#milkymist IRC log for Thursday, 2012-05-03

kristianpaulxiangfu: do i need to apply all werner patches?01:56
kristianpaulMorning btw !01:56
xiangfukristianpaul, yes. apply all werner patches.01:57
kristianpaulgee01:57
kristianpaulno matter order?01:58
sb0what patches?01:58
kristianpaulhi01:58
sb0rtems?01:58
kristianpaulyes01:58
sb0some of them are already merged01:58
sb0use the quilt file, it should be up to date01:58
kristianpaulah where is it?01:59
Action: sb0 bothered some NASA engineers today about RTEMS bugs01:59
xiangfukristianpaul, : http://projects.qi-hardware.com/index.php/p/wernermisc/source/tree/master/m1/patches/README02:00
sb0should be in the patch repository... and the readme file tells you how to run quilt02:00
kristianpaulgood02:00
kristianpaulsb0: and?? ;-) what they said?02:00
kristianpaulxiangfu: ah tricky thats another directory from yday u pointed..02:01
kristianpaulah, fool me not reading werner one :)02:01
sb0it's basically that everyone's using it...02:02
kristianpaulhe, well long time since no compile rtems ....02:03
kristianpaulokay compiling now.. gn802:11
kristianpauldid nasa bother you about using milkymist soc in coming missions? ;)02:12
wolfspraulnasa is using rtems or milkymist?02:14
wpwraksb0: how happy are they with it ? :)02:18
wpwrakwolfspraul: hehe, we wish :)02:19
sb0they're using rtems and the milkymist dram controller (the old one)02:19
sb0actually I saw their board with the memory controller today :)02:20
sb0they're using some freaky space-qualified DRAM chips. hadn't seen DRAM like this before.02:20
sb0they're packaged in something between a QFP and a ceramic PGA, about 1cm high02:21
sb0http://www.electronicdesign.com/files/29/5765/figure_01.jpg02:25
wpwrakwow. looks fancy :)02:28
wpwrakwhen the aliens catch voyager, they'll think our computers are very bulky and slow, which will make them over-confident. their invasion fleet won't stand a chance.02:29
GitHub64[board-m1] adamwang pushed 1 new commit to master: https://github.com/milkymist/board-m1/commit/cb81048d2a0057e347a0b6b564665818d1c8abf003:36
GitHub64[board-m1/master] 1. split FPGA_P[1..2].sch into FPGA_BANK[0..3].sch and EXPANSION_RESET.sch - Adam Wang03:36
GitHub30[board-m1] adamwang pushed 1 new commit to master: https://github.com/milkymist/board-m1/commit/bff801f46e59f1a0aeba6ed65b3e1ca820e25a4903:41
GitHub30[board-m1/master] split FPGA_P[1..2].sch into FPGA_BANK[0..3].sch and EXPANSION_RESET.sch - Adam Wang03:41
GitHub79[board-m1] adamwang pushed 1 new commit to master: https://github.com/milkymist/board-m1/commit/e3a19800e3ac1adf0832f6578a94c39242ff0d4703:45
GitHub79[board-m1/master] remove FPGA_P[1..2].sch - Adam Wang03:45
cladamwwpwrak, xiangfu could you git pull board-m1 to see if ERC without any err. thanks. :-)03:46
xiangfuok03:49
wpwrakhmm, the LibDir change breaks my setup04:04
wpwraklet's fix that ...04:04
wpwrak0 troubles04:05
cladamw"LibDir" ?04:06
wpwrakthis one :)04:06
GitHub39[board-m1] wpwrak pushed 1 new commit to master: https://github.com/milkymist/board-m1/commit/0d28a62ddedd2331c9c6702ad1c57fa011c01e4104:06
GitHub39[board-m1/master] r4/m1.pro: keep ../../kicad-libs/components in LibDir - Werner Almesberger04:06
cladamwif i git pull, will my LibDir then doesn't find ?04:07
wpwrakno no. you'll be fine04:10
wpwrakwhen adding the local ../components, you accidently removed the ../../kicad-libs/components04:10
wpwraki put it back04:10
cladamwhmm ? i didn't manually add the local in m1.pro text file but used eeschema. :)04:12
cladamwwpwrak, i tried to embellish the rules wiki to do such better schematics same as gta02-core. hope you like. Now they should be easier to more readable. :-)04:14
cladamwthe wiki should continue to add, set common rules whenever we think rules are good for read. :)04:16
wpwrak(edit m1.pro) yeah, kicad can sometimes do strange things to that file04:25
wpwrak(rules) excellent ! let's see ...04:25
