[FFmpeg-devel-irc] IRC log for 2010-08-10

[irc at mansr.com]

[00:40:50] <mchinen> if a line exceeds 80 characters and I can't see how to break it down without breaking one of the indentation rules, is it better to use a temporary variable or leave it hanging over?
[00:44:59] <bcoudurier> hi guys
[00:45:25] <mchinen> hi boudurier
[00:45:41] <mchinen> oops, bcoudurier :)
[00:48:36] <saintdev> peloverde: so how do i read the bottom window in aacx?
[00:52:32] <bcoudurier> hey
[01:05:27] <Dark_Shikari> cool, if I completely dump CQM support from ffh264, I can eliminate an add and a shift in dequant
[01:05:33] <Dark_Shikari> and I can convert the multiplies to use MUL16
[01:05:53] <Dark_Shikari> and the quant data to be int16_t
[02:37:01] <peloverde> saintdev: mdct coefs on a log scale
[02:37:58] <saintdev> the blue lines? the red line is global gain.
[02:38:31] <saintdev> then how does the window line work? is that the window shape in the frequency domain?
[02:44:21] <peloverde> lines that match the windows are scalefactors in the mdct domain
[03:29:59] <Dark_Shikari> well, my ffh264 modification is 28% faster than trunk ffh264 now.
[03:30:33] <Dark_Shikari> it just happens to break...
[03:30:48] <Dark_Shikari> cabac, bframes, weightp, multiref, deblocking, 8x8dct, subpel, inter partitions, interlacing, and CQMs
[03:32:20] <Dark_Shikari> though that is tested on a rather high bitrate clip.  lemme try a lower bitrate one.
[03:32:30] <Compn> we dont need any of that crap anyways
[03:32:31] <Compn> :D
[03:43:21] <Dark_Shikari> up to 34% at 2mbps 800x600
[03:50:08] <Compn> what if you just turn off those features , i wonder how much speed increase there is :P
[03:57:26] <j0sh> Dark_Shikari: sounds like it'd be useful for things like embedded videoconferencing
[04:25:25] <saintdev> peloverde: would you mind a quick review?
[05:33:46] <KotH> moin
[05:34:50] <pJok> servus!
[05:53:09] <pJok> mornings jai :)
[05:54:02] <jai> good morning pJok :)
[05:54:02] <kshishkov> goda morgnar alla manniskor
[05:54:25] <pJok> god morgon, kshishkov :)
[06:03:14] <cartman> moin
[06:39:44] <saintdev> peloverde: aacx seems to display the same scalefactors for all the blocks in a short sequence
[08:38:09] <CIA-93> ffmpeg: mstorsjo * r24754 /trunk/libavformat/rtpenc_xiph.c: rtpenc_xiph: Clarify that num_frames shouldn't ever get larger than max_frames_per_packet
[10:21:22] <Dark_Shikari> siretart: an ubuntu 10.10 alpha3 user is reporting  http://ffmpeg.pastebin.com/b6xUDtZJ
[10:21:31] <Dark_Shikari> Did you go and perform debian crippling to the ubuntu packages again? =p
[11:05:18] <Dark_Shikari> wow, mru, smlabb is a 4-argument instruction??
[11:05:21] <Dark_Shikari> since when does arm have those
[11:05:38] <Dark_Shikari> fuck it I need to get a beagleboard
[11:08:32] <twnqx> you like slow computers?
[11:12:47] <kshishkov> well, standard MLAs take at least three arguments
[11:13:30] <kshishkov> for example, standard ARMv4 MLA has four arguments as well
[13:14:14] <Tjoppen> libswscale/rgb2rgb.c: In function ?palette8tobgr16?:
[13:14:14] <Tjoppen> libswscale/rgb2rgb.c:234: error: implicit declaration of function ?bswap_16?
[13:14:25] <Tjoppen> am I the only one getting this?
[13:14:34] <Tjoppen> made clean and distclean
[13:15:55] <spaam> its only you in the whole world
[13:17:12] <Tjoppen> did some more cleaning out and making without -j4
[13:17:14] <Tjoppen> kaffe
[13:18:00] <spaam> works here
[13:18:23] <J_Darnley> If you're using git, did you remember to update libswscale?
