00:03 AeroNotix: I get about 9-15fps on the lowest (custom) with some extra settings turned down, at 1280x1024 and on the highest settings I get 4fps
00:03 AeroNotix: But changing the pstate definitely helps
01:39 mooch2: mwk: i have a question about the nv3 card id in pmc
01:39 mooch2: what should it be for the win9x driver to accept it?
02:07 dboyan: imirkin, i'm now working on rcp in codegen/lib, on gk110. denorm result handling and sched data calculation is yet to be done
02:08 dboyan: I only assign a "safe" sched data for most of the code
02:09 dboyan: about denorm, the blob seems to be extra careful, using really complicated logics with rounding.
02:10 dboyan:suspects if he really knows ieee754 rounding
06:33 dboyan: imirkin, imirkin_: https://github.com/dboyan/mesa/tree/fp64-rcprsq1
06:33 dboyan: Finally finished drcp for gk110
15:56 Bleh: Hi are there any support members online?
16:01 imirkin: dboyan: that's great. since all that stuff is still fresh in your head, please add a LOT more comments to the code. ideally with pseudocode at the top in the comments.
16:04 dboyan: imirkin: okay, i'll do that
16:04 MarcinWieczorek: Hello. My Xorg hanged for 1.5h... http://sprunge.us/hdCG
16:05 MarcinWieczorek: What do you think guys?
16:05 imirkin: MarcinWieczorek: sounds like some kind of pushbuf desync, or other sync error type of thing
16:08 MarcinWieczorek: imirkin: what can I do about it?
16:08 MarcinWieczorek: The whole xorg has been not responding but unhung eventually
16:09 MarcinWieczorek: this happens when the machine is idling (mouse/keyboard)
16:11 gnarface: MarcinWieczorek: maybe its just the compositor
16:13 imirkin: MarcinWieczorek: absolutely nothing.
16:22 dboyan: imirkin, i started looking at rsq, the code that blob generates is a lot longer
16:22 dboyan: and there was an instruction unrecognized by envydis
16:24 dboyan: how can i know what that instruction is doing?
16:28 imirkin: dboyan: use nvdisasm
16:28 imirkin: dboyan: and feel free to fix up envydis :)
16:28 imirkin: the gk110 ISA definition is the least filled out
16:29 imirkin: i try to add to it when convenient, but sometimes laziness overpowers me
16:29 imirkin: here's a handy way to operate nvdisasm:
16:29 imirkin: perl -ane 'foreach (@F) { print pack "I", hex($_) }' > tt; nvdisasm -b SM35 tt
16:29 imirkin: and then paste 32-bit hex-encoded dwords, and then hit ^D
16:30 dboyan: okay thanks
16:31 imirkin: nvdisasm is part of one of the nvidia cuda tools packages
16:50 imirkin: skeggsb: the commit id reference in https://github.com/skeggsb/linux/commit/5112abc6a43372b2c1934717ed9bf132a427c46c is wrong
16:51 pmoreau: Noooooooo: NVIDIA is working on OpenCL 2.0! I must hurry up to try to get Nouveau on 2.0 before them! … (though, I really doubt I will manage that)
16:52 karolherbst: imirkin: hum... it goes into the nouveau repository though
16:52 karolherbst: pmoreau: pls ;)
16:52 karolherbst: we just support it, and fix bugs later
16:52 karolherbst: and to be quite frank: who would notice anyway ;)
16:52 imirkin: karolherbst: right, but it needs to be in the regular repo
16:52 pmoreau: :-D
16:53 imirkin: skeggsb: also, since the commit it's fixing isn't upstream yet (right?), perhaps just squash it into the commit that broke things? or reorder? and drop the stable tag?
16:55 karolherbst: +1
16:59 imirkin: pmoreau: builder still broken i guess?
16:59 imirkin: i thought you said you fixed it :p
17:31 RSpliet: pmoreau: you could go for OpenCL 2.1 straight away :-P
17:32 imirkin: i somehow doubt he'll be adding SPIR 1.0 support
17:35 RSpliet: there is that... and a lack of compatibility with the OpenCL 2.0 memory consistency model for Kepler and older I bet
17:36 pmoreau: imirkin: It did build properly on Monday, so it should be fine.
17:36 pmoreau: I have reduced the frequency from every day to every Monday, since it gets little use.
17:37 pmoreau: 2.1 requires SPIR support? I thought it was only SPIR-V support that was required.
17:50 sinatosk: hi, anyone to help me with something?
17:51 sinatosk: *anyone here
17:51 sinatosk: possibly help anyway
18:38 MarcinWieczorek: Sorry that I left, it hung again and I had no time to recover it
18:40 imirkin_: pmoreau: RSpliet: i think CL 2.0 might have required SPIR 1.0. not sure. perhaps it was just introduced at the same time.
18:43 pmoreau: imirkin_: Ah, could; I’ll double check.
18:43 pmoreau: Though, I might be able to feed SPIR back to LLVM and have it converted to some SPIR-V. :-)
18:43 librin: imirkin_: anything new on the Civ6 thingie?
18:44 imirkin_: librin: no... that's weird.
18:44 imirkin_: it dies in a draw
18:44 imirkin_: in edgeflag stuff
18:44 imirkin_: leads me to believe there's some memory corruption going on
18:44 imirkin_: which makes the claim that the edge flag is enabled
18:45 librin: wew
18:45 imirkin_: if you can make a trace that repros the issue, that'd be helpful
18:52 librin: imirkin_: m'kay, but that's ain't happenin' until tomorrow
18:55 imirkin_: librin: alternatively, if you can make a new NV50_PROG_DEBUG=1 output, that might be enough for me to track down which shader causes it
18:59 librin: imirkin_: again, tomorrow
18:59 imirkin_: yep
21:35 duttasankha: Hi I am new to nouveau and I needed some help regarding nouveau working principle. Is anybody there to help me out. Thanks.
21:49 duttasankha: clear
21:54 mooch2: mwk: are you there?
22:15 mwk: mooch2: yes
22:19 mooch2: mwk: for riva 128's pmc id, what id does the win9x driver expect?
22:19 mwk: uhhh
22:19 mwk: no idea
22:19 mooch2: the one i got from envytools doesn't work lmao
22:19 mooch2: fucc
22:19 mwk: FWIW the PMC.ID on my NV3 is 0x00030110
22:20 mwk: and on NV3T it's 0x00030122
22:20 mooch2: welp, i tried 0x00030101
22:20 mooch2: which is an earlier revision lmao
22:21 mwk: try mine?
22:21 mooch2: okay