**** BEGIN LOGGING AT Wed Dec 12 02:59:58 2012
Dec 12 04:20:49 <jtmitch> Good evening
Dec 12 04:21:29 <jtmitch> I was curious if anyone familiar with using the GPIO's is watching? Not sure I understand when you do or don't use a pullup/down resistor
Dec 12 04:27:23 <jtmitch> nobody eh? Is this room busier at any particular time?
Dec 12 04:33:03 <Russ> ...you use one when the application calls for it
Dec 12 04:34:00 <jtmitch> such as?
Dec 12 04:34:12 <jtmitch> needing to maintain a logical high or low state?
Dec 12 04:34:15 <Russ> when you need a pull-up or pull-down resistor
Dec 12 04:35:03 <jtmitch> I realize it's a noobish question, but a little more clarity is all I was looking for
Dec 12 04:35:26 <jtmitch> Never had someone ask you a question and it makes you suddenly not quite sure of what you know for a second?
Dec 12 04:35:30 <Russ> well it's a logic input, you can't leave it floating
Dec 12 04:35:44 <jtmitch> thanks
Dec 12 04:35:50 <jtmitch> that affirms my prior knowledge
Dec 12 04:36:07 <Russ> If you aren't sure what a pull-up or pull-down resistor is, or what its for, the information is super easy to find
Dec 12 04:36:14 <jtmitch> does the beagle board (I dont' have or use one) have internal pu/pd resistors?
Dec 12 04:36:28 <Russ> iirc, yes
Dec 12 04:36:35 <Russ> https://en.wikipedia.org/wiki/Pull-up_resistor
Dec 12 04:37:07 <thurbad> yes it can be configured as either
Dec 12 04:46:37 * mranostay tosses free virtual Leffe's in the channel
Dec 12 04:48:23 * mranostay fastballs one to Russ 
Dec 12 04:54:51 * Russ wasn't even looking
Dec 12 04:54:58 <Russ> (6 minutes later)
Dec 12 04:56:28 <mranostay> why does LinkedIn new profile have a "Personal Details"
Dec 12 04:57:06 <mranostay> why would i want HR to know questions they can't ask like my age and marital status
Dec 12 07:29:14 <panto> hi
Dec 12 07:31:06 <dm8tbr> good moaning
Dec 12 07:50:29 <mranostay> panto: did it get balder in here!?!?
Dec 12 07:50:36 * panto kicks mr
Dec 12 07:50:38 * panto kicks mranostay 
Dec 12 07:50:41 <panto> ugh
Dec 12 07:53:49 <LetoThe2nd> charming as always :=
Dec 12 07:54:16 <mranostay> yeah i rarely hear that one
Dec 12 07:54:49 <mranostay> but thank goodness for alcohol :P
Dec 12 07:55:23 <LetoThe2nd> mranostay: you're welcome.
Dec 12 07:55:42 <mranostay> LetoThe2nd: you are much less bald
Dec 12 07:55:52 <mranostay> _much_
Dec 12 07:58:08 <LetoThe2nd> \o/
Dec 12 07:58:20 <mranostay> |o|
Dec 12 07:58:23 <mranostay> /o/
Dec 12 07:58:27 <mranostay> \o\
Dec 12 07:58:32 <mranostay> -o-
Dec 12 07:58:34 <panto> but not for long...
Dec 12 08:00:41 <LetoThe2nd> mranostay: something for you to sing along: http://www.youtube.com/watch?v=nRnmpC98xkQ
Dec 12 08:02:01 <mranostay> LetoThe2nd: or http://www.youtube.com/watch?v=7Yg8MuaWMT0
Dec 12 08:02:27 <LetoThe2nd> hrhrhrhr
Dec 12 08:02:48 <mranostay> need to go to Boston for a Patty's Day
Dec 12 08:03:04 <mranostay> and completely not remember a minute of it...
Dec 12 08:04:12 <KotH> JIHAD!
Dec 12 08:04:44 * mranostay breaks a beer bottle over KotH 
Dec 12 08:05:13 * KotH pulls the trigger on his bomb belt
Dec 12 08:06:23 * mranostay pops on Dropkick Murphys list
Dec 12 08:07:29 <LetoThe2nd> mranostay: i was on the very same tour as http://www.youtube.com/watch?v=xEzvM0h0zNc just on sunday ;)
Dec 12 08:10:33 <mranostay> great coding music
Dec 12 08:11:40 <mranostay> i think my coworkers love my earbuds blaring some heavy metal today.. i had to drown out the java talk some how :)
Dec 12 08:15:16 <KotH> java talk?
Dec 12 08:20:50 <mranostay> KotH: i work with java monkeys
Dec 12 08:22:33 <KotH> why would you do that?
Dec 12 08:22:42 <KotH> and dont tell me you're young and need the money
Dec 12 08:24:46 <mranostay> KotH: i'm not in java monkey team thank god
Dec 12 08:25:02 <KotH> you are still near to them
Dec 12 08:25:10 <KotH> the germs are everywhere!
Dec 12 08:25:17 <mranostay> no i'm young and needed "free" relocation
Dec 12 08:25:34 <LetoThe2nd> KotH: germ knödels?
Dec 12 08:25:55 <KotH> mranostay: you're still in the us... so it didnt work out very well... did it?
