#milkymist IRC log for Thursday, 2011-04-07

roh720p itself means only 'resolution'00:03
rohthere is some us stuff which uses YUV on analog rca jackets with something similar to 'sync on green'00:05
rohso its only 3 rca jacks. analog.00:05
rohused mainly in the us and asia. in europe we just used rgb and scart with csync on cvbs pin00:05
rohwhich had the feature that the tv doesnt need to do any conversion. basically the signal directly goes to the amps on the tube-pcb, sync to the deflection circuit00:05
lekernellooks great00:06
rohyet the limited bandwith on an rca jack never gave proper picture. and in tft/dlp times its just a intermediate format which gets digitized, buffered and scaled anyhow.... on dvi/dlp/hdmi you usually even let the sender do the scaling and send the optimal format to the screen00:06
rohthe scalers on gpus are usually also much better than the ones in the display controllers (there are still some in there, but they need to be cheap)00:07
lekerneli know. we'll also add your beloved hdmi at some point, don't worry :)00:07
rohlekernel: nice cable.. now you just need to find out what pins it wires how. this only works with graphics cards which do more than 'normal vga'00:08
lekernelbut this has a bit less priority than cvbs because 1. I have rarely seen HDMI projectors 2. it needs a little PCB mod to route a HDMI connector to the FPGA00:08
lekernelcommon projector formats are 1. VGA 2. CVBS/S-Video 3. HDMI/DVI, in decreasing order00:10
lekernelpins 1/2/3 are RGB, so we could definitely use such a cable to connect to an unmodified M100:16
lekernelassuming we sort out the possible DAC resolution problem00:17
rohwell.. if the dacs are dc coupled it could work already00:17
lekernelyou need DC coupling for VGA to work already, no?00:18
lekernelalso, can we use the cable as a 'natural' oversampling low pass filter? :-)00:21
rohlekernel: dunno.00:24
rohi am just sure that we need dc coupling to simulate the sync in the dac signal00:24
lekernelthat should be OK00:24
rohthen i would buy such a cable and just try how good it works/the picture looks00:24
rohmy guess is one looses a bit of dynamic with a 8bit dac due to the amount in voltage the sync already 'eats'00:24
lekernelyou loses 26% of the dynamic range, but 140MHz oversampling would buy you 3 times the resolution00:25
rohhm. well.. i guess that calls for rediscussion after trying out00:26
rohhm. fsck.00:44
rohshouldnt have done that00:44
kristianpauland yes i had troubles with a vga spliter !!!00:50
kristianpauli'll post details soon..00:50
kristianpaultroubles = video signal distorcionated/malformed00:51
kristianpauli'll post image soon..00:51
Action: kristianpaul still at dorkbot00:51
Action: roh is counting hundreds of acryllic spacers00:52
aw_kristianpaul, what's the possible trouble?00:58
kristianpaulaw_: the video technician told me about power and frecuency01:30
kristianpauli also cant run mm1 video on a big plasma screen01:31
aw_kristianpaul, related to design of video? or video format.01:31
kristianpaulaw_: vga video, dunno01:31
kristianpauljust was a tought from the guy01:31
aw_kristianpaul, hmm...got it.01:31
larsclekernel: if i use the 800x600 mode from the rtms fb driver, the image is displaced by 100 pixels or so to the left02:19
kristianpaulaw_: sorry03:50
kristianpauli'll drop a mail saturday i hope, when more time i have.. 3 days workshop is energy consuming..03:52
kristianpaulbtw like 4 guys asked me about milkymist price and avaliabillity03:52
aw_kristianpaul, no sorry. :-)03:53
kristianpaulah i dint tried last msd..03:54
aw_kristianpaul, AFAIK, current only have one M1 available. others are the RC3 ones later. :-)03:54
aw_I'll try latest msd today. :-)03:54
kristianpaulaw_: http://en.qi-hardware.com/wiki/File:Mm1vgaoutaftersplitter.JPG04:27
kristianpaulokay bed time04:28
aw_kristianpaul, oah...using splitter..good. cu nice night!04:29
xiangfukristianpaul: n804:29
wolfspraulkristianpaul: I guess this is a bug you are reporting? http://en.qi-hardware.com/wiki/File:Mm1vgaoutaftersplitter.JPG05:21
