08:46imirkin: wow, we don't fail so much of CTS with (at least) a g84: https://hastebin.com/konitecivu.css
08:46imirkin: i expect the gs5 failures are totally bogus (i.e. test ran when it shouldn't have)
08:47imirkin: gonna try to figure out what's up with these
13:26pmoreau: I’m getting timeouts when trying to edit the wiki (getting to the edit page works, but trying to save, cancel, or doing anything, will result in a timeout).
15:06pmoreau: karolherbst: I updated the IL MR with the comments you made, and replied for one of them (+ rebased on latest master).
15:06karolherbst: yeah, I saw that already :)
19:10imirkin: ok, down to 1 real failure with CTS, i think. neat.
19:28imirkin: and one hw failure, i think, which causes srgb decoding to happen on the wrong side of texture swizzling
19:29imirkin: and one weird one which i'll track down... probably test issue
21:03imirkin: looks like we don't do so horribly on GTF tests either ... http://paste.debian.net/hidden/63950487/ ... need to investigate what's going wrong.
22:16karolherbst: imirkin: are you running the 3.3 tests on tesla?
22:16karolherbst: mhhh, I guess the 3.3 list is complete enough actually
22:16imirkin: (coz that's what i have plugged in)
22:16imirkin: well, GTF is pretty much exclusively pre-GL4 tests
22:17karolherbst: but like the 3.2 test list is quite short
22:17karolherbst: the non gtf one
22:18karolherbst: but I guess that's kind of where they change stuff or so
22:21imirkin: yeah, dunno. i just ran KHR-GL33.* and GTF-GL33.*
22:29karolherbst: imirkin: ohh btw, we had a nouveau talk on fosdem
22:29imirkin: yeah, saw mention on phoronix
22:30karolherbst: it was all quite a bit on short notice, as it was decided this week to have one
22:34imirkin: hrmph. looks like we don't properly compute GL_PRIMITIVES_GENERATED when xfb is disabled.
22:34imirkin: (on g84 ... works fine on pascal)
22:37imirkin: i'm going to start putting stuff in trello
22:37imirkin: with investigation results
22:37imirkin: for when i don't fix things immediately
22:37imirkin: in the nouveau-cts board
22:54imirkin: i really hate the tests that just say "fail"
23:10imirkin: hrm. reducing max attribs *way* down makes it work. are we just off by a factor of 2 there or something? wtf...
23:11imirkin: huh. very odd. blob driver uses GL_MAX_VERTEX_ATTRIBS_ARB == 16 across the board.
23:11imirkin: on fermi/kepler/maxwell too.
23:13HdkR: yep yep, it supports 16 :D
23:15imirkin: yeah. actually we also show 16.
23:15imirkin: something odd is going on here.
23:15imirkin: MAX_INPUTS may not be what i think it is.
23:18imirkin: ok yeah. with 15 it passes, with 16 it fails.
23:18imirkin: but it should be 16... hm.
23:19imirkin: we do some questionable opts for small draws, perhaps those use the extra attrib
23:27imirkin: with the 16th attrib, we assign instanceid to 0x100, and fetching from a[0x100] presumably fails.
23:27imirkin: because the input space is, again presumably, only 0x100 in size
23:32imirkin: hrmph. doesn't seem like there's a way to do this
23:32imirkin: maybe they got an exception.