10:17karolherbst: imirkin: what's missing for nv4x OpenGL 3.x support btw? xfb?
10:18karolherbst: or multiple random bits?
10:25karolherbst: uhhh.. nv4x is totally buggy
14:47karolherbst: with one of the cards my system doesn't even boot
15:52imirkin: karolherbst: what's nv4x missing for GL3? basically everything required for GL3. start with integers.
15:53imirkin: GL3 is not a practical goal. it can barely do GL2.1
16:24maccraft: imirkin: ok i set it up and it runs beatifully :D
16:24maccraft: it - reverse prime stuff
16:27maccraft: the only issue i am seeing is that composition makes everything lag
16:27maccraft: but it is xfwm being xfwm
16:29imirkin: reverse prime copies are async with the rest of the flow
16:31karolherbst: imirkin: yeah.. I figured. But we do have lowering for ints in mesa
16:32karolherbst: anyway.. I was just wondering
16:32karolherbst: the biggest issue was just, that both cards I have are just causing massiv issues
16:33karolherbst: the one the system boots with, but all connectors are marked as being connected, the other just makes the kernel go oops very very earyl
16:33karolherbst: once I got some message and mce complained about a broken system
16:33karolherbst: or so
16:34imirkin: karolherbst: every featured that was added in GL3 is missing. it's a DX9.0c board...
16:34imirkin: GL3 is basically the DX10 featureset (or beginnings of it)
16:34imirkin: also missing required texture formats iirc
16:35imirkin: not to mention integer textures
16:35imirkin: karolherbst: https://people.freedesktop.org/~imirkin/glxinfo/#p=compat&v=Mesa%2017.0.0
16:36imirkin: it's only missing ... just about everything :)
16:36imirkin: it technically does support RGBA32F rendering, with various restrictions that made it not worth implementing
16:39karolherbst: imirkin: do you still have nv4x gpus you can plug into your system? I am wondering if 5.3/5.4 is just broken or if I am just very unlucky
16:39karolherbst: or.. we can't have those GPUs as secondary ones
16:40imirkin: which one are you having trouble with?
16:40imirkin: fwiw at least the FX 3450 *really* wants to be the primary GPU (and is generally not super-compatible with a secondary pcie gpu)
16:41karolherbst: I have a nv41 and nv43 afaik
16:41imirkin: the FX 3450 is a nv42
16:41karolherbst: the one which kind of worked was GeForce 6600 GT
16:41karolherbst: yeah... maybe I have a nv42 then :)
16:41imirkin: yeah, so that one's pretty finicky, at least in my system
16:41imirkin: make sure you plug in the aux power too
16:41karolherbst: any idea why?
16:42imirkin: i was just happy it worked at all :)
16:43imirkin: iirc it didn't even get past POST with it in a secondary PCIe slot
16:49loonycyborg: primary/secondary in which sense? If mobo got 2 pci-express slots how does it determine which card is primary?
16:49karolherbst: imirkin: well, it even started the kernel for me :) but yeah.. the kernel totoally failed to do anything
16:50karolherbst: the intel GPU in the CPU is the main one ;) decided by the firmware
16:51loonycyborg: And if you have cpu without builtin gpu pci express one will be primary?
16:51karolherbst: dunno.. probably yhe highest one
16:51imirkin: yeah, dunno. i just have 2 pcie slots
16:52imirkin: iirc it just doesn't work in the second one. or maybe only if it's the only one plugged in, i forget
16:52karolherbst: usually one slot gets its lane provided by the GPU
16:52karolherbst: and I guess this will be the main one
16:52loonycyborg: Well at least in my system I have phenom cpu, I could potentially replace it with fusion apu
16:52loonycyborg: I'm curious if I did so
16:52loonycyborg: which gpu would be used for early boot?