Dec 12 08:26:06 <KotH> LetoThe2nd: i'm not the knödel type
Dec 12 08:26:15 <KotH> LetoThe2nd: rather iskender kebab, or karni yarik
Dec 12 08:26:23 <KotH> LetoThe2nd: or manti
Dec 12 08:26:47 <LetoThe2nd> germknödel != knödel
Dec 12 08:27:16 <mranostay> KotH: i've lived in bumfuck parts of the US before this :)
Dec 12 08:27:25 <mranostay> panto will confirm :)
Dec 12 08:29:15 <mranostay> KotH: US is a big place only two to three real places to work
Dec 12 08:34:35 <KotH> LetoThe2nd: so ka
Dec 12 08:34:39 <KotH> LetoThe2nd: didn't know that
Dec 12 08:35:51 <KotH> LetoThe2nd: still, rather dibi yandi :)
Dec 12 08:46:19 <mranostay> KotH: i hope you know you are getting us all on watchlists :)
Dec 12 08:46:34 <tasslehoff> koen: is angstrom-v2012.12-yocto1.3 ready for the not-so-general public? some issues making stuff build.
Dec 12 08:47:14 <koen> tasslehoff: it should be
Dec 12 08:52:27 <tasslehoff> koen: http://pastebin.com/nSiMkB7U, *could* be my toolchain/sdk recipe, but I'm not sure
Dec 12 08:53:23 <tasslehoff> ouch. it *is* me
Dec 12 08:58:21 <tasslehoff> koen: hm, no. meta-toolchain-qte doesn't build either
Dec 12 09:26:52 <KotH> mranostay: that's the goal! :)
Dec 12 09:32:06 <Russ> http://www.engadget.com/2012/12/11/ti-audi-jacinto-5-infotainment-system/
Dec 12 09:32:07 <Russ> umm...
Dec 12 09:32:12 <Russ> someone should tell audi
Dec 12 09:32:41 <Russ> oh, it's a dm
Dec 12 09:33:33 <panto> Russ, and qnx...
Dec 12 09:37:33 <koen> ah, jacinto
Dec 12 09:38:14 <koen> tasslehoff: I am currently working on getting the base images working well, I haven't looked at SDKs and such yet
Dec 12 09:38:17 <keesj> the car looks weird
Dec 12 09:38:56 <tasslehoff> koen: console-image also fails for me with the same dependency chain
Dec 12 09:56:09 <tasslehoff> koen: do I need some linaro-toolchain installed?
Dec 12 10:24:54 <yegorich> Hi! I'm trying to change resolution on am335x via fbset. The newest arago kernel provides fb_set_par facilities and it seems as if lcd controller would do the right thing and monitor reports proper resolution, but the picture is wrong. It seems that X doesn't switch it's resolution. Any idea?
Dec 12 11:13:06 <LoseAws> hello everyone, I have a problem, I wanna set wifi in the beagleboard with android2.1, but when I click the wifi, the putty window show me this, init cannot execve(&#39;/system/bin/wpa_supplicant&#39;) permission denied
Dec 12 11:13:38 <LoseAws> how can I correct it
Dec 12 11:22:16 <keesj> LoseAws: check the file is present (eg ls /system/bin) executable ls -l /system/bin/wpa_supplicant, try running it from the command line , check the logs e.g. logcat -d and the kernel logs.
Dec 12 11:28:39 <LoseAws> keesj: thank you for ur answer, i had tried to run it from commond line, it showed the same message, and I can't check the log cuz the device is not here
Dec 12 11:29:41 <LoseAws> it's excutable as well
Dec 12 11:29:45 <keesj> you need to figure out if it's the file is that not executable or something else.
Dec 12 11:30:59 <keesj> if it is something else probably it's probably either related to the kernel (e.g. missing some modules) or to the capabilities given to the process. in booth cases it's best to instrument the code.
Dec 12 11:32:30 <keesj> so you need to figure out if it starts properly (on normal linux I would use strace or similar) but in android I don't think you have that. just start it using the gdbclient
Dec 12 11:34:37 <keesj> something like I written here: https://github.com/keesj/gomo/wiki/AndroidGdbDebugging (that's all I can and want to help with) good luck.
Dec 12 11:35:09 <LoseAws> well, i ll check it tomorrow, thx for ur answer, and thx again
Dec 12 13:59:25 <manta> hi
Dec 12 13:59:44 <manta> i need serious help regarding a beagleboard xm
Dec 12 14:01:05 <manta> so i've been tampering a bit with GPIO pins trying to make a working keypad on it
Dec 12 14:01:52 <KotH> .o0(#beagle, serious business)
Dec 12 14:01:59 <manta> anyway, my roommate thought it'd be a good idea to scare me, so my hand slipped and accidentally shorted some pins on the GPIO header
Dec 12 14:02:45 <manta> now all the poor thing does is two lit up LEDs (D5 and D14) and a 0x00 on serial
Dec 12 14:02:56 <manta> is there any chance to "unbrick" it?
Dec 12 14:03:12 <KotH> change the part that's fried
Dec 12 14:03:25 <_troll_> can you unfry an egg?
Dec 12 14:03:46 <KotH> _troll_: with the right chemicals, of course
Dec 12 14:05:21 <manta> it's most likely the DM3730, right?
Dec 12 14:06:27 <KotH> i dont know what you are measuring, nor do i know the schematics
Dec 12 14:06:31 <KotH> it can be anything
Dec 12 14:06:46 <KotH> most likely the part that is directly connected to the gpio pins, but not necessarily
Dec 12 14:07:03 <manta> and that part is the DM3730
Dec 12 14:07:10 <KotH> the only way to know is to debug your board using a dmm and an oscilloscope
Dec 12 14:11:53 <thurbad> if you removed ALL the solder and still nothing but 0x00 on the serial prognossis is not good
Dec 12 14:12:52 <thurbad> or were you moving wire around on a live board?