[13:21:43] <Tjoppen> I'm using git-svn. so.. hm
[13:22:45] <Tjoppen> svn up libswscale updated some files. let's see what happens
[13:24:24] <Tjoppen> success \o/
[13:40:31] <Tjoppen> *sip*
[13:41:47] * kshishkov suspects Tjoppen does a favourite thing of most Nordiska people
[13:42:23] <Tjoppen> yep. was in finland this weekend. kahvi \o/
[13:42:36] * kshishkov was in finland a year ago
[13:43:16] <Tjoppen> I was at assembly. I forgot to ask if any ffdevs were there though
[13:43:26] <kshishkov> unlikely
[13:43:52] <wbs> Tjoppen: /me was
[13:44:06] <spaam> \o/
[13:44:13] <kshishkov> Finnish FFmpeg dev subgroup was all present then :)
[13:44:36] <spaam> how many are they? 1 ?
[13:44:49] <Tjoppen> wbs: scener ticket?
[13:44:49] <kshishkov> yes
[13:44:59] <wbs> Tjoppen: yeah
[13:45:20] <Tjoppen> cool. we probably ran into eachother :o   we failed to get our demo finished though
[13:45:34] <wbs> probably, yeah. :-) what group are you in?
[13:45:36] <kshishkov> spaam: FFmpeg has strong Swedish branch though
[13:45:46] <lu_zero> Dark_Shikari: you want aa beagleboard or any arm-board would do?
[13:45:53] <twnqx> Tjoppen: going to alternative party? :X
[13:46:08] <spaam> kshishkov: thats why you like sweden? :)
[13:46:19] <twnqx> (am i the only one who skips assembly because of the stupid "no alcohol" policy...)
[13:46:33] <Tjoppen> indeed. stupid guards..
[13:46:43] <kshishkov> lu_zero: any Cortex-based board should be good for him, BBs are just most accessible, I think
[13:47:09] <Tjoppen> alt sounds interesting.. I'll have to run it by the other three
[13:47:14] <kshishkov> Tjoppen: I heard about Systembolaget too and it's not in Finland
[13:47:30] <Tjoppen> wbs: .feil
[13:47:40] <kshishkov> spaam: no, it's just a wonderful country with nice weather and good food
[13:47:46] <wbs> Tjoppen: oh, don't think I've heard of you
[13:47:47] <lu_zero> Dark_Shikari: I could ask genesi to send you an efika
[13:48:23] <Tjoppen> you probably wouldn't have. we just sort of got together a few entries for dreamhack winter 2009 and decided to go to assembly too
[13:49:02] <Tjoppen> won 1st place in useless utility, 3rd in demos (4th if you count disqualified crew entry)
[13:49:42] <wbs> ah, I see
[13:50:09] <Tjoppen> kshishkov: well, they're only closed on sundays. same as alko
[13:52:28] <kshishkov> Tjoppen: still, it's rather good system. Look at Russia for comparison
[13:52:28] <Tjoppen> the finnish guy behind us suggested we go to a party that's between wasa and helsinki next month. forgot what it's called though. more geared toward sauna/drinking and less serious compos
[13:53:00] <wbs> streem, in tammerfors?
[13:53:10] <Tjoppen> that rings a bell
[13:53:33] <Tjoppen> ah, yes: http://www.mattcurrent.org/stream10/index.html
[13:54:55] <wbs> ah, that's how they spell it, yeah
[13:55:09] <Tjoppen> kshishkov: the male median age is something like 50, right?
[13:55:43] <Tjoppen> considering the price of vodka is less than 1/20th of sweden..
[13:56:50] <lu_zero> uh
[13:56:56] <lu_zero> that pricy/cheap?
[13:57:08] <kshishkov> yes
[13:58:08] <kshishkov> I've never bothered with prices but looks like standard 0.5l bottle of vodka still costs less than 1 Euro in .ua
[13:58:30] <Tjoppen> dworek vodka is 180 SEK (20 EUR). subtract taxes and you end up around 5-10 SEK
[13:58:50] <Tjoppen> for 70 cl
[14:00:52] <Tjoppen> wbs: what group are you in btw?
