00:03 imirkin: phi u32 %r7827 %r7814 %r12659 %r12577 %r12459 %r12445 %r12391 %r12089 %r11829 %r11815 %r11801 %r11790 %r11779 %r11768 %r11757 %r11746 %r11735 %r7825
00:03 imirkin: that ... sounds like a recipe for disaster
00:03 imirkin: i somehow doubt that a BB has *that* many inbound edges
00:04 imirkin: BB:11 (4 instructions) - idom = BB:8, df = { BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 BB:9 }
00:05 imirkin: urgh. need to figure out what's happening there. i was so hoping i could live out my life without understanding what the dominator frontier is =/
00:05 airlied: sounds star trek
00:05 koz_: imirkin: What're you compiling?
00:07 imirkin: koz_: https://bugs.freedesktop.org/show_bug.cgi?id=101708
00:07 koz_: imirkin: Ah. Deepest sympathies - that stuff is intense.
00:08 imirkin:wishes a compiler expert would come and help nouveau
00:10 airlied: imirkin: you might get sb then :-P
00:10 imirkin: airlied: well, i'd say calim was pretty expert
00:10 imirkin: which got us nv50_ir, which was WAY ahead of all the other efforts afaik
00:11 koz_: imirkin: I'm afraid I don't have much to contribute there.
00:11 imirkin: unfortunately it then stagnated and got ignored by the other projects, which have since caught up, and in some areas exceeded it
00:11 imirkin: i have neither the time nor the expertise to keep up, esp with the finer points
00:12 imirkin: [like RA. or building the SSA dominator tree.]
00:14 mangix: so maxwell2+ vbioses are signed
00:14 mangix: i happen to have a modded one
00:14 mangix: wonder how he did it
00:17 airlied: imirkin: the problem with being WAY ahead is nobody can understand it :-)
00:17 airlied: like in theory SB is way ahead as well, in practice adding features without having the original designers brain content is pretty hard
00:18 imirkin: airlied: can't speak for sb, but with nvir, it's actually been generally fine. there are just occasional "core" bugs which kinda require understanding how those core bits work
00:18 imirkin: which are generally based on the various SSA papers and whatnot, i'm just not super-familiar with that stuff
00:19 imirkin: anyways, i see what's going on with those phi nodes -- seems fine actually. there are a ton of break's that all point to the end of the while loop
00:19 imirkin: so it makes sense for that BB to have a ton of inbound edges
00:19 airlied:read some of the papers, I even saw a book in the office once, I still couldn't figure out one of the classic SSA hazards in sb :-P
00:20 airlied: expecting someone else to come along and pick up nvir and understand it a deep enough level to use it somewhere else without calim being available it a pretty large ask
00:20 imirkin: yep
00:21 airlied: nearly as bad as using LLVM as the basis for a gpu compiler ask :-P
00:21 imirkin: oh grrr... i know this bug.
00:22 imirkin: oh wait. this is a bug i added in one of my WIP things
00:22 airlied: hmm flu symptoms and efifb, should make for a good day
00:26 imirkin: i thought efifb was nice
00:27 airlied: imirkin: it's simple :-)
00:28 imirkin: simple == nice
00:28 airlied: the underlying firmware is more the problem :-P
00:29 imirkin: heheh
00:30 airlied: and I have to understand caching modes like uncached-minus :-P
00:40 imirkin: is that even a thing? (understanding those modes, that is... not the modes themselves)
00:41 airlied: imirkin: yeah it's like travelling back in time to when I understood mtrees
00:41 airlied: mtrrs
00:42 imirkin: for me - an alternate universe :)
00:57 imirkin: erm... oops. i think we've just been messing up RA bigtime since we added compute shaders =/
00:58 imirkin: embarassing.
01:00 imirkin: none of which explains why it's failing with 64 regs. but at least it'll work on nvf0+
01:08 imirkin: WTF. who thought THAT was a good idea... 16-wide register block?
01:08 imirkin: how can that even happen
01:08 imirkin: no wonder it can't RA with just 64 regs...
01:09 imirkin: oh no. that's bytes. more reasonable.
02:08 mangix: huh
02:08 mangix: running cmp on stock vs. modded bios doesn't reveal many differences
02:53 rhyskidd: karolherbst: https://github.com/envytools/envytools/commit/26fd72574b1a49f312696c8580dab895b8fdad8f#diff-ff8ba7c89c7f2d69f1f5ad5afe7d32caR1045 did you ever see the down_off or up_off values as being signed?
02:53 rhyskidd: is causing warnings when they are used with the bios_u16() function -- it expects an unsigned variable
02:54 rhyskidd: you can see the new compiler warnings here: https://travis-ci.org/envytools/envytools/jobs/252148198#L825
04:24 nan0s7: So I'm new to using this driver (usually go for the glx one), is there a simple way to be able to control the fan speeds and such? I'm trying to find it
07:03 hopland: Hey folks :) Is there a version of the nouveau kernel drivers that support fan control (or power control) for Pascal (1070/NV134 (GP104)?
