02:27karolherbst: imirkin: anything I could try out regarding the memory issue?
06:42pq: Echoing anything to 'pstate' file on nv96 returns ENOSYS. Am doing it wrong, or is it simply not implemented enough to be even worth attempting? Wiki says it should be MOSTLY.
06:43pq: 4.1.3, fwiw
07:21Tom^: pq: "#define ENOSYS 78 /* Function not implemented */" sounds like it :/
07:27karolherbst: pq: it does not work for all cards
07:27karolherbst: for example it doesn't work for fermi either
08:06 < ml|> Hello, any recommendations on which video card to get?
08:14specing:would like to know the same
08:14specing: preferrably something that can fit into normal size (one PCI-e slot height, normal length)
08:16 < ml|> Need to replace a faulty Quadro NVS 295. Perhaps will just the same card again.
10:01imirkin: pq: support for those early cards got dropped at some point
10:02imirkin: pq: iirc 3.12 was the last kernel that supported reclocking them
10:12RSpliet: imirkin: should fs-op-assign-mult-mat2-float (and co) be discarded as wrong?
10:12imirkin: RSpliet: huh? those should pass
10:12imirkin: RSpliet: did i break them?
10:12RSpliet: it fails, but because the signedness of 0 is incorrect when a negative value is multiplied with 0
10:13RSpliet: which, according to the shader_precision specs... well "In general, correct signedness of 0 is not required."
10:13imirkin: oh, the shader precision tests... those are mostly bs
10:13imirkin: there are patches to remove a lot of them
10:13imirkin: the reason they fail is due to the fma uncertainty
10:13imirkin: i.e. a*b+c <-> fma(a, b, c)
10:14RSpliet: in this case it's very specifically because -1.00000000e+00 * 0.00000000e+00 ~= -0.00000000e+00
10:15RSpliet: which... according to the shader_precision spec is all right, but the shader marks it wrong
10:19imirkin: ah ok
10:20imirkin: don't think i saw that failure, but i have a bunch of precision-specific patches locally
10:20imirkin: [i mean to piglit]
10:26RSpliet: this one is interesting too: Failed to compile fragment shader piglit/generated_tests/spec/glsl-4.00/compiler/built-in-functions/textureQueryLod-sampler1DArray.frag: 0:14(20): error: no function with name 'textureQueryLod'
10:31imirkin: there's some disagreement about whether it's supposed to be textureQueryLOD or textureQueryLod
10:31imirkin: it would appear that mesa only has the former, but the glsl 4.00 spec calls for the latter
10:32imirkin: RSpliet: send patches ;) should be an easy one, see src/glsl/builtin_functions.cpp
10:34imirkin: RSpliet: but first double-check what glsl 4.50 says
10:37RSpliet: imirkin: checking the specs just raises more questions :-D
10:37RSpliet: "Available only in the fragment shader"...
10:37RSpliet: while in piglit, it only succeeds when it's inserted in the "vertex" shader
10:38RSpliet: oh wait, no, that's because the test is supposed to fail there
10:40RSpliet: but yes, in GL_ARB_texture_query_lod it's capitalised differently than in the GLSL 4.0 specs
10:54karolherbst: imirkin: do you still need a mmt ?
13:21 < ml|> See that the NVS 510 is shown for the NVE0 family (Kepler). Can one assume the NVS 310 would work as well, since it is not listed? Or being not listed means it is not supported?
13:27 < ml|> Or even the NVS 315?
13:28glennk: nvs 510 is a gt 630
13:33 < ml|> glennk: Thanks for that info.
13:34 < ml|> Trying to find an affordable card, for general use and perhaps some gaming for the kids.
13:36glennk: that card is probably around the same performance level as an intel hd 4000-4600
13:38 < ml|> Thought of rebuilding that system, but that would a lot more than just getting a video card.
13:38 < ml|> *would cost...
13:49glennk: ml|, does the card have to be low profile?
13:49 < ml|> glennk: Nope
13:53imirkin: ml|: esp among the low-end cards, there's a tremendous amount of overlap between different chips with the same marketing name but diff chipsets
13:54imirkin: ml|: look at pci.ids to identify which is which, not the wiki page
13:54imirkin: karolherbst: yes please!
13:54glennk: what cpu is in the machine ml| ?
13:56 < ml|> glennk: AMD Athlon(tm) II X3 445 Processor
13:57glennk: ok, so its just about fast enough to drive a newer gpu without too much bottlenecking
13:57glennk: also probably big enough power supply to handle all but the highest end cards
13:57 < ml|> imirkin: Yes, have noticed this. Makes looking for a card a bit harder.