[14:00:58] <lu_zero> 19 for 70cl here apparently
[14:01:04] <lu_zero> 19euro
[14:01:30] <kshishkov> 2.5 Euro in some Ukrainian internet-shop
[14:02:04] <kshishkov> (and they tend to have prices higher than in retail)
[14:12:16] <Tjoppen> oh, and crinkler ftw. I'm going to give making a 4k game a try. a local company is holding a gamedev competition
[14:33:26] <jai> Tjoppen: you went to assembly eh. nice
[14:33:45] <jai> i thought all the cool kids were using kkrunchy
[14:37:17] <Tjoppen> hm. seems to be for 64k rather than 4k
[14:37:44] <Tjoppen> I'll have to compare them
[14:37:51] <jai> Tjoppen: ah, ok
[14:38:27] <jai> Tjoppen: though both are for PE
[14:38:46] <Tjoppen> indeed. crinkler replacess link.exe
[14:39:04] <jai> hmm, i guess noone writes 4k intros on linux anyway
[14:39:04] <Tjoppen> dunno about kkrunchy though
[14:39:21] <jai> kkrunchy is a "traditional" exe packer
[14:39:35] <jai> like UPX etc
[14:39:45] <Tjoppen> I see. it seems able to make use of various stuff the compiler outputs though
[14:41:07] <Tjoppen> crinkler can do neat stuff like truncating float constants and placing code/data in the PE header
[14:41:42] <Tjoppen> uses 100 megs of RAM for decompression by default :o
[14:41:53] <jai> whoa
[14:42:08] <jai> but that's normal for intros i guess
[14:42:33] <kshishkov> "daddy, is 4096 bytes of code much or not? - it depends whether we write in assembly or C#"
[14:43:12] <jai> i remember some conspiracy 64ks taking up > 2gb ram during init
[14:43:47] <Tjoppen> well, that's probably the program itself, not the decompressor
[14:44:37] <jai> of course, this was the playback engine generating the models,textures etc
[14:45:01] <jai> i meant 100mb for decompressing the code is ok in the bigger picture :)
[14:45:06] <kshishkov> a: \n mov ax, 10000000h \n call malloc \n jmp a
[14:47:08] <jai> kshishkov: but where are the pretty pictures :)
[14:47:31] <Tjoppen> a: push ax \n jmp a
[14:47:46] <Tjoppen> should give you a nice popup
[14:48:59] <kshishkov> some stuff like plasma can be generated quite easily in few instructions
[14:49:44] <jai> rotozoom too
[14:49:59] <jai> maybe even a basic tunnel
[14:50:53] <jai> Tjoppen: planning to include an audio track in your 4k?
[14:51:05] <Tjoppen> maybe. I hear 4klang is good
[14:51:35] <wbs> Tjoppen: i'm not in any group myself actually, but I hang around with the immersion/inversion guys
[14:51:47] <Tjoppen> ah, ok
[14:51:55] <wbs> Tjoppen: and did the platform stuff for a dekadence demo for symbian s60 a few years ago
[14:52:19] <wbs> and an individual mobile demo a few years ago
[14:52:37] <jai> Tjoppen: yep
[14:52:46] <Tjoppen> main goals atm are: get GL window up, poll keyboard via winapi
[14:55:15] <Tjoppen> one of the judges is in our group though, so it might be a bit unfair to enter
[14:57:07] <jai> Tjoppen: why do you need keyboard input?
[14:58:58] <wbs> jai: usually they require you to quit the demo when esc is pressed, or something similar
[14:59:46] <Tjoppen> indeed. but in this case I had a simple game in mind
[15:00:17] <jai> right, k
[15:00:21] <Tjoppen> even 256b's check for esc :)
[15:00:49] <jai> i thought you had a cool framestep mode :)
[15:02:21] <Tjoppen> good idea :)
[15:10:56] <BBB> Dark_Shikari: I've got a corei7 now *blink*
[15:11:20] <kierank> a mac core i7?
[15:11:48] <wbs> BBB: congrats. a bit faster than the previous, I assume :-)
[15:13:37] <BBB> kierank: yes
[15:13:42] <BBB> wbs: oh, boy, incomparable
[15:13:54] <BBB> although network is a little slow, so I'm still running updates
[15:14:46] <wbs> I love the feeling of speed when upgrading. too bad most machines lag down a bit after being in use for some time and the harddrive is filled with weird sdks and whatnot
[15:23:01] <BBB> downloading 2GB of XCode SDK, 1GB of updates and more...
