Open in new window / Try shogun cloud
--- Log opened Thu Nov 28 00:00:43 2013
-!- iglesiasg [] has joined #shogun00:33
-!- mode/#shogun [+o iglesiasg] by ChanServ00:33
-!- iglesiasg [] has quit [Quit: Leaving]01:37
shogun-buildbot_build #631 of nightly_default is complete: Failure [failed test notebooks]  Build details are at
-!- Saurabh7 [~Saurabh7@] has quit [Remote host closed the connection]05:14
-!- sonne|osx [] has joined #shogun05:51
-!- sonne|osx [] has quit [Client Quit]05:51
-!- sonne|osx [] has joined #shogun05:51
-!- sonne|osx [] has quit [Quit: sonne|osx]06:21
-!- lisitsyn [~lisitsyn@] has quit [Quit: Leaving.]08:45
-!- besser82 [~besser82@fedora/besser82] has joined #shogun09:42
besser82sonne|work, wiking: Guten Morgen  :D09:42
besser82wiking: are you able to login into obs, today?  http://build.opensuse.org09:45
@wikingbesser82: noup... it does not complain about user/pass being wrong or not existing09:51
@wikingit just processes the login information09:51
@wikingand then nothing09:51
besser82wiking: mhhhh....09:52
@wikingi get back to the same page09:52
@wikingand  Sign Up | Log In  is still there09:52
@wikingsuse(r) ;09:52
@wikingand i can keep on requesting login09:52
@wikingbut nothing happens it's the same09:53
besser82mhh... sounds weird....09:53
besser82I'll check that with me SUSE buddies  ;)09:55
@wikingheheh i'm sorry but this is just wrong :)09:56
-!- new_lido [~walid@] has quit [Ping timeout: 272 seconds]09:57
besser82wiking: you can login now  :D10:05
besser82wiking: and now you are a member of science:shogun-project  ;)10:06
besser82wiking: maintainer actually10:06
@wikingbesser82: what was the problem?10:07
-!- lisitsyn [] has joined #shogun10:07
besser82wiking: someone changed the way user are created on osc....10:08
besser82wiking: now new users need to login first using the osc-client to be created on obs  :(  Sh17 h34|)5   XD10:08
besser82^^ those obs developers10:09
besser82lisitsyn: plz, give me your novell-credentials on pm, so I can activate your login, too ;)10:09
lisitsynbesser82: activate like?10:10
-!- new_lido [~walid@] has joined #shogun10:11
lisitsynbesser82: I mean I can log in :D10:11
besser82lisitsyn: activate like you can login to OBS10:11
lisitsynisn't it enough?10:11
lisitsynah let me try to login to obs10:11
besser82lisitsyn: and I can add you as a maintainer to science:shogun  ;)10:12
lisitsynbesser82: ohkay so what do you mean by credentials? :D10:12
besser82lisitsyn: novel-account  user/passwd10:13
lisitsynbesser82: okay let us change it later then :D lisitsyn/123abc10:14
besser82lisitsyn: kk, thanks!10:15
besser82lisitsyn: done && added  :D10:16
lisitsynbesser82: thanks, can I change it now?10:16
lisitsyn123abc which was published online is a bit unsecure10:17
besser82lisitsyn: sure, go and change that high-level-security-passwd  ;)10:18
lisitsynbesser82: I am curious if street address 'Shitfactory' is ok for them10:18
lisitsyntelephone number 777 haha10:19
@wikingbesser82: patch?10:21
besser82wiking: patch, what?!?10:25
besser82lisitsyn: Why not?  Mine is General T'So Alley  ;)10:26
besser82wiking: I'm puzzled....10:27
@wikingbesser82: the crazy cmake patch10:28
besser82wiking: I'm still working on that...10:28
besser82wiking: getting the Modules build and testsuite run from build-tree10:28
-!- sonne|osx [~sonne@] has joined #shogun10:43
besser82sonne|osx: Guten Morgen  ;)10:45
sonne|osxbesser82: to you too :)10:45
sonne|osxbesser82: I read you do 100000 times compile speed ups?10:45
sonne|osxmuch appreciated :D10:45
