IRC logs for #openrisc Saturday, 2014-06-28

--- Log opened Sat Jun 28 00:00:54 2014
mor1kx[mor1kx] skristiansson pushed 1 new commit to master:
mor1kxmor1kx/master 1762954 Stefan Kristiansson: cappuccino/lsu: let stores go through to dcache when disabled...00:25
olofk_stekern: Cool stuff06:06
olofk_What's the coremark for a single CPU?06:14
stekernolofk_: under Linux, 1.66 coremark/MHz07:00
stekernolofk_: what's the status on your de0 nano fixes btw?07:23
stekernI've got some changes I'd like to add, but they clash with yours so, should I:07:25
stekern1) just do my changes and ignore you07:25
stekern2) pick up the clashing patch and fix my comments and commit that and then do my changes07:26
stekern3) wait for you, because you've just fixed them07:27
olofk_Wait for me sounds like the worst idea ever08:25
olofk_Do your stuff and I'll rebase when I have the time08:26
stekernok, cool, thanks08:38
olofk_But if any of my patches are helpful, feel free to change and apply them09:05
stekernyeah, I think I have something that need wb_intercon regeneration, so it's mostly related to that09:06
stekernand the only comment that needed some change there was the size from dec->hex09:08
juliusbhello from ehsm!09:40
juliusbstekern: that's some impressive coremark!09:41
juliusbdual core linux?09:41
juliusbit took less than 10 minutes from saying hello sb0 to us discussing lm32 vs or1k ;)09:44
stekernjuliusb: yup, dual core linux09:51
juliusbway cool. I will quote those numbers toay09:55
juliusbyou can watch live as I do it if you like
stekernat what time?09:58
stekernand, will there be non-live recordings available?09:58
juliusbummm, 2:30 Germany time09:58
juliusbalmost certainly09:58
juliusbbut in about 2.5 hours I think. It's just coming up to midday here09:58
stekernok, I'll probably be afk then :(10:05
juliusbwell, will make a shout-out anyway in case you are10:06
juliusbis a XILINX variable required in your path when running FuseSoC for the atlys (or other Xilinx target boards?)10:10
stekernyes, you need to do the source /path/to/ISE/ dance10:10
juliusbwell, I didn't have /path/to/ISE/bin/lin in my $PATH, maybe that's all that is required?10:11
juliusbyes, that appeared to be all that was reqired10:12
juliusbgah! license file required to build for atlys with ISE14.6?!
juliusbBloody Xilinx, honestly...10:22
juliusbis it possible I'm not using the right ISE? Or even then, do I require a license?10:24
stekernyou always need license files10:29
stekernbut, you shouldn't have /path/to/ISE/bin/lin in your path, the source /path/to... handles that10:39
juliusbobviously haven't used ISE on this machine in a while (ever??)11:33
juliusbwell that's less painless to do than I remember. just download a file and point LM_LICENSE_FILE at it11:47
juliusbthey're not even really node locked anymore, what's the point Xilnix?11:48
poke53281blueCmd: I tried to chroot into the debian environment. Unfortunately I get a segmentation fault after a while.11:50
poke53281I tried it with your Linux version:
poke53281According to your debian it should work with or1ksim. Is this true?11:53
juliusbI seem to recall there was a nice diagram done of the debugging possibilities of OR1K systems12:02
juliusbolofk_: I found this one by trawling the archives:
juliusbit's a good picture12:02
juliusbdid you ever do another spin or did it ever go up in print anywhere?12:03
olofk_juliusb: I haven't done anything more with that pic, but I had in mind to do a blog post where I could use it to explain some concept. Haven't done that yet12:15
juliusbwell, do you mind if I use it in the presentatino?12:16
juliusbI think it's the goods12:16
olofk_Sure. Use it12:17
olofk_I tuned into the live stream now. You're up in 15, right?12:17
* olofk_ will make popcorn12:18
juliusband thanks12:18
olofk_Blast them away with the OpenRISC greatness, and always remember why we are doing this ;)12:18
juliusband money12:18
juliusbpreferrably groupies with money12:18
olofk_Break a leg12:24
olofk_Woohoo!! It's starting12:31
olofk_Is that mor1kx explanation an afterthought? :)12:33
sb0juliusb, no, we discussed making yet another most probably failing open softcore vs. fixing the existing ones12:41
sb0juliusb, and that chisel is an academic toy12:42
sb0the or1k gcc is not upstream12:46
sb0the lm32 gcc has bugs12:47
