03:09RSpliet: woo, AMD finally has a beta OpenCL 2.0 toolkit
03:10RSpliet: ... oh that beta has been around for a lot longer
03:10RSpliet: well, still...
07:21imirkin: calim: yeah. and it's wrong for the halfz case. and i did try to remove it iirc, with no effect
07:22imirkin: calim: i'm wondering if the GL_EQUAL is subject to some sort of quantization issue... dunno
07:23imirkin: i think i did check with the specific value it was using, 0.98, and it didn't seem like that would happen
07:42imirkin: calim: also, it works on every other gpu, and it works with nvidia blob drivers, so more likely a nouveau bug.
08:00cyndis: Hardware video encode/decode on tegra are not currently supported upstream :( on Tegra210 (X1) the blocks changed completely from MSENC/VDE to NVENC/NVDEC. I think the latter two should be easier to support, though they still userspace code to use them.
08:01imirkin: cyndis: i think the decoding blocks on desktop gpu's are called MSDEC -- those are supported by nouveau
08:02imirkin: cyndis: although maxwell switched it up yet-again, and we haven't caught up
08:02cyndis: yeah, I guess we don't have that on Tegra.
08:02imirkin: but maybe the thing in the X1 is the same thing in GM107+... that'd be really convenient
08:03cyndis: it won't be inside the gpu anyway
08:03cyndis: they are separate host1x clients
08:03imirkin: coz after you RE a video decoding engine, it kinda takes the will to live out of you, and if not, definitely the will to RE a second vdec engine.
08:03cyndis: but if they are accessed similarly that would certainly be useful
08:04cyndis: yeah.. :)
08:05cyndis: i wonder what userspace code google uses with them
08:05imirkin: so mlankhorst and i are out. hopefully some third sucker will come in and figure out how maxwell works.
08:05RSpliet: imirkin: can't we bribe some of the CedarX RE'ers to do it?
08:06imirkin: RSpliet: unlikely. they've also RE'd a vdec engine ;)
08:06imirkin: RSpliet: you, on the other hand.... :)
08:08mlankhorst: shrug i would do it if we had the firmware and fifo stays the same..
08:09imirkin: mlankhorst: oh of course. but i'm sure they switched it all up.
08:09imirkin: mlankhorst: working out all the formats is what sucks the life out of you
08:09imirkin: at least what sucked the life out of me
08:10mlankhorst: we have automation now :P
08:10mlankhorst: at least for some of the parts
08:11imirkin: yeah, i never got that going for some reason
08:37mlankhorst: yeah its meant to work with mmt but the format changes
08:53imirkin: mlankhorst: well, mmt is a lot more awesome now... the functionality could be built-in
08:53imirkin: mlankhorst: or rather, demmt
08:54imirkin: there's already logic to decode tic/tsc, shaders, etc. easy enough to throw the vdec stuff in
08:56imirkin: we should probably do that to find the remaining h264 issues
09:48Shibe: what if the reason noevau is so slow is that for some reason it isnt using the dedicated gpu at all?
09:48Shibe: like in some benchmarks it's even 6.5 times as slow
09:48Shibe: that can't be possible, even if the driver were not so good
09:49imirkin_: 6.5x as slow as what?
09:52mlankhorst: because boot speed gpu is slow :p
10:03Shibe: mlankhorst: what do you mean
10:05imirkin_: those appear to be graphs
10:05imirkin_: still no clue what you're comparing.
10:07Shibe: imirkin: proprietary vs noevau
10:08imirkin_: which gpu
10:09Shibe: how can it be possible that a different driver is sometimes half the speed, sometimes even 6 times
10:09Shibe: and thats framerate btw
10:10imirkin_: quite simply, as you can see
10:10imirkin_: ... or you can tell me what GPU you have and i might be able to provide some specifics
10:10Shibe: well I have a gt 740m and I don't really use noevau due to performance
10:11Shibe: but those benchmarks were from this http://www.phoronix.com/scan.php?page=article&item=nouveau_april_2013&num=3
10:11imirkin_: that's what... GK107 probably?
10:11imirkin_: first off, it boots to its lowest pstate. that means all the clocks are waaaaaay low
10:11Shibe: gk208 sadly :(
10:11imirkin_: secondly, it's also at 2.5GT/s instead of 5 or 8 GT/s
10:11imirkin_: well, gk208 is the same deal
10:12Shibe: slower than gk107 unfortunately
10:12imirkin_: depends :)
10:16imirkin_: all things being equal, the chips are probably equally powerful. but things never are equal... ram speeds, pcie lanes, etc.
