00:00imirkin_: ajax: the event processing issue looks like it wasn't wholly resolved in xorg 1.19 - Fervi is having issues with DRI2 and nouveau on a G94, kernel 4.9.0.
01:21skeggsb: pmoreau: i've pushed a patch which will hopefully fix your g80 rotation issue
01:34skeggsb: pmoreau: actually, confirmed. i can reproduce if i use kde to do it, xrandr/gnome both perform the rotation stuff in a different way which avoids the bug
12:01bloblo: I have gtx 770 4 gb oc, i want know how i can test lastest changes nouveau
12:01bloblo: https://nouveau.freedesktop.org/wiki/InstallNouveau/ the method section: Out-of-the-tree compilation
12:01bloblo: is work ?
12:03bloblo: after the "nouveau.ko" created, where i need move ?
12:05bloblo: And for lastest changes i neeed use this: git clone https://github.com/skeggsb/nouveau.git or other newer available ?
12:05bloblo: thank you
12:42ajax: imirkin_: yeah, i've heard. was staring at the dri2 code a lot yesterday.
14:02Fervi: Can somebody check something? Trigger Rally on Nouveau, do you have little glitches there?
14:34Fervi: If I want to report a bug (3D); i should choose X.org -> Driver/nouveau -> Version?? (git?)? It is technically for 2D support only
14:35imirkin_: mesa -> Drivers/DRI/nouveau
14:38imirkin_: don't worry about version
14:38imirkin_: if you want a realistic chance at the bug being fixed, include an apitrace
14:39imirkin_: which reproduces the issue
14:39imirkin_: note that there remain some unfixed issues on nv50, which i'll be honest - i stared at for a LONG time - and couldn't figure out wtf we were doing wrong.
14:40Fervi: just in one game sprites and textures are glitched
14:41Fervi: but backgronds working good ;)
14:42Fervi: but on apitrace video it works very good (like in simply screen recorder)
14:42Fervi: but maybe logs can help you
14:42imirkin_: er what - recording a trace with apitrace fixes the issue?
14:43Fervi: no, but when I watch video from there (and for example recorded by simplyscreenrecorder) all in video is fine
14:43Fervi: but when i play - it isn't (I also recorded it by camera)
14:44imirkin_: well, i'd need a way to repro in order to debug
14:44imirkin_: i don't have the game in question (or any games, really), so i mostly just work off traces
14:44Fervi: yhm :P
14:45Fervi: i have warnings in ApiTrace with glPushMatrix
15:07imirkin_: Fervi: you might want to xz -9 the trace
15:07Fervi: yhm, ok
15:07imirkin_: also do you plan on hosting the file for a while? if not, you might want to put it up on google drive or some other semi-permanent location
15:07imirkin_: i don't even have a nv50-era gpu plugged in atm
15:08Fervi: So maybe i should upgrade my GPU :P
15:08imirkin_: and it can be months in between gpu config changes for me
15:08imirkin_: because i hate rebooting
15:09imirkin_: is this a free/free-to-play game?
15:09Fervi: yep, FLOSS
15:09imirkin_: oh. you probably should have led with that :)
15:10imirkin_: hmmmmmmm ... is the thing you recorded representative of what it really looks like?
15:11Fervi: From camera how it's looks like
15:11imirkin_: [sometimes cameras can be misleading, that's why i ask]
15:11Fervi: but all desktop recording - record perfect
15:12imirkin_: i wonder if it's some DRI3 issue. that would suck.
15:12Fervi: I can downgrade to DRI2 and check it xd
15:13imirkin_: you had issues with dri2
15:13imirkin_: you can check it without any restarts though - LIBGL_DRI3_DISABLE=1
15:13imirkin_: (in your environment)
15:15Fervi: yep, it works
15:15Fervi: so something with DRI3
15:16imirkin_: GL_STACK_OVERFLOW in glPushMatrix(mode=GL_MODELVIEW)
15:16imirkin_: iirc our stack is moderately deep... i guess not deep enough?
15:18imirkin_: src/mesa/main/config.h:#define MAX_MODELVIEW_STACK_DEPTH 32
15:49imirkin_: Fervi: i'm kinda curious if reclocking works for your G94
15:49hussam: imirkin_: hi. out of curiosity, which distribution do you use for development? I would like to eventually switch to a non-rolling release distribution.
