52:45] <BBB> \o/
[18:52:54] <BBB> now to make sure I didn't cripple performance
[18:52:56] <mru> none of those r0foo names are real
[18:53:01] <Dark_Shikari> ask Bugmaster about r0mp
[18:53:06] <Dark_Shikari> I know it has a specific purpose
[18:53:11] <Dark_Shikari> but only he knows win64
[19:17:41] <CIA-63> libswscale: ramiro * r32211 /trunk/libswscale/swscale-test.c:
[19:17:41] <CIA-63> libswscale: swscale-test: move conversion from ref to source back to doTest()
[19:17:41] <CIA-63> libswscale: The source format parameters are kept in static variables and conversion from
[19:17:41] <CIA-63> libswscale: ref to source is only made when any parameter changes.
[19:17:41] <CIA-63> libswscale: ramiro * r32212 /trunk/libswscale/swscale-test.c: swscale-test: always use SWS_BILINEAR to convert from ref to src
[19:17:42] <CIA-63> libswscale: ramiro * r32213 /trunk/libswscale/swscale-test.c:
[19:17:43] <CIA-63> libswscale: swscale-test: take input from reference file
[19:17:43] <CIA-63> libswscale: If the CRC from the src->dst conversion matches a reference, it is not
[19:17:44] <CIA-63> libswscale: necessary to perform a dst->yuva420p conversion and check the SSD.
[19:17:44] <CIA-63> libswscale: ramiro * r32214 /trunk/libswscale/swscale-test.c: indent
[19:17:45] <CIA-63> libswscale: ramiro * r32215 /trunk/libswscale/swscale-test.c: swscale-test: allow specifying src/dst pixel formats
[19:27:21] <BBB> blegh, I hate gcc sometimes
[19:27:46] <BBB> (in this case gcc does something iffy that is actually good for performance but leads to a horrible asm function, which I can't easily cleanly replicate in yasm)
[19:45:39] <CIA-63> ffmpeg: stefano * r25108 /trunk/libavdevice/v4l2.c:
[19:45:39] <CIA-63> ffmpeg: Cosmetics: simplify layout of fmt_conversion_table, make it shorter
[19:45:39] <CIA-63> ffmpeg: and more readable.
[21:32:34] <CIA-63> ffmpeg: stefano * r25109 /trunk/libavutil/pixdesc.c:
[21:32:35] <CIA-63> ffmpeg: Cosmetics: use x ? y : z construct and merge intialization and
[21:32:35] <CIA-63> ffmpeg: definition for the val variable in av_read_image_line(). Simplify.
[21:33:07] <mru> damn, I hate when people mess commit stuff mid-patch review
[21:33:30] <mru> saste: ping
[21:40:47] <CIA-63> ffmpeg: mru * r25110 /trunk/libavutil/pixdesc.c:
[21:40:47] <CIA-63> ffmpeg: pixdesc: use 8-bit accesses when possible in av_read/write_image_line()
[21:40:47] <CIA-63> ffmpeg: This fixes out of bounds accesses for big endian formats and should be
[21:40:47] <CIA-63> ffmpeg: a little faster.
[21:40:47] <CIA-63> ffmpeg: mru * r25111 /trunk/libavutil/pixdesc.c: pixdesc: indent
[21:45:57] <MIH1406> Hi, Can I use FFmpeg in my closed source application?
[21:46:15] <mru> ask your lawyer to read you the license
[21:46:58] <MIH1406> I do not have alower
[21:47:05] <mru> but we do
[21:47:13] <MIH1406> a lawyer*
[21:47:14] <mru> so watch your step
[21:47:32] <MIH1406> I am watching
[21:48:49] <MIH1406> I thought that I can use any LGPL-licensed applications in my closed source (property) applications
[21:49:02] <mru> if you follow the rules, yes
[21:49:46] <MIH1406> Let suppose I am using the LGPL parts only. Can I use them in my application and but it?
[21:50:02] <MIH1406> and sell it*
[21:50:13] <mru> if you follow the rules layed out in the license, yes
[21:50:48] <mru> now I don't know your app, nor am I a lawyer, so I'm not qualified to answer any such questions
[21:51:25] <MIH1406> I am just asking, I do ot have start any app. I am just dream of an app!
[21:51:47] <iive> IANAL in short, provide your modifications, allow every user to replace the ffmpeg with one compiled from your sources.
[21:51:48] <MIH1406> I want to create a GUI for using FFmpeg to convert media only.
[21:52:17] <mru> have you read the legal section on our website?
[21:52:36] <mru> there's some advice there
[21:52:52] <mru> but if you're uncertain, you should talk to a real lawyer
[21:53:14] <BBB> http://ffmpeg.org/legal.html specifically
[21:53:44] <BBB> what's amazing is that one particular loop unroll (add16) leads to a 25% speed increase of that particular code
[21:53:54] <BBB> that was the only function where gcc was faster
[21:53:59] <MIH1406> what if I am in a country that does not have a low for cracking or violating software licenses?