Dec 12 14:14:22 <manta> i attached a keypad i made
Dec 12 14:14:34 <manta> what shorted a gpio to ground
Dec 12 14:15:59 <manta> does anyone know on what potential VBAT_MAIN should be?
Dec 12 15:05:33 <cezane_> Hi
Dec 12 15:05:51 <cezane_> did anyone already work with java rxtx lib running on beagleboard/angstrom?
Dec 12 15:06:03 <cezane_> I have an application using the java rxtx lib to communicate through serial port, but after some time running, the reading of serial port 'freezes'
Dec 12 15:06:08 <koen> jkridner_: you'll need to replace the kernel, but give http://dominion.thruhere.net/koen/angstrom/beaglebone/v2012.12/ a try
Dec 12 15:06:18 <cezane_> I already set the flow control to 'none', but is still freezing...
Dec 12 15:06:39 <cezane_> if anyone can help me, please see the code here: http://pastebin.com/UyeyGjZt
Dec 12 15:10:32 <keesj> cezane_: try instrumenting your code (like actually doing something when an exception is thrown)
Dec 12 15:12:21 <KotH> cezane_: or write a small c snipped that reads the serial port ad infinitum and see whether this blocks
Dec 12 15:12:38 * KotH would guess that the problem is java and not the kernel or hardware
Dec 12 15:12:53 <cezane_> hi, kessj... did you see my code here: http://pastebin.com/UyeyGjZt ?
Dec 12 15:13:15 <keesj> yes , perhaps http://embeddedgeeks.wordpress.com/2012/02/26/arduino-beagleboard-serial-communication-using-java-rxtx/ is better to follow
Dec 12 15:14:05 <KotH> wtf...
Dec 12 15:14:15 <KotH> 150 lines to just handle a serial port?
Dec 12 15:14:22 <KotH> and that using a lib?
Dec 12 15:14:25 * KotH is astonished
Dec 12 15:14:27 <keesj> your code looks like http://www.slideshare.net/daltoncezane/how-to-communicate-java-application-via-serial-port-on-beagleboard-running-angstrom#btnPrevious it looks like code from a c programmer trying to do some java :P
Dec 12 15:14:56 <mdp> stupid C programmers
Dec 12 15:17:02 <thurbad> cezane_:  what are you doing on 114.. reitinitalizing your result, and using string concatenation is slow
Dec 12 15:19:18 <cezane_> keesj, this slideshare is mine :) but I am a java programmer and get this code at internet... maybe from a c programmer...
Dec 12 15:19:19 <jacekowski> java sucks
Dec 12 15:19:25 <jacekowski> it would take 3 lines of code in C
Dec 12 15:20:01 <cezane_> thurbad, do you think the problem can be this concatenation?
Dec 12 15:20:36 <thurbad> I just wandered back in.. dunno what the actual problem is
Dec 12 15:20:44 <thurbad> just stating an observation
Dec 12 15:21:22 <keesj> but just ignorring the errors and exceptions is just asking for trouble. first fix that problem
Dec 12 15:22:20 <thurbad> agreed... even printing a stack trace is better than nothing
Dec 12 15:23:23 <thurbad> handling an exception with an empty block is generally a bad idea
Dec 12 15:27:00 <thurbad> also unless you're running 2.6.32 or older you don't want ttyS2
Dec 12 15:32:19 <cezane> I'm back... my connection had been closed...
Dec 12 15:38:08 <KotH> thurbad: why?
Dec 12 15:49:11 <Estevao> hi,  my beaglebone ethernet is not working. When I use the command dmesg | grep PHY, it shows "PHY 0.00 not found PHY 0.01 not found". Someone can help me?
Dec 12 15:51:59 <thurbad> KotH ttyS2 changed to ttyO0
Dec 12 15:52:34 <thurbad> assuming that's what cezane was going for
Dec 12 15:58:02 <KotH> ok
Dec 12 15:58:04 <KotH> thanks
Dec 12 16:02:34 <mranostay> morning trolls
Dec 12 16:03:34 <cezane> sorry, thurbad... I did not understand...
Dec 12 16:03:35 <KotH> günaydin mranostay
Dec 12 16:03:56 <cezane> at beagleboard, the serial port is ttyS2
Dec 12 16:04:03 <thurbad> what kernel are you running?
Dec 12 16:04:10 <KotH> cezane: he answered my question why using /dev/ttyS2 is a bad idea for kernels later than 2.6.32
Dec 12 16:05:25 <cezane> I am using 2.6.32... --> "Linux beagleboard 2.6.32 ... ... armv7l unknown"
Dec 12 16:06:34 <Estevao> hi,  my beaglebone ethernet is not working. When I use the command dmesg | grep PHY, it shows "PHY 0.00 not found PHY 0.01 not found". Someone can help me?
Dec 12 16:07:11 <cezane> strange is the fact that this java application runs ok at a x86 machine, without freeze, but when I put to beagleboard, the problem occurs after some time...
Dec 12 16:11:30 <thurbad> what /is/ the problem?