besser82sonne|osx: not that much, but ~600% on rebuid with ccaches10:46
lisitsynhow's that?10:46
besser82sonne|osx: and ~70% without10:46
lisitsyn*just curious*10:46
besser82lisitsyn: besser82-magick(tm9  :-P10:46
sonne|osxbesser82: so when can I haz it?10:47
besser82sonne|osx: as soon as the swig-module are building and testuite runs from build-tree  ;)10:47
besser82sonne|osx: I think during this week10:47
besser82sonne|osx: currently I'm working to get the modules build with blazing speed-up10:48
sonne|osxbesser82: so long way still - swig is not so easy to get to work...10:48
sonne|osxbesser82: and we always needed the stupid -Dfoo on g++10:48
sonne|osxand siwg10:48
sonne|osxfor some reason it totally ignored our lib/config include attempts10:48
sonne|osxbesser82: modules == the interface right?10:49
besser82sonne|osx: there's ways to get around that  ;)  If you know the deep internals of cmake....10:49
besser82sonne|osx: right modules= lang-interface for py,...10:49
sonne|osxbesser82: no I mean when you call swig <somefile.i>10:49
sonne|osxand somefile.i includes the config.h10:49
sonne|osxthen for some reason the defines are not known in there10:49
sonne|osxand don't ask me why not10:50
besser82sonne|osx: broken / dirty-hacks cmake-scripts is 90% of your problems....10:50
besser82sonne|osx: broken = it does build, but you ship stuff around of cmake's internal bookkeeping....10:51
besser82lisitsyn: the most speed-up is gained by splitting the BIG glob of *.cpp into several obj-lib and later linking them back to one single big libshogun10:52
besser82lisitsyn: the next big speed-up is from moving the headers out of sources  mv src/shogun/*/*.h ---> includes/shogun/10:53
* sonne|osx likes that10:53
sonne|osxsound like x-mas :)10:54
besser82sonne|osx: actually the whole libshogun builds in < 1:30 mins without ccache10:54
besser82sonne|osx: rebuilds from ccache are ~25 sec.10:55
sonne|osxbesser82: libshogun rebuilds?10:55
sonne|osx25s is due to the linking steps?10:55
besser82sonne|osx: like doing a commit and build that new checkout with ccaches10:56
besser82sonne|osx: yes, the longest part is linking10:56
sonne|osxbesser82: ahh ok and if you just recompile w/o change?10:57
besser82sonne|osx: linking actually takes ~12 secs  / 50%10:57
besser82sonne|osx: that's done in less 15 secs.10:57
besser82sonne|osx: times without configure-stage....10:57
besser82sonne|osx: as: cmake -DMY_SETTINGS ...10:58
besser82sone|osx: pure `make all`10:58
sonne|osxI can live with that :)10:59
besser82sonne|osx: is there any specific reason not to use arc4random on linux?11:00
@wikingbesser82: ?11:01
besser82wiking: !11:01
@wikingbesser82: arc4random is just seeding the prng11:01
besser82wiking: so it is superfluous on linux i guess?11:02
besser82wiking: because of other function in stdlib11:02
besser82wiking: doing the same11:03
@wikingwe have our own prng11:03
@wikingwe just need a good seed for the prng11:03
@wikingand that is by arc4random if possible11:03
@wikingif not then /dev/[u]random11:03
besser82wiking: so we should use that on linux as well11:03
@wikingyep anywhere it's possible11:03
@wikingjust as it is now in cmake11:04
besser82wiking: that way it would be skipped on linux and is used on bsd / oscrap only11:04
@wikingbesser82: why?11:04
besser82wiking: which on linux always report not found11:05
-!- sonne|osx [~sonne@] has quit [Quit: sonne|osx]11:05
besser82wiking: because on linux one would need to look-up libbsd and run the check against that....11:05
