03:32rhyskidd: Lyude: Do you have access to one of those Chamelium testing boards?
10:53RSpliet: Integer dot product (aka. the neural net op)?
11:53karolherbst: fun, doing a piglit run and wondering why the machine crashes. Then I noticed, Nouveau wasn't even loaded
11:57pmoreau: Good job!
11:57pmoreau: Though, to be honest, similar things happened to me as well.
11:59karolherbst: well, it is a Coffee Lake CPU, so the intel driver seems to be still a bit buggy
12:04karolherbst: pmoreau: is there some review you still need for some clover patches?
12:04karolherbst: wondering when we can actually merge those
12:04pmoreau: Good questions, I need to have a look at it. :-D
12:05karolherbst: would be good to merge those things first, so that the other stuff makes more sense
12:05pmoreau: I was wondering whether to put the SPIR-V validation thing directly in clover/spirv/invocation.cpp, rather than in clover/api/program.cpp and move them to clover/spirv/invocation.cpp in the second series.
12:06karolherbst: I think the spirv specific code should stay in clover/spirv
12:07karolherbst: I don't really like the fact, that the interface is different for each backend anyway :/ but this mistake was made befor your spirv work already anyway
12:07karolherbst: when I have some time I might even cleanup clover generally
12:08pmoreau: I’ll change patch 16 then.
12:09pmoreau: Anyway, I need to go back to classes, ttyl
16:46Lyude: rhyskidd: mhm!
16:47Lyude: I wrote the igt support for them
16:47Lyude: well, the initial stuff. an intel intern helped a lot as well :)
17:38karolherbst: imirkin: for atomics I have to set the mask to 0x1 though
17:42imirkin_: probably a good idea.
17:42imirkin_: i don't think it matters tho
17:42karolherbst: and it seems the hardware is unhappy about 0xf in a few other places
17:42karolherbst: well, I got hangs
17:43karolherbst: I coulg imagine that for textures it actually is important
17:43karolherbst: I mean tex instructions
17:43imirkin_: for tex it's definitely extremely important
17:43imirkin_: i meant for atomics tho
17:44karolherbst: I will take a deeper look tomorrow
17:44imirkin_: as always... just do what tgsi does.
17:45karolherbst: the spilling fails I saw with kepler1 should be real bugs though. I remember seeing them somewhere else
17:46karolherbst: we just fail spilling at values of 64 bits or more
17:48karolherbst: it is usually okay if your access starts at the .x component, but if it starts at something else, it breaks
17:48karolherbst: anyway, gtg
21:33rhyskidd: Lyude: oh great! i think i will take you up on that. that hardware looks really helpful for validation. was Paul K the intern?
21:39Lyude: rhyskidd: yep! :), keep in mind you might want to make sure what you need to test will probably need to be supported in IGT since I don't really have much else setup
22:10mupuf: rhyskidd: before working on the chamelium support and testing for nouveau, we need to have a way to get a CRC and/or frames out of our crtcs
22:11mupuf: this would unlock a lot of IGT tests