[21:54:05] <BBB> I'm still not quite sure why a loop unroll is so much faster there
[21:54:19] <BBB> (as opposed to other places where it has no measurable effect and makes the code far bigger)
[21:54:21] <mru> MIH1406: I see you're in saudi arabia
[21:54:32] <mru> MIH1406: if you're not careful you'll get 30 lashes or something
[21:54:33] <MIH1406> Yes
[21:54:41] <MIH1406> lashes?
[21:54:52] <BBB> :-D
[21:54:57] <mru> or perhaps a light stoning
[21:55:11] <MIH1406> Lash, Afghanistan, a place in Afghanistan
[21:55:36] <mru> whip-lashes
[21:56:42] <MIH1406> These are not in Saudi Arabia in real life!
[21:56:47] <mru> your country is often seen as barbaric in the west
[21:56:55] <mru> but what do I know, I've never been there
[21:57:43] <MIH1406> It depends on the definition of "barbaric"
[21:58:00] <mru> anyway, regardless of the local law, it would be polite of you to do as we wish regarding the license
[21:58:33] <MIH1406> I respect licenses
[21:58:39] <mru> good
[21:58:43] <MIH1406> I am using Linux because that
[21:59:09] <mru> start by reading the page BBB pointed at
[21:59:09] <MIH1406> Many (Most) People here do not even know that (cracking) is illegal
[21:59:36] <MIH1406> all shops in saudia arabia are selling Windows Ultimate for less than 10 $
[21:59:47] <MIH1406> I wish I am a lawyer for Microsoft
[21:59:50] <mru> still more than it's worth imo...
[22:00:00] <MIH1406> I will read it
[22:00:03] <MIH1406> Thank you
[22:00:26] <MIH1406> By the way, are there any other applications like FFmpeg?
[22:00:38] <mru> I don't think there's anything quite like it
[22:00:42] <MIH1406> A fully LGPL-ed one
[22:01:17] <MIH1406> what about a property one?
[22:01:26] <MIH1406> what about a property library?
[22:01:40] <CIA-63> ffmpeg: stefano * r25112 /trunk/libavutil/pixdesc.c: Apply minor simplification in av_read_image_line().
[22:02:07] <kierank> no there is nothing similar that is proprietary
[22:02:13] <mru> I hardly consider that simpler...
[22:04:10] <MIH1406> Back to lashes, I am with the Islamic rules, they are Excelant but they are not applied 100% in real life and this is very sad! (To say I am Islamic Country, Without following Islamic Rules!!!)
[22:04:11] <MIH1406> Goodbye
[22:04:55] <kierank> the rules are not excellent if you are at the receiving end of a punishment
[22:09:31] <MIH1406> kierank: If someone killed your kid, is it enough to put him in a prison for some years?
[22:10:40] <MIH1406> Ask a parent who lost their kid by a gun of a crimnal do not answer this while you are sleeping in a safe place with you kid between you and your love!
[22:11:00] <MIH1406> Sorry this is not a place to discuss this
[22:11:01] <MIH1406> bye
[22:11:04] <iive> i prefer that rule: do to the others the same thing what you want to be done to you.
[22:11:41] <MIH1406> have you lost a kid, loved person, a father?
[22:11:49] <kierank> an eye for an eye for an eye makes everyone go blind
[22:12:33] <iive> MIH1406: you are trying to invoke emotional response. Actually there is a study that lesser punishments lead to less violent crimes.
[22:13:16] <iive> if a criminal kills somebody during a robbery, he is already doomed, so he is more likely to kill everybody in sight.
[22:15:14] <MIH1406> "you are trying to invoke emotional response" No I just want to let you think about the lose
[22:16:24] <MIH1406> http://www.islamonline.net/servlet/Satellite?pagename=IslamOnline-English-Ask_Scholar/FatwaE/FatwaE&cid=1119503544052
[22:16:57] <MIH1406> Christian friend who asked me why punishment in Islam like stoning to death, cutting the hand off, etc., is so cruel and inhumane?
[22:17:35] <kierank> you should take that to another channel
[22:17:45] <iive> indeed.
[22:18:19] <MIH1406> bye
[22:18:38] <BBB> bye
[22:19:02] <kierank> do you plan to leave this time after saying bye
[22:19:30] <MIH1406> Dispute does not lead to hate
[22:19:44] <MIH1406> OK?
[22:24:38] <BBB> uhm... that was a little weird
[22:25:15] <BBB> I wonder if I should've brought up that I went to that anti-mosque gathering yesterday near ground zero (not because I do or don't support it, but just to see what kind of people go to that kind of stuff, "sniff real american society")
[22:25:39] <BBB> oh, and because the dutch mp Wilders was there too, of course :-p
[22:26:00] <J_Darnley> He was?