besser82wiking: arc4random is a bsd-stdlib thing....11:06
@wikingand that's only available if u haev libbsd installed?11:06
besser82wiking: yes, at least on linux11:06
@wikingthen leave it as it is11:06
@wiking/dev/urandom is fine as well11:07
besser82wiking: kk...11:07
besser82wiking: But I could have added the check in new cmake with few lines11:07
@wikingnah not necessary11:07
@wikingurandom should be there in linux11:07
@wikingand that's good enough as well11:08
besser82wiking: which is better arc4random or urandom?11:08
@wikingdoesn't matter really11:08
@wikingas we call it once...11:08
besser82wiking: as you say....11:08
@wikingor say 2-3 times max11:08
@wikingand in that aspect both of them are good enough seeds11:08
besser82wiking: alrighty!11:08
besser82wiking: on the other hand libbsd is cheap by means of check && size...11:09
besser82wiking: it's ~60 kbytes and has pkg-conf on board11:09
@wikingbesser82: really that's just unecessary11:10
besser82wiking: ok11:10
-!- sonne|osx [] has joined #shogun11:21
-!- sonne|osx [] has quit [Quit: sonne|osx]11:44
-!- new_lido is now known as new_lido_afk12:43
-!- iglesiasg [] has joined #shogun13:09
-!- mode/#shogun [+o iglesiasg] by ChanServ13:09
-!- sonne|work [] has quit [Read error: No route to host]14:04
-!- iglesiasg [] has quit [Quit: Leaving]14:05
-!- sonne|work [] has joined #shogun14:44
besser82sonne|work: I have some good news  :D16:03
besser82sonne|work, wiking, lisitsyn: Actually I improved compile time ~8-times with da modules and without ccaches  :D16:03
besser82sonne|work, wiking, lisitsyn: in values that is 15 Minutes with current HEAD and on my improvement / changes ~2 minutes (using -j8)16:04
lisitsynbesser82: cool16:05
sonne|workerrm? whats the magic behind this?16:05
besser82sonne|work, wiking, lisitsyn: with da ccache it takes < 40 secs and modules enabled16:05
besser82sonne|work: extreme optimizations16:05
lisitsynbesser82: like? ;)16:06
lisitsynthis would be very important16:06
lisitsynbecause currently development is like blocked :D16:06
besser82sonne|work, lisitsyn: external header guards and all defines put into config.h and used from there16:06
lisitsynwith compile times16:06
sonne|workbesser82: and you are sure that the defines from config.h are used?16:07
besser82sonne|work: absoluteliy16:07
sonne|workbesser82: what did you do that the defines from config.h are being used?16:08
besser82sonne|work: included them properly in the $swig_file.i16:08
lisitsynnot sure what you mean by that16:08
besser82sonne|work: and autogenerating some of that .i16:09
besser82sonne|work: like having $  ----> $swig.i16:09
besser82sonne|work: and making cmake: configure_file( in out )16:10
besser82no external defines needed anymore  :D16:10
besser82lisitsyn: what exactly?16:11
lisitsynbesser82: so what did you put to config.h?16:11
sonne|workthat really sounds super cool - I will have a thorough look once your patch is ready16:11
besser82lisitsyn: basically every define done by cmake16:12
besser82sonne|work: will push my first stage to my gh-fork during this eve16:13
lisitsynbesser82: but why does it change stuff that significantly?16:13
lisitsynare defines that slow?16:13
besser82lisitsyn: defines on the compiler-cli actually slow down gcc significatly and shred ccaches useless....16:14
lisitsynbesser82: so it matters a lot whether I put 20 defines or 40 defines to gcc while compiling some .o?16:14
sonne|workwhere do you learn such stuff?!16:14
besser82lisitsyn: for one file no, but for a 20 or more yes.  Explicitly when using ccache16:15