10:17Shibe: thank you for the explanation
10:18imirkin_: nouveau does have experimental support for reclocking, although on my gk208, it doesn't seem to actually end up changing the memory speed
10:53Hauke: it looks like my 2560x1440 screen only exposes 2048x1152_60, 2560x1440_60 and some smaller resolutions
10:54Hauke: my gtx560 ti is not capable of doing 2560x1440_60, but 2560x1440_56 works with the blob
10:54Hauke: I even increased the max clock in nouveau to the same value the blob uses
10:54imirkin_: Hauke: did you try the @56hz thing with nouveau and the hack i gave you?
10:54Hauke: imirkin_: yes, but then it goes to 2048x1152_60 or stays there
10:55Hauke: but it looks like the virtual screen is at 2560x1440, at least I am missing something at the buttom und right side
10:55imirkin_: so you manually added the @56hz mode?
10:55Hauke: I used this: xrandr --newmode 2560x1440_56 225.000 2560 2608 2640 2720 1440 1443 1448 1481 +hsync -vsync
10:55imirkin_: hmmmm ok
10:55Hauke: and then xrandr --addmode ...
10:56imirkin_: and then you switched to that mode and it stayed at 2048xbla?
10:56imirkin_: that means that something still denied you that mode =/
10:57Hauke: yes the screen wnet black for ~1 sec and then it stayed at 2048x1152 and the right side and the buttom of the screen are not displayed
10:57imirkin_: yeah, so diff parts of modesetting disagreed
10:57imirkin_: check for other instances of 165000
10:58Hauke: I checked the driver and there are some parts in the driver where some bits are set when it is > 165000
10:58imirkin_: looks like exec_clkcmp tries to use the dual-link conf for >165mhz
10:58Hauke: Yes that was also my assumtion
10:58imirkin_: in drm/nouveau/nvkm/engine/disp/nv50.c
10:58Hauke: is there something special needed for > 165000, or just for dual link?
10:58imirkin_: (i think that the gf110.c one doesn't apply to you... that one's only for gf119)
10:59imirkin_: sorry, dunno :)
11:00Hauke: currently in a tel conference
11:00imirkin_: a boring one, i take it
11:02RSpliet: isn't that inherent to teleconferences
11:04imirkin_: Hauke: file a bug with all the relevant info though (including a link to the mmiotrace). otherwise it'll all get forgotten.
11:05imirkin_: Hauke: alternatively, if your gpu has a dual-link dvi output just use that :)
11:06Hauke: the monitor does not have dvi
11:06Hauke: are there dual link dvi to dp or hdmi adapters
11:06imirkin_: and your gpu doesn't have DP i presume
11:06imirkin_: they'd have to be active
11:06Hauke: yes 2x dvi + 2 mini hdmi
11:06imirkin_: hdmi/dp don't have enough pins
11:48Hauke: back again
11:53Hauke: I will check the places with > 165000 if they are actually checking for dual link
12:07Hauke: imirkin_: no I am using 2560x1440_56 with nouveau
12:07Hauke: so it works
12:07Hauke: I remved every check with 165000
12:08Hauke: imirkin_: http://pastebin.com/tuRhNiE0
12:08Hauke: my hack
12:09Hauke: I will clean this up
12:09Hauke: imirkin_: how do I know if my connection is dual link or not?
12:10Hauke: the clock is a pretty bad indication
12:10glennk: was the dvi cable silly expensive?
12:11Hauke: the monitor does not have dvi
12:11Hauke: only hdmi and dp
12:11Hauke: are there adapters from dual link dvi to dp or hdmi?
12:12glennk: hdmi and dp don't have dual link afaik
12:12Hauke: so making hdmi work is my only option with this gpu and this monitor
12:13glennk: hmm, hdmi 1.3 does have dual link, but is rare apparently
12:13Hauke: yes but is support higher clock rates
12:14Hauke: dvi and hdmi 1.2 are going till 165 mhz and hdmi 1.4 goes up to 340 mhz
12:14Hauke: and this nvidia gpu supports up to 225mhz
12:15Hauke: in recent versions of the blob they are also allowing up to 225 mhz
12:39imirkin_: Hauke: if it's hdmi, it's single-link
12:39imirkin_: hdmi plain doesn't do dual-link
12:40imirkin_: neither does dp. obviously either of them will happily do it with an active adapter
13:34Hauke: imirkin_: when an activa adapter is in use the gpu sees it still as a non dual link conection, is that correct?
13:35imirkin_: Hauke: yeah
13:35imirkin_: Hauke: it basically puts up a fake monitor, receives the signal, and translates into the output one
13:35imirkin_: as opposed to passive, which just hook wires up cleverly
13:43pmoreau: mwk: --^ I tried to do the renaming we talked about in my EVO pull request (#11).
14:51nfk: a fun educational video mishap i thought was worth sharing (btw, fun YT channel, if you ask me): https://youtu.be/XhU-nNiAgtI?t=4m52s