Dec 12 16:12:39 <thurbad> Estevao: not ignoring.. .just don't know much about the beaglebone... sorry
Dec 12 16:13:13 <thurbad> Estevao: maybe if you posted your bootlog it would get some attention
Dec 12 16:19:06 <cezane> thurbad, do you have experience with java also? my problem is that my application is freezing when it communicates through serial port: it starts to read some data and, after a time, it stops... my code is here: http://pastebin.com/UyeyGjZt
Dec 12 16:23:48 <cezane> I'm back again
Dec 12 17:24:16 <halil> does anybody worked with webcams on beaglebone? I searched web and ask colleges but no help. I have select timeout error with black image read from camera.
Dec 12 17:25:12 <halil> *colleagues
Dec 12 17:30:37 <jpastore> hi. i am a total noob. i was hoping someone can help with a simple test. i want to hook up a thermistor to the beagle bone and sample output to record a timestamp and the voltage. i have no idea where to begin.
Dec 12 17:38:20 <jpastore__> sorry i got disconnected. any help would be greatly appreciated.
Dec 12 17:45:35 <Tartarus> back
Dec 12 17:57:48 <jacekowski> jpastore: you can't just hook it up directly
Dec 12 17:58:06 <jacekowski> as far as i know beaglebone has no way to sample anything anlogue
Dec 12 17:58:23 <jacekowski> so you need some measuring circuitry and then send it to beaglebone
Dec 12 17:58:54 <jacekowski> hmm, i may be wrong
Dec 12 17:59:03 <jacekowski> but you still need measuring circuitry
Dec 12 18:00:20 <jacekowski> jpastore: you need that thermistor and another resistor in series, connected in between ground and 1.8V and midpoint has to go to AIN
Dec 12 18:29:40 <georgem> Yay, just got my bone booting 3.7.0 with ptxdist
Dec 12 18:30:38 <mranostay> time to drink?
Dec 12 18:31:05 <georgem> Sounds like as good as an excuse as any :)
Dec 12 18:48:15 <mdp> I'm in!
Dec 12 18:48:42 <mranostay> i told the bouncer to keep you out
Dec 12 18:48:53 <mranostay> i want my bribe back!
Dec 12 18:49:36 <mdp> tough...go play words with enemies instead
Dec 12 18:49:39 <jpastore> jacekowski, i found an article. but it does not use a resistor. also the path to the ain does not exist. i have ubuntu loaded on the beaglebone. before i receive criticism it was the only method to get certain drivers working.
Dec 12 18:50:04 <mranostay> mdp: 60 points! :P
Dec 12 18:50:13 <jpastore> following the article with meter i see variance in temp. but i cannot find the file to reference ani1
Dec 12 18:50:23 <mdp> mranostay, yes, nice ;)
Dec 12 19:39:21 <ds2> Hmmmm
Dec 12 20:06:00 <mranostay> _troll_: nice post
Dec 12 20:13:51 <_troll_> mranostay: which one?
Dec 12 20:14:40 <mranostay> the beiber one
Dec 12 21:02:43 <unsolo__> hi guys
Dec 12 21:03:05 <unsolo__> i was wondering if someone could lend me a hand on some simple neon gcc inline asm
Dec 12 21:03:40 <unsolo__> a simple example such as a sqrt of a non vectorized double should be fine
Dec 12 21:04:10 <unsolo__> which means we need to stick a scalar into a vector then extract it back out
Dec 12 21:05:32 <unsolo__> anyone up for giving me some pointers ?
Dec 12 21:09:49 <unsolo__> vsqrt.f64 %P0,%P1 : "=w"(retval) :"w"(input) seems to work..
Dec 12 21:09:56 <unsolo__> but thats a vfp version of it
Dec 12 21:11:23 <_troll_> there isn't a neon sqrt
Dec 12 21:12:18 <_troll_> with the right flags gcc will generate scalar vsqrt for sqrt() calls
Dec 12 21:12:21 <_troll_> no need for inline asm
Dec 12 21:12:57 <_troll_> -fno-math-errno -fno-signed-zeros or so should do the trick
Dec 12 21:17:38 <unsolo__> _troll provided you do not have a messed up libm and want to replace it with something fast and glorious
Dec 12 21:18:03 <_troll_> you're writing your own libm?
Dec 12 21:18:08 <unsolo__> jupp
Dec 12 21:18:14 <_troll_> why?
Dec 12 21:18:18 <unsolo__> fun
Dec 12 21:18:21 <_troll_> what's wrong with the existing ones?
Dec 12 21:18:27 <unsolo__> slow
Dec 12 21:18:39 <unsolo__> not inline
Dec 12 21:18:41 <unsolo__> etc
Dec 12 21:19:00 <_troll_> I already said gcc will inline most simple math functions if you give it the right flags
Dec 12 21:19:12 <_troll_> -fno-math-errno is absolutely required for this to happen
Dec 12 21:19:19 <unsolo__> cos sin log atan etc are not very inline
Dec 12 21:19:28 <_troll_> nor should they be
Dec 12 21:19:37 <_troll_> on x86 they are since x86 has instructions for those
Dec 12 21:19:44 <_troll_> arm doesn't have sin/cos instructions
Dec 12 21:19:56 <_troll_> so you need a fairly complicated sw implementation
Dec 12 21:20:11 <_troll_> it ends up large enough that inlining is not a good idea
Dec 12 21:20:16 <unsolo__> i have a 6 madd -> 12.1 digits precision version of cos
Dec 12 21:20:28 <unsolo__> thats not whats in libm
Dec 12 21:20:34 <_troll_> for what range of inputs?
