IRC logs for #openrisc Sunday, 2015-01-11

--- Log opened Sun Jan 11 00:00:50 2015
ysangkokpoke53282: there were some permissions errors in the new basefs.json, i fixed them myself and committed01:41
olofkpoke53282: Do I need to build a separate binutils for each libc?08:43
poke53282olofk: I guess not. There might be configure options, which one libc needs and the other not.09:41
poke53282MAy I ask you what you are trying to do?09:41
poke53282Another virtual machine image for the website?09:41
poke53282ysangkok: I can't remember, but I guess, that I took the permissions from my archlinux distribution.09:43
poke53282Is there a tool, which checks for the correct filesystem permissions?09:43
stekernhmm, seems like the lto problem is that it exchange a l.lbs from a var that is 0xff with a l.addi with the constant -109:47
stekernwhich of course isn't the same09:47
olofkpoke53282: I'm partly trying to understand how these things fit together and partly just trying to install stuff on a fresh machine to update the build instructions if needed09:48
stekernthe lbs will be 0x000000ff and the l.addi 0xffffffff09:48
poke53282Stekern: wow, I didn't know, that lto is kind of doing this.09:50
stekernyeah, me neither, I've been quite ignorant to what lto does until now09:52
poke53282but wait.09:55
poke53282r[((ins >> 19) & 0x7C)>>2] = ((ReadMemory8(paddr|0)|0) << 24) >> 24;09:55
poke53282This is my code.09:55
poke53282and loading a -1 would be 0xffffffff09:55
stekernsorry, typo09:55
stekernl.lbz09:56
stekernotherwise everything would be in order, as you say09:56
poke53282Yes09:56
poke53282So I would look for a wrong signed or unsigned char.09:57
poke53282Or update to 4.9.2 ;)09:57
stekernl.lbz   r6,0x0(r5)      r6         = 000000ff09:57
stekernl.addi  r6,r0,-1        r6         = ffffffff09:57
stekernfrom or1ksim dumps09:58
poke53282yes, I understand09:58
poke53282and l.lbz is the correct one.09:58
stekernyes09:58
poke53282the result is still correct if the register is treated as 8 Bit in the compiler. Then it does not matter.09:59
poke53282So, this optimization strategy might still be valid.10:00
stekernyes, and this might be the assumption that is done10:01
stekernbut it's then used in blueCmd's atomic stuff that shifts and ands around stuff10:01
stekernand that's probably opaque to the lto logic10:02
poke53282Ok, then just two assumptions interfere here The one of blueCmd and the one of gcc.10:02
stekernso, I would be inclined to try to find a way to tell that no assumptions can be made on the ingoing arguments to blueCmd's stuff, if that's possible10:03
poke53282sounds good10:05
stekerntime to dig myself out from the house, there was obviously some sort of snowstorm here tonight10:10
poke53282And we had yesterday 22°C. The hottest day in winter since the start time for recording10:14
poke53282stekern: How do you update your Linux kernel with the official rep? Which git command?11:40
poke53282git pull https://github.com/torvalds/linux.git gives me conflicts11:41
stekernlike that and then resolve the conflicts12:09
stekernthey are usually straight forward12:09
poke53282Ok, done12:25
ysangkokpoke53282: you must be back in canada? it's not 22 degrees in the Germany i'm in :P12:43
poke53282Nope, Germany. I heard it in the radio.12:50
stekernmaybe the right way to handle this is actually to treat the incoming argument as if the other parts of it than the ones that are to be used can contain whatever garbage13:05
stekernat least the failing test is fixed by doing so...13:07
poke53282Sounds reasonable.13:21
ysangkokpoke53282: ah ok :P13:26
ysangkokpoke53282: it seems buildfs is missing some packages that are shipped in the sysroot13:26
ysangkokpoke53282: is this intensional?13:27
poke53282no, it is not updated yet.13:29
ysangkokpoke53282: you know wget can follow links and read the directory listing itself?13:29
stekernblueCmd: are you up to context switching into some gcc atomic stuff?13:30
poke53282well, the listing has also a lot, which is not included.13:30
ysangkokah ok13:30
stekernblueCmd: http://pastie.org/982535513:31
stekernthat's the patch I have, it's in "cmpxchg_mask"13:32
poke53282ysangkok: Wait one second. I update buildfs13:32
stekernand the idea is that you don't make assumptions about the %4 register, but rather mask out the interesting bits13:33
stekernI just re-ran all the atomic tests and they all pass now13:34
stekerns/interesting/uninteresting13:36
poke53282ysangkok: updated. The filesystem which is built, should be identical with the one in jor1k-sysroot13:36
poke53282at least I hope so.13:37
ysangkokgreat, thanks13:39
poke53282forgot two packages :)13:47
poke53282ysangkok: finally Identical except of the src folder.13:54
poke53282I had to update jor1k-sysroot json file too.13:54
blueCmdstekern: wsup?17:03
stekernblueCmd: read the log? ;)17:05
blueCmdyeah, i was unable to scroll17:06
blueCmdgot myself a new keyboard now so here goes17:06
stekernI have done a minor change to "cmpxchg_mask", and wanted get some input on that17:06
stekernI can still see all I said in my irssi window, how small screen do you have?17:08
blueCmdhm, still cannt scroll - oh well17:08
blueCmd15 lines or so17:08
stekernhttp://pastie.org/982535517:08
blueCmdyes, i see the hilighta17:08
blueCmdbut nothing more :p17:08
blueCmdsorry17:08
blueCmdbut adding another mask seems harmless enough17:10
blueCmdso if it helps / fixes stuff - i have nothing against it. is it for input sanitation?17:10