lisitsynbut ain't that sound stupid for gcc?16:15
lisitsynbesser82: is there any define that is changed each compilation?16:15
besser82sonne|work: self-studies  :D16:15
lisitsynis that the thing that makes ccache useless?16:15
besser82sonne|work: but i can give enterprise trainings for some $$$$  XD16:16
besser82lisitsyn: for ccache every cli-define is basically a fresh compile16:16
sonne|workbesser82: better mentor some gsoc student doing such magic :)16:17
lisitsynbesser82: why?16:17
sonne|workbesser82: that would explain the slowdown16:17
besser82lisitsyn: because it doesn't cache cli-define16:17
lisitsynbesser82: essentially what you did looks like that stdafx.h shit from visual studio16:17
lisitsynbut I'd never expect it matters that much..16:17
besser82lisitsyn: Visual Studio?!? Who uses that?!?16:18
lisitsynbesser82: welcome to cruel world16:18
lisitsynI have to compile some project in VS even though I develop it in linux16:18
besser82lisitsyn: from gcc-pov cli-defines need expensive checks for possible collisions with included headers...16:18
lisitsynbesser82: well that explains *a bit*16:19
lisitsynbut expensive?16:19
besser82lisitsyn: use mingw  ;)  It's about da 10-times beddah u know?16:19
lisitsynbesser82: why?16:19
besser82lisitsyn: because M$ doesn't gain any $$$ for that  ;)16:20
lisitsyntarget os is windows why to run it using that shit? :)16:20
besser82lisitsyn: mingw will produce native ddl16:20
besser82lisitsyn: or exe16:20
lisitsynthat's way too fanatic for me, I have VS here to compile, thats easy and works16:21
lisitsynand hey VS is quite advanced IDE actually, its debugger is very powerful ;)16:21
lisitsynwe'd have to live with that but there is nothing free that is as good as it is16:22
besser82lisitsyn: expensive, because gcc preferes the cli-define over header-define and preprocessor needs to override that into every header. after that gcc needs to check whether all is still sane, ......16:22
besser82lisitsyn: use mingw from ECLIPSE  :D16:23
lisitsynbesser82: but that's like operating on a few sets of strings right?16:23
besser82lisitsyn: unfortunately no...16:23
besser82lisitsyn: much more complicated....16:23
lisitsynwell okay, probably I overlook some additional complexity here16:24
besser82lisitsyn: and the worst is: it bypasses ccaches16:24
lisitsynbut looks easy, any typical application does more advanced stuff and it doesn't take 15 minutes you know :)16:24
besser82lisitsyn, sonne|work: with the right setup I think, i can gain another 150% - 200% ; like getting it about da 10-times faster without ccaches16:25
besser82lisitsyn, sonne|work: and limiting the speed with ccaches to the output rate of your tty16:25
sonne|workthis will save a few trees :)16:26
besser82lisitsyn, sonne|work: so one could build the whole stack with all(!!) modules within < 2 minutes on recent hw16:26
besser82lisitsyn: ever build libreoffice from sources?!?16:27
besser82lisitsyn: it's da 5|-|1716:27
sonne|workit is what?16:28
besser82sonne|work: shit16:29
besser82sonne|work: |_1ZZ4|2|)57`/|_3 you know?16:30
besser82sonne|work: lizzardstyle  ;)16:31
besser82sonne|work: I usually use that to `crypt` swears in irc  ;)16:32
besser82sonne|work: as iglesiasg, he can read that, too  :D16:32
besser82lisitsyn: actually one can make compilation of every code blazingly fast with just some small, but additional work / steps16:34
@wikingok but basically the downplay of this is better use clang ;)16:35
lisitsynbesser82: yeah what about clang16:35
lisitsyndoes it help with normal compiler?16:35