wpwrakon MiscControl.sch, the DMX zeners look a bit confusing because of the package.04:33
wpwrakhah ! and the pin names in the oscillator are in a 40 mil font ;-)04:35
wpwraknot sure if you want to change that, though. the whole component looks very nicely drawn. changing the font size may mess this up.04:35
wpwrakthe USB series resistors (R136, R137, etc) are very close. how about moving the resistor value into the symbol ?04:37
wpwrakhmm, at 60 mil, that looks bad, too :(04:37
wpwrakin Audio.sch, the connectors have ground looping up. that looks a bit odd. it may be clearer to just make ground cross the other pins close to the connector. like on J104:43
wpwrakthat's all i found at a quick glance :)04:47
wpwrakoverall, it looks very good. congratulations ! even the FPGA doesn't look so intimidating anymore ;)04:48
wolfspraulxilinx is pushing their zynq chips very much12:13
wolfspraulthose are chips that have for example two cortex a9 arm cores + 85k cells fpga fabric + ddr3 + spi/nand/nor...12:14
wolfspraulis this of interest to us/milkymist?12:14
wolfspraulwhat if some of those chips are more easily available and/or cheaper than fpga-only chips?12:15
Fallenouwell it's of interest if we want to replace all the open source ip cores by proprietary hard-burned ip ^^12:16
Fallenouit could certainly increase performance, but will decrease the added value of the project12:17
Fallenou(and 85k cells is really small)12:17
wolfspraulthe only reason I bring up the discussion is to unite the community :-)12:25
wolfspraulso we have to be clear on what we want or don't want, and for what reason12:25
wolfspraulluckily I think we have a relatively coherent philosophy here, which is:12:26
wolfspraul#1 make it work now and do something useful now12:26
wolfspraul#2 reuse as much as possible of the milkymist tech we have now, and let us grow that further12:26
wolfspraul#3 get the highest performance and most easily available chip for the least cost (per performance)12:26
wolfspraulthe choice of the slx45 (speed grade 2 I think) for m1 was pretty good, even in hindsight12:28
wolfspraulone could debate about slx45 vs slx75, but nobody uses those cells right now anyway, so slx45 is/was good12:28
larsci have a few zynq boards on my desk and they are nice, because you have the flexibility of a fpga with the processing power of a dual core arm.12:37
larscbut i don't think they are the right fit fore the milkymist project12:38
wolfspraulwhich chips do you have on those boards?12:38
larscs/fore/for12:38
larscnot sure12:38
wolfspraulwhy not right fit?12:38
wolfspraulit's early now, we also have to see about availability and prices12:39
wolfspraulfpga makers are known for extensive list of part numbers with few chips actually generally available :-)12:39
larscbecause they come with a bunch of hardcores12:39
wolfspraulyeah12:41
wolfspraulmaybe we could run the VJ app on Intel's "Next Unit of Computing" boards right away :-)12:42
larscexactly12:42
wolfspraulso we think that the extra cores and IP distract from the beauty and simplicity of what Milkymist is about12:43
wolfspraulthat's just a tough thing to sell potentially, if someone says "why not use that chip and have a much easier life"12:43
wolfspraulbut I agree my feeling is it would be a huge distraction12:44
wolfspraullarsc: thanks for your thoughts on this! since you already play with those chips :-)12:45
larscif you happen to be at Xfest by chance there is a demo I build at the ADI booth, showing a zynq running ubuntu with 1080p video out12:49
wolfspraulnice13:13
wolfspraulbut not very likely, don't even know what Xfest is?13:13
wolfspraulthis one? xfestmodesto.com ?13:14
sb0a trade show organized by xilinx13:14
wolfspraulhah, and I found this xfest music festival :-)13:15
Hodapphm, odd, there is an Xfest every year through my hometorn13:19
Hodapps/hometorn/hometown/13:19
Hodapppeople don't go for the music, they go because of the weed and the bare breasts from girls who want to earn beads13:19
