00:08imirkin: mwk: any good ideas on how to write a 64-bit value to a buffer from a GR macro?
00:08imirkin: mwk: i'd been using the query thing for writing 32-bit values, but i can't seem to figure out how to operate it to write a second 32-bit value adjacent to the first one
00:08imirkin: i think it might require 128-bit address alignment
00:09imirkin: or i'm a total retard - a distinct possibility.
00:09imirkin: and i don't think i can use any of the p2mf/etc stuff from within that macro
00:12imirkin: the only other thing that comes to mind is to use the CB data stuff for those writes... seems like that could backfire though.
00:20imirkin: oh ugh. i guess i might have to write a firmware method for it? that'd be most sad.
00:22imirkin: huh. the nvidia macro contains a 0x9194 (37268) value. what are the chances that was supposed to be 32768 :)
00:22airlied: imirkin: I'd take a bet on that :-P
00:23imirkin: these macros are the definition of unreadable. esp without a use-case =/ no way to know what the args mean...
00:24imirkin: airlied: well, in similar positions it also contains 0x8866 and 0x8867, so it's not within the realm of possibilities that they really did mean that weird value.
00:25imirkin: that macro does make judicious use of serialize, which i did not... what are the chances that nvidia knows something i don't...
00:25hakzsam: imirkin: trying to fix 64bits qbo?
00:25imirkin: hakzsam: yeah. but i don't have a trace of nvidia doing it.
00:26hakzsam: mmh I should have one
00:26hakzsam: the question is where :)
00:26imirkin: for 64-bit?
00:26imirkin: i've already tried fixing my macro in the past, but it totally didn't work =/
00:26hakzsam: yeah, I think I traced blob one day
00:26imirkin: need to find where my patch is for that...
00:27imirkin: i wonder if there's a way to grep through branches
00:27imirkin: oh, nope, there it is. found it.
00:27hakzsam: maybe thisone
00:28imirkin: [my patch that is]
00:28imirkin: hakzsam: thanks
00:28imirkin: You don't have permission to access /~hakzsam/mmt
00:28hakzsam: fucking permission of valgrin
00:28imirkin: annoying right?
00:29hakzsam: try again?
00:33imirkin: interesting ok
00:34imirkin: so looks like they do it exactly the way that i did it when it didn't work
00:34imirkin: which suggests that there's still some kind of critical little difference between our impls ;)
00:54imirkin: GAH. apparently one of the macro instruction variants isn't as super-duper-supported as one might hope.
00:56imirkin: mwk: mov $r6 (add $r6 0x4) -- that doesn't work. sticking the 0x4 into a reg and adding those works fine.
01:34hakzsam: imirkin: ah you figured ? :)
01:35imirkin: hakzsam: yeah. i saw the blob macro missing an obvious optimization opportunity, figured maybe they weren't being *total* idiots
01:35imirkin: anyways, it still doesn't TOTALLY work
01:35imirkin: among other things, GL_ANY_SAMPLES_PASSED fails for 64-bit (there's a TODO in the code)
01:36imirkin: as does GL_COMPUTE_SHADER_INVOCATIONS for which we've done nothing useful
01:36hakzsam: I remember that one
01:36hakzsam: blob seems to use a crazy macro IIRC
01:36imirkin: which iirc doesn't work ;)
01:37hakzsam: well, we can lie for the number of invocations :)
01:37hakzsam: just compute it in sw, that's it
01:37imirkin: or just return 0
01:37imirkin: (sw might be trickier for indirect)
01:37hakzsam: yeah, but one piglit will failt if you return 0
01:37hakzsam: it already fails actually but heh
05:39milesrout: This nouveau code is fascinating
14:25mmfood: Hello, I'm new to this IRC thing so forgive me if I'm doing this the wrong way. But I am looking too troubleshoot why my laptop with a nvidia 650M card is freezing when I plug in / unplug my hdmi cable. It is only happening in my freshly installed fedora system. I hade no problems with this on previous installs using the same install media.
14:27mmfood: I guess I can reinstall again and hope that it works better, but have no guarantee that it will, and I have spent quite som time setting this up so I'd rather find out the cause and maybe learn something useful on the way.