13:58 < ml|> glennk: If I recall correctly, it has at least a 300W PSU.
14:00 < ml|> It currently has a NVS 295, that has worked well for years. Just recently it has started to fail.
14:02 < ml|> Probably cause kids playing games on it, is my guess. Thought of just getting another since they can be found really cheap. But perhaps might be a good time to get something better.
14:05glennk: i'll throw in something like a radeon R7 260 as a card that should be well supported by the open drivers
14:05karolherbst: imirkin: tests/spec/arb_tessellation_shader/execution/vs-tes-tessinner-tessouter-inputs.shader_test ?
14:05glennk: imirkin can probably fill you in on an equivalent geforce one
14:07imirkin: karolherbst: yep
14:08imirkin: ml|: definitely go for amd radeon if you're looking to use open drivers
14:09imirkin: karolherbst: also let me know if that test fails on the blob, in which case the trace will be considerably less interesting :)
14:09karolherbst: it passes
14:10karolherbst: ohh right older driver :/
14:10 < ml|> Hmm, radeon would be easier or similar to using nouveau driver; in terms installing and using? On Debian here BTW.
14:10karolherbst: what has to be done, that mmt works with newer drivers? :D
14:10imirkin: karolherbst: one needs to figure out why it's not working, and adjust the code so that it works. ;)
14:11 < ml|> Have only ever used nvidia cards.
14:11imirkin: ml|: radeon would be the same level of difficulty of using nouveau, but it'll have the advantage of working much better.
14:12imirkin: it has actual paid developers supporting it, and those developers have access to documentation
14:12imirkin: hard to compete with that
14:13 < ml|> imirkin: I see, will look for a card to what glennk recommended then or? Thanks
14:13 < ml|> Yes, indeed; did not know that.
14:14karolherbst: imirkin: http://www.filebin.ca/29t8uJq01ttB/foo.mmt.xz
14:15imirkin: great thanks!
14:16imirkin: and can you confirm this passes?
14:21karolherbst: yeah it passes
14:21karolherbst: as I said already ;)
14:21karolherbst: will go now anyway
14:23karolherbst: imirkin: by the way, did skeggsb pushed or mailed anything regardint eh cleanups? If not, it would be nice to have a branch I can start the work on top. Doesn't need to be ready or anything, but I doubt, that there will much inside the pci subdev :)
14:23imirkin: not that i'm aware of
14:27karolherbst: okay, then I just will wait a little longer
14:27karolherbst: by the way: I looked over the hack he wrote and saw, that the blob is doing stuff a bit different (different amount of writes), maybe I will dig into this then
14:28imirkin: afaik he no longer has access to the relevant hw
14:29imirkin: so any help is appreciated. i think he fired off a question to nvidia, but it's questionable whether they'll respond
14:29imirkin: they haven't responded at all to my latest questions
14:29imirkin: the last one i got a response to was re the shader header, which was a pretty good doc but i had a bunch of follow-up questions that went unanswered
14:29imirkin: a bunch of stuff in there also directly contradicts what i've observed
14:30karolherbst: I see
14:30karolherbst: I think the blob just writes 0x0 into the reg until something happens
14:30karolherbst: otherwise I can't explain why a different amount of writes changes anything
14:30karolherbst: writing only 1 0x0 is the wrong thing todo btw
14:31karolherbst: my system hard reseted after it
14:31imirkin: and you say this from direct knowledge of the hw? :)
14:31 < ml|> What name does the radeon R7 260, as go by. Do not see it here... http://www.x.org/wiki/RadeonFeature/
14:31imirkin: ml|: search for "radeon decoder ring"
14:31 < ml|> Ok, Thanks
14:31karolherbst: and with hard reset I mean like 3 LED blinking (roll, numpad and the other thing)
14:31karolherbst: and after 10 seconds it turned the power off
14:32 < ml|> Haha, the first link is to the one above :)
14:32karolherbst: imirkin: nah, I tried it out with only 1 write
14:32imirkin: ml|: well it should be on there
14:33imirkin: ml|: it's probably a bonaire or something
14:33imirkin: r7 260 is on the "BONAIRE, KABINI, MULLINS, KAVERI, HAWAII" line. and i happen to know that hawaii is high end, and kaveri/mullins are igp's... i think kabini is too.