wolfspraulah yes, now we know where lars is going13:20
HodappHave you guys seen the tiny laser projectors? I just met the guy who developed the driver boards on the Microvision ones, and I guess they're $150 - $300 or so depending on where you find them and what version; new one is 50% brighter but in human vision terms that's really not much13:21
wolfspraulI have one right here13:22
HodappWas thinking of this with the video someone posted of a Milkymist set up with a projector in a very impromptu way13:22
Hodappas the Microvision can run from battery or from USB power13:22
wolfspraulfor a while I thought this might be good with m1, but nowadays I believe in large & cheap monitor screens instead13:22
wolfspraulthe laser is cool though, but the big issue is brightness13:22
Hodappyeah...13:22
HodappI am going to pick up a couple and play around13:23
wolfspraulit would be great to open more of this design, galvo control etc.13:23
wolfspraulpick up a couple?13:23
wolfspraulmaybe you can pickup a couple m1 too?13:23
Hodapplol13:23
HodappI'm thinking about it. If I am able to meet up with a couple local friends and practice some VJing stuff, I may get one.13:23
wolfspraulI *thought* (guessed) about the company behind a little - microvision13:23
wolfspraulplease do13:23
wolfsprauland my feeling is that this product is more or less a proof-of-concept for them13:24
wolfspraulthey are not serious about entering consumer electronics in a big way, they just do this product to ease their investors pain over many years of investing13:24
wolfspraulthat's my feeling13:24
wolfspraultheir real market (and maybe exit) strategy seem to be some vertical markets13:24
Hodappthe man who made the board purposely made it hackable and broke out all of the relevant pins right underneath the battery and told them it was for testing purposes (which it sort of was)13:24
wolfspraulis this documented somewhere?13:24
HodappI'll talk to him about it13:25
wolfspraulI am mostly interested in the electric design and firmware that controls the galvos and lasers13:25
HodappI'm interested in knowing how fast you can modulate the laser because I'm told that the resolution is largely a software limitation13:25
wolfspraulor rather not me, but marcan our openlase friend13:25
Hodappopenlase?13:25
wolfspraulyeah13:25
HodappI will talk to Bill about this13:25
Hodappall I know now is that it's FPGA based and it's his design13:26
Hodapphe is on IRC - as wjsteele - but rarely13:27
HodappI have had the Milkymist thesis paper sitting on my desk waiting to be read but have just been so busy :| that and Kinko's screwed up and printed it on card stock and so it's an inch thick13:27
wolfspraulI think the laser stuff needs a lot more investment13:28
wolfspraulor, of course, some really big 'open' push by someone that drives it forward that way, but I don't see that happening right now13:28
Hodappperhaps, if I go to grad school, I'll focus on something like this...13:50
wolfspraulget an m1 first that gets you in the right direction13:51
Hodappbah. I need to get some sleep first13:51
wolfspraulcan't beat that. same here... n813:51
wolfspraulthanks for bringing up the laser!13:51
HodappI have interest in computer graphics and in programming, but my undergrad was in EE13:51
wolfspraulI like that little thingie...13:51
Hodapphowever, I'm kind of EE-clueless and I don't presently work much in EE except for when my job involves image & signal processing13:52
Hodappor someone needs a person who can solder or read a schematic13:52
Hodappwolfspraul: this OpenLase project is very interesting14:08
wpwrak(chips combining CPU, memory, and FPGA) sounds very attractive for an FPGA-Ya.14:43
Hodappwpwrak: what's this about?14:43
wpwrakBUT ... if this can't be treated as a reasonably generic bundle, then it probably isn't worth it. (i.e., if it's all proprietary cores and things, blocking the way to, say, llhdl)14:43