stekernthe point is that it seems to be so, that you can't assume that the data in %4 is loaded as a 8-bit value17:10
stekernand where the testfailure occurs, lto change a l.lbz (that will load 0x000000ff) to a l.addi -117:12
blueCmdmakes sense, probably possible to constrain it17:12
blueCmdah hah, thats a bit naughty of it.17:12
stekernthat was my first thought, but on the other hand, that lto optimization indicates that you should only *pick out* the 8-bit value from the incoming data17:13
blueCmdthis should be outside the critical section anyway, so as long as 4 is marked as clobbered (maybe even early clobbered) we should be fine17:13
blueCmdthis might be a problem with the other atomics as well, you might want to check that17:15
stekernyeah, that's the only thing I'm worried about, and that should probably be added17:15
stekernyou mean the fetch_and_<op_name>_mask?17:16
blueCmdthanks for debugging this. really awesome17:16
blueCmdyeah, i dont recall how many  _mask functions there are17:17
stekernI think that's the only one except the one I already changed17:17
stekernbut you're right, looks like that needs the same treatment17:17
poke53282ysangkok: Did you try your pull request with the video files?17:27
ysangkokpoke53282: i tested it with a 55mb squashfs that i also mounted17:28
ysangkokusing overlayfs :D17:28
ysangkokpoke53282: is there a problem?17:28
ysangkokit only needed 4 chunks to mount :D17:29
ysangkoki didn't experiment with the chunk size... but maybe that should be configurable too.. if we add support for giving random json fs descriptions in the query string, people can make more use of it since they know what chunk sizes work best with their data17:30
poke53282ysangkok: For some reason, he don't activate the lazy file mode for one file.17:38
poke53282I added "lazy":1 at the orion mpg file17:39
poke53282It works17:41
ysangkokah good17:41
ysangkokusually i have problems cause it caches the metadata also17:41
poke53282well, I have a button, which clears the cache17:41
ysangkokso i open the developer console (it has "disable cache" checked) while loading the basefs17:41
ysangkokpoke53282: you're saying it works, but also that it doesn't work for one file... should i test again or not?17:42
poke53282Now it works. My first test failed. Don't know why.17:43
poke53282Doesn't matter17:43
ysangkokhmm ok17:43
poke53282You can also use the ata device for the squash filesystem.17:51
ysangkokhow ?17:52
poke53282I used js/worker/dev/ata.js17:53
poke53282The function SetBuffer17:53
poke53282The ata device can also work with the whole buffer.17:54
poke53282But I had a version, whcih worked with small junks on the server.17:54
poke53282I never used it, because it was slow.17:55
poke53282This was before I started the virtio/9p filesystem.17:55
ysangkokhmm ok17:55
poke53282In principle the Lazyuintarray can be used as well for this device.17:56
mor1kx[mor1kx] bandvig pushed 1 new commit to withfpu: https://github.com/openrisc/mor1kx/commit/8b646f0f0fd3c5ce4d48731f85a6d261c5cae9c318:03
mor1kxmor1kx/withfpu 8b646f0 Andrey Bacherov: Size optimization by replace 64-bit adder to 48-bit one in 2nd stage of multiplier18:03
ysangkokwell there aren't many disadvantages to the lazyuint8array i guess... cause even if access behaviour is random, it will always calculate the same chunk boundaries. so the chunks can be cached18:04
poke53282no disadvantages. I just wanted to tell you, that there is an alternative if you just want to mount one big file.18:10
olofkstekern: Just finished Book of Unwritten tales today. You got to check it out. Best adventure game I've played in ages20:07
LoneTecholofk: on to the critter chronicles then?20:17
olofkLoneTech: Yes! I got both as part of a humble indie bundle20:18
olofkSeems like Book of Unwritten tales 2 is coming along as well. There is an alpha version on steam20:18
LoneTechyay (:20:18
stekernolofk: you really don't want me to work on these gcc test fails, do you?21:31
olofkstekern: Book of unwritten tales is your reward for a clean regression test suite21:32
stekernyou're buying then too? ;)21:33
stekernI've got the back to the future game to play too21:33
stekernI can play that on highest settings on my laptop, something I can't on my ws that has a crappy graphics card21:35
olofkstekern: You're happy with the zenbook?21:40
stekernyes, the only bad thing I can say about it is the lack of dedicated pgup/pgdn/home/end21:45
stekernI've got multi-touch gestures working with touchegg too, hint if you haven't tried.21:48
stekernon the touch-pad21:48
stekern115 gcc errors now, was 400+ the first run21:49
olofkstekern: Interesting. I remember seeing touchegg but didn't really understand how it fit into the rest of the system21:54
stekernyou just run it and define your gestures, I'm using the defaults22:06
stekernyou have to turn off the builtin synaptic gestures (like two finger scroll)22:07
olofkAhh... maybe it was a bit work to build for gentoo, so I gave up22:07
stekernI especially like the three-finger move window gesture22:08
olofkDoes it work with all window managers?22:12
stekernit should, but I recall reading that gnome have some own that might clash with it22:13
stekernbut the only thing that happens then is that you just can't overrule them22:13
stekernor was it unity...?22:14
olofkxfce doesn't have anything like that built-in so that shouldn't be a problem22:14
olofkfor me22:14
stekernI'm running plasma 5 on this22:14
--- Log closed Mon Jan 12 00:00:52 2015

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