Dec 12 21:20:39 <unsolo__> all
Dec 12 21:20:44 <_troll_> unpossible
Dec 12 21:20:59 <unsolo__> well you do some range fixing
Dec 12 21:21:03 <unsolo__> but thats a single div
Dec 12 21:21:20 <_troll_> that sounds very wrong
Dec 12 21:24:22 <unsolo__> troll if you do a static inline cos(vector x) method of this  and unroll a loop some im quite sure it will be insane http://www.ganssle.com/approx/approx.pdf
Dec 12 21:28:45 <unsolo__> _troll_: i compared my cosf implementation against the one in libm on my corei7 and it performed more or less 1:1
Dec 12 21:28:58 <_troll_> so what's the problem with the libm one?
Dec 12 21:29:05 <unsolo__> well knowing my i7 has instructions for it.
Dec 12 21:29:39 <unsolo__> for my case its simply not compiled with vfp neon.
Dec 12 21:29:55 <_troll_> why don't you compile it that way then?
Dec 12 21:30:03 <_troll_> seems much simpler than rewriting it all
Dec 12 21:30:39 <unsolo__> a i dont need all of it
Dec 12 21:30:59 <unsolo__> b i can cut back on precision as i see fit to reduce computation
Dec 12 21:31:31 <_troll_> I still think you should compile the usual libm with the right flags and only then decide if it's too slow
Dec 12 21:31:47 <unsolo__> hehe i could
Dec 12 21:32:01 <unsolo__> but then i wouldnt have learned all that i have learned :)
Dec 12 21:32:29 <_troll_> are you doing this to learn stuff or to complete a task?
Dec 12 21:32:34 <unsolo__> and i still think a unrolled vectorized cos sounds fun
Dec 12 21:32:39 <unsolo__> _troll both
Dec 12 21:33:50 <_troll_> those goals are often at odds
Dec 12 21:34:20 <_troll_> you do use a ready-made compiler rather than writing your own, right?
Dec 12 21:34:28 <unsolo__> problem is that im stuck with a codesourcery toolchain(dont ask ) and it lacked libm
Dec 12 21:34:41 <_troll_> invalid excuse
Dec 12 21:34:41 <unsolo__> well a vfp / neon libm
Dec 12 21:34:51 <_troll_> you can always build it from source
Dec 12 21:34:59 <unsolo__> indeed
Dec 12 21:36:16 <unsolo__> but now it all works
Dec 12 21:36:29 <unsolo__> and i first had to prove that this was the reason i was struggeling
Dec 12 21:36:47 <unsolo__> so i needed a wrapper just to count number of times each method got called
Dec 12 21:39:49 <unsolo__> btw how good is gcc at re arranging asm .. (eg how much can one gain by inlining methods)
Dec 12 21:40:34 <_troll_> inline asm wreaks havok with the gcc scheduler
Dec 12 21:40:35 <panto> unsolo__, depends
Dec 12 21:40:50 <_troll_> sometimes the result is slower than not having the asm at all
Dec 12 21:40:58 <_troll_> even if you don't get that special instruction
Dec 12 21:41:05 <panto> _troll_, it's not so bad
Dec 12 21:41:08 <unsolo__> _troll there is a difference between asm and static inline in c
Dec 12 21:41:21 <_troll_> panto: yes, it is
Dec 12 21:41:25 <_troll_> unsolo__: wtf does that mean?
Dec 12 21:41:36 <panto> you have to check out the asm output & verify
Dec 12 21:41:54 <_troll_> panto: yes, and sometimes you notice things got worse
Dec 12 21:41:54 <panto> for some cases I've used it for, it was a win
Dec 12 21:42:02 <_troll_> yes, frequently it is a win
Dec 12 21:42:08 <_troll_> but _sometimes_ it's a loss
Dec 12 21:42:15 <panto> well, that's why they pay us the big bucks!... snortle
Dec 12 21:42:17 <_troll_> you _must_ benchmark and check
Dec 12 21:42:30 <panto> listen to the _troll_
Dec 12 21:42:57 <unsolo__> _troll_: indeed my experience is that the more you inline and unroll the faster things get untill you start hitting cache issues
Dec 12 21:43:35 <_troll_> the trick is knowing when to stop
Dec 12 21:43:44 <_troll_> branches are really very cheap on modern cpus
Dec 12 21:43:44 <unsolo__> nevah
Dec 12 21:44:03 <unsolo__> true
Dec 12 21:44:22 <panto> fwiw, if you don't _need_ to use simd, or some weird instruction, static inline functions without asm can work fine
Dec 12 21:44:25 <unsolo__> but branching also does not leave the compiler to stick things inbetween
Dec 12 21:44:38 <unsolo__> panto: thats what im doing most of the tmie
Dec 12 21:44:44 <_troll_> I recommend unrolling until you can eliminate most pipeline stalls in the loop or until the loop is at least 8 instructions or so
Dec 12 21:45:11 <unsolo__> wouldnt the neon benefit from > 8 instructions in a blow
Dec 12 21:45:21 <panto> instructions are cheap, it's the memory accesses that kill you
Dec 12 21:45:37 <_troll_> yes, and instructions reside in memory
Dec 12 21:45:56 <panto> for tight code, you're going to be in icache anyway
Dec 12 21:46:06 <panto> we're not talking about jitted monsters
Dec 12 21:46:37 <_troll_> unless you can unroll a loop completely, don't exceed 16 instructions
Dec 12 21:46:47 <_troll_> (unless a single iteration needs more, of course)
Dec 12 21:46:54 <unsolo__> example lets say you do 6000 cos calls every 10ms and two of these are adjecent so that a1=cos(x) a2=cos(y) the benefit in this case of a inlined cos is indisputable of course a vectorized one would be even better
Dec 12 21:47:03 <_troll_> if you do that you're insane
Dec 12 21:47:38 <_troll_> calculating two cosines in parallel would of course be nifty
Dec 12 21:47:43 <_troll_> but rather hard to do
Dec 12 21:47:54 <unsolo__> not with the code link i gave you
Dec 12 21:48:01 <unsolo__> it will do it just fine
Dec 12 21:48:10 <unsolo__> with > double precision if you need it
Dec 12 21:48:11 <_troll_> if it were easy it would already be the default
Dec 12 21:48:41 * unsolo__ will write it up for the _troll some other day.. remind me to do that kristian.jerpetjoen@gmail.com
Dec 12 21:49:23 <unsolo__> it should also be possible to do sin tan atan acos etc with same methods
Dec 12 21:49:34 <_troll_> interleaving the calculation of two cosines or whatever is trivial of course
Dec 12 21:49:49 <unsolo__> vectorized they can execute in paralell
Dec 12 21:49:55 <unsolo__> no interleaving
Dec 12 21:50:10 <_troll_> you can't fully vectorise it
Dec 12 21:50:15 <unsolo__> and the interleaving i expect the compiler to do if the cos is inlined
Dec 12 21:50:24 <unsolo__> _troll let me try!
