IRC logs for #openrisc Friday, 2014-03-21

--- Log opened Fri Mar 21 00:00:29 2014
analognoiseSo what's stopping the movement to a new portal?00:09
blueCmdanalognoise: nobody doing it :P00:11
blueCmd - yeah, I just forked archweb and played some with the colors00:44
blueCmdmight work00:44
blueCmdsomething is weird with IPC in either 1) or1k with qemu or 2) glibc. (or more, but I think those are the most probable)01:31
blueCmdfakeroot-sysv just hangs, fakeroot-tcp seems to work though01:31
stekernblueCmd: what's IPC?03:47
stekernI'm guessing inter-process communication in this context03:53
stekernI *really* hate theater of the living arts03:54
analognoiseERROR: Failed to run the simulation.03:58
analognoiseFailed to run Icarus Simulation03:58
stekernsounds like failure04:02
analognoiseERROR: Can't find core 'wb_bfm'04:03
analognoiseHaving a hell of a time :)04:03
stekerngive me some more context and I might be able to help04:03
stekernblueCmd: (reading backlog about portal) I agree about a new portal. But, the last effort to do that ( ended in a big fight and the idiocy of having several mailing lists are a result of that04:10
analognoiseI would give you more context, but I'm not at all sure where the error(s) ile04:34
analognoiseFor some reason it installed on my other machine just fine04:34
analognoiseThey're both running Crunchbang (Debian based) in Virtual machines04:35
stekernmore context could be: what you actually are doing (reading the backlog I saw running fusesoc/orpsoc-cores and the wb_bfm gave that away) and/or paste the command you are trying to run + the output04:40
analognoiseI'm trying to run the example to check the installation (fusesoc sim wb_bfm)04:46
analognoiseanalognoise@crunchbang:~$ fusesoc sim wb_bfm04:46
analognoiseERROR: Can't find core 'wb_bfm'04:46
stekerndid you add a fusesoc.conf file to your build directory?04:55
stekern(and I see you are trying to run it in your home dir, you might not want to do that)04:55
stekernfusesoc.conf should contain something like this:04:58
stekerncores_root =../orpsoc-cores/cores04:58
stekernsystems_root =../orpsoc-cores/systems04:58
analognoiseAh, but if I cloned from github and did the autoreconf -i...04:59
analognoiseI was just following the steps04:59
analognoiseShould I have done it differently if I cloned from Git?05:00
stekernthe 'autoreconf -i' is related to the ./configure step05:00
stekernthe fusesoc.conf is related to running the installed fusesoc05:00
stekernso, no, your installation is fine05:01
stekernyou just have to setup your 'runtime' environment correctly now05:01
analognoiseI have fusesoc.conf in ~/05:01
stekerneven though that works, I would advise to run it in a seperate directory05:02
stekernbecause it will scatter some files in there05:02
stekernif you don't care polluting your home dir, it doesn't matter though05:02
stekernso... with the fusesoc.conf and similar lines to what I pasted in it, does it still give the error?05:04
analognoiseERROR: Error: Failed to register cores root05:05
analognoiseERROR: ../orpsoc-cores/cores is not a directory05:05
stekernmmm... *similar*05:06
stekernyou don't have orpsoc-cores in /home/orpsoc-cores, do you?05:06
trsohmersI'm new to the OpenRISC project, so what is the status of the ASIC hardware implementation?06:10
trsohmersand hi analognoise06:13
stekerntrsohmers: if you refer to the donation project started by orsoc, it's pretty much stalled06:20
trsohmersstekern: Yea I noticed. So is the majority of the costs just to get into a fab house?06:20
stekernI'm not betting any money on it ever being unstalled neither06:20
stekernif you mean fabrication costs by "get into a fab house", then yes06:22
stekernthe designs already exists06:23
trsohmersHave you looked into alternative fabrication techniques?06:23
stekernalternative to what? as far as I know, all techniques are expensive06:24
stekernbut I'm no expert in ASIC fabrication06:25
trsohmersSo I have not had experience with IC fabrication, but I have worked with electron beam lithography for MEMs systems06:25
trsohmersit is shit slow in comparison to any photolithography technology, but you can (slowly) dry etch a wafer with transistors06:26
trsohmersDo you know who was involved in the ASIC project, and what they had synthesized?06:27