wolfsprauldid it work at all? A picture like this one with a functioning m1 in the background would be cool :-) http://en.qi-hardware.com/wiki/File:IMG_3638.JPG05:22
lekernellarsc: press "auto setting" on your screen?06:31
lekernelkristianpaul: how do power and frequency of the video signal (?!) relate to driving big plasma screens?10:11
lekernelmilkymist at tacheles techno party next wednesday :)12:26
kristianpaullekernel: i'm not saying it was a tought/comment from the guy was setting up video13:00
kristianpaulwolfspraul: it worked with out the splitter13:01
wolfspraulkristianpaul: yes, but do you think 'support' for this splitter is a missing feature/bug?13:04
wolfspraulit's something worth remembering maybe, we need an issue tracker :-)13:05
lekernelimo it's just a problem coming from the splitter13:05
wolfspraulkristianpaul: if it worked, do you have a nice picture that shows it in action? or even video?13:05
lekernelthe m1 puts out perfectly standard VGA13:05
wolfspraullekernel: fine, but even then we should document it. let's keep an eye on splitters :-)13:05
wolfspraulthe end users only cares whether it works or not, not who is to blame, or who could or could not fix it.13:06
wolfspraulso the best is if we document well13:06
wolfspraulthen no surprises...13:06
wolfspraulto me it's a good finding, I registered 'splitters' as something to watch out for13:06
kristianpaulwell, it worked(http://en.qi-hardware.com/wiki/File:IMG_3638.JPG) afaik just that pic from me, but i'll ask others pics..13:06
kristianpaulalso this was a quick talk i just had 10 minutes to setup and talk..13:07
wolfspraulah ok. that pic looked similar to the bug :-)13:07
kristianpaulno much time for take pictures..13:07
wolfspraulany feedback on milkymist?13:07
wolfspraulwhat was your talk about?13:07
kristianpaulmilkymist in 7 minutes13:07
kristianpaulpeople dont want to quit the milkydrop just to edit the patch13:08
kristianpaulthey found it like a lack of real time13:08
kristianpauli mean13:08
kristianpaulinterrupt perfomance, edit patch and run again was the critic13:08
wolfspraulwhich is a fair point, and I think sebastien thought about a second vga interface many times already13:09
wolfspraulwe will not introduce it in rc3 though13:09
kristianpaulhey, why the last pic looks bad? it is running a patch as usual13:09
kristianpaulyes sure, just saying what other commeted13:09
wolfspraulmaybe I just didn't get it13:09
wolfspraulno worries I am not complaining, I was just confused with the splitter bug and that pic13:09
lekernelkristianpaul: yes, otoh you can prepare many patches in advance and switch them in no time, e.g. with a midi controller13:10
kristianpaulyes i told that lekernel13:10
wolfspraulthe pic is great. if that is in fact the intended visual effect, then even better!13:10
wolfspraulI just didn't know, that's why I ask.13:10
kristianpaulbut pople want *real* *time* editing of patches13:10
wolfspraulnow I know :-)13:10
kristianpaulwell i got that impresion13:10
lekernelkristianpaul: some do, some others don't...13:10
kristianpaulalso asked for OSC support13:11
kristianpaullekernel: he, sure ;)13:11
lekernelOSC is already built in13:11
kristianpaulhow they use it?13:11
kristianpaulplease dont RTFM now :-)13:11
lekernelwell, there's no documentation for it yet :)13:11
lekernelso it would be "RTFS" instead :)13:11
lekernelbut in more friendly terms13:12
lekernelyou have parameters in the patches iosc1, iosc2, etc.13:12
lekernelwhich are floating point values you can set through the OSC interface over UDP13:12
lekernelyou also have a MIDI sink (i.e. MIDI transport over OSC) which gets merged with the rest of the MIDI messages coming from the DIN13:13
wolfspraulkristianpaul: for that pic 3638, can you give me the exact event name and location?13:13
lekernelthe possibility to switch patches with an OSC command13:13
wolfspraulin case I want to mention it in the news (the pic is nice!)13:13
kristianpaulevent labsurlab at mamm13:13
wolfspraulname of event, location, date13:13