16:53karolherbst: imirkin: maybe the GPU just needs to be POSTed otherwise it does dumb stuff
16:55imirkin: yeah, i have zero additional info =/
18:22karolherbst: imirkin: btw, there are patches on the ML for a nice feature :p
18:23karolherbst: I had no chance to give a proper review either though :D
18:25karolherbst: might be even that one patch is in the ML review list.. uff
19:07imirkin: karolherbst: yeah, will have a look
19:07imirkin: i was never a huge fan of how some of that input/output stuff was done
19:07imirkin: but also didn't hate it enough to change anythign
19:07karolherbst: well, shader cache is a good reason to change it :)
19:08imirkin: the general idea seems reasonable, but i haven't looked carefully
19:09karolherbst: yeah.. I also have to take a look at the patches. I am just worried of some annoying corner cases piglit doesn't check
19:09karolherbst: imirkin: are you an admin on the ML btw?
19:09imirkin: i see patches 1..8 though
19:10imirkin: which do you think is missing?
19:10karolherbst: 2/8 as well?
19:10karolherbst: mhh, I don't
19:10imirkin: PATCH 2/8] util/blob: Add overwrite function for uint8
19:10karolherbst: and the archive broke as well
19:11karolherbst: when searching for the patch I see it now as well
19:11karolherbst: email filters
19:11karolherbst: it's not a nouveau patch :)
19:11imirkin: it was cc him rather than to you
19:12karolherbst: my filter for nouveau is a bit more complicated than just checking for the ML
19:12karolherbst: and I am stupid to have looked at the nouveau ML as well... it's all there on mesa-dev
19:13imirkin: btw, if you need a list of stuff that's broken on nv4x, lmk... can put something together
19:20karolherbst: it's not like I need a list, I was just checking how well the nv4x cards work I've got, and it simply turned out that nothing really works using them as secondary GPUs :/
19:21imirkin: afaik it's just the FX 3450
19:21imirkin: i have a nv44 which works just fine in a second slot
19:22karolherbst: well, the other GPU does kind of work, but glxgears totally killed the machine and all connectors are marked as connected although there was no display attached
19:23imirkin: try an older kernel, just in case?
19:23karolherbst: I tried it with 5.4 and 5.3.. but yeah, maybe I should check with even older ones
19:24imirkin: on 5.0, a tnt2 works fine
19:36MML_: last year this was posted on the GSOC page for Xorg
19:36MML_: Maxwell Accelerated Video Decoding:
19:36MML_: Description: RE the video encoding component of Kepler chips and produce sufficient documentation for an open source implementation (while reusing blob firmware). A stretch goal would be to integrate something into nouveau to be able to drive the engine, and a toy user-space application to encode a video.
19:37imirkin: that's the wrong subject or wrong description
19:37MML_: RE the mechanism for video decoding in VP6 (the video engine iteration used in GM10x chips). Implement support for driving the engine.
19:37MML_: but I cannot seem to find VP6 or it's code specifically
19:37MML_: or the retail version
19:38imirkin: that would be the RE aspect of the project
19:38karolherbst: MML_: we have a more complete list of projects here btw: https://www.x.org/wiki/SummerOfCodeIdeas/
19:38karolherbst: under "Nouveau (Open Source NVIDIA driver)"
19:39MML_: that is what led me here
19:40karolherbst: so the idea is: you have no documentation and no code, only the propriatary driver and need to black box reverse engineer the stuff _if_ you want to write the code as well or do whatever kind of reverse engineering and write proper documentation so that others can implement it
19:41karolherbst: decompiling and stuff comes with a legal risk, so we don't do that as developers at all
19:42MML_: but then Reverse Engineer?