besser82wiking: the comeback is: better use brains on planning a project's build-sys.  :-P16:36
@wikingbesser82: noup16:36
@wikingbesser82: sorry but that gcc is not handling well .cpp and .h in the same place16:36
@wikingit's really gcc specific16:36
besser82lisitsyn: I think there should be some speedup, too...16:36
@wikingas u told yesterday16:36
besser82wiking: which compiler is more common && versatile? clang or gcc?16:37
@wikingbesser82: nowadays?16:37
@wikingi would say llvm16:37
lisitsynI'd say clang is better now16:37
@wikingi mean sorry but since gcc went gpl316:37
besser82wiking: what's wrong about that?16:37
@wikinga lot of entities out there went on developing llvm16:37
@wikingit's not that it's wrong or not16:38
@wikingit's just the facts16:38
@wikingllvm was nowhere when gcc was 4.216:38
@wikingand gpl216:38
@wikingand suddenly when gcc went gpl316:38
@wikingllvm boomed16:38
@wikingsonne|work: pure coincidence of course :)16:38
besser82wiking: but actually in most enterprise envs there is gcc16:38
besser82sonne|work: there no connection between that  :-P16:39
@wikingbesser82: u can roll llvm an a toaster16:39
besser82wiking: noup.  My 6502 doesn't like llvm  :(16:39
besser82wiking: but handles well with gcc  :-P16:40
besser82wiking: the comeback: if you need 8-bit you can't use llvm  :(16:41
besser82wiking: and with the right build-sys setup gcc perform as good as llvm / clang...16:42
@wikingbesser82: noup16:42
@wikingbesser82: have u really checked benchmarks of gcc vs clang?16:42
@wikingsorry man but numbers are saying differently16:42
besser82wiking: no, because i heavily use gcc16:42
@wikingbesser82: and hence benchmarks doesnt matter i guess...16:43
besser82wiking: that's the redhat way for at least the next 2 yrs...16:43
@wikingok i really dont want to go into a relegious discussion16:43
besser82wiking: me neither; it's just the redhat-way for me, not more not less16:43
lisitsynallah versus jesus haha16:43
@wikingbesser82: but redhat is just one player amongs so many16:44
besser82wiking: when redhat switches over to clang / llvm in fedora 24 - 26, i'll start using that  ;)16:44
@wikingbesser82: none of us is on redhat's payroll apart from maybe you16:44
lisitsynas llvm is younger I'd say you are compiler-muslim, wiking16:45
@wikingso we really dont give too much damn about rh or any other distrib16:45
@wikinglisitsyn: heheheh16:45
lisitsynand besser82 is like compiler-orthodox-christian16:45
@wikingbesser82: and dont get me wrong... it's a problem that we dont give a damn16:45
besser82wiking: it's all ok  :)  Use whatever you want  ;)16:45
besser82wiking: btw. i'm not on rh-payroll, too  ;)16:46
@wikingbut let's be honest, we all do this semipartime16:46
@wikingso there's best efford16:46
sonne|workfrom what I've seen is that gcc generates faster code than clang...16:46
@wikingsonne|work: depends. if u use openmp of course16:46
lisitsynI measured same thing for tapkee16:46
besser82sonne|work: some actually say the different...16:46
lisitsynclang was faster than gcc16:46
sonne|workthis might not hold for long though16:46
@wikingas clang is a shit with anything openmp related by far16:46
lisitsynand intel c++ compiler was faster than both haha16:46
sonne|workno it was not16:47
besser82lisitsyn: no even close16:47
lisitsynbesser82: ehm have you done my experiments or what? :D16:47
@wikinglisitsyn: let's start using intel c++ and optimmize everything for that :D16:47
lisitsynI just say I got that numbers with my simple experiment haha16:47