Dec 12 21:50:28 <_troll_> of course
Dec 12 21:50:31 <unsolo__> email me and ill see what i can do
Dec 12 21:50:34 * panto let's unsolo__ try 
Dec 12 21:51:09 <panto> *lets
Dec 12 21:51:17 <unsolo__> :)
Dec 12 21:51:51 <_troll_> in many cases you're better off using a precalculated lookup table
Dec 12 21:52:12 <unsolo__> this has a higher precision
Dec 12 21:52:19 <unsolo__> and its close to just as fast
Dec 12 21:52:19 <_troll_> depends
Dec 12 21:52:34 <unsolo__> but log2 and exp is a different tale
Dec 12 21:52:36 <_troll_> if the input values are known to be from a small set
Dec 12 21:52:44 <unsolo__> at least for the mantissa parts
Dec 12 21:52:44 <_troll_> you can include all in the table
Dec 12 21:52:49 <_troll_> that's how any decent fft workss
Dec 12 21:52:59 <unsolo__> hihi
Dec 12 21:53:03 <_troll_> damn, this keyboard bounces
Dec 12 21:53:24 <unsolo__> on the ps3 i wrote a idct 8x8 (butterfly ) that took 181 instructions to complete..
Dec 12 21:53:39 <unsolo__> which is a specialized fft
Dec 12 21:54:15 <_troll_> I know the 8x8 idct quite well
Dec 12 21:54:19 <unsolo__> it would suprize me if its not faster to write a fft in neon than solving it using lookup
Dec 12 21:54:37 <_troll_> the neon fft _includes_ a lookup table
Dec 12 21:54:44 <unsolo__> and iirc using the neon you can still execute some other instructions
Dec 12 21:55:06 <panto> the lookups will dirty your cache though
Dec 12 21:55:14 <_troll_> uh?
Dec 12 21:55:26 <unsolo__> the L1 will have to poll some from L2 perhaps..
Dec 12 21:55:37 <unsolo__> but if the table is within a reasonable size it will reside in L2
Dec 12 21:56:03 <_troll_> it obviously depends on the size of the fft
Dec 12 21:56:04 <unsolo__> but linear approximation for non linear functions introduces an error..
Dec 12 21:56:05 <panto> which is fine if you're not doing anything else but run the algorithm
Dec 12 21:56:24 <_troll_> for an 1k single-precision fft, the lookup table needs only 4kB
Dec 12 21:56:28 <_troll_> that's well within L1
Dec 12 21:56:40 <_troll_> there's no approximation
Dec 12 21:56:53 <unsolo__> thats a huge table
Dec 12 21:56:55 <unsolo__> ;)
Dec 12 21:56:59 <_troll_> no it's not
Dec 12 21:57:19 * unsolo__ is off to go home and sleep and celebrate that i got this piece of s working ;)
Dec 12 21:57:21 <_troll_> "large" is more than half L1
Dec 12 21:57:26 <_troll_> "huge" is more than half L2
Dec 12 21:57:52 <unsolo__> good thing there is multiple contextes working in L2 ;)
Dec 12 21:58:29 <_troll_> if you're dealing with very large, sparse FFTs there are other tricks of course
Dec 12 21:59:24 <unsolo__> libblas
Dec 12 22:23:09 <djlewis> are you sure you want to click this button?
Dec 12 22:23:35 <djlewis> clicking this button might make system unusable.
Dec 13 01:45:23 <beaglenewb> hello everyone, I had a question if anyone has a few minutes?
Dec 13 01:45:43 <mranostay> *ahem* topic
Dec 13 01:45:48 * Bisqwit points at topic
Dec 13 01:45:52 <prpplague> beaglenewb: you just used your one question
Dec 13 01:45:59 <beaglenewb> rofl
Dec 13 01:46:03 <prpplague> beaglenewb: now you have to wait 24 hours for another one
Dec 13 01:46:33 <beaglenewb> topic
Dec 13 01:46:46 <beaglenewb> no idea what that means?