14:27imirkin: most likely DE-related... it gets a hotplug signal and does something sad
14:28mmfood: desktop environment?
14:29mmfood: what logs should I be looking at?
14:29imirkin: is your laptop freezing btw?
14:29imirkin: or are there just no further display updates?
14:29imirkin: but other than that, everything works fine?
14:30mmfood: well I can move the mouse but I clicking does nothing
14:30imirkin: can you ssh in?
14:30mmfood: i can switch tty
14:31imirkin: so ... not frozen.
14:31mmfood: if you say so, not sure about the terminology here
14:31imirkin: i gtg, but dmesg + xorg logs should provide a bit more visibility into your setup
14:31mmfood: ok, thanks for taking the time mate!
14:31imirkin: "frozen" implies "system is fubar"
14:32mmfood: ok, where is dmesg btw? can't find it in /var/log and the command doesn't work, is it included in fedora 25?
14:46karolherbst: mmfood: the command is called dmesg
14:46karolherbst: and it should work
14:46mmfood: actually this time it worked
14:47mmfood: been looking over it now but I'm not sure what to look for to be honest
14:47mmfood: 5.083597] nouveau 0000:01:00.0: iccsense: Unknown sensor type 30, power reading disabled
14:48mmfood: that one looks suspicious
14:49mmfood: [ 1683.897793] [drm:drm_wait_vblank [drm]] *ERROR* Unsupported type value 0xe9, supported mask 0x7400003f
14:49mmfood: I have loads of these
14:50mmfood: [ 1363.377660] uvcvideo 1-1.3:1.0: Entity type for entity Extension 4 was not initialized!
14:51mmfood: [ 1363.377677] uvcvideo 1-1.3:1.0: Entity type for entity Processing 2 was not initialized!
14:51mmfood: [ 1363.377680] uvcvideo 1-1.3:1.0: Entity type for entity Camera 1 was not initialized!
14:51mmfood: and then there is these
17:25japele: I have a question on detection of memory amount into fermi based card.Somebody have time?
17:26imirkin_: just ask...
17:34japele: ok tanks, I have an optimus laptop with a geforce GT 630M and the nouveau driver detect 1024 MB of VRAM, but my card have 2Gb of dedicated memory(detected correctly by Nvidia driver).Any suggestion?
17:35imirkin_: that's very surprising
17:42japele: TTM report 4GB of available graphics memory for the kernel zone and 2GB for the dma32 zone, is normal?
17:43imirkin_: i'm guessing you're on a 32-bit kernel?
17:46japele: No, 64-bit and this situation is unchanged from kernel 3.2 thath is the first version used on this laptop.
18:11japele: I have seen this commit: https://github.com/skeggsb/nouveau/commit/4e49f8011f1c97e4db04427b0c559eee5e97da67 can my problem be similar?
18:11imirkin_: mind pastebinning your dmesg?
18:34japele: This is the nouveau dmesg: http://pastebin.com/VCTr6dxQ and this is the nvidia dmesg: http://pastebin.com/3SphDBgm
18:37imirkin_: japele: and what makes you believe you really have 2GB of vram?
18:51imirkin_: japele: grab a copy of envytools
18:52imirkin_: and then run
18:52imirkin_: nvapeek 100800; nvapeek 22438; nvapeek 22554
18:53imirkin_: and tell me what it produces
18:53imirkin_: oh, and also:
18:53imirkin_: nvapeek 10f20c
18:57japele: imirkin: I'm compiling...must run with nouveau loaded?
18:57imirkin_: mmm... no
18:57imirkin_: but the GPU does have to be powered on
19:04japele: I powered on the gpu with bbswitch and the only registry that return a value is 22438 with the result: 00022438: 00000001. All other return three dots.
19:05imirkin_: even 10f20c?
19:05imirkin_: also grab 11020c
19:06imirkin_: ("..." means 0 btw)
19:06japele: both 0
19:07imirkin_: ok. that's very odd.