[15:23:02] <BBB> bla
[15:28:30] <Honoome> BBB: who's worse? I got to re-install VS2k8 hoping that the WinSDK 7.1 will notice its presence
[15:28:41] <Compn> BBB : ugh, xcode :V
[15:30:29] <kshishkov> Compn: XCode is not bad
[15:30:39] <kshishkov> even though it uses gcc
[15:31:35] <kshishkov> Honoome: and don't forget to install the latest service pack for VS as well
[15:31:48] <Honoome> doesn't microsoft update take care of that?
[15:32:22] <Compn> after experiencing microsoft update and macosx update, i'd have to say macosx update is much better
[15:32:43] <Compn> but i hate both of them... :)
[15:32:46] <Honoome> I actually don't use the latter that much
[15:33:01] <Honoome> three macs at home, plus three more in service at a customer's...
[15:33:07] <Honoome> better have the update dmgs around
[15:35:04] * kshishkov bought MacMini only for PowerPC inside
[15:38:05] <Compn> its kinda sad, i think osx has the power to beat windows hands down, but apple doesnt seem interested in doing that
[15:38:33] <kshishkov> maybe it's for the best
[15:38:59] <kshishkov> (and maybe it's more effective this way too)
[15:39:30] <Honoome> it's kinda sad, apple actually provides free software improvements (cups, anybody?) and actually shows how you can do commercial/proprietary and free sw without risking too much, and people seem to think worse of them than, say, intel and their poulsbo driver...
[15:40:18] <Honoome> [yeah I'm of the school of "beat them with their own weapons" fights :P]
[15:41:18] <Compn> Honoome : well , i think you will never please some of the free software people unless all proprietary is gone
[15:41:43] <Honoome> which always bring me to ask them what kind of dishwasher/washing machine they use
[15:41:59] <Compn> ?
[15:42:44] <kshishkov> Compn: iive may be the only one who REs their firmware (and replaces it with GPLed one ;)
[15:42:46] <Honoome> most recent dishwashers and washing machines are electronic
[15:43:15] <Honoome> and as such they have a non-indifferent firmware in them, without which they simply don't work...
[15:43:25] <kierank> i would like to hack the firmware on my washing machine
[15:43:43] <kierank> because you can't choose a wash time under 1hr
[15:43:43] <Honoome> a freetard with a modern washing machine with a proprietary firmware is going to be fun to watch mirror-climbing
[15:43:47] <Compn> oh!
[15:43:47] <kierank> you can have 1hr-24hrs
[15:43:58] <Compn> i use old washing machines ahhaha
[15:44:03] <kierank> as if people would want a 24hr wash cycle
[15:44:23] <Honoome> kierank: indeed there would be quite a few cases where I also wouldn't mind being able to play with mine
[15:44:29] <Honoome> or with my espresso coffee maker...
[15:44:43] <Compn> you could hack it to make tea? :P
[15:45:08] <kshishkov> maybe he can hack it to brew coffee
[15:45:25] <BBB> Compn: xcode is nice
[15:45:34] <Compn> so you say bbb
[15:45:45] <Compn> but if there is a problem version of gcc in the toolchain... well.
[15:45:48] <BBB> it's nice to have a useful IDE around my files so I can browse around and so on
[15:46:33] <kshishkov> Compn: there's always some problem with GCC
[15:46:48] <Compn> kshishkov : did you switch to icc ?
[15:46:59] <Honoome> Compn: I would hack it so that I could ask for it to make my coffee when I leave the study
[15:48:01] <Compn> networked coffee maker? mmm
[15:49:10] <Compn> so no one has the old dial based washers anymore?
[15:54:51] <kshishkov> Compn: no, but I'm not RMS to praise GCC for being the best compiler
[15:55:30] <kshishkov> Compn: my grandmother has two
[15:56:34] <KotH> she must be rich then :)
[16:05:34] <BBB> where and how do I download the fate samples?
[16:05:58] <BBB> I'm sure there's some doc/fate.txt or so
[16:08:29] <KotH> rsync samples/something/or/other/fate-suite
[16:11:14] <BBB> isn't there a tarball?
