02:22shpx_: will using nouveau impact performance if I use my GPU for deep learning? If it does, is the propriteray driver good, or should I just use windows?
02:35imirkin: presumably you'd want to use opencl, which is currently unsupported by nouveau. afaik nvidia blob drivers offer the same featureset on linux as on windows (minus d3d)
02:56mew007: it's me, again
04:48radiofree: jm116 is the real expert though
04:48radiofree: wrong window sorry
07:15mlankhorst: anyone mind if I add a /tests to envytools?
07:15mlankhorst: and re-use some code from intel-gpu-tools
07:26RSpliet: mlankhorst: what kind of tests? :)
07:27mlankhorst: vblank tests
07:29RSpliet: if it's very nouveau specific I guess /tests is about as all-right as /nva
07:30mlankhorst: I want to make some kind of small testsuite :p
07:37mlankhorst: and probably take intel's testsuite as a base because I'm too lazy to make functions like test_assert myself..
07:38RSpliet: the sort of regression tests?
07:39mlankhorst: just some basic tests like for vblanking counters
07:41mwk: mlankhorst: how would that be different from hwtest/?
07:43mlankhorst: I test the drm interface, rather than the hw registers directly
07:58mlankhorst: plus the test i want to add is going to test pageflipping with a system suspend, can't work with hw regs directly :P
10:13obiwood: Hi, I'm a newbie and I'm wondering if anyone has worked on it or that its even possible to activate/code/work on the 3dvision on geforce cards under linux, thanks
10:33linas: imirkin, I'm still seeing the TTM Failed to find memory space messages ...
10:35buhman: obiwood: doesn't look like that's in-scope for nouveau
10:36buhman: obiwood: looks like that's mostly just application-level playing with the shutter glasses
10:38buhman: https://github.com/bobsomers/3dvgl/blob/master/src/stereo_helper.h#L137 in particular
10:39leat1: does vdpau work with nouveau?
10:40buhman: leat1: http://nouveau.freedesktop.org/wiki/VideoAcceleration/
11:15obiwood: ok thank you buhman
11:59linas: kernel 3.18.1 gives me this: [ 3.785494] nouveau E[ DRM] GPU lockup - switching to software fbcon
11:59linas: and after that X11 never really comes up.
12:46dbacc: hey, there. I'm using the nouveau driver together with bumblebee and a three-monitor setup. Basically, this works fine with the proprietary driver and almost does with nouveau. Unfortunately. The third screen is not "stable", i.e. from time to time gets off and soon on again. In addition there are some blue doots (also flickering on semi-dark parts of the screen (see dark blue dots: d)
12:47dbacc: and hints?
12:48RSpliet: dbacc: your link is either invalid or directupload requires registration prior to downloading files
12:48RSpliet: would you mind hosting that in a public place?
12:56dbacc: ah wait, now it should work: http://www.directupload.net/file/d/3859/76trh2fg_jpg.htm
12:57RSpliet: that's funky... kind of looks like a poor HDMI cable issue tbh
12:57RSpliet: are you sure it's connected properly? maybe we're driving it wrong (although that's a little unlikely)
13:07dbacc: sry, broke down my x-server and took the chance to recheck the monitor configuration with windows. There it is absolutely fine. I just realized that these blue dots move arround during start of kde. However, should not be a connection issue, right?
13:13dbacc: is there some maximal resolution supported by nouveau? http://paste.ee/p/pAy6G
13:16dbacc: hmm, happens independently from screen size. I'm just using the problematic screen with the same problems.
13:17dbacc: I used intel-virtual-output to create the virtual screens. Maybe the problems start there in conjunction with nouveau? However these problems are not present with the proprietary driver
13:27dbacc: no more ideas? :)
13:57dbacc: hm, i'm getting MMIO read of 0x00000000 FAULT at 0x002100 error at boot. Could this be related?
14:02pmoreau: Read fault on 0x2100 seems weird!
14:03pmoreau: Hum of 0x0
15:08RSpliet: imirkin: could you take a peek at http://hastebin.com/inafazusop.coffee ?
15:08RSpliet: There's a couple of small changes I still want to do, but it's a pretty invasive patch so I'd appreciate your feedback :)
15:36imirkin_: RSpliet: i hate the pass name, and i'm not 100% sure what coalesceValues does... but in general it seems reasonable
15:36imirkin_: RSpliet: also, you don't need the ~0xf on the chipset. and there's a #define somewhere for GF100_ISA or something
15:37imirkin_: eventually i'm going to nuke chipset and switch it to SM10/SM20/SM30/SM35/SM50
15:37imirkin_: since that's all anyone checks for
15:37imirkin_: but... i'm so lazy... and it's just code churn; i.e. the current stuff works fine
16:08RSpliet: imirkin_: for the pass name I'm open for suggestions
16:13RSpliet: I mean, technically it's const folding, but very limited scope, and in shader language not a const either... I'm bad with naming
16:13imirkin: RSpliet: how about NV50MAD? dunno
16:16imirkin: RSpliet: yeah... maybe like NV50PostRASpecial