stekernI don't, and I think it only got to the stage of begging for money06:28
trsohmersAny idea how to find out? I'm not really knowledgable on the HDL and design side, but I'm familiar with semiconductor fab06:29
trsohmersbasically, if I have a GDSII file, I can fab anything down to ~6nm06:32
trsohmersbut yield would be shit at that resolution, so it would probably be more effective at ~22-28nm06:33
stekernyou probably have a better chance getting HDL coders among the active people in this channel that are interested in collaborating with you06:34
trsohmersShould I hang around here, or is there a better place/person I can look at/talk to?06:35
stekernhang around here, all the core openrisc devs are here06:36
analognoiseYeah and I've gotten my questions answered. It's actually pretty awesome. (Just chiming in)06:37
analognoiseAlthough I'm not sure wtf is the problem right now :)06:37
stekernand there are the mailing lists, for more elaborate questions and ideas:
trsohmersI don't have a copy of any Cadence or other EDA software that to my understanding would generate the GDSII files, so I'm wondering if the ASIC people got to that stage06:39
analognoiseElectric VLSI can do that, and it's free06:39
analognoiseThat'd be a hell of a layout project06:40
analognoiseBut the parameter extraction would be questionable at advanced nodes06:40
analognoiseThere's no reason to go to advanced nodes for something this size though - save money, older process.06:40
trsohmersWell, the same machine can do anything down to 6nm06:41
trsohmersI have free (within reason when it comes to amount of time) access to the ebeam machine06:41
trsohmersbut using a larger process would cut down on the time it takes to etch significantly06:42
analognoiseThere's a reason production chip manufacturing isn't at 6nm yet. A slew of them, actually.06:43
stekerntrsohmers: another problem is that the fabs have their own logic libraries, which they charge money for06:43
stekernlike RAMs and such06:43
analognoiseLogic libraries and, more importantly, verified models of their process.06:43
analognoiseThat's really what you're paying for.06:43
stekernyes, that too06:43
analognoiseAny dope can lay out a 6T register cell and pattern it. But how does that nice layout actually look on the process you're going to spend 1-2 million dollars for one run doing?06:45
analognoiseUnless you do a multi project wafer06:45
analognoiseWhich I've thought is a better place to start06:45
analognoiseBut it's still a little pricey for an individual.06:45
trsohmersThe major part of fabrication is the lithography step, which I've done in other scenarios (not IC fab)06:46
trsohmersThe most expensive actual cost for the fab using photolithography is the creation of the masks, and maintenance for the optical equipment06:47
trsohmersDoing direct etching is a significant time increase, but if you are doing a very small run, it's relatively easy06:48
trsohmersAt least from my understanding06:48
analognoiseI don't know if there are published layouts for the OR1K designs. That'd be interesting...06:49
trsohmersI haven't found any, so I came here to see if anyone knew about anything06:50
analognoiseWell it'd be a hell of a lot of fun to build them.06:53
analognoiseThat's for sure.06:53
analognoiseYour process will determine what types of blocks you can build, then your parasitic extraction will tell you how the device should actually be performing06:53
analognoiseWe'd have to work out a clock tree and all that jazz without fancy tools06:54
analognoiseI'm down. Anyone else?06:55
trsohmersI'm not sure how much I can help on that side of things, but I would love to learn along the way06:55
trsohmersBut I'm obviously up for it on the fabrication side06:56
analognoiseI love analog shit (it doesn't get more brutal than that) but I make my bread with HDL. I've only been seriously considering OR1K for a production design for a short period of time, but I like the way it ties things together so far (when it works - it's fine on my other machine, damn it)06:58
analognoiseI'd say go for something like the AK-47 of designs - not cutting edge at all, but very cheap and effective. As a proof of concept. Also badassery.06:59
