08:13imirkin: skeggsb_: yay, thanks for fixing that libdrm thing!
09:04imirkin: skeggsb_: perhaps also look into the corruption that appears to happen with stk 0.9?
09:10imirkin: pmoreau: for your portal issue, can you try making TargetNV50::isOpSupported return false for OP_PRERET and seeing if you get the same fail on your nvac?
09:12imirkin: pmoreau: er, also would have to fix up NV50LegalizePostRA::visit i guess =/
09:14imirkin: [er hm... that shouldn't matter... only matters for functions, which aren't a thing in tgsi quite yet]
12:39imirkin_: skeggsb_: i think i know why glamor is messed up on nouveau -- if no glamor, you never go through the DRI2ScreenInit logic. i dunno how it's supposed to be done for dri3, but without it, we don't end up passing screen->dri2->base.version >= 3 and so GLX_ARB_create_context_profile is never exposed
13:00pmoreau: imirkin_: Ok, I'll give it a try :)
13:01pmoreau: (And then I'll probably disappear again for some days)
13:02imirkin_: pmoreau: i think i realized it wasn't going to make a difference =/
13:03pmoreau: Oh, well, I shouldn't take that long to check
13:04pmoreau: brb, need to reboot
13:09imirkin_: skeggsb_: yeah, i guess you have to do the dri2 thing for glamor. how about nuking glamor support from nouveau and refusing to load for cards that don't have exa?
13:48tobijk: is it "feed longstanding issues into bugzilla" day? :D
13:49pmoreau: Why longstanding?
13:49tobijk: heard of that a few month ago if i remember correctly
13:50pmoreau: Oh right, it's just I have really little time for a few weeks/months now, and imirkin wanted a bug report to be open, so... :)
13:51imirkin_: tobijk: you have a G96 right? could you test to see if you see the corruption too?
13:52pmoreau: So, le's try with the modifications
13:55tobijk: imirkin_: no sorry its a G86
13:56imirkin_: tobijk: good enough
13:56tobijk: ke will do that later on :)
13:56imirkin_: tobijk: use glretrace -w to get the final image
14:01pmoreau: Ah ah ah!! I had forgotten to load Nouveau and was wondering why libGL was searching for swrast... --"
14:02airlied: skeggsb_: you going to release libdrm 2.61 today?
14:03pmoreau: imirkin_: No changes
14:04imirkin_: pmoreau: as expected :)
14:05pmoreau: Oh wait, I forgot to install it and just built it
14:14pmoreau: imirkin_: I should be able to test on another G96 mid-June - I just remembered I had a discrete G96 at home.
14:15imirkin_: pmoreau: ok. that's around when i might be able to test with one as well.
14:17pmoreau: Ok, this time I did install, and still no changes
14:56imirkin_: tobijk: do you happen to remember what trace i noticed the 3-wide spill issue with?
15:04tobijk: imirkin_: i'm afraid, no :/
15:08tobijk: imirkin_: oh wait i reproduced it with civilization: BE
15:09imirkin_: tobijk: oh awesome
15:09imirkin_: tobijk: can you provide the shader that causes the crash (NV50_PROG_DEBUG=1 should print it)
15:09imirkin_: that's not a valve game is it?
15:10imirkin_: Publisher: 2K, Aspyr (Mac, Linux)
15:10imirkin_: o well
15:28tobijk: [28431.002708] nouveau E[ PFIFO][0000:01:00.0] write fault at 0x0003e50000 [PTE] from GR/GPC0/PROP_0 on channel 0x007f8b2000 [CivBE]
15:28tobijk: [28431.002713] nouveau E[ PFIFO][0000:01:00.0] PGRAPH engine fault on channel 5, recovering...
15:28imirkin_: not the error i had in mind :p
15:29tobijk: yeah now my system is going somewhere not good, gonig to reboot ~_~
15:38skeggsb_: airlied: yeah, assuming it's safe for other drivers, i haven't looked if there's other changes yet
15:38skeggsb_: airlied: also, i'm not sure if i've got correct permissions on fd.o yet for it, i usually get someone else to run the release script for nouveau ddx even..
15:39skeggsb_: imirkin_: yeah, that's a serious consideration, i've been meaning to post to the ML about a proposal for something similar
15:39imirkin_: skeggsb_: which thing are you responding to? i have like 20 outstanding questions :p
15:40skeggsb_: imirkin_: but going further, and not loading on all boards that can be fully supported by -modesetting with glamor (if built against server 1.17 with modesetting+glamor, >=nv50 && no config option to opt into exa)
15:41imirkin_: skeggsb_: dunno about that... not against it, but definitely a bit weird.