[22:29:55] <iive> BBB: now i'm curious what happened...
[22:46:04] <BBB> J_Darnley: yeah, he was
[22:46:57] <BBB> it was big news in the netherlands and it was (in 1-2 sentences) in the new york times frontpage article
[22:47:01] <BBB> so not a small deal
[22:47:25] <BBB> but it was fun, a bunch of old, white people talking trash about muslims and tolerance
[22:47:52] <BBB> and two blocks from there, a bunch of teenagers who haven't seen the world yet talking equal trash about how a mosque there will solve world hunger and terrorism and stuff
[22:48:00] <BBB> I wasn't sure which of the two was worse
[22:48:05] <BBB> but it was pretty awful :-p
[22:52:06] <mru> well, that fixed that crash on sparc/bsd
[23:02:31] <mru> BBB: http://www.apenotmonkey.com/2010/08/25/building-a-community-center/
[23:02:49] <mru> also http://www.apenotmonkey.com/2010/08/23/mosque-controversy-for-conspiracy-theorists/
[23:08:54] <Dark_Shikari> BBB: those people are hilarious
[23:09:01] <Dark_Shikari> especially given that
[23:09:02] <Dark_Shikari> a) it isn't a mosque
[23:09:05] <Dark_Shikari> b) it's not near ground zero
[23:09:48] <Dark_Shikari> then again, it's just the generic US media distraction of the week
[23:09:58] <Dark_Shikari> where the corporate media invents a controversy to distract people from actual issues
[23:25:55] <iive> Dark_Shikari: looking at the politics in your country it would seem people are interested in mosque, gay marriage and abortions. I wonder if these could be combined in a single story...
[23:26:29] <Dark_Shikari> iive: there is a small vocal constinuency of people who are interested in whatever the TV tells them to be
[23:26:32] <Dark_Shikari> the rest don't care
[23:26:39] <Dark_Shikari> And yes, they can be combined.
[23:28:12] <iive> Dark_Shikari: i'm more puzzled by the proportions in is blown out on TV.
[23:28:33] <Dark_Shikari> 09:09 <@Dark_Shikari> where the corporate media invents a controversy to distract people from actual issues
[23:28:36] <kierank> for people in the uk s/TV/Daily Mail
[23:28:46] <Dark_Shikari> it's the concept of an invented controversy
[23:28:58] <Dark_Shikari> For many people, it's dangerous to talk about actual issues, for fear you might be proven wrong.
[23:29:05] <Dark_Shikari> Therefore, you invent a controversy that doesn't exist and talk about that instead.
[23:29:21] <Dark_Shikari> it's simliar to the concept of a "pundit"
[23:29:39] <Dark_Shikari> opinionist: writes/talks about his opinions.  Congratulates politicians when they do what he wants, insults them when they don't.
[23:29:53] <Dark_Shikari> Pundit: picks a target, and always disagrees with that target.  No matter what the target does, he did the wrong thing.
[23:30:13] <Dark_Shikari> That is, if the target has to pick between A and B, and picks A, he should have done B.  But if he had picked B, the same pundit would have said he should have done A.
[23:30:26] <iive> sounds like obama.
[23:30:33] <Dark_Shikari> yes.  he's a common target
[23:30:35] <Dark_Shikari> example
[23:30:56] <Dark_Shikari> a while back, there was an invented controversy over obama not speaking to the Boy Scouts, and instead speaking to some other group on the same day
[23:31:03] <Dark_Shikari> But of course, suppose he had gone and spoken to the boy scouts
[23:31:11] <Dark_Shikari> the same pundits would have raked him over the coals for "indoctrinating american kids"
[23:31:20] <Dark_Shikari> And we know they would, because they did a few years ago when bush did it, too.
[23:31:48] <iive> well, their brown uniforms doesn't help either.
[23:31:56] <Dark_Shikari> it's all about invented controversy and punditry, because people are scared to talk about anything of substance.
[23:33:38] <iive> "people are scared" this is never good sign.
[23:34:11] <kierank> it is also easier to appeal to people's emotional side than to appeal to their rational side
[23:34:48] <iive> ok, maybe we should stop talking about fox news.
[23:35:29] <kierank> talk about the software equivalent, that is xiph
[23:44:09] <iive> n8 ppl.
[23:53:48] <BBB> Dark_Shikari: I think you're correct re: media, and that's a very unfortunate thing
[23:53:50] <BBB> :(
[23:53:53] <BBB> ohwell
[23:53:57] <BBB> just don't watch foxnews
[23:57:43] <ohsix> irony: 16:35 < kierank:#ffmpeg-devel> talk about the software equivalent, that is xiph