kristianpaulwas just before dorkbot talk started13:14
lekerneland finally, you can print text messages on the screen using another OSC string message13:14
kristianpaulmamm = museo de arte moderno de medellin13:14
wolfsprauldate was April ? 6?13:14
lekernelyou can try easily with the 'oscsend' utility from liblo13:14
kristianpaulyup wolfspraul13:14
wolfspraulis there a url for the event?13:14
lekernelkristianpaul: do you have pics with the video-in patches?13:14
kristianpauli think mm1 should support the splitter somw way, well or do more test, but all laptops around dint had problem with splitter13:15
lekernelkristianpaul: http://en.qi-hardware.com/w/images/5/54/Mm1vgaoutaftersplitter.JPG is blurred as well outside of the screen13:15
kristianpauli think there was just one including mm113:15
kristianpaullekernel: i know13:15
lekernelkristianpaul: what was the problem exactly?13:15
kristianpaulvide-in not tested yday13:16
kristianpaulimage was partially viewved13:16
kristianpauli mean it was kinda zoomed13:16
lekernelkristianpaul: please demo the video-in patches. they're really at the next level :)13:16
kristianpauli will, i hope flash mm1 with last firmware today..13:17
lekernelkristianpaul: weird13:17
kristianpaullekernel: indeed13:17
wolfspraulhow long did m1 run at the event in total?13:17
wolfsprauljust during your talk, or for some longer period?13:17
kristianpaulno more than 1 hr13:17
lekernelfrom my knowledge splitters are simple amplifiers... can't really see how that could do that13:17
kristianpaulbefore talk13:17
kristianpauland during13:17
wolfspraulok then I say 'demoed'13:17
lekernelcould be some murphy-esque interaction between splitter, projector and signal integrity problems13:17
lekernelkristianpaul: what flickernoise/soc versions are you using atm?13:18
lekernel0.2/RC2 (dec2010 msd) already support the video-in13:18
kristianpaulsure it tested video in just dont have the best setup up for a demo13:19
lekernelah, you're having another demo later on?13:20
kristianpauli will do more tests today i hope, and a meeting with a local VJ group saturday13:20
kristianpauli have a big screen for me all day13:20
lekernelah, cool!13:20
kristianpaulbut the mm1 just never got to work, the screen said no signal detected13:20
lekernelyou could even invite people to stand in front of the cam :)13:20
lekernelwhat screen is that?13:21
kristianpaulbut i plug a small screen at works..13:21
kristianpauli'll take specs today13:21
kristianpaulnoo time yday13:21
lekernelfirst time I hear about screen signal detection problems...13:21
lekerneli've plugged mm1 to all sorts of screens and projectors already, without a single issue13:22
lekernelis the cable ok?13:22
kristianpaulok sure !, i tested with other laptops, and NO problem13:24
kristianpauli need leve, i'll bnack ramdomly in next horus13:24
CIA-8milkymist: Sebastien Bourdeauducq master * rd8ddaca / (4 files in 4 dirs): Ethernet: workaround for permanent RX FIFO overflow bug - http://bit.ly/hE7Hdj17:25
mwallelarsc: do you still facing problems on real hw?17:39
CIA-8rtems-milkymist: Sebastien Bourdeauducq master * r5456116 / (3 files in 2 dirs): Ethernet: use library CRC - http://bit.ly/f2BGDr17:40
larscmwalle: nope17:51
larscnow i'm facing problmes in qemu... qemu crashes as soon as the kernel has finished loading the initrd image17:52
larscPC is 0 for some reason17:52
larscbtw. all the got relocations seem to be against soft-math functions http://pastebin.com/gzv9wyTh17:54
mwalleon real hw too?17:54
larscmwalle: not that i can see17:54
mwallenot good17:55
larscbut i'm maybe using an outdated version of qemu17:56
mwallelarsc: do you have applied the gcc-fix-lm32-libgcc.patch?17:57
larsci've applied all your patches17:57
mwalleis __PIC__ undefined?17:58
mwallehave a look at libgcc/config/lm32/_ashrsi3.S17:58
larsc__PIC__ is probably defined then18:01
larscbtw. the latest qemu build fails with make[1]: *** No rule to make target `../ptimer.o', needed by `qemu-system-lm32'.  Stop.18:02
mwalleofficial one?18:03
larscthe milkymist branch18:03