14:34glennk: the wikipedia pages are also good for general gpu decoder rings
14:34glennk: https://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units and https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units
14:34 < ml|> imirkin: Yes, just found that on that page. Was looking at only the top portion of the page. Which only shows... R100 R200 R300/R400 R500 R600/700 Evergreen N.Islands S.Islands1 C.Islands V.Islands
14:35imirkin: it's a sea islands card (aka C Islands)
14:35 < ml|> Sorry, for the noise and Thank You guys for all this info.
14:35imirkin: you can probably get more info in #radeon
14:36 < ml|> Indeed on my way ;)
14:36 < ml|> glennk: Bookmarked, Thanks.
15:31Mittttens: is there any way i can help nouveau in the effort to add reclocking to maxwell?
15:31Mittttens: i'm pretty terrible at C and I don't know much about reverse engineering :s
15:31RSpliet: then at the moment no... I'm afraid we need a talented kernel dev with time on his/her hands for that
15:32imirkin: meh, kernel is optional. but definitely someone with the time, hardware, and tenacity
15:33Mittttens: is there anything else i could do then?
15:33RSpliet: become one of those? :-P
15:34Mittttens: what about decoding the VBIOS
15:34RSpliet: VBIOS is mostly well understood
15:34Mittttens: or is there any info that i can dump about my card that would help
15:35Mittttens: s/help/be helpful
15:35imirkin: Mittttens: what card do you have? a GM107?
15:35Mittttens: gtx 770
15:35RSpliet: well, sure, MMIOtraces in combination with your VBIOS *could* help a potential person who does have time, skills and persistence
15:35Mittttens: a gk104 (?) I think
15:35RSpliet: but nothing will happen until someone matches that profile :-P
15:36imirkin: Mittttens: oh, that one should work mostly ok
15:36imirkin: Mittttens: except you probably can't reclock to the highest level
15:36imirkin: but mid-level should mostly work
15:37imirkin: Mittttens: boot with nouveau.pstate=1
15:37imirkin: and you should be able to use the mid-level pstate
15:37Mittttens: the mid-level pstate for reclocking, that is?
15:37imirkin: highest level will most likely hang the gpu, but not necessarily
15:38Mittttens: i would just at that after the "rw" right
15:39Mittttens: so this will likely give me a significant performance increase, no
15:39Mittttens: and power consumption decrease
15:40imirkin: power consumption increase
15:40imirkin: and performance increase
15:47Mittttens: i thought reclock means a power consumption decrease since it can lower the clock when it's not doing anything imirkin
15:47Mittttens: or did i already have reclock and not know it
15:47imirkin: Mittttens: reclock is changing clock speeds
15:48imirkin: by default we don't allow any clock speed changes, so you get what you get coming out of the vbios
15:48imirkin: for kepler cards, that tends to be the lowest power state
15:48Mittttens: and without reclock aren't you just stuck at a medium clock speed
15:48Mittttens: this is a maxwell card
15:48imirkin: gk104 = kepler
15:48imirkin: the k should give it away
15:49Mittttens: oh woops
15:49imirkin: anyways, nouveau.pstate=1 enables (experimental) code to allow the user to switch between pstates
15:49imirkin: but it's all manual
15:49Mittttens: so without doing anything it'd be exactly the same
15:50Mittttens: oh well i couldn't see myself doing that all the time
15:50imirkin: yeah, echo'ing a value into a file is hard work before and after playing a game.
15:51Mittttens: what do value(s) and what file?
15:52imirkin: cat /sys/class/dri/card0/device/pstate
15:52imirkin: it'll say things like 07: 0a: 0f: for example
15:54Mittttens: i don't have a /sys/class/dri/
15:54Mittttens: there's a /sys/class/drm/card0/device
15:54Mittttens: but no /sys/class/drm/card0/device/pstate
15:55imirkin: er yes, sorry, drm
15:55imirkin: did you boot with nouveau.pstate=1?
15:57Mittttens: oh, no
15:57Mittttens: i thought you meant it'd already be there
17:30erwin2: I have problem which I can't solve .... after ``xrandr --setprovideroffloadsink nouveau Intel'' I have in `DRI_PRIME=1 glxinfo' all looking good, no EEs in Xorg.0.log, nothing in dmesg but I always see only black window
17:31imirkin: erwin2: you need to use a compositor
17:31erwin2: i have kernel 3.11.10-29-desktop (suse 13.1)
17:31imirkin: that's also a pretty old kernel
17:32erwin2: I've read the thread about that at freedesktop bugzilla and probably not understanding how ... it doesn't work in kde and it doesn't work in twm no matter if I have enabled in kde effects and so
17:32erwin2: imirkin: can you please elaborate that compositor part ? what compositor ? o.O
17:32erwin2: it seems that it could work ...
