00:08pmoreau: skeggsb: (Probably) unsurprisingly, the first bad commit is “drm/nouveau/mmu: remove support for old backends”
00:10skeggsb: on the contrary, i'd be *very* surprised if it was that commit
00:10skeggsb: and what're you referring to anyway? the segfault?
00:13pmoreau: The BUG_WARN and ZETA I was talking about a bit earlier
00:14pmoreau: skeggsb: ^
00:14skeggsb: those only happen during bisect?
00:14skeggsb: you didn't mention them before that
00:14skeggsb: then i don't care, can't really fix bisect bugs :P
00:15skeggsb: as long as they don't happen on master
00:15pmoreau: True enough
00:15pmoreau: Will have to do another bisect tomorrow, it’s getting kind of late here.
00:15skeggsb: what about this segfault though? did you track down where that happens?
00:16pmoreau: Besides what’s printed in dmesg, no.
00:17pmoreau: I don’t think I have debug symbols for nouveau_dri.so around to figure out where it comes from within it.
02:05rhyskidd: Lyude: you asked about the meaning of nvidia's SLCG acronym
02:07rhyskidd: from commit messages on two nvidia employee kernel patches, it stands for "Second Level Clock Gating"
02:08rhyskidd: see https://marc.info/?l=linux-tegra&m=145797155901123&w=2 and upstream https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=497fbe24987bd24ee271c67c212ec681995188b6
11:16RSpliet: rhyskdidd: good find
12:22karolherbst: mupuf: there?
12:22karolherbst: mupuf: I would like to talk about that EVoC application robclark forwarded to us
13:08imirkin: Lyude: afaik some patch went in to fix dri3 which is what was triggering some of the "death while resizing firefox" failures
15:38karolherbst: imirkin: I suppose you don't know much about vulkan, do you?
15:39imirkin_: not much. some.
15:39karolherbst: asking because there might be somebody interested in an EVoC doing vulkan stuff
15:39karolherbst: but this requires a lot of figuring things out before actually accepting
15:40karolherbst: I doubt you will have time for that, I asked just in case you are interested nethertheless
17:13nyef``: ... Did I see something about MCP89 yesterday?
17:14imirkin_: someone was complaining about something
17:15nyef``: Power management and memory clocking?
17:16imirkin_: something like that
17:16imirkin_: or at least that was tangentially related
17:16nyef``: Fair enough. I know of an HDMI audio bug on MCP89, and suspect a hotplug bug (possibly DPort only).
18:02Lyude: imirkin_: oh nice; I'm guessing by dri3 you mean libdrm? or are you just referring to dri3 in the kernel driver
18:03karolherbst: Lyude: I am sure he means Dri3 on a DDX level
18:03karolherbst: because that what you usually refer to as dri2 or dri3
18:04karolherbst: and the kernel stuff involved
18:15karolherbst: imirkin_: is there any benefit in using getSSA over getScratch before converting to SSA in codegen?
18:15karolherbst: or is there just the warning about redefining a ssa value?
18:25imirkin_: karolherbst: i'm honestly not 100% sure
18:25imirkin_: i never quite figured that ou
18:25imirkin_: that said, i'd really advise against redefining getSSA() values
18:25imirkin_: not 100% sure that it matters
18:25imirkin_: might be just that stupid warning
18:25imirkin_: Lyude: i was talking about a mesa patch to its dri3 impl
18:25imirkin_: (which i think has landed)
18:26imirkin_: which was causing issues in all the drivers
18:45Lyude: imirkin_: ahhh
18:50Lyude: imirkin_: any idea which particular patch?
18:55imirkin_: search for dri3 patches landing in the past few days
18:55imirkin_: (git log --grep dri3)
18:58Lyude: ahh, I think I see the one you're talkign about
21:01pmoreau: skeggsb: I just finished the second bisect and “drm/nouveau: switch over to new memory and vmm interfaces” “transforms” the ZETA errors in the segfault.
21:02karolherbst: imirkin_: double inputs use two locations in nouveau, right?
21:03karolherbst: well, vs inputs that is
21:03karolherbst: and I guess the same goes for outputs
21:05karolherbst: that stuff is weird
21:06imirkin_: yes, hw only supports 32-bit inputs
21:06imirkin_: i don't remember the literal slot being important at the hw level
21:06imirkin_: it's just N sequential 32-bit inputs
21:06karolherbst: decl_var shader_out INTERP_MODE_FLAT dvec3 a (VARYING_SLOT_VAR0.xyz, 0, 0)
21:06karolherbst: decl_var shader_out INTERP_MODE_FLAT double b (VARYING_SLOT_VAR1.z, 2, 0)
21:07karolherbst: the problem is here, that the driver_location is 2 for the second output
21:09karolherbst: this looks wrong anyway
21:10karolherbst: second thing should be decl_var shader_out INTERP_MODE_FLAT double b (VARYING_SLOT_VAR1.z, 1, 0) actually
21:11karolherbst: or wait... I think I messed something up here
21:11karolherbst: I convert that store_output against b to export u64 # o[0xa0] %r14d
21:11karolherbst: but it should have been export u64 # o[0x98] %r14d
21:20glisse_: pmoreau: the vmm patch are already queued ?
21:24glisse_: oh i missed that
21:28pmoreau: glisse_: Indeed they are :-)
21:29glisse_: yeah for some reasons i thought Ben was holding them on
21:30pmoreau: “My precious vmm patches... we will keep you safe, oh yes, we promise!”
22:38Lyude: hey on https://nouveau.freedesktop.org/wiki/CodeNames/ I see the 860M being listed as both kepler and maxwell; is that correct?
22:44pmoreau: Seems correct
22:45Lyude: hm, should probably put a note on there clarifying that
22:45Lyude: had someone try to use my patchset only to learn they had a maxwell card, not kepler :P
22:46Lyude:will fix it in a bit
22:46pmoreau: Are you going to fix it by implementing Maxwell support? O:-D
23:07Lyude: jfyi: e5ff036c6751c39ee008ca7db47b3ce4d7a38a15 ("st/dri: Add support for BGR[A/X]1010102 formats.") breaks nouveau on wayland really badly
23:07Lyude: at least for this kepler card
23:07Lyude: imirkin_: ^
23:12Lyude: damn, looks like that bugfix for the firefox stuff isn't the one that we're looking for here
23:22msbs: my nouveau seems to crash the whole time when i start to playback videos
23:23msbs: the error: nouveau 0000:01:00.0: Xorg: error fencing pushbuf: -16
23:23msbs: the video is freezing and i cant do anything. i cant minimize it, nothing. the computer itself seems to still work