08:07 liamdawson: Hi, I can't get display output when using KMS with my NV110/Maxwell device (GTX970). I was hoping to run Wayland on my desktop.
08:07 liamdawson: last boot log: https://paste.ubuntu.com/p/RtybktNCZc/
08:08 liamdawson: can confirm I can boot if I set `nouveau.modeset=0`, but then I get the wrong resolution on my monitor, and can't log into a wayland session
08:08 liamdawson: (actual behaviour I experience is Ubuntu splash pops up very briefly, then the screen goes black and goes into power-saving mode after a few seconds)
08:11 liamdawson: (uname: Linux uluhtc 4.15.0-29-generic #31-Ubuntu SMP Tue Jul 17 15:39:52 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux)
08:40 liamdawson: interesting, I could boot with modeset if I removed my 4K second display first, then I connected it later and it all works
10:28 karolherbst: liamdawson: interesting
10:28 karolherbst: liamdawson: is this HDMI or DP?
11:55 RSpliet: liamdawson: your kernel params contain video=1920x1080. You'll probably want to remove that
11:57 karolherbst: RSpliet: that shouldn't cause the EVO timeout though
11:58 RSpliet: True. Anything between 4.15 and 4.18 touching/fixing EVO problems?
12:00 karolherbst: I have still my patches but those only fix if you select interlaced modes with DP when the GPU can't do that
12:00 karolherbst: but I think we hade some reworks there
12:00 RSpliet: A few patches in the run-up to Volta support apparently; liamdawson: make sure you test with a more recent kernel as well (4.17)
16:01 karolherbst: mhh, on the PIOR we also got EXT_TMDS10BPC_ALLOWED
16:01 karolherbst: MAX_PIXELS3TAP444/MAX_PIXELS3TAP422
16:17 pendingchaos: imirkin, karolherbst: would extending nv-report.py to also print changes in only affected programs be interesting?
16:25 pendingchaos: another idea is an option to print affected shaders that are most affected or smallest
16:37 karolherbst: pendingchaos: yeah
16:37 pendingchaos: the first idea, second idea or both?
16:38 karolherbst: I think the ordering stuff isn't as important
16:38 karolherbst: still helpful
16:38 karolherbst: I think we should also report the deepest loop depth
16:39 karolherbst: and for kepler the dual issueing rate
18:34 pendingchaos: karolherbst: any idea what "shared programs" means in nv-report? "all programs"?
18:57 karolherbst: pendingchaos: yes
18:58 pendingchaos: thanks
18:59 karolherbst: pendingchaos: it would be nice to have that dual issue report though, as I have some patches to improve it, but if you don't want to add it, I will propably do it myself. There is also dual issueing for maxwell, but we don't enbale it yet
19:00 karolherbst: should allow us to gain up to 20% more perf as well
19:00 karolherbst: or maybe more on maxwell, dunno
19:00 karolherbst: on kepler it is quite significant overall
19:01 karolherbst: we could probably just copy what maxas is doing here
19:02 karolherbst: and maybe evevn improve it on kepler with the knowledge we got there
19:15 karolherbst: https://gist.githubusercontent.com/karolherbst/171e642b0ce0fafbebf8dcc236ac9fe5/raw/12f709bdffcb7cc279981c5cc6e9bbf1ff17f175/gistfile1.txt
19:15 karolherbst: on kepler it was quite alot...
19:15 karolherbst: but kepler was special
19:16 karolherbst: well and fermi
19:16 karolherbst: SM 2.1 to 3.7 was able to do that fancy dual issueing thing
19:16 karolherbst: 2.0 is GF100 and GF110
19:16 karolherbst: wondering if it really doesn't work there
19:17 karolherbst: pendingchaos: I guess I take care of that myself. I just got interested in doing that
19:21 pendingchaos:nods
19:23 karolherbst: anyway, doens't seem like we get 20% more on maxwell with that... more like 5%
19:23 karolherbst: at least those float/int conversions are there, so this will be nice
19:26 pendingchaos: 5% improvement is nice
19:29 karolherbst: yeah well
19:29 karolherbst: on kepler we got a lot more from this
19:30 mooch2: i mean, 5% improvement IS still beyond the margin of error :/
19:30 mooch2: there's probably a bug somewhere keeping it from being any higher tho
19:31 mooch2: and it could be in the hardware itself ;)
19:32 pendingchaos: where do I post patches for shader-db? mesa-dev@lists.freedesktop.org?
19:33 pendingchaos: actually #dri-devel might be better for this
19:38 karolherbst: pendingchaos: do you know NV50_PROG_CHIPSET?
19:39 karolherbst: mooch2: only fermi/kepler had real dual issueing
19:40 mooch2: ah :/
19:40 pendingchaos: karolherbst: maybe I did? I do now anyway
19:40 mooch2: that figure makes sense then
19:45 karolherbst: ../nouveau_shaderdb/gputest_pixmark_piano/7.shader_test - type: 1, local: 0, shared: 0, gpr: 50, inst: 3680, bytes: 33648, dual_issue: 0.246467 :)
19:45 karolherbst: on gk104
19:45 karolherbst: ./nouveau_shaderdb/gputest_pixmark_piano/7.shader_test - type: 0, local: 0, shared: 0, gpr: 7, inst: 19, bytes: 176, dual_issue: 0.421053
20:56 karolherbst: perl is really a terrible language
20:57 nyef: The pathologically eclectic rubbish lister? Yes. Yes, it is.
22:09 karolherbst: nice, got a st 0x0: (st 0x0 yl wt 0x10)
22:15 karolherbst: uhm
22:15 karolherbst: imirkin: GALLIUM_HUD=help is broken
22:15 karolherbst: I get dozens of "this_is_not_the_query_you_are_looking_for"s
22:18 karolherbst: ohh uhm
22:18 karolherbst: we don't expose that stuff on pascal