mwallei have one fix regarding system calls18:04
mwallethat is not in my repo yet18:04
mwalleyou could try that too18:06
larschunk#3 doesn't apply, against which branch is the patch?18:08
mwallemh my working one here18:09
mwallei try to merge upstream and milkymist into master18:09
mwallebtw, does anybody know how to avoid duplicate commits you get if your patch is merged upstream (and the commitmsg was changed?)18:11
lekernelmwalle: got an idea what's happening when the board suddenly freezes and gdb shows it's executing the debug ROM?18:11
lekernelsome exception? how to know more?18:12
mwallehow shows gdb its executing the debug rom?18:12
lekernelas soon as it connects, it shows PC is in the debug rom18:13
lekernelwell, actually past it18:13
lekernel(gdb) set remote interrupt-on-connect on18:13
lekernel(gdb) target remote /dev/pts/6018:13
lekernelRemote debugging using /dev/pts/6018:13
lekernel0x1f18e2a4 in ?? ()18:13
lekernellooks pretty bad...18:13
mwallemhhh doenst look like a valid PC18:13
lekernelra is messed up too.. 0x1f18e82c18:14
mwalledid you debug before?18:16
lekernelno, just booted and connected GDB when it froze18:16
mwalleis this a new freeze (introduced by the gdb stub?)18:17
lekernelin fact, I did that a couple of times already after 'random' freezes, and everytime I got that weird PC18:17
lekernelit's an old bug I'm just tracking down now using the new gdb tool :)18:17
mwallex/x 0x1f18e2a4 ?18:17
mwallesome valid opcode?18:18
mwallemh there was some fmt modifier to decode that instruction, wasnt it?18:18
mwallex/i 0x1f18e2a418:19
mwallelooks like a loop ;)18:20
mwallemaybe we should enable BUS_ERRORS18:20
lekernelbus error?18:21
mwallebut then you would have to signal invalid addresses on the WB18:21
lekernelyou mean generate a bus error when the cpu tries to access an invalid location?18:21
mwalletheres some CFG_BUS_ERRORS or sth like that in the lm32 core18:21
mwallewhich raises an exception18:22
lekernelso it would stop the execution as soon as it's crashing... yeah18:22
lekernelwe should have an "invalid instruction" exception too18:22
larschm, interesting: gcc/Makefile:TARGET_LIBGCC2_CFLAGS =18:22
larscgcc/Makefile:echo TARGET_LIBGCC2_CFLAGS = '$(TARGET_LIBGCC2_CFLAGS)' >> tmp-libgcc.mvars18:22
larscyet libgcc.mvars contains TARGET_LIBGCC2_CFLAGS = -fPIC18:22
mwalleyep, well you would have to have some special exception handling for data and instruction bus errors, too (eg branch to debug rom18:23
mwallelarsc: not for me18:25
larscah. found it18:25
larscwell maybe not18:25
larscbut i'm using lm32-*-linux instead of -uclinux18:25
larscand there is a t-linux which sets TARGET_LIBGCC2_CFLAGS to -fPIC18:26
mwallelarsc: mh there are some other option wich are set if you have *uclinux* triplet iirc18:35
mwallelekernel: mh dunno if thats worth the effort18:37
lekernelwe'll have to fix those bloody intermittent crashes at some point...18:38
lekernelfortunately they do not seem to happen when rendering, only when using the gui18:38
lekerneland they're quite rare18:39
CIA-8rtems-milkymist: Sebastien Bourdeauducq master * r0c0d5bc / c/src/lib/libbsp/lm32/shared/milkymist_networking/network.c : Ethernet: check received frame CRC - http://bit.ly/ik4VxK18:40
mwallelekernel: whats behind the ra address?19:00
mwallea call?19:00
lekernelhm I rebooted19:00
lekernelwill check next time19:00
mwalleif its rom its the content hasnt changed ;)19:03
mwallei would look at my own, but i have a newer rom ;)19:03
CIA-8milkymist: Michael Walle master * r02cdac9 / (cores/monitor/rtl/gdbstub.rom software/gdbstub/gdbstub.c): gdbstub: fix max packet size reporting - http://bit.ly/gktZAV19:37
lekernelmwalle: thanks19:37
lekernelthis time crash @ 0x1000029c ...20:26
mwallemh this sounds like the gdbstub was already executing when BREAK was received20:27
mwallelarsc: lekernel: i updated my qemu repository. please note that there is no milkymist branch anymore.20:30
mwalleonly master which is synced with upstream20:30
Fallenoumwalle: all milkymist stuff is upstream now ?20:32