Dec 13 01:47:23 <Bisqwit> It may be shown as a title bar on your irc window, or you may type /topic to retrieve the topic (which may be rendered in a status window or something)
Dec 13 01:49:01 <beaglenewb> TOPIC
Dec 13 01:49:04 <beaglenewb> damnit
Dec 13 01:49:11 <beaglenewb> i did /topic
Dec 13 01:49:20 <beaglenewb> gave me insufficient arguments
Dec 13 01:49:22 <agmlego> ( don't ask to ask )
Dec 13 01:50:19 <prpplague> beaglenewb: Please read http://beagleboard.org/chat for a guide on how to ask questions
Dec 13 01:51:27 <beaglenewb> yup
Dec 13 01:51:28 <beaglenewb> reading it now
Dec 13 01:51:30 <beaglenewb> thanks
Dec 13 01:58:29 <Bisqwit> "/topic #beagle" on some encumbered clients, then
Dec 13 02:01:01 <Russ> wow, did someone just compare udev/systemd to rape of genocide?
Dec 13 02:01:57 <Russ> er, and, not of
Dec 13 02:02:09 <prpplague> Russ: hehe reading the same g+ post?
Dec 13 02:02:17 <thurbad> was hitler mentioned too?
Dec 13 02:02:23 <beaglenewb> ...
Dec 13 02:02:24 <beaglenewb> lol
Dec 13 02:02:24 <thurbad> :P
Dec 13 02:02:31 <Russ> escalated quickly
Dec 13 02:02:33 <bones_was_here> well if they can justify the merge, i guess that means systemd must be *fucking perfect* ;)
Dec 13 02:02:51 <Russ> does anyone know what the problem is with libkmod and libblkid?
Dec 13 02:03:39 <Russ> bones_was_here, I would imagine it saves the developers a bunch of headache and saves users a bunch of headaches from breakage
Dec 13 02:04:36 <bones_was_here> i would hope so, except that i only found out about the merge because of complaints :p
Dec 13 02:04:41 <bones_was_here> so i'm not confident :p
Dec 13 02:06:27 <bones_was_here> i figure that when debian starts shipping systemd as their default stable init, that will mean it's ready to merge with udev or other key components :)
Dec 13 02:07:11 <bones_was_here> but currently even unstable is sticking with pre-systemd udev...
Dec 13 02:08:25 <Russ> but why?
Dec 13 02:08:29 <Russ> because it's 'icky'?
Dec 13 02:08:39 <Russ> is there some licensing issue?
Dec 13 02:09:29 <bones_was_here> because debian calls itself the 'universal operating system', meaning it needs to work better than sysvinit across a massive amount of use cases, including running a bsd kernel with GNU userland
Dec 13 02:09:59 <Russ> right, but the post-systemd udev still works without systemd
Dec 13 02:10:19 <Russ> and as far as I know, udev doesn't work on bsd anyway
Dec 13 02:10:53 <bones_was_here> bsd has its own equivalent, and (afaik) it's not compatible with systemd
Dec 13 02:11:27 <Russ> I'm really confused here
Dec 13 02:11:32 <CareBear\> bones_was_here : try openorc
Dec 13 02:11:34 <CareBear\> openrc sorry
Dec 13 02:11:41 <Russ> the post-merge udev works with or without systemd
Dec 13 02:11:59 <Russ> but because the repo contains the icky code that doesn't work on bsd, debian can't use it?
Dec 13 02:11:59 <bones_was_here> CareBear\: i'll try openrc when using gentoo, or when debian uses it as default init ^_^
Dec 13 02:12:16 <Russ> for a package that doesn't build on bsd anyway?
Dec 13 02:14:40 <bones_was_here> Russ: distros like debian and gentoo that try to offer support for much greater flexibility and range of hardware platforms than most, can't switch something as critical as the init system as soon as its working ok on basic desktops
Dec 13 02:15:37 <Russ> again, why does going to a newer udev version mean switching the init system?
Dec 13 02:15:59 <bones_was_here> dunno :)
Dec 13 02:16:30 <bones_was_here> but apparently it is causing problems for a number of package maintainers (else i would not even have heard of the merge)
Dec 13 02:16:53 <bones_was_here> i'm sure i can only think of some of the reasons
Dec 13 02:22:55 <bones_was_here> but as with most major changes, the fast/new desktop/laptop focused distros will do it first, the ones that aim to support 10k+ packages in almost any combo on almost any hardware (especially ones with alternate kernels) will take years of testing and bug reports before they're happy enough with the quality
Dec 13 02:24:48 <Russ> why is the newer version of udev a major change?
Dec 13 02:24:57 <Russ> because it builds out of the systemd tarball?
Dec 13 02:25:03 <bones_was_here> i dunno, i don't maintain a udev package for any distro
Dec 13 02:25:05 <CareBear\> why is this on #beagle?
Dec 13 02:25:25 <Russ> I just did a clone of the repo and built a udevd binary on my ubuntu system with zero problems
Dec 13 02:25:42 <bones_was_here> ext4 for example - took years for the conservative distros to adopt (justifiably too)
Dec 13 02:25:45 <bones_was_here> so?
Dec 13 02:26:02 <Russ> it isn't a major revision change for udev
Dec 13 02:28:04 <bones_was_here> you just proved it builds in your ubuntu pc, but what about that old SPARC64 box still chugging along in a DC that needs its 40 daemons to come up in the right order?
Dec 13 02:28:19 <Russ> udevd doesn't bring up daemons
Dec 13 02:28:28 <bones_was_here> no but systemd does
Dec 13 02:28:35 <Russ> we aren't talking about systemd
Dec 13 02:28:41 <Russ> we are talking about udev
Dec 13 02:29:14 <bones_was_here> still, was it tested on all those kinds of use cases?