19:42karolherbst: we have tools to trace what commands the nvidia userspace libs are sending to the kernel driver and we have tools to trace what the kernel drivers sends to the GPU
19:43karolherbst: by tracing it and interpreting what everything means, we can deduct what userspace has to do
19:43karolherbst: more or less
19:44imirkin: the maxwell vp6 thing ought to be fairly similar to vp3/4/5, just in a combo engine rather than spread out
19:44imirkin: i suspect it's not a ton of effort
19:44imirkin: although figuring out how to do anything is an immensely steep learning curve
19:45imirkin: not to discourage you or anything :)
19:45karolherbst: what's the deal with kepler though?
19:45imirkin: for encode?
19:45imirkin: never even looked at it
19:45karolherbst: yeah, I missed that bit
19:46imirkin: vp6 decode should be the msdec stuff, just in a combo engine rather than bsp/vp/ppp
19:46karolherbst: probably some subtile stuff changed... I even have the right GPU I think
19:47karolherbst: ohh most of 2nd gen maxwell is also vp6
19:47karolherbst: just gm206 is vp7
19:47imirkin: and i bet gm206 is a minor iteration
19:47karolherbst: adds h.265 and vp9
19:48karolherbst: speaking of "minor"
19:48karolherbst: vp6 is some kind of hybrid host-gpu decoding stuff
19:48karolherbst: at least according to wikipedia
19:49karolherbst: seems like you need some additional shaders to do h.265
19:59imirkin: maybe they moved ppp into shaders
20:00imirkin: or maybe it's just for h265
20:00karolherbst: it's just for h265
20:01imirkin: i know nothign about h265, thankfully
20:01karolherbst: according to wikipedia: "Similar to feature set D but added support for decoding H.264 with a resolution of up to 4096 × 4096 and MPEG-1/MPEG-2 with a resolution of up to 4080 × 4080 pixels. GPUs with VDPAU feature set E support an enhanced error concealment mode which provides more robust error handling when decoding corrupted video streams. Cards with this feature set use a combination of the PureVideo hardware and software
20:01karolherbst: running on the shader array to decode HEVC (H.265) as partial/hybrid hardware video decoding."
20:01karolherbst: and the next one (VP7): "Introduced dedicated HEVC Main (8-bit) & Main 10 (10-bit) and VP9 hardware video decoding up to 4096 × 2304 pixels resolution."
20:04karolherbst: mhh and gm108 doesn't even seem to have decoding engines
20:04karolherbst: same with gp108
20:04imirkin: yeah, it's missing on many mobile chips
20:05karolherbst: might be worth mentioning it on the wiki
20:05imirkin: fused off on a bunch of G86M's too :)
22:10Manoa: imirkin, is thare any way to change some settings to not do that ?
22:11imirkin: no idea what "that" is
22:11Manoa: <imirkin> Manoa: yes, when you switch away from an X server, that also freezes all X calls made by clients, which generally means that they will hang.
22:11Manoa: <imirkin> it doesn't run "in the background" in that case
22:11imirkin: oh, then no.
22:11Manoa: ok thank :)
22:20Manoa: about the other discussion, do the graphics drivers have significant improvement in performance from vectorization ?
22:21imirkin: G80 and later are all scalar
22:22Manoa: I thought so :) only pixel stuff like pixman should give significant improvement :)
22:57imirkin: (it's all highly vectorized, but largely behind the scenes ... as presented in the shader ISA, it's scalar)
23:11Manoa: it a smart design choise :)
23:14Manoa: it realy amazing drivers, so far all the games I tested all running perfectly
23:15Manoa: some tiny things like the dual X server and GALLIUM_MSAA not working but that is not a big problem
23:16Manoa: latence is realy good, bether than windows by far
23:18Manoa: this realy is big surprise for me, I didn't expected that, 2.5 years ago it was total fail (wine problem) but now it running everything perfectly
23:21Manoa: not only games and interactivity things, but also movies: no stuttering no lags even when not synced to refresh the movies are perfectly smoothed
23:40imirkin: GALLIUM_MSAA was removed a long while back
23:40imirkin: it's basically unsupportable
23:45imirkin: anyways, glad that things are working semi-ok for you