[16:11:21] <BBB> I'm doing wget -r now
[16:11:24] <BBB> I'm sure that's not ideal
[16:11:35] <KotH> which you'd like to download each time something is updated?
[16:11:40] <BBB> just once
[16:11:52] <KotH> you can use ftp too, if you want
[16:11:55] <KotH> or http
[16:12:02] <KotH> though, i'd prefere if you'd use rsync
[16:13:58] <BBB> rsync -r samples.mplayerhq.hu/fate-suite doesn't do anything
[16:14:24] <KotH> it does, it just doesnt tell you :)
[16:14:27] <KotH> use -Pvr
[16:14:50] <BBB> rsync -r rsync://samples.mplayerhq.hu/fate-suite/ fate-suite/ ?
[16:15:27] <BBB> "@ERROR: Unknown module 'fate-suite'"
[16:15:36] <KotH> rsync -Pvr samples.mplayerhq.hu::samples/fate-suite .
[16:16:18] <BBB> ah, it's doing something
[16:16:18] <BBB> thanks
[16:16:22] <KotH> (whether there is a / or not at the end of a path is significant)
[16:17:14] <KotH> aha.. you're working on a mac :-)
[16:17:24] <KotH> oh..
[16:17:32] <KotH> i forgot, add -t too
[16:17:40] <BBB> too late
[16:17:46] <BBB> it's already running
[16:17:50] <KotH> so the next time you rsync it doesnt have to crc check all files
[16:17:56] <KotH> doesnt matter
[16:18:09] <KotH> the nice thing about rsync is, that you can always resume :)
[16:18:32] <BBB> ok, -t added
[16:18:41] <BBB> don't ban my ip please :-p
[16:18:58] <KotH> that's why i said to use rsync :)
[16:19:23] <KotH> the rsync users are usually legitimate, so there is less chance that i'd ban you by accident ;)
[16:20:47] <KotH> BBB: is it me or do you get very little speed?
[16:22:35] <BBB> I'm downloading like 100 things at a time here, so that particular download might be a little slow, yes
[16:22:51] <KotH> with a uni connection?
[16:22:56] <KotH> 0_o
[16:23:11] <BBB> seems so
[16:23:29] <KotH> does cornell live on dial up or what?
[16:23:51] <KotH> on the other hand, the sysadmins there seem to think that icmp echo is evil...
[16:24:23] <kierank> KotH: get ipv6 ;) - usually since nobody's using it ipv6 is blazing fast
[16:24:41] <KotH> kierank: huh? how come ipv6 is faster than ipv4?
[16:25:38] <kierank> some places use separate lines for v4 and v6
[16:25:44] <kierank> and the v6 lines are less congested
[16:27:16] <lu_zero> Honoome: there are opensource washing machine available
[16:27:29] <lu_zero> and apparently they are _quite_ eco-friendly
[16:27:32] <KotH> you mean, not only are we running out of tubes, but the tubes are also clogged?
[16:27:38] <kierank> yes
[16:27:40] <Honoome> lu_zero: where are they supposed to be found?
[16:27:44] <kierank> too many trucks in the tubes
[16:27:56] <KotH> apropos.. i should ask my isp for ipv6 connectivity.. natively
[16:28:48] <lu_zero> oswash.org I think
[16:29:04] <lu_zero> artist+arduino+some kind of washing machine
[16:37:31] <CIA-93> ffmpeg: aurel * r24755 /trunk/libavformat/nutdec.c: get rid of MAX_STREAMS limit in nutdec
[16:50:39] <CIA-93> ffmpeg: aurel * r24756 /trunk/Changelog:
[16:50:39] <CIA-93> ffmpeg: 2 lines spacing between versions in Changelog should be enough for readability
[16:50:39] <CIA-93> ffmpeg: and may avoid some confusion for patch submitters
[16:59:50] <wbs> jenk: you did the rtsp server in perl, right?
