IRC logs for #openrisc Sunday, 2015-03-29

--- Log opened Sun Mar 29 00:00:42 2015
DarkFoxpoke53281: Just found a quote from you while searching up openrisc and alpine - you've made a suggestion for running alpine on openrisc. Just curious to if you had made any progress on that? :)02:42
DarkFox<http://juliusbaxter.net/openrisc-irc/%23openrisc.2015-01-19.log.html>02:42
DarkFoxJust noticed dates... poke53281 I'm gonna guess not yet - but I am sure interested. :P02:43
poke53281DarkFox: Unfortunately no.09:07
poke53281It is still on my todo list.09:07
poke53281unfortunately, alpine doesn't support cross compiling. Therefore I wanted to fix qemu-user-mode first.09:08
poke53281Obviously I haven't done it.09:08
DarkFoxpoke53281: Fortunately; alpine isn't a black box and can always be improved :D13:06
DarkFoxpoke53281: Until we are both online for a chat - I guess I can only wish you luck here.. I'd like to work on this problem and learn more about openrisc and porting for it - so if there is anything that I can do; I'd be interested. :-)13:07
DarkFoxQuestion for the community: Is jor1k still the fastest or1k emulator? o.0 "It seems that at the moment jor1k is the fastest emulator" - <http://openrisc.2316802.n4.nabble.com/OpenRISC-The-reason-why-QEMU-is-such-a-slow-OpenRISC-emulator-td4641631.html>13:18
DarkFoxwhich is not acceptable. (That was 2013)13:18
DarkFox(Paste fail "which is not acceptable" was meant to be in the quote..)13:19
* DarkFox wonders how difficult it should be to compile a rust written RTOS (like <http://zinc.rs/>) for OpenRISC. LLVM appears to work fine - while rust uses LLVM extensively. I may hold back until I can download an image (hoping for alpine) for openrisc for such development purposes.13:40
olofkDarkFox: or1ksim is the classic simulator. It should be faster than jor1k (although I'm not sure if that's the case)16:28
olofkpoke53281: What's going on with the qemu patches btw? Is everything upstreamed yet?16:32
daliasbtw does the simulator do cycle-exact or at least approximate timing?16:43
dalias(motivation of the question is whether it can be used to evaluate relative performance of different implementations of a function for or1k)16:44
dalias(for example if we wanted to compare different memcpys)16:44
olofkdalias: Don't think it's cycle accurate, but you can use a verilated RTL model for that16:46
olofkNot nearly as fast, but works well enough to boot linux16:47
olofkBut for something like memcpy, you would probably have to factor in memory latency as well.16:48
daliasyeah16:54
daliasi was just wondering if there were good methods for doing these kinds of performance evaluations without real hardware16:54
olofkdalias: Standard event-based simulators will be most accurate, but not fast enough. (I gave up trying to boot Linux after ~24 hours). Cycle-accurate such as verilator is probably a good compromise if you want an accurate model of the CPU + caches which is fast enough to do a bit more16:59
stekernyeah, and you can easily add memory latencies towards the block ram if that's required17:01
daliasbtw i don't think i'll have a chance to, but if anyone is interested in doing optimized versions of some of this stuff for musl/or1k i'd be happy to look at it :)17:04
daliasi wonder if there's any reasonably-figured "level of difference" (e.g. 2x faster) on the emulators where it's overwhelmingly likely that the faster version would be faster on real metal too17:05
daliaswith qemu-arm i've had pretty comparable 'relative performance' measurements vs real arm hardware17:06
olofkstekern: Do you have any bandwidth for upstreaming Linux patches?19:19
olofkpoke53281: I can't find any of the busybox-1.23* patches listed in http://pastie.org/10056824 here https://github.com/sabotage-linux/sabotage/tree/master/KEEP19:20
stekernolofk: I've got bandwidth, but I'm still waiting on Jonas to sign my GPG key23:54
--- Log closed Mon Mar 30 00:00:44 2015

Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!