IRC logs for #openrisc Tuesday, 2015-10-13

--- Log opened Tue Oct 13 00:00:24 2015
blueCmdpoke53281: I got virtio-net working, wasn't so hard when I figured out how to the Virtio MMIO bus actually works :)03:17
blueCmdpoke53281: with virtio-net the lock-ups are much rarer, but alas still there03:17
blueCmdI did manage to get a full gcc + g++ testsuite run though (well, only 2 timeouts that i manually verified)03:18
blueCmdit looked pretty good, gonna upload the full logs later03:18
blueCmd for now shows what works/doesn't work for me03:20
-!- orsonmmz|away is now known as orsonmmz07:29
poke53281blueCmd: two gcc.c-torture tests fail. This gets worse and worse.07:59
poke53281musl or glibc?08:00
poke5328199.93% success. RISC-V does only achieve 99.6%08:02
-!- Netsplit *.net <-> *.split quits: blueCmd08:45
-!- orsonmmz is now known as orsonmmz|away14:09
latifhi all!18:16
latifquick question: what is the performance result of mor1kx for coremark?18:17
stekernlatif: from old notes, it looks like last time I ran it I got 1.76 coremark/MHz18:24
latifOk thanks Stekern.18:25
latifI have seen conflicting numbers on the web.18:25
latifSo I had to ask here.18:25
blueCmd_stekern: 3 hours to run the testsuite with simexec btw18:38
stekernnice one18:38
blueCmd_can easily be halfed by running gcc and g++ at the same time18:38
blueCmd_but then I get lockups :P18:38
blueCmd_I need to figure out how to figure out what's doing those lockups18:39
stekernare you only running against qemu18:39
blueCmd_yes, qemu-system-or3218:39
blueCmd_need an exact mirror of the filesystem for simexec to work, so 9p works very well. should be possible with NFS as well, but I don't want to mess with that18:40
blueCmd_(nfs would be if I wanted to run or1ksim)18:40
blueCmd_or 9p over network I guess would work as well18:40
stekernwe could add 9p support for or1ksim too... but yeah18:41
blueCmd_but anyway, if qemu-system can be made reliable we gain a lot - everyone that wants to test or1k can very easily boot up an or1k system18:41
blueCmd_apt-get install qemu; ARCH=openrisc make; tar -xf initram.tar.xz; qemu-system-or32 --a-lot-of-flags; done18:42
stekernthe lockups sounds like there's some issue with scheduling18:44
stekernhow is the timer implemented in qemu?18:44
stekernand interrupts?18:45
blueCmd_I'll have to dig into it, but I've suspected interrupts as well. Timers is an interesting idea.18:53
blueCmd_given that ethoc caused seemingly more lockups than virtio-net it seems logical that something around those lines might be a cause.18:54
latifHi blueCmd18:54
blueCmd_I did run a few load tests locally in the console without any network traffic to/from the system and that doesn't seem to lock it up18:54
blueCmd_latif: hello!18:54
latifI have a question.18:54
latifI have a problem with mor1kx on atlys board18:55
latifI couldn't get it work even the simplest thing18:55
latifHow do you make the FPGA .bit file? Do you use fusesoc?18:56
blueCmd_You can, but you don't need to18:56
latifOr you build by makin your own project with xilinx18:56
latifSo, you tested fusesoc ?18:56
blueCmd_I usually build my own projects (but don't tell olofk :P)18:57
blueCmd_latif: I have, quite a while ago though18:57
latif:D ok you got it18:57
latifI have done lots of things with v218:57
latifBut couldnt get it runnig with v3 (mor1kx)18:58
latifI am wondering might it be related with fusesoc18:58
latifCan you send me your .bit file for atlys board?18:59
latifblueCmd: What do you use for boot? SPI Flash?19:00
latifblueCmd: another question, are you developing the ubuntu port for or1k?19:03
blueCmd_latif: sorry, need to run to lunch - but there are other nice(er) people in this channel that I'm sure can help you19:08
latifok. another time maybe. Have a good lunch19:08
stekernlatif: so, are you only unable to get spi thing running under orpsocv3?19:21
stekernerr, fusesoc I mean19:22
stekernand you are having some troubles with or1200 in orpsocv2?19:22
stekernjust to sanity check, you could try to just drop mor1kx in orpsocv2, or have you already tried that?19:23
latifHi stekern19:23
latifI haven't tried to make my own project without using fusesoc. If that is what u r asking19:24
latifHow could I drop mor1kxx under v219:24
stekernbasically, just copy that part of the top file19:25
stekernand make sure the .v files are included in the build19:25
latifI understood what u mean19:26
latifBut are they that compatible v2-v3?19:26
latifWhich are the highest top files that are compatible between v2 and v3?19:27
latifLast time I checked they are not compatible19:27
stekernwell, you probably have to change the signal names where mor1kx is instantiated, but it should be easy to figure out by looking at the differences between or1200 in the two of them19:28
stekern... I meant that you should only copy the mor1kx instantiation from the fusesoc atlys top file to the orpsocv2 top file19:28
latifOk I understood. Let me give you the exact RTL file names.19:30
latifor1200_top.v is the top file of the openrisc core19:31
latifThe ports of the module defined in this file is not compatible with the top file of mor1kx.19:32
latifmor1kx.v for mor1kx19:33
latifAvalon bus is used in mor1kx. And wishbone bus is used in or1200. So just changing the port names is not enough.19:36
latifstkern: wouldn't you agree?19:36
olofkAlright! FuseSoC can now extract filesets from ipxact files and use them in the .core files21:28
GeneralStupidolofk: what do you think about "integration" e.g. add a "System" to the bus by a command21:29
GeneralStupidolofk: do you know "Qsys"? From Alterra21:29
olofkNot completely sure this is the way I want it in the end, but the current idea is that you can specify an ipxact component and then src_files = file1.v @foo file2.v to insert filesets from the ipxact file21:30
olofkGeneralStupid: God yes. That tool gives me nightmares21:31
GeneralStupidolofk: Ok, it's like most complex gui tools are...21:31
olofkGeneralStupid: If you mean block-level design when you say integration, current plan is to take advantage of IP-Xact to do that21:32
GeneralStupidolofk: but the idea, to configure a core and add (graphically) co processors or stuff is great.21:32
GeneralStupidolofk: that looks great!21:33
olofkYes, I agree, but I don't want that to be a responsibility of FuseSoC. It's way too complex to handle there21:33
olofkCheck out Kactus2 if you want to test an open source IP-Xact tool21:33
olofkTime to sleep now21:33
andrzejrolofk, I wonder if you could write a short tutorial on ipexact+fusesoc. This looks like a useful flow but it is rather confusing.21:33
GeneralStupidme neither. But it would be nice to have something like "add stuff" so you dont have to generate the "bus" by hand21:33
GeneralStupid(BTW: FuseSoC makes openrisc usable without that much pain, its a great  tool)21:34
GeneralStupidandrzejr: isnt it always confusing if you try to make a really complex task easy to use? :D21:37
GeneralStupidandrzejr: it these things would 'just work' That would be great... My experience tells me: "you will always have to do stuff by hand" -.-21:37
andrzejrI like code generators. But the have to get the job done and I must understand what they do and how to use them.21:48
GeneralStupidi like generators as long they work...22:25
--- Log closed Wed Oct 14 00:00:25 2015

Generated by 2.15.2 by Marius Gedminas - find it at!