stekernI'd be the first to agree on that07:01
trsohmersFrom what it seems like, whoever was working on this OR1K ASIC didn't publish much of anything07:01
trsohmersSo I guess this should be the OpenOpenRISC ASIC? ;)07:01
analognoiseFull custom ASICs are...fucking hard. They might have just gotten overwhelmed.07:02
trsohmersMost likely. I was secretly hoping they got a lot of the design finished and just couldn't get to fab07:03
trsohmersI suppose the likelyhood of that is slim, as nothing seems to have been released07:04
analognoiseI'd agree07:07
analognoiseI see fusesoc.conf in the fusesoc main folder - is it also being copied somewhere else & then referenced?07:12
trsohmersorpsoc is now fusesoc?07:14
trsohmersolofk: Ping07:16
analognoiseAlright yall I gotta pass out07:18
analognoiseI'll catch everyone tomorrow07:18
_franck_web_trsohmers: yes, orpsocv3 is now fusesoc08:16
trsohmersHi _franck_web_08:17
trsohmersYea... are you knowledgable of any attempts to make a OpenRISC ASIC?08:18
_franck_web_not at all08:19
trsohmersor know anyone in here that does08:22
_franck_web_sorry no08:23
jungmastekern: i tried it again, configure: error: no usable dependency style found08:54
stekernjungma: ok, but at what stage does that happen?09:07
stekernanother hour wasted on connecting a reset signal with wrong polarity09:48
jungmastekern: at the stage "Build gcc"09:57
stekernjungma: could you post the config.log somewhere?10:02
_franck_web_../or1k-gcc/configure: line 4808: g++: command not found10:41
stekernyes, you'll need that10:41
jungmayes, i discoverd that also, so the dependecy list should be updatet10:41
jungmafor centos it is gcc-c++10:42
jungmanot g++10:42
jungmadifferent package name10:42
stekernok, feel free to add a note on the wiki, somewhere around the other dependencies10:42
_franck_web_I'm using a centos right now and I have g++ in my PATH10:43
_franck_web_which version do you have ?10:43
jungmaits only the package name10:43
jungmathe executable is still named g++10:44
_franck_web_ah sorry :)10:44
jungmaand it is necessary to run this:10:44
stekernI think the dependency list stems from the time when gcc wasn't built with g++10:44
jungmacd ../or1k-gcc10:44
jungmaI have a new list now... i will post it later into the wiki10:45
blueCmdstekern: yes, IPC. stekern: ah I see (re: portal)11:01
blueCmdwow is really bad :)11:04
blueCmdA guy emailed me and asked me to fix it11:04
blueCmdjeremy_bennett: you spoke that you will have some time to dedicate to GCC. I'm very much interested in upstreaming it as well, so any help you can give in the effort would be appriciated. I don't know if olofk is up for another scan of the revision systems, but that would be super helpful.11:07
jeremy_bennettblueCmd: It won't be me, it will be a new member of staff who joins us at the end of June.11:21
jeremy_bennettHe'll be able to support you then. Thanks for your work on this.11:22
blueCmdjeremy_bennett: yes, I didn't mean you in person :) that's great!11:48
blueCmdjeremy_bennett: so, I emailed you about this but maybe it got lost - now when binutils is heading upstream containing o1knd references, would you consider it to be included for upstream config.guess/config.sub?11:53
blueCmdstekern: olofk: my board designer wants a name for the board de0_something - I'm thinking de0_orpsoc but, is it named orpsoc? or fusesoc?11:56
blueCmdor is just fusesoc the builder?11:56
stekernyes, fusesoc is just the builder12:56
stekernI think that orpsoc-cores fits more into what is orpsoc12:59
stekernand if you want to take the name literally (Openrisc Reference Platfrom System On Chip), what are here is the core of orpsoc:
maxpalnHi, been a busy few weeks so no chance to check-in13:24
_franck_web_hi, good to see you13:25
stekernmaxpaln: wb13:26
maxpalnThings are progressing very well though - the memory controller is pretty much done (I've had precious little time to test it in HW but simulation testing is pretty much done) and I am nearing completion on everything else13:26
maxpalnI have a question though - slightly out of the ordinary13:26
maxpalnmy boss has asked me to do a short presentation to the other FAEs on the ORPSOC platform on Monday13:26
maxpalnI have (traditionally) left it to the last minute to do the prep13:27