wpwrakHodapp: the idea of making another nanonote, but this time FPGA-based14:44
Hodappoooh14:44
wpwrakhard not to like that idea, eh ? in all its crazy glory ;-)14:45
Hodappwhy does https://en.wikipedia.org/wiki/Ben_NanoNote mention the Milkymist?14:45
wpwrakdunno. perhaps marketing was at work ? :) mentioning qi-hw there may be better.14:47
kristian1aulFPGA-Ya indeed15:39
kristian1aulperhaps 85k cells is not enought for a soc, but for somethine else15:39
kristian1aulso not us and the soc..15:52
wpwrakexactly. if you already have an ARM core, you only need to synthesize the peripherals15:56
wpwrakbut i suspect that simply lack of openness would spoil that fun15:56
kristianpauloh yeah, propietary bus interface.. etc15:57
kristianpaulan aprouch like the one from SIE could work for a YA, but.. to be honest a dual fpga or dual core aprouch now that there is a mmu in progress15:58
Fallenouin progress is the word :p15:59
Fallenoureally work in progress15:59
kristianpaulbut it works, no?16:00
Fallenoutranslation of virtual addresses to physical ones through TLB is working yes (at least 15 test cases are passing)16:00
kristianpauljust missing Linux support i want to meant with "in progress" btw :-)16:00
Fallenouoh no16:00
Fallenouit misses other things16:00
kristianpaultoo many?16:00
Fallenouexception handling is work in progress (in a good shape though)16:01
Fallenouand ITLB is still completely missing16:01
Fallenouand it will be a tough task :)16:01
kristianpauli see16:01
kristianpaullarsc: your demo include some fancy gateware implemented on the 85k cells?16:07
sb0wpwrak, also those ARM cores typically come with lousy bus interfaces16:13
sb0things like wishbone probably look too simple and therefore lack a feeling of job security16:14
sb0you need things with "advanced" in the name and tons of bloated complications16:15
kristianpaulagreed :)16:15
kristianpaulwell, still a Linux friendly with fgpa and recofigurable cores but thats other part of the history16:16
kristianpaulperhaps xilinx already look at it?16:17
wpwraksb0: if you starve marketing, they will starve engineering in retaliation :)16:17
wpwraksb0: but i'd be mainly interested in the core speedup, which should be enormous16:18
sb0let's make a LM32 ASIC then16:18
kristianpaulis alredy no?16:18
kristianpaullets ask lattice for samples ;-)16:18
kristianpaulor you mean whole ASIC SoC by instance?16:18
wpwrak(asic) hmm, i see lots of chips and lots of pins :-(16:18
kristianpaulhehe16:19
wpwrakkristianpaul: that would mean losing the configurability16:19
kristianpaulthats why i point double fpga solution16:19
sb032 bit address, 32 bit data, 4-5 control signals16:19
sb0you can also multiplex address/data16:19
sb0not too inefficient when using burst16:20
sb0that's around 40 pins...16:20
wpwrak40 sounds manageable ...16:20
sb0also if you have just the LM32 and that bus i/f, the thing should be rather risk-free and painless.16:21
sb0but it can only be used with a FPGA/CPLD16:21
wpwrakthe advantage over using something non-LM32 would be that you could optimize the bus for low pincount, because you know you have the FPGA on the other side to take care of the rest16:22
sb0I wonder what silicon area we'd be talking about exactly16:22
wpwrak(only FPGA) yes, exactly16:22
wpwrakan interesting approach. the core is something we wouldn't want to change too frequently anyway16:23
sb0I was thinking about including some optional peripherals (simple sdram controller, lcm, ...) so the asic can be used standalone in e.g. the ya, but it increases difficulty and risk of a totally nonworking chip16:24
kristianpauldrasko mentioned about an LM32 asic time ago http://lists.milkymist.org/htdig.cgi/devel-milkymist.org/2010-September/000860.html16:28
kristianpauloh, do we have an lcm core already? :-)16:42
rohkristianpaul: any idea how fast a lm32 with mmu can be clocked?20:55
rohmy question goes into the direction of 'how fast can we hope to get a linux kernel on that platform'20:55