[17:00:03] <jenk> yeah
[17:00:27] <wbs> jenk: I tried to use it - broadcasting to it seems to work, but when trying to play it back, I get only this as a reply to SETUP:
[17:00:30] <wbs>  << RTSP/1.0 200 OK
[17:00:32] <wbs>  << CSeq: 3
[17:00:34] <wbs> that is, no Transport: line
[17:00:36] <wbs>  << Session: 2
[17:01:04] <jenk> interesting
[17:01:06] <jenk> what client/server?
[17:01:14] <wbs> ffmpeg as client
[17:01:16] <jenk> are you transmissing an ANNOUNCE and RECORD?
[17:01:27] <wbs> yeah, those work just fine, this is for the client that tries to play it back later
[17:01:42] <jenk> i've tested with vlc and it worked more or less OK
[17:01:54] <jenk> what transport header does SETUP send?
[17:01:57] <jenk> when you make the request
[17:02:08] <wbs> Client listener: >> SETUP rtsp://cone:8554/test.sdp/streamid=0 RTSP/1.0
[17:02:09] <wbs> Got method SETUP
[17:02:09] <wbs> Client listener: >> Transport: RTP/AVP/UDP;unicast;client_port=5004-5005
[17:02:09] <wbs> Client listener: >> CSeq: 3
[17:02:09] <wbs> Client listener: >> Session: 2
[17:02:45] <jenk> that certainly looks chill
[17:02:58] <jenk> are you putting the streamid in manually?
[17:03:02] <wbs> yeah, with vlc it seems to work, more or less
[17:03:30] <jenk> try not adding the streamid to the uri
[17:03:38] <wbs> I don't add it manually, the rtsp layer in libavformat does
[17:03:49] <jenk> ok
[17:03:52] <wbs> you could try ffplay yourself :-)
[17:04:20] <wbs> hmm, yeah, you seem to reply in the same way to vlc, too
[17:04:25] <wbs> but it doesn't mind if there's no transport lines
[17:04:58] <jenk> hrmz
[17:06:02] <wbs> according to the rfc, it's required
[17:06:30] <jenk> yeah, it should be present
[17:07:54] <jenk> what if you add $self->add_resp_header('Transport', $transport); to line 93 of Client/Connection.pm
[17:08:01] <jenk> looks like that may be missing
[17:10:10] <wbs> I added the server_port as in Source/Connection.pm, too - seems to work just fine
[17:14:17] <jenk> what change did you make exactly?
[17:14:22] <jenk> i wanna add these
[17:15:36] <wbs> http://pastebin.org/466213, but it doesn't really seem right, the server_ports in the reply doesn't match the ports from where the packets actually are sent
[17:15:48] <wbs> which makes wireshark unable to parse the rtp packets as rtp, too
[17:17:44] <wbs> don't have time to look at it any more now, but I'll check it out more later
[17:19:10] <jenk> it already adds the ports on line 189
[17:28:13] <CIA-93> ffmpeg: cehoyos * r24757 /trunk/libavformat/utils.c:
[17:28:13] <CIA-93> ffmpeg: Fix mem leak when trying to open a non-existing image file.
[17:28:13] <CIA-93> ffmpeg: Patch by Przemys?aw Sobala, psobala wp-sa pl
[20:26:21] <Dark_Shikari> mru: ping, questions about instruction latency on arm
[20:27:19] <mru> pong
[20:27:56] <Dark_Shikari> 1) which instruction is the 32x32->32 multiply?
[20:28:09] <Dark_Shikari> 2) how does its latency and throughput compare to smulbb and smlal?
[20:28:17] <Dark_Shikari> er, not smlal.  smlabb.
[20:28:40] <mru> it's all in the trn
[20:28:45] <mru> trm
[20:28:47] <roxfan> just mul, if you don't need saturation
[20:29:06] <Dark_Shikari> what is a "trm" and where can I find it
[20:29:09] <mru> plain arm, not neon?
[20:29:14] <Dark_Shikari> plain arm
[20:29:21] <mru> infocenter.arm.com
[20:29:34] <Dark_Shikari> well, http://hilbert-space.de/?p=66 says they're wrong
[20:29:53] <Dark_Shikari> also, is the apple A4 an A9 ripoff or A8 ripoff?
[20:30:09] <roxfan> why "ripoff"?