maxpalnI wondered if there is any sharable PPT material with general overview content that I could include13:27
_franck_web_juliusb_ might have this13:28
_franck_web_and it should be online somewhere13:28
maxpalnotherwise, I hope the vacation went well - you must have only just returned if my memory serves me well13:28
maxpaln_franck_web_ - great, I'll see if I can find it13:28
_franck_web_stekern went in vacation far away :)13:29
_franck_web_(I just did a search with "dropbox" on the IRC logs)13:31
maxpaln:-) great - nice idea, I'll have a look at that13:31
stekernthen there's probably the orpsocv3/fusesoc/orpsoc-cores slides from this
stekernif you ask olofk about them13:32
maxpaln...more good suggestions. olofk: do you have any slides you think could be relevant? Top level stuff is fine13:33
maxpalnFWIW does anyone have a meaningful or otherwise metric of performance for the latest iteration of the ORSOC that I can include in material I am presenting on Monday?15:35
blueCmdmaxpaln: It's 500 times slower on 'openssl bench' than my i5 laptop :)15:38
blueCmdA very scientfic statement15:38
maxpalnI'm having that one :-)15:38
blueCmdthat's 50 MHz, mor1kx, de0_nano ($79)15:39
maxpalncan I have more detail on the i5 spec too please ;-)15:39
blueCmdsure, let me boot it up15:40
maxpalnonly kidding - although 500 times slower for something that is probably running at 1/500th the clock rate is not bad15:41
maxpalnooops, bad maths - 1/50th the clock rate15:42
maxpalnworse, but still pretty impressive15:42
maxpalnespecially if you have several cores running on the i5!15:42
blueCmdit's 2.6 GHz, i5 3320M15:42
blueCmdbut it's not very scientificly done15:42
maxpalnLOL - I might still include it to see if everyone is awake!15:43
blueCmdI just ran the same thing on my laptop and the or1k thing15:43
blueCmd500 times slower is quite good. We don't have any optimized assembler in openssl (as x86_64 does), atomic operations are done via expensive syscalls (shouldn't matter much since it's single threaded, but still) and the CPU is not doing out of order execution15:44
maxpalnagreed - plus Intel have quite a few engineers working on theirs!15:45
blueCmdstekern: crack-md5_5.0a-9.3_or1k.deb - now we can do fun stuff!15:48
blueCmdthat will be useful16:15
stekernah, yes, crack = fun, that I've heard. Especially when combined with pipes16:20
blueCmdand printer drivers16:21
stekerncombining crack with printer drivers, that I haven't heard of, but I guess anything goes when it comes to drugs16:22
stekernah, now I get it:
stekernmaxpaln: you could quote a coremark score of 88 for mor1kx @50MHz on de0 nano16:26
blueCmdso what I'm doing is building all debian packages, the first pass (the ones that do not depend on anything but essential)16:29
blueCmdwhich means we now have 'cdbackup' available16:29
maxpalnstekern: thanks16:30
blueCmdfor all your FPGA-based backup needs16:30
stekernthat's cool, and reminds me that I should test it on real hw16:41
stekernperhaps hacking together some ise support to fusesoc and an atlys port is next on the agenda to fulfill that16:42
analognoisehowdy all.16:46
blueCmdanalognoise: hello!17:03
blueCmdstekern: that would be awesome :)17:04
blueCmdanalognoise: what are you up to? any cool ideas your hacking on?17:09
analognoiseblueCmd:Too many projects, never enough time. :) I'm verifying the datapath for a space program  currently, so I beat my face against Verilog all day. I want to get conversant with the  Openrisc materials so I can use it as an embedded soft processor, then I'd like to see if we  can get some movement to lay one out.17:17
blueCmdspace! space is cool :)17:17
analognoiseSpace is pretty cool, but it is weak from a technology standpoint. hahaha17:18
analognoiseUnless you're doing radio/satellite comm stuff, which is intense.17:18
analognoiseWe use the absolute LATEST in 1990's technology.17:19
analognoiseWhat about you?17:19
blueCmdI'm currently building all debian packages.17:28
analognoiseThat's awesome! I have no bloody clue how to do that17:29
analognoiseThat'll really get the work out there17:30
analognoiseSo you're doing these for openrisc? Any way I could help?17:35
blueCmdanalognoise: well, my build server is doing that17:37
blueCmdit's just a slow process really17:37