sb0the or1k llvm is not upstream, and the inline asm lacks features (that might have been done and need merging)12:47
sb0mor1kx is bloated12:47
sb0the lm32 verilog code is a tad ugly12:48
sb0all this stuff could be rewritten in migen12:48
sb0it's not like there isn't work to do on the existing stuff...12:48
olofk_I LOVE the ecosystem picyure :)12:48
sb0and yeah, or1k arch has flags and delay slots that could be nice to remove12:48
sb0and heck, I could even spend some m-labs money on getting some of this done. we're not groupies, we just want to have excellent tech.12:50
olofk_Hi, I'm watching :)12:51
* olofk_ applauds as well12:57
juliusbsb0: I paraphrased13:44
juliusbsb0: I thought that ecosystem picture could have been laced with more humor but I couldn't think of more13:45
juliusbbut I hope you noticed the "software" label very close to a load of bugs ;)13:45
juliusbwhoops, ecosystem comment for olofk_ I meant13:46
sb0juliusb, seriously, forget academia. or just name one successful EDA tool or IP core they released.13:48
sb0all they can do is buy zynq boards13:48
juliusbMIPS and SPARC came out of academia13:48
sb0yeah, right. last time I checked, you needed a virtex-5 110 board to run opensparc, and it'll still be slow13:52
juliusbthat implementation came out of industry, the ISAs came out of acaedmia13:54
sb0I said 'IP core'. not ISA.13:54
sb0and every single actual core coming out of academia is not usable13:55
juliusbwell,  I mentioned RISC-V today in terms of it being a good-looking ISA13:55
juliusbthere is a core available, but it's written in Chisel, and I'm not the biggest fan of that, something in Verilog would be good, but anyway, these were observations, not statements of intent13:56
juliusbI will drink a beer13:56
juliusbthat is a statement of intent13:56
juliusband a true one13:56
olofk_sb0: I'm inclined to agree here. There are a lot of concepts, ISAs, new languages and stuff coming from academia, but I can't think of a single core actually :)14:10
-!- rfajardo is now known as rschmidlin15:11
stekernsb0: re mor1kx bloat, with a fresh release out and me poking around the source for a bit lately I think now might be a good moment to poke at that17:05
stekernone thing I have in mind for your typical use case is to go over mor1kx_ctrl_cappuccino.v and make some things optional, some might be 'mandatory' by the spec, but at the same time very specific implementations might not be interested in them17:07
stekernfor instance some things like the existance of the npc (which is a rather large mux) is only of high importance if you have a debug unit present17:10
stekern*of the npc spr17:10
olofk_stekern: If you only execute one instruction, you can probably throw away the complete npc18:37
stekernolofk_: the point is, the npc spr is only used for exactly one thing, the spr register18:42
stekerninternally 'the next pc' is a complete different story18:43
stekern...and since it's 'disallowed' to read the npc spr from software, it's really just a du thing19:09
olofk_I'm starting to wonder what options we would have to do more invasive changes to the spec19:11
stekernbut since there's sw that we care about around that disobey that rule, so it's been left in as unconditional default19:11
olofk_Like marking mandatory things as optional19:11
olofk_ah ok19:11
olofk_Out of curiousity, what software is cheating?19:11
stekernour newlib/libgloss code19:12
olofk_Only that?19:12
stekernI think so, I've been meaning to fix that forever19:13
stekernbut people tend to try to use the npc spr from sw all the time, _franck_ did it just a week ago19:13
stekern(marking mandatory things as optional) we have to remember that it's one thing to be "or1k spec compliant" and for people to want to use an or1k ISA core that can be configured to bend some rules19:15
stekernmandatory things should of course always be default on, but that doesn't mean that someone might want to opt out things because they simply don'y need them for a specific use case19:16
olofk_All very true19:23
stekernall the overflow stuff is one other thing, it's mostly pretty useless for most use cases19:26
stekernI think most of that is optional in mor1kx already though, and I can't remember how much, if any, of it is actually mandatory19:27
stekernhmm, mibuild seems to miss the -detail map option as well20:28
--- Log closed Sun Jun 29 00:00:55 2014

Generated by 2.15.2 by Marius Gedminas - find it at!