19:07imirkin_: i wonder if the gpu isn't properly "on"
19:12hakzsam: imirkin_: http://www.phoronix.com/scan.php?page=news_item&px=Nouveau-Civ6-Mesa-17.1
19:12imirkin_: hakzsam: yep
19:14japele: imirkin: Do you want i reboot with all modules blacklisted and re-try?
19:14imirkin_: japele: no... maybe try it with nvidia loaded
19:22japele: imirkin: ok, I'm sorry but I'm very late to work,I can find you on next moment?
19:22imirkin_: people always in here.
19:25japele: imirkin: Tank you very much and compliments for your hard work. See you
20:11Tom^: imirkin_: is that something you would want uh "tested/debugged" further? the civ crash
20:12Tom^: since seemingly it was on a 780ti which im on heh
20:14Tom^: or is it not related to 780ti at all?
20:15RSpliet: Tom^: You're the best person to find out :-P
20:16Tom^: yea well i suspect im gonna crash just as him, but if the issue is a more uh "general" one. i dont think my skills would help much further hehe
20:18RSpliet: Tom^: well the good thing is that you have marginally more time than Michael. If you can, it'd be interesting to see if 1) This happens when not changing clocks, 2) you could try obtaining an apitrace for others to debug. Since it crashes instantly, shouldn't take a lot of debugging I bet
20:18Tom^: sure, il scrap together some info then
20:19RSpliet: haha sorry, didn't mean to guilt you into it, but could be very useful :-) Thanks!
20:25Tom^: 20.7gb this will take a while to dl, heh
20:25Tom^: oh well, il just leave it running while i sleep
21:29gregory38: karolherbst: hello. I did a couple of speed optimizations on latest git.
21:30gregory38: maybe soc will be a bit faster. However you need to use nvidia blob
21:31gregory38: nouveau is limited by the cpu (will be fixed by glthread) and then it would remain some gpu perf issue
21:31gregory38: However due to non-standard float VU emulation will suck forever (and the way the game is written)
21:32gregory38: By the way, you can use VU cycle rate and/or MTVU
21:34imirkin_: Tom^: i suspect it's due to the threads stuff. afaik that game uses threads heavily.
22:19japele: imirkin_: hello, I did the tests: with nvidia loaded nvapeek fail with pci init failure(with/without root rights),without drivers loaded only 22438 return 1 all other is 0, with nouveau 00100800: 00000001, 00022438: 00000001, 00022554: 0, 0010f20c: 00000400, 0011020c: 00000400, 0010f20c: 00000400. What do you think about it?
22:20imirkin_: 0x400 is what ... 1024. and i'm pretty sure that's in MB. and you only have 1 part, according to those regs. so ... yeah. dunno.
22:21imirkin_: you could do a mmiotrace of nouveau loading to see if it's getting a different result somehow
22:21nyef```: nouveau, or the blob?
22:21imirkin_: thanks. i meant blob.
22:25imirkin_: nyef```: how go your 3d adventures?
22:33nyef```: imirkin_: Been taking a break to have some compiler adventures. And some knitting adventures. Currently planning to spend some time sorting out an initial new patch series next weekish, though it may end up getting pushed off a bit.
22:33imirkin_: ah ok
22:33nyef```: And I've ordered a couple of video cards to cover the gaps in my DISP lineup, and should have my paws on them the week after next, but I have yet to obtain a computer to put them in.
22:34imirkin_: wow, dedication!
22:36nyef```: I just found the phoronix article on the HDMI 3D stuff today.
22:36imirkin_: best not to pay too much attention to that site... lots of trolls
22:39nyef```: Yeah, I'm not planning to seriously lurk their forums or anything at this point.
22:39nyef```: ("Lurk harder.")
22:45nyef```: Anyway, I'm just heading out to do a bit of shopping and find dinner and whatnot.
22:45imirkin_: see ya
22:46japele: imirkin_: it's strange right? In this case the mmiotrace must be done with nouveau driver instead of nvidia, right? Something else can help?
22:47imirkin_: japele: no, mmiotrace nvidia, to see where it's getting its different result from
23:57japele: imirkin_: mmiotrace is done, but after launching X it became huge, is normal?
23:57imirkin_: should be like 100MB
23:58imirkin_: compresses very nicely with xz -9