blueCmdbut I appriciate your offer. I'll let you know if I think of anything17:37
analognoiseNo worries17:39
trsohmersGoodmorning folks19:45
trsohmershey analognoise19:45
trsohmersolofk: Are you here?19:51
trsohmersor jeremy_bennett19:57
trsohmersjuliusb_: Are you Julius Baxter?19:57
blueCmdtrsohmers: yes, he is :)19:59
trsohmersJust found a powerpoint by Julius and Jeremy about the process of making an OpenRISC ASIC19:59
trsohmersblueCmd: Do you know anything about the OpenRISC ASIC development?20:02
blueCmdtrsohmers: well, I know there is none - sadly20:04
stekernwell, there are several ASICs with openrisc implementations on them. None that are community driven nor "open" though.20:13
trsohmersAre they all corporate/for profit projects?20:16
analognoiseIt's time we change that.20:21
blueCmdanalognoise: well, sure - we just need cadence / mentor to sponsor the utilites and a fab to do the same - or kickstart something20:34
blueCmdstill, it would probably need a lot of work to make or1k worth doing an ASIC of. just putting it on the chip and hope the performance would increase enoguh to make it worth doesn't seem to be a good idea to me :)20:35
stekerntrsohmers: there's an openrisc processor on allwinner a31 SoCs, a couple of MP3 chips uses openrisc, it has been featured in samsungs DTV SoCs and some more20:36
stekernall corporate projects, where they presumably needed an embedded processor20:37
blueCmdstekern: it would be weird if they didn't20:37
stekernyes M)20:37
blueCmd"Oh, so why do we have this OpenRISC thing here?"20:37
blueCmd"I don't know, I thought you needed it!"20:38
stekernyou never know in the corporate world ;)20:38
blueCmddebian has a _lot_ of weird packages20:38
blueCmdI just built jpnevulator - god knows what it does20:39
stekerna thing or two lost in the translation in a multinational company, and anything can happen ;)20:39
blueCmd"Minimize risk?" "Yes, yes, openrisc!"20:39
trsohmersblueCmd: Well, I can probably help on the fab end20:44
blueCmdcool! trsohmers: what do you do?20:44
trsohmersI've done some work on MEMS fabrication using direct etching tech... Electron beam and focused ion beam primarily20:45
trsohmersthe electron beam lithography machine I have access to can technically etch feature sizes as small as 6nm20:46
blueCmdhah, that's cool :)20:52
blueCmdI've only done full custom stuff for 350nm20:53
trsohmerstransistors have been made as small as 9nm on this machine, but not with good yield AFAIK20:54
trsohmersSo the smallest realistic run with good yield would be around 22nm, and it would take forever to make (E beam is very very *very* slow compared to photolithography)20:55
trsohmersbut you wouldn't have to have every layer done on the ebeam... only the smallest feature layers20:56
trsohmersit would probably still take a full work day at least to do a 8 inch wafer thoug20:56
analognoiseblueCmd: We could use Electric VLSI.21:01
analognoiseAlthough if we were serious, we'd have to sign NDAs for some process, which can really only be done as a corporate entity. For volume production.21:01
trsohmersfor volume, yes. but a wafer a day with a small enough die size isn't horrible21:02
analognoiseFor a multi-project wafer at MOSIS there might be a set of files available21:02
analognoiseIt is if you're paying for electricity for an ebeam machine...21:03
analognoiseI think doing the layout/simulation and then trying to get kickstarter backing on a MOSIS run might be the best option.21:04
trsohmersLuckily I am not paying anything for the ebeam machine, as long as I am not taking up a ridiculous amount of time on it21:04
trsohmersI could technically buy out time on it for ~$10k a month, but I wouldn't be able to use it for production runs (only prototyping)21:05
analognoiseSo we'd have to get a corporate account to get into MOSIS21:07
analognoiseMight want to set up a coprorate entity for that.21:08
analognoiseBah...getting off the topic of openrisc.21:08
analognoiseTrying to explain IC layout to people in #electronics22:20
analognoiseI still think this is worth doing.22:20
blueCmdstekern: - let me know if you want us to build one for you :)23:45
--- Log closed Sat Mar 22 00:00:31 2014

Generated by 2.15.2 by Marius Gedminas - find it at!