17:33imirkin: erwin2: xcompmgr or something else
17:33erwin2: at least no errors are here and even glxgears prints 10 times more frames per second (with black screen) with DRI_PRIME=1
17:33imirkin: that's coz it's not vsync'd
17:34erwin2: can you please show me the example command how youd run glxgears with DRI_PRIME=1 and with using xcompmgr ?
17:35imirkin: just run xcompmgr -- should be enough.
17:37erwin2: indeed, it works now o.O
17:37erwin2: erwin2: perfect, thanks !!!!! I'd never install molestatory nvidia blob ...
17:39imirkin: erwin2: well, don't celebrate too soon... nvidia blob delivers a lot more perf
17:39imirkin: (mostly because it's able to increase clocks up to spec)
17:40erwin2: nouveau: kernel rejected pushbuf: Invalid argument
17:40erwin2: nouveau: ch0: krec 0 pushes 1 bufs 0 relocs 0
17:40erwin2: hmmm .... well, it freezes now with this ...
17:41erwin2: but it doesn't show black screen :D
17:41imirkin: well, you might consider updating your software
17:42erwin2: I've seen it hour ago, I checked modinfo and loaded nouveau with :: cat /sys/module/nouveau/parameters/vram_pushbuf
17:42imirkin: you have a year or two old kernel
17:42imirkin: and i'm guessing the 3d library isn't much newer
17:42erwin2: so you think/know, that my current (freezing issue) could/is caused by that ?
17:43imirkin: but what i do know is that no one is going to spend time debugging issues that even might be fixed in the current version of things
17:43erwin2: imirkin: this is my business laptop where everything works perfect so [probably I'd be glad if I won't upgrade
17:43imirkin: sounds good
17:47erwin2: imirkin: anyhow, I'd appreciate if I'd be able to find some evidence or something what will tell me the reason why it is doing .. if the known workaround for yet unknow reason will be upgrading what needs to be upgraded I will ofc upgrade
17:47imirkin: erwin2: sorry, we have *nowhere* near the level of support to provide that sort of info
17:48erwin2: ^^ last thing
17:48erwin2: can you share your opinion on that ?
17:49imirkin: that just means the gpu is hung
17:49imirkin: you probalby also got a GPU lockup notification in dmesg
17:51erwin2: i have not in dmesg anything other related to nouveau
17:51erwin2: if I dount count an initialization block
17:51imirkin: hm odd
17:52erwin2: this is the rest http://susepaste.org/view/raw/3670750 ... pyrogenesis is one game I tried only for a test
17:58imirkin: yeah i dunno. *so* many things have been fixed since 3.11 that it's impossible to count
17:58erwin2: fair enough
17:58erwin2: thanks for your time and the input
18:00erwin2: now only last and un-related question when I'm here ... I never cared for a performance, you mentioned something about perf at the beginning ... can you just tell me, how big is the performance degradation with nouveau compared to nvidia blob ?
18:00erwin2: say on my nvs 4200m
18:01erwin2: wild guess ofc
18:02erwin2: and q #2, okay, I can upgrade kernel in suse instantly .. we have some well supported repo with recent kernels, but not that much the other stuff, do you think, that upgrading just kernel could resolve my issue >
18:06imirkin: dunno, kernel's a big part of it
18:10erwin2: hmm ... found yet another error http://susepaste.org/view/raw/56390916 ... anyhow, I'll go to try updating the kernel first
18:13imirkin: erwin2: actually it looks like your card only has a single perf level?? that's odd. but if that's true, then nouveau should be around 60-80% of blob perf
18:15erwin2: card here is really low-end
18:15imirkin: that has little to do with it
18:16imirkin: most fermi gpu's, and esp most laptop gpu's have multiple perf levels
18:16erwin2: > /sbin/lspci -nn | grep 4200
18:16erwin2: 01:00.0 VGA compatible controller : NVIDIA Corporation GF119M [NVS 4200M] [10de:1056] (rev a1)
18:18erwin2: I'll be glad if upgrading kernel fixes my issues and won't cause me another issues since for me here on this laptop everything works and worked in past .. I have by default blacklisted nouveau, never needed it
18:19imirkin: you can also keep on not using it :)
18:20erwin2: I want to use it because in this stupid laptop is dvi hardwired with nvidia only
18:30imirkin: ah yeah, that's annoying
18:32imirkin: well, chances are if you stick to just using the dvi screen and don't offload 3d to the nvidia gpu, even that old kernel could work ok