| --- Log opened Thu Jun 12 00:00:30 2014 | ||
| olofk | Crap! I Alt-F4'd my simulation | 07:25 |
|---|---|---|
| olofk | Oh well. At least it came far enough to start NFS | 07:26 |
| olofk | Not sure I need NFS in Icarus. Should probably make some changes to that kernel | 07:26 |
| olofk | Are there any plans to do something about arch/openrisc/kernel/vmlinux.lds.S in the kernel? Do any other arches have the same issue? | 07:33 |
| olofk | Ahh.. I see that other arches have ifdef's controlled by a CONFIG_. Should we do something similar? Either a or1k/or32 checkbox, a list of valid names or free text | 07:55 |
| olofk | a list would perhaps make the most sense if we want to have or1k, or1knd and or32, and perhaps some future sub arch or 64-bit versions | 07:56 |
| olofk | Please note that I don't really have a fucking clue about these things, so comments are appreciated :) | 07:57 |
| LoneTech | olofk: I'm not even sure what issue you refer to | 08:43 |
| stekern | olofk: yes, we should do something about it | 08:50 |
| olofk | LoneTech: We can't link the kernel with the or1k- toolchain | 10:55 |
| LoneTech | ah | 10:56 |
| olofk | Hmm... so far in the boot I have executed ~38M instructions, and spent 10% of them inside of memset | 12:43 |
| olofk | So optimizing that to write words instead should increase the performance ~7.5% (not sure on the math here) | 12:44 |
| olofk | What the hell... there's even a nop in there | 12:45 |
| olofk | In the inner loop | 12:45 |
| olofk | Are the memset routines from the kernel or from gcc or newlib? | 12:46 |
| LoneTech | I would have guessed gcc, as I don't recall the kernel ever using anything of newlib | 12:47 |
| _franck__ | http://juliusbaxter.net/openrisc-irc/%23openrisc.2014-05-15.log.html | 12:56 |
| _franck__ | look at stekern answer | 12:57 |
| _franck__ | olofk: this is for you ^ | 12:57 |
| _franck__ | I started something but I switched to something else. I compiled musl memcpy, disassembled it and started to optimize it in order to put the result in the kernel | 12:59 |
| olofk | Yeah, I remember that we have talked about it before | 13:10 |
| stekern | ~70 patches left in my rebase experiment... | 13:12 |
| olofk | The current one doesn't seem very complicated though. Just shuffling around a few instructions to remove the nop would make it ~20% faster (stalls not taken into consideration) | 13:13 |
| olofk | Looks like my simulation stopped now | 13:17 |
| olofk | Poor Icarus | 13:17 |
| olofk | Oh well. It got through almost ~39M instructions at least | 13:18 |
| olofk | Stole the memset from microblaze instead. Let's see how that turns out :) | 13:56 |
| olofk | stekern: Exactly what are you doing? Do we have 400 patches in the OpenRISC tree that haven't made it upstream? | 14:02 |
| stekern | olofk: 740 | 14:17 |
| stekern | but at least ~670 of those have made it to upstream, but in another form than they were committed | 14:19 |
| ysionneau | what are you going to do? create a new branch from top of upstream and rebase the not yet upstreamed patches onto it? | 14:21 |
| stekern | that's what I've done, yes | 14:25 |
| ysionneau | then you will say "please don't develop anymore on the old dev branch but on the new one rebased upon upstream" ? | 14:26 |
| * ysionneau is curious about the workflow | 14:26 | |
| stekern | no, this is just my own tree, I'm just doing it because I wanted to get an overview of the differences | 14:27 |
| stekern | everyone should develop against upstream anyway | 14:28 |
| ysionneau | and it's just you in your local development cycle who diverged from upstream? | 14:28 |
| stekern | what jonas will do with his tree, that's up to him | 14:28 |
| stekern | no, there's all kind of stuff there | 14:29 |
| ysionneau | I mean, do you have an "official" repo for the openrisc community with the openrisc kernel? | 14:29 |
| stekern | old and 'new' | 14:29 |
| ysionneau | which you from time to time rebase on Linus' tree | 14:29 |
| stekern | jonas send pull-request to Linus from his repo at git.openrisc.net | 14:30 |
| stekern | ...but it's another branch than the master branch | 14:31 |
| ysionneau | so basically everyone commit in master branch in git.openrisc.net | 14:31 |
| ysionneau | then jonas keeps some other branch up to date to send to Linus | 14:32 |
| ysionneau | like "for-linus" | 14:32 |
| ysionneau | ah there is stefan/linux and jonas/linux | 14:33 |
| stekern | yes ;) | 14:54 |
| _franck__ | "everyone should develop against upstream anyway" ---> totally agree | 14:56 |
| stekern | ...but currently there are a bunch of patches scattered around that you need | 14:58 |
| _franck__ | that too bad | 14:58 |
| _franck__ | and that shouldn't be the case | 14:59 |
| stekern | I think there are only two that _really_ matters | 14:59 |
| _franck__ | so they _really_ should be pushed | 15:00 |
| _franck__ | :) | 15:00 |
| stekern | and the other jonas posted to lkml in 2011, but got no response | 15:00 |
| _franck__ | so we have an unmaintained upstream kernel. stekern you should volunteer to be an openrisc kernel co-maintainer | 15:29 |
| -!- Netsplit *.net <-> *.split quits: zama | 19:02 | |
| -!- Netsplit over, joins: zama | 19:03 | |
| stekern | _franck_: http://git.openrisc.net/cgit.cgi/stefan/linux/log/?h=openrisc-3.15-rebase | 20:01 |
| stekern | try that and see if you have the same problem as you had with vanilla upstream | 20:02 |
| -!- Netsplit *.net <-> *.split quits: Amadiro | 20:18 | |
| -!- Netsplit over, joins: Amadiro | 20:18 | |
| --- Log closed Fri Jun 13 00:00:32 2014 | ||
Generated by irclog2html.py 2.15.2 by Marius Gedminas - find it at mg.pov.lt!