[20:30:09] <mru> A4 is a8
[20:30:11] <cartman> Dark_Shikari: 9
[20:30:12] <cartman> err 8
[20:30:14] <cartman> pfff
[20:30:18] <mru> not even a ripogg
[20:30:19] <cartman> next version might be 9
[20:30:21] <mru> ff
[20:30:24] <roxfan> it's just some custom samsung chip
[20:30:24] <Dark_Shikari> mru: HAH
[20:30:25] <Dark_Shikari> lol
[20:30:28] <Dark_Shikari> fruedian slip ;)
[20:31:05] <mru> no, android softkbd
[20:31:06] <Dark_Shikari> mru: you can explain probably better than that list though, at least to someone who isn't very familiar with the cortex
[20:31:12] <Dark_Shikari> as the restrictions and dual issue stuff is a bit weird
[20:31:24] <Dark_Shikari> so anyways, compare mul, smlabb, and smulbb.
[20:31:33] <Dark_Shikari> and what's "bb"?  "byte byte"?  it's a 16-bit multiply.
[20:31:44] <mru> botyom
[20:31:48] <mru> t
[20:32:15] <roxfan> http://infocenter.arm.com/help/topic/com.arm.doc.ddi0344k/ch16s02s03.html
[20:32:26] <Dark_Shikari> they're not even listed on the multiply instructions page
[20:32:34] <Dark_Shikari> I know, I'm there
[20:32:46] <roxfan> it's SMLAxy
[20:32:47] <Dark_Shikari> I see no "bb"
[20:32:55] <roxfan> x and y can be t or b
[20:33:02] <Dark_Shikari> ah
[20:33:05] <Dark_Shikari> er, what about "bb"?
[20:33:08] <Dark_Shikari> oh
[20:33:09] <Dark_Shikari> I see
[20:33:23] <Dark_Shikari> wait.  a normal multiply is only two cycles?
[20:33:25] <Dark_Shikari> I thought it was 4?
[20:33:35] <roxfan> maybe on older cores
[20:33:51] <Dark_Shikari> Wait, why does smulbb even exist then?
[20:33:55] <Dark_Shikari> it's not even faster than normal multiply.
[20:34:06] <Dark_Shikari> why do a 16x16->32 multiply if it's the same speed as 32x32->32?
[20:34:15] <Dark_Shikari> Or is this the part where they're lying and it's really one cycle faster?
[20:34:32] <roxfan> so you don't need to shuffle half-words around
[20:35:09] <Dark_Shikari> er... why would you have to shuffle?
[20:35:17] <Dark_Shikari> if you have two 16-bit ints, it's just as fast to do mul as smulbb
[20:35:19] <Dark_Shikari> according to that
[20:35:34] <roxfan> what if you have them packed?
[20:35:51] <Dark_Shikari> then you use simd
[20:35:55] <roxfan> this way you don't need to zero out high parts
[20:36:07] <Dark_Shikari> anyways, mru, the question is whether I should use MUL16 for my multiplies, or *.
[20:36:10] <roxfan> well, smal predates simd
[20:36:22] <Dark_Shikari> I worry it may make gcc dumber for scheduling
[20:36:26] <Dark_Shikari> because it can't schedule inline asm
[20:36:33] <roxfan> and mixing arm and neon can have some issues
[20:36:46] <Dark_Shikari> relevant code:
[20:36:47] <Dark_Shikari> block[*scantable]= MUL16(level[i], qmul[*scantable]);
[22:26:36] <mru> http://bit.ly/bFlcbf
[22:27:02] <kierank> link without bitly crap: http://blogs.arm.com/multimedia/why-opencl-will-be-on-every-smartphone-in-2014/
[22:27:14] <Dark_Shikari> "the killer app for opencl is graphics"
[22:27:15] <Dark_Shikari> IRONY
[22:27:25] <saintdev> lol
[22:28:00] <saintdev> GPUs are well suited to graphics no wai!Q!
[22:39:56] <mru> Dark_Shikari: anyway, the 16-bit multiplies are faster than the manual says, at least in some cases
[22:40:32] <mru> that blog post you linked is correct
[23:11:41] <BBB> mru: can I create a shortcut so that make fate-wmavoice tests wmavoice-7k, 11k and 19k alltogether?
[23:11:57] <BBB> like you did for vp
[23:11:59] <BBB> vp8