kristianpaulnope sorry20:59
kristianpauli'm total ignorant about mmu21:00
kristianpaulfor my self a computer memory should be a flat space, just like the internet should be ;)21:00
kristianpauli guess so far clocks as usual21:00
kristianpaullets wait Yann elaborate more about this topic later :)21:01
kristianpaulhe, a jornada 720 ram run by 50mhz and cpu is 200mhz21:01
kristianpauldisplay is 640x32021:01
kristianpaulinteresting21:01
kristianpaulperhaps an altera fpga could clock milkymist at that speed..21:02
rohok.. then lets forget that stuff and go to proper integrated cores21:03
rohthat low resolutions and clocks makes it unsellable in my eyes21:03
kristianpaulall varies roh , it depends what porpuse you found for it21:05
kristianpaulas everything else21:05
rohkristianpaul: i was mostly thinking about the nn ya discussion22:31
wolfspra1lok, the rough consensus seems to be to skip the Zynq chips for now23:32
wolfspra1lwhen they are available with pricing etc (q1/2013) we can take another look23:32
wolfspra1lI somehow feel it would be a big distraction, but it's important that the Milkymist community is more or less united about this issue - we are small enough23:33
wolfspra1lbtw, there have been similar attempts by other fpga makers in the past, and eventually they all fizzled23:34
wolfspra1lnow the next round23:34
wolfspra1lI think it's also very risky from Xilinx, marketing-wise. their customers may like the arm cores so much that in the next product iteration, they kick that convoluted fpga fabric out :-)23:35
wolfspra1lbut I'm sure they have this carefully thought through and aligned with their major customers23:35
wpwrakthey have the ARM choice already today. seems low-risk for xilinx.23:46
wpwrakfor most customers, the difference would probably be that they don't need to many chips23:47
wpwraki like sebastien's idea of having just the core in an asic23:47
wolfspraulwhere do they have the arm choice today?23:52
wolfspraulyou mean the customers?23:52
wpwrakyes23:54
wolfspraulthat's my point. I don't know the exact relationship between Xilinx and its customers, but I think it's fairly tight.23:54
wpwrakyou once mentioned yourself that everyone you talked to said to use a regular SoC for the core and the FPGA for "special tasks"23:54
wolfspraulI do think the fpga fabric is in the middle of this relationship. A Mercedes s-class supposedly has over 10 xilinx fpgas inside, for example.23:55
wpwrakwow :)23:55
wolfsprauland I think their customers look from the perspective of the ISE design suite23:55
wolfspraulnow, with Zynq, Xilinx takes a big risk. not saying it's bad, businesses have to take risks.23:55
wolfspraulbut those chips boot and act like ARM chips first of all23:55
wolfsprauland Xilinx comes out with an entirely new design suite23:55
wolfspraulI think they want to reposition the entire company and how this company (xilinx) is perceived by its customers23:56
wpwrakah, you mean the R&D investment on xilinx's side23:56
wolfspraulnot an fpga company, but a "chip" company23:56
wpwrakwell, let's see how that goes :)23:56
wolfspraulincluding arm cores (of course, best toolchain, best portability, best this that), programmable fabric, more linear functionality/mixed signal, etc.23:56
wolfspraulyes sure, as I said I am *guessing* only from reading the web23:57
wpwrakkinda like cypress with psoc 3-5. i wonder what happened to them ... checking ...23:57
wolfspraulof course customers decide23:57
wpwrakwow. seems that they are available now (cypress)23:58
wolfspraulas for "make lm32 asic", sure that's a nice plan but just theory. I rather go by where we are today practically.23:59
wolfspraul"making chips", what is that? and why? how do we compete? of course it's "cool", so?23:59
wolfspraulI see the entire logic board just as 2nd level IC packaging nowadays :-)23:59
--- Fri May 4 201200:00

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