#milkymist IRC log for Saturday, 2011-04-09

larscmwalle: do you have an crti.o in your lm32 toolchain?00:29
larschm... now i have a toolchain and a rootfs, but nothing happens after "VFS: Mounted root (ext2 filesystem) readonly on device 1:0."02:22
larscmwalle: what is your make version?02:29
larscmwalle: i removed lm32-softemu/config-* and now it works....02:35
larscqemu still crashes though02:36
larscmwalle: as soon as irqs are enabled02:39
larschm... ok that was due to a half applied patch02:40
larscbut it still crashes when entering usermode02:40
larscIN: PC=002:41
larscie=00000003 (IE=1 EIE=1 BIE=0) im=0000000b ip=0000001002:41
larscit looks like an interrupt within a systemcall03:40
larschm the syscall handler should increase the return address before calling eret, yet this seems not to happen in qemu04:24
larschm... eret is called from _restore_irq_frame_and_return and not from _restore_and_return_exception which explains why ea is not incremented05:15
larschm, or maybe not05:17
vr1est-ce que milkymist est une bonne plateforme pour traiter du son ?12:10
lekernelça peut se faire oui... quel genre de traitement?12:11
vr1de la reco12:12
vr1(analyse statistique)12:12
vr1g un proto qui fct bien en C/Linux sur un x86_6412:13
vr1j'aimerais l'embarquer, je me demande si le milkymist tient la charge12:13
lekernelavec un accélérateur fpga, il n'y a sûrement pas de problème. mais il faudra le développer.12:14
vr1c-a-d le code en C pur tiendra pas la charge ?12:15
lekernelaucune idée12:15
vr1le CPU est equivalent à un ARM7 ou ARM9 ?12:16
lekernelplutôt arm712:16
vr1ok je vois12:16
vr1la possibilité de faire un bout de code FPGA pour accélerer est intéressante12:17
vr1si on developpe un soft sur la plateforme, quelle en est la license ?12:17
vr1je veux dire: est-on obligé d'opensourcer ?12:18
lekernelle design FPGA est largement sous GPL, pour le soft C il n'y a pas de restriction de licence12:18
lekerneldonc si tu modifies le design FPGA et que tu le redistribues, en général oui12:19
lekernelceci dit, une licence ARM pour faire la même chose coûterait dans les 100k¬, donc choisis :)12:19
vr1est-il possible de dériver le HW pour en lever les composants non vidéo ? (dans le cas d'une production)12:19
vr1je me suis mal exprimé12:20
vr1est-il possible d'enlever certains composants du design HW ?12:20
vr1par exemple enlever les composants vidéo12:20
vr1pour faire une BOM moins cher dans le cas d'une prod12:21
vr1et est-ce autorisé ?12:21
vr1par ex plateforme de son uniquement12:21
lekerneloui, c'est parfaitement autorisé12:23
vr1quelle est la BOM que l'on peut espérer pour une telle plateforme ? juste une idée de prix approximative12:24
lekernely compris la commercialisation de ton design, tant que les licences GPL et CC-BY-SA sont respectées12:24
lekernelle PCB/schéma est en CC-BY-SA ...ceci dit, le design FPGA (en GPL) est assez portable également12:25
lekernelil tourne aussi sur plusieurs cartes xilinx/altera12:25
vr1salut les gars12:39
wpwraklekernel: i hope i haven't just added a 2 month delay to MM1 progress ;-)14:20
lekernelwpwrak: ?14:20
wpwraklekernel: see the qi-hw mailing list ;-)14:20
wpwraklekernel: well, or maybe don't :)14:20
lekernelthe robot?14:21
wpwraklooks cool, doesn't it ? and it also looks kinda doable ...14:22
lekernelheard of boston dynamics?14:22
lekernel(big dog etc.)14:22
wpwrakkinda scary that device. looks a bit too "alive"14:23
wpwrakwhat i like about the rotopod is its symmetry - all the six axes are identical. so you don't need to design each axis separately, like in traditional CNC machines.14:28
lekernel!karma test18:26
CIA-8milkymist: Sebastien Bourdeauducq newmac * r99be230 / (10 files in 3 dirs): minimac2: TX working in simulation - http://bit.ly/fcr6dh18:34
larscmwalle: something is wrong with the scall code generation. http://pastebin.com/3KVHpa7t there is only an scall instruction at 40017c5819:54
larscVFS: Mounted root (ext2 filesystem) readonly on device 1:0.20:12
larscHELLO WORLD20:12
larscKernel panic - not syncing: Attempted to kill init!20:12
mwalleis scall broken?20:37
larscmwalle: i reverted d53066a5578945748dd7d8099c96de13f9a69211 and now it works again20:39
mwallelarsc: could you please paste the code arount 40017c58?20:43
larscwell, there is one scall plus a few mv20:46
mwalledo you know what happens?20:49
larscnot really20:50
larscfor some reasons scall exceptions are also created for the lines following the real scall20:51
mwallemh so why doesn't this happen with test_scall.S?!20:52
mwallemake check20:54
mwallelarsc: could you upload your kernel and initrd?20:58
mwallemh i think i dont need it21:00
mwallehappens for me too21:00
mwallemh ok, works again ;)21:02
mwallelarsc: i just uploaded the fix21:04
mwallenarf, commit contains opengl 'fix'..21:13
mwallelarsc: i guess you already pulled it :)21:18
larscfell free to rebase21:20
TS-Labsi have a question about Navre AVR clone21:34
TS-Labsas it is said in description, the classic avr instruction set is supported. but when reading the sources I see that the MUL's aren't implemented21:38
mwallei guess it isnt needed by us. so it may not be implemented21:42
mwalleesp. since multipliers needs lots of ressourecs21:42
TS-Labsyeah, ok )21:42
mwalleor its done in software21:43
TS-Labsin fact, 8bits by 8 multiplier ate ~110 LE's on Acex I Altera21:43
TS-Labsno, just * in verilog )21:44
TS-Labscombinatorial logics21:45
TS-Labs4 or 5 summators of different sizes21:45
mwalleso its likely that this will be the critical path :)21:45
TS-Labsyeah )21:45
TS-Labsi used 40 mhz, worked ok, but i didn't measure its latency21:46
lekernelmultiply isn't part of the 'classic' instruction set22:50
lekernel(and btw modern synthesizers can pipeline '*' in verilog)23:02
larscmwalle: I finally managed to get OpenWrt to build bootable images :)23:38
wolfspraullarsc: !23:57
--- Sun Apr 10 201100:00

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