lisitsynwhy do you argue16:47
besser82lisitsyn, wiking: who pays the $$$$ for lincense?16:48
lisitsynit is free for open source16:48
lisitsynbut doesn't matter16:48
besser82lisitsyn: i didnt want to argue  :D16:48
besser82lisitsyn: but real speed is archived with intel's proppy c[++]-lib and such....16:49
lisitsynbesser82: yeah as they know their cpus better16:49
lisitsynthat's perfectly clear16:49
besser82lisitsyn: with other means there wouldn't be much benefit with intel's cc16:50
besser82wiking, lisitsyn: actually every cc has it's pros and cons....16:50
@wikingoh well neverming i'm back to my sweet jvm :D16:50
lisitsynhaha yeah16:50
lisitsynjvm is much better anyway16:50
besser82wiking, lisitsyn: but only the openJDK one  ;)16:51
besser82wiking, lisitsyn: who needs proppy stuff  :-P16:51
@wikingbesser82: of course i'm using oracle jdk :D16:51
lisitsynbesser82: no openjdk is the worst one16:51
lisitsynwho needs slow java :D16:51
@wikingopenjdk is just a fucking mess16:51
sonne|workit is like emacs vs. vim16:51
sonne|workor fortran and C indexing16:52
lisitsynoh emacs vs vim is solved16:52
sonne|work0 vs 116:52
lisitsynthey are both outdated16:52
@wikinglisitsyn: :D16:52
sonne|worklisitsyn: you are too young!16:52
@wikingsonne|work: noup he is actually right :D16:52
lisitsynsonne|work: ohh it was true but I am getting older and older16:52
besser82sonne|work: let's collect some fundings and buy lisitsyn a C64 for christmas  :D16:53
lisitsynsome day I'll catch you16:53
lisitsynwho needs old shit, come on16:53
sonne|worklisitsyn: in my grave in my grave!16:53
lisitsynsonne|work: one day we both become useless grandpas :D16:53
@wikinglisitsyn: yeah i mean it's sooo fucking cool that you can run both of them on toaster... but come on, who the hell wants to develop on a toaster :P16:53
@wikingfor sure somebody now will raise their hand :P16:54
besser82wiking: is C64 more smart than toaster?!?16:54
@wikingbesser82: depends which  toaster16:54
@wikinglisitsyn: how much? :)16:55
lisitsynwiking: first time!16:55
@wikinglisitsyn: we've got like 2cm on monday16:55
@wikinglisitsyn: ah u r then slow :)16:55
sonne|workwiking: I would love to see a demo on a toaster16:55
lisitsynwiking: yes stoopid slow winter16:55
@wikinglisitsyn: you are soooooo day before yesterday's news :D16:55
sonne|worknow the Siberian guys become awake!16:56
lisitsynsonne|work: finally I'll get more energy16:56
besser82wiking: these guys are bored much too often  :)16:59
-!- lisitsyn [] has quit [Quit: Leaving.]16:59
besser82sonne|work, wiking:  ^^ current compile-time for libshogun from ccache  :D17:43
besser82sonne|work, wiking:  with serial make && gcc17:44
sonne|workbesser82: excellent!17:44
besser82sonne|work: expect ~30 secs for the build with ALL lang-modules  ;)17:45
besser82sonne|work, wiking:  ^^ without ccache, serial make, gcc17:47
-!- besser82 [~besser82@fedora/besser82] has quit [Ping timeout: 252 seconds]18:05
-!- lisitsyn [~lisitsyn@] has joined #shogun18:21
-!- besser82 [] has joined #shogun18:22
-!- besser82 [] has quit [Changing host]18:22
-!- besser82 [~besser82@fedora/besser82] has joined #shogun18:22
-!- sonne|osx [~sonne@] has joined #shogun18:46
-!- sonne|osx [~sonne@] has quit [Quit: sonne|osx]19:01
-!- new_lido_afk is now known as new_lido_bsy21:44
-!- new_lido_bsy [~walid@] has quit [Ping timeout: 246 seconds]21:51
-!- sonne|osx [] has joined #shogun22:43
sonne|osxbesser82: time machine already invented?22:48
-!- sonne|osx [] has quit [Quit: sonne|osx]22:50
--- Log closed Fri Nov 29 00:00:45 2013