07:30 pmoreau: hopland: Sadly no, we still need signed firmwares from NVIDIA to do that.
07:34 hopland: pmoreau: poor nan0s7 has been trying to figure it out for some hours now ^^; didn't find anything on freedesktop that hinted that the lack of firmware would also result in a lack of power control... though it is pretty obvious now that I think of it
07:37 pmoreau: Well, it's only the case since Maxwellv2, it hasn't been the case forever.
07:38 hopland: pmoreau: Yeah, and besides: nouveau has made great strides the last couple of years. Reverse engineering at it's finest :)
07:39 pmoreau: "Little hope of reclocking becoming available for GM20x and newer GPUs as firmware now needs to be signed by NVIDIA to have the necessary access." : It could be more explicit that that includes controlling the fan.
07:39 pmoreau: Though, if you are stuck on the lowest clock and can't reclock, why would you want to change the fan speed.
07:41 hopland: pmoreau: long lasting GPU intensive processes require constant vigilance from the fans :P
07:41 hopland: At least if you want the GPU to have an average or longer lifespan..
07:45 pmoreau: Does the GPU get warm with those, on Nouveau with no reclocking?
07:48 hopland: pmoreau: apparently the GK104 is nerfed on nouveau :/ from what nan0s7 told me the nouveau drivers running glxgears gave an average of 3900FPS, while the prop nvidia drivers gave 28000FPS - so I'm guessing no. But it's not my GPU, so I'm in the clear :P
07:49 pmoreau: hopland: There is no dynamic reclocknig on Nouveau, so unless he manually reclocked his GK104, he won't be able to match NVIDIA's driver performance.
07:50 hopland: Ah, ofc - he probably has to "hack" the firmware on the GPU then, if I'm not mistaken
07:50 hopland: Without power or dynamic reclocking, that is
07:50 pmoreau: GK104 does not need special firmwares to reclock, just a >=4.10 kernel.
07:51 hopland: oh? is it in /sys/debug/kernel somewhere?
07:51 pmoreau: Yes: "Experimental support for manual performance level selection (also known as "reclocking") on GM10x Maxwell, Kepler and Tesla G94-GT218 GPUs. Starting Linux 4.5, available in /sys/kernel/debug/dri/0/pstate"
07:52 pmoreau: cat that file to see which performance levels are available, and echo the performance level to the file to switch that perf level
15:00 mwk: who'd have expected
15:03 mwk: well that's the ugliest PoS in hwtest so far, and I only got to compute POS
20:01 tobijk: hey there airlied, imirkin: new day new bug: when i have no external screen attached to my hdmi my system OOPSes and hangs shortly after, is this one known? https://hastebin.com/kumajetipa.vbs
20:02 tobijk: this is with the bugfix applied from yesterday
21:55 imirkin_: tobijk: oh fun, that was just reported on list too
21:55 imirkin_: tobijk: can you bisect that? you'll need to keep applying my fix at every bisect step too, unfortunately
21:56 imirkin_: tobijk: actually it was a slightly different bug that was reported
21:56 imirkin_: but highly related
21:56 imirkin_: also in nouveau_display_suspend, also in the vblank thing
21:59 tobijk: imirkin_: yeah will bisect it tomorrow, just thought the cause would be known already as well :)
22:00 mwk: hey, anyone here knows if OpenGL texture matrix has a glEnable or is always-on?
22:01 imirkin_: tobijk: ok, and feel free to reply to Mike Galbraith's email on nouveau@ as well
22:01 imirkin_: mwk: texture matrix?
22:01 imirkin_: you mean glEnable(GL_MODELVIEW) that sort of thing?
22:01 mwk: imirkin_: one of many encient things replaced by vertex programs
22:02 imirkin_: right, but which thing? btw, a good reference of the options available at: http://docs.gl/gl2/glEnable
22:02 tobijk: imirkin_: yep i will have a look on it
22:02 mwk: glMatrixMode(GL_TEXTURE)
22:02 mwk: that one
22:04 imirkin_: mwk: afaik no
22:04 mwk: alright
22:04 imirkin_: mwk: may want to ask in #dri-devel as there are more GL people there
22:05 mwk: right, good plan
22:05 mwk: well, nv10 T&L hardware makes slightly more sense if there is no such enable in GL, but that's not exactly strong ecidence...
22:07 mwk: ISTM NV10 stuffs random crap into transformed TXCs if you enable texgen without enabling texture matrix, or if you disable bypass without enabling perspective-aware texture coords
22:07 mwk: and by "random" I really mean non-deterministic
22:07 mwk: so I suppose they just disable the matrices as an optimization if they happen to be == I and user hasn't enabled texgen
22:32 imirkin_: could be
22:32 imirkin_: probably a semi-common case