15:42imirkin_: skeggsb_: among other things, i think that's likely to lead to a lot more hangs as we end up using "more" GL things
15:42skeggsb_: we don't gain anything from using our ddx over -modesetting, and -modesetting is kept up to date with drmmode changes (among other things), for example
15:42imirkin_: skeggsb_: also you lose the self-containedness of the nouveau ddx, which imo is kinda nice
15:43skeggsb_: well, the code wouldn't go away
15:43imirkin_: right, ofc
15:44imirkin_: should probably double-check the xv situation on glamor -- lots of people use that...
15:44skeggsb_: Xv is implemented in glamor, I do wonder if it's got the stuff setup to interact with vdpau correctly though
15:45imirkin_: correctly, yes... but efficiently?
15:49skeggsb_: *shrug* it's stuff that should be looked into first, of course
16:01tobijk: imirkin_: you want the compiled output of the failing shader as well?
16:03imirkin_: tobijk: i want the tgsi
16:03imirkin_: so that i can feed it to nouveau_compiler
16:03eFuchs: hmm, ipv6 ist instabil hier
16:04eFuchs: sorry wrong channel
16:04tobijk: go for iv4 :P
16:06tobijk: imirkin_: http://hastebin.com/ufipegojew.avrasm
16:06imirkin_: awesome thanks
16:13imirkin_: tobijk: added to the bug
16:13tobijk: btw tracing the portal bug wont work
16:14tobijk: as soon as a portal opens
16:14imirkin_: i just want to know if it repros...
16:15tobijk: dont know if its tha problem, it looks really wrong
16:15imirkin_: and what gpu
16:16imirkin_: pmoreau's screenshot that "inside" the portal was all messed up
16:16tobijk: g84 and im on it
16:16tobijk: no its all over the screen
16:18tobijk: just a snapshot of the pixelwars
16:19tobijk: portal is on the left where the rays are coming from kind of
17:02tobijk: imirkin_: btw that thing up there drives the nouveau kernel part crazy
17:03imirkin_: tobijk: that's with a recent mesa?
17:03imirkin_: portal had some issues back in like mesa 10.1 days
17:03imirkin_: or 10.0
17:04imirkin_: looks like the vertices being submitted are all wrong
17:04tobijk: is a fairly recent snapshot (2 days ago)
17:04imirkin_: can you check if this is a new issue? e.g. check with 10.5 or 10.4 or something?
17:05tobijk: ke, btw is this a problem on nvc0+ as well?
17:05imirkin_: shouldn't be
17:05imirkin_: and pmoreau said it worked fine on his nvac as well
17:05tobijk: ah right
17:06imirkin_: do you have an old apitrace maybe?
17:07tobijk: mh let me check, in that time you may have some fun with the dmesg: https://homepages.thm.de/~tjkl80/dmesg_portal.txt all knid of nice erros
17:08imirkin_: tobijk: looks like you're running out of vram
17:08imirkin_: which in turn leads to variuos awesomeness
17:10tobijk: i dont have an old trace :/
17:11imirkin_: ouch, 128MB of vram
17:14tobijk: it really starts to get problems when the first portal opens, thats doing something evil, no doubt :D
17:14imirkin_: yeah, like running out of vram :)
17:15imirkin_: i dunno why nouveau really wants all this junk in vram in the first place
17:16tobijk: as long as we have plenty, its just faster...
17:16imirkin_: right, but why does it *require* it
17:22tobijk: oh now i know why the vram is running out "textures on high"
17:24tobijk: imirkin_: looks fine to me now -> https://homepages.thm.de/~tjkl80/portal_2.png
17:24imirkin_: except with G80, where various annoyances, but with G84+ it should be equally easy to have all this junk in vram and gart
17:24imirkin_: does the "event horizon" look right?
17:25tobijk: kook at the picture
17:25imirkin_: that seems wrong to me
17:26tobijk: border too big? the inside just shows the selfprojection just fine, maybe i need an example
17:26imirkin_: yeah, the inside seems fine
17:26imirkin_: pretty sure it wasn't in pmoreau's screenshot
17:26imirkin_: but the border of the portal seems way too wide
17:27tobijk: maybe its just the low texture thing
17:27imirkin_: or alpha blending is messed up
17:37tobijk: you may be right, thats an extra effect that builds up when a new portal opens
18:52imirkin_: skeggsb_: have you looked at https://bugs.freedesktop.org/show_bug.cgi?id=90276 ? it got bisected to one of your disp changes... i think i've seen a bunch of various "pdisp hangs" issues too
23:30pmoreau: imirkin_: If I didn't mess my test, it seems it worked in Mesa 10.1