Dec 13 02:29:27 <CareBear\> stop trolling now
Dec 13 02:29:32 <Russ> no minor release of udev has that kind of testing
Dec 13 02:29:40 <Russ> why would 183 be any different?
Dec 13 02:30:01 <CareBear\> or continue trolling, but somewhere else :)
Dec 13 02:30:57 <bones_was_here> CareBear\: please revise your definition of trolling, the popularly accepted one is that the troll is trying to elict overly emotional responses
Dec 13 02:31:01 <Russ> CareBear\, udev is pretty integral to bb
Dec 13 02:31:50 <bones_was_here> and while i'm aware many people have become overly emotional about this issue, i believe currently we are attempting a rational discussion rather than a flamewar ;)
Dec 13 02:31:54 <CareBear\> bones_was_here : another popular one is that the troll wants to maintain a discussion without actually saying anything
Dec 13 02:32:07 <CareBear\> I see no flaming
Dec 13 02:32:10 <bones_was_here> CareBear\: is that the case here?
Dec 13 02:32:22 <CareBear\> it looks like it to me
Dec 13 02:32:31 <prpplague> hehe funny, if you google "jason kridner" and click images, you get a bunch of pictures of koen and gerald
Dec 13 02:32:41 <bones_was_here> really? which troll started the discusion and then didn't say anything further?
Dec 13 02:33:01 <CareBear\> bones_was_here : you're writing plenty, but not saying much
Dec 13 02:33:12 <CareBear\> if that makes sense
Dec 13 02:33:39 <Russ> CareBear\, I'm really trying to understand the reason for the fork here, bones_was_here seems like a good person to ask
Dec 13 02:33:47 <CareBear\> in particular, the talk about sparc64 makes no sense "was udev tested on all those kinds of use cases"
Dec 13 02:34:02 <CareBear\> Russ : I wouldn't try that :)
Dec 13 02:34:09 <CareBear\> Russ : but all right..
Dec 13 02:34:32 <bones_was_here> i'm not a good person to ask, i'm just trying to demonstrate that there might be real reasons for more conservative distros not to adopt the newly merged udev and systemd
Dec 13 02:35:13 <CareBear\> you're doing a very good job I'm afraid
Dec 13 02:35:42 <bones_was_here> unfortunately those distros are often portrayed as unnecessarily backward 'stick in the muds' who don't adopt tech 'because they dont like it' rather than for real reasons
Dec 13 02:36:10 <CareBear\> *shrug* I don't know what the generalization brings
Dec 13 02:36:18 <CareBear\> talk about concrete problems
Dec 13 02:36:26 <Russ> I'd be more convinced if I'd seen any real reasons or if there was an argument as to why newer versions of udev are 'new tech'
Dec 13 02:36:36 <bones_was_here> CareBear\: well perhaps you can do a better job, or link to an article that does a good job, instead of calling me a troll
Dec 13 02:38:14 <CareBear\> Russ : it is about fear that udev will become too tightly coupled to systemd in the future. rather than waiting for that to happen, some people forked and in some sense start over after removing parts of the code
Dec 13 02:38:27 <CareBear\> it's utterly useless imo
Dec 13 02:38:38 <bones_was_here> well for a short time udev was very new tech indeed, because it tried to implement firmware loading in a way the kernel did not support, fortunately this was fixed but only after linus got emotional
Dec 13 02:38:47 <Russ> so they lose all stake in the official upstream version, seems pretty counter productive
Dec 13 02:38:52 <CareBear\> of course
Dec 13 02:39:13 <CareBear\> Russ : I'd say that is on purpose
Dec 13 02:39:18 <Russ> bones_was_here, I'm familiar with the firmware loading issue
Dec 13 02:39:23 <CareBear\> Russ : because they don't "like" upstream
Dec 13 02:41:46 <bones_was_here> personally i'm not qualified to comment on this issue, i can see advantages and disadvantages, but not the whole picture, which is why i refer to the decisions of maintainers of large, conservative distros, as they are likely to have (or be in the process of getting) 'big picture' understanding
Dec 13 02:42:05 <Russ> if you don't know, then don't say
Dec 13 02:42:31 <bones_was_here> well, you'd better shut up too then ;)
Dec 13 02:48:57 <mdp> uh oh, you could get banned from #raspberrypi for "shut up"
Dec 13 02:49:01 <mdp> easy there!
Dec 13 02:49:57 <Russ> damn, that's strict
Dec 13 02:50:00 <Russ> (banned again)
Dec 13 02:50:33 <Russ> hmm..wonder if 'duck rape' will get SpeedEvil banned
Dec 13 02:51:00 <_troll_> Russ: duck rape is quite real
Dec 13 02:51:10 <_troll_> happens more often than you probably want to know
Dec 13 02:51:13 <Russ> I know, don't go on youtube and search that
Dec 13 02:52:21 <ka6sox> welcome to the Beaglboard and #raspi Peanut Gallery channel...
Dec 13 02:53:42 <bones_was_here> everyone sucks at communicating, myself included, imo freenode should just kline *!*@* and be done
Dec 13 02:54:05 <_troll_> bones_was_here: systemd is a complete clusterfuck
Dec 13 02:54:23 <_troll_> and they're hard at working making sure udev will become unusable without it
Dec 13 02:54:36 <_troll_> lennart is "looking forward to" the day that happens
Dec 13 02:54:46 <ka6sox> kill kitty, kill! kill!
**** ENDING LOGGING AT Thu Dec 13 02:59:58 2012