mwalleFallenou: yes20:33
mwallelekernel: btw http://lists.nongnu.org/archive/html/qemu-devel/2011-04/msg00504.html20:33
mwallethere are only two missing commits atm, one which stops the vm instead of a cpu_abort and one which fixes debug exception handling20:34
lekernelgood :)20:37
lekernelwe should also add support for the higher video resolutions, it didn't work last time i tried (segfault)20:37
lekernelshould be as simple as reading the hres and vres registers and resizing the window (ignoring the other video timing registers)20:37
lekerneli'll have a look at that if you don't before me20:37
Fallenoumwalle: ok so I will remove the "checkout the milkymist branch" from the wiki :)20:38
Fallenousince all is in master now20:39
mwalleFallenou: yeah20:39
mwallelekernel: theres already some resizing code20:39
mwalleif you change hres/vres registers20:39
lekernelha? different bug then... didn't investigate yet... but it appeared when I tried to switch to 1025x76820:40
lekernel#{[{#~! why are the pc->board transfers 16ko/s... with very low lm32 cpu utilization20:41
mwallelekernel: is there a way to get the flickernoise binary out of the fbi file20:42
lekerneldd the header off20:42
mwallehow long is it?20:42
lekernel8 bytes20:43
mwallei can see any ELF tags in the fbi file20:45
mwalleah because its .ralf20:48
larscmwalle: i still get the same error building qemu20:55
mwalleis there a CONFIG_PTIMER in default-configs/lm32-softmmu.mak?20:56
larsci can fix it by prepending the obj file with hw/20:59
larscbut then i get the same error for sd.o20:59
mwalledo you build in-tree?20:59
mwallecould you try a out-of-tree build?21:00
larsc`make -C qemu-lm32`?21:02
mwalle../qemu/configure --target-list=lm32-softmmu21:02
CIA-8rtems-milkymist: Sebastien Bourdeauducq master * r14cce88 / c/src/lib/libbsp/lm32/shared/milkymist_networking/network.c : Ethernet: copy-free RX - http://bit.ly/flWS5m21:04
lekernelanyone wants my /opt/rtems4.11 with all the flickernoise libs compiled?21:05
larscmwalle: nope, same error21:05
mwallelekernel: yes :)21:06
mwallelarsc: grml21:06
mwallewhats your configure line?21:06
lekerneluploading in http://www.milkymist.org/3rdparty/ ... should be done in ~30min21:09
lekernelsize 83725495 bytes21:09
larscmwalle: well `../qemu/configure --target-list=lm32-softmmu`21:09
mwallelarsc: http://paste.debian.net/113329/21:09
lekernelethernet seems pretty stable now... except that FTP uploads are slow (but downloads aren't)21:11
lekernelstrange bug..21:11
mwallemaybe it'll work for me too ;)21:11
lekerneland there are still some rx fifo overflows too :(21:12
lekerneleven when doubling the on-chip buffering.. if you ping flood with the max packet size, you get a few21:12
lekernel(this no longer crashes the board though :)21:13
lekernelmaybe I could fix that by making the ethernet core DMA to FML21:13
rofl0rlekernel, which mtk files specify how stuff is drawn on the screen ? i wanna try to get it to work on directfb21:13
lekernelbut the problem is the L2 cache should be flushed for retrieving packets then21:13
lekernelrofl0r: see fb.c in flickernoise, you just pass MTK a 16bpp framebuffer21:14
lekernelfor some reason the board sends duplicate TCP acks and messes up with the TCP window21:17
lekernelonly for receiving data - board->PC is flawless21:18
mwallelekernel: vgafb resizing should be fixed21:29
mwallelarsc: any success?21:45
mwallereally strange21:48
lekernelqemu has a profiler?22:14
lekernelmwalle: yup, hires works now22:18
lekernelsp             0x8cf628c14780890823:04
lekernelra             0xffff0fbf-6150523:04
lekernelPC             0xffff35dc-5174823:04
kristianpaulokay, lets flash finally the MM1 :D23:07
lekernelhmm multi-slot RX seems to have fpga bugs23:35
wolfspraullekernel: multi-slot RX?23:45
lekernelyes, programming multiple DMA reception buffers into the (softcore) ethernet controller23:46
lekernelit seems that when switching quickly to another buffer, some data gets corrupted23:47
--- Fri Apr 8 201100:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!