15:50imirkin_: Fervi: if you don't mind potentially hanging your system, mind mounting debugfs and playing with pstate?
15:50imirkin_: hussam: i use gentoo.
15:50hussam: imirkin_: ok.
15:51imirkin_: i believe it's what one might call a "rolling release distribution", but there's some amount of qualification before a package is marked stable, esp for "important" packages.
15:51Fervi: mount -t debugfs none /sys/kernel/debug/
15:51Fervi: This is it?
15:51imirkin_: Fervi: yes.
15:51hussam: yes. thank you.
15:52imirkin_: Fervi: assuming you have debugfs built. if not, you'd have to enable it in the kernel.
15:52imirkin_: Fervi: then pastebin `cat /sys/kernel/debug/dri/0/pstate'
15:59Fervi: i'm not sure if this is working (and i doing it correct)
16:01imirkin_: Fervi: ok, so it boots to the lowest perf level
16:01imirkin_: Fervi: try "echo 0f > /sys/kernel/debug/dri/0/pstate" - note this might hang your box.
16:01imirkin_: if it doesn't, cat the file after, and pastebin again
16:02karolherbst_work: are there any fermi gpus which doesn't?
16:02imirkin_: karol: "which doesn't"?
16:02imirkin_: Fervi: woohoo! it worked!
16:02imirkin_: RSpliet: --^ good job :)
16:03karolherbst_work: ohh this is a tesla
16:03imirkin_: Fervi: things should go faster now :)
16:04Fervi: I think you talking about performance
16:05Fervi: so should i test trigger rally again?
16:05imirkin_: Fervi: sure :)
16:05Fervi: still glitching. Is there a option to force disabling VSync (or enabling)?
16:06Fervi: ofc. i can use DRI2 to Trigger Rally, but it is nice to fix it for everyone
16:06imirkin_: vblank_mode=0 will force it off
16:07Fervi: vblank_mode=0 and vblank_mode=1 works worse :D
16:07imirkin_: there's 0, 1, 2, and 3
16:07imirkin_: tbh i don't remember what they all do. 0 is super-disable though.
16:08Fervi: 2,3 looks the same in this game
16:08karolherbst_work: some application just manage to mess vsyncing up
16:12imirkin_: Fervi: anyways, if your gpu heats up or you want to save power or whatever, echo 03 > .../pstate
16:12imirkin_: this flips between fast and slow, basically.
16:12Fervi: my GPU is too cold :D
16:13imirkin_: but at the high pstate you should get perf that's comparable to blob (probably in the 50-80% range)
17:05karolherbst_work: skeggsb: ping on power budget and g92 pcie patches
20:10karolherbst: mupuf_: got dynamic reclocking working again with minmal amount of interrupts :)
20:11karolherbst: which means, the PMU actually only requests a recock if it's actually smart to request one with having all calcuations on the host
20:12karolherbst: only the silly memory thing is missing
20:25karolherbst: RFC: https://github.com/karolherbst/nouveau/commit/a4bef14260b198c307034fa58039400efd9c1ff1
20:32karolherbst: skeggsb: and also ping again on my other reclocking patches: https://github.com/karolherbst/nouveau/commits/stable_reclocking_kepler_v6
20:42karolherbst: I am wondering, maybe those PMU counters are also configured through the vbios
22:02karolherbst: meh, if it's in the vbios, it is in a different form than in the registers
22:03Fervi: Hm, I wonder if someone writes an open vbios for nouveau? P
22:05imirkin_: that wouldn't really be achievable... the vbios has a ton of board-specific stuff.
22:05imirkin_: and there's a million different boards
22:45glennk: imirkin_, fwiw i don't think any GL implementation ever has supported more than 32 deep matrix stacks
22:49imirkin_: glennk: yeah, 32 seems like a lot. they're probably just forgetting to pop
22:51glennk: i remember crashing a sun box by doing exactly that long ago
23:14glennk: wonder why such newly written code would be using such old GL api
23:22imirkin_: i don't think it's newly-written code...