10:23 karolherbst: ahh found that thing again
10:23 karolherbst: filesystem fuzzing: https://events.linuxfoundation.org/sites/events/files/slides/AFL%20filesystem%20fuzzing%2C%20Vault%202016_0.pdf
10:23 karolherbst: I was thinking of doing the same for nouveau
11:09 karolherbst: ohhh, I could run the nouveau_compiler with that :O
11:09 karolherbst: that will be fun
11:14 karolherbst: I am sure I will find lots of RA issues. Allthough I have no clue how good afl will be at generating complex enough shaders through code instrumentation
11:30 Tom^: karolherbst: uhm why or rathere where would that create a .csv file?
11:32 karolherbst: in the same directory
11:32 Tom^: i got gpu test in /usr/bin
11:32 Tom^: so i doubt that :p
11:33 karolherbst: uhhh
11:33 karolherbst: no clue?
12:00 karolherbst: :O
12:00 karolherbst: those results
12:00 karolherbst: afl is awesome
12:00 karolherbst: 10 unique crashes within 25 secs found
12:04 karolherbst: imirkin: do you mind if I send you 77 tgsi files which all produce different crashes?
12:04 karolherbst: :D
13:02 t-ask: karolherbst: is there any of your reclocking branches working with 4.7?
13:02 Tom^: t-ask: yes
13:02 karolherbst: all my branches should work with 4.7 except if they have 4.6 in the name
13:03 t-ask: karolherbst: well I can't compile some except 4.6 ones ,)
13:04 karolherbst: seems like you run 4.6 then
13:04 t-ask: karolherbst: which one should I try the best?
13:04 t-ask: Arch-4.7.1 here
13:05 Tom^: t-ask: https://aur.archlinux.org/packages/nouveau-kepler/
13:06 Tom^: t-ask: and ofc, pacman -Q linux; uname -r , sort of have to matchup. or you got a pending linux update waiting for reboot.
13:06 Tom^: karolherbst: yea no idea, but gputest does not produce .csv
13:06 karolherbst: Tom^: no idea how that works with system wide installs
13:06 t-ask: Tom^: btw I'm also running d39
13:06 Tom^: karolherbst: i moved the bin to ~ still didnt :p
13:07 karolherbst: did you run it like I told you?
13:07 Tom^: yap
13:07 karolherbst: there should be one then beggining with _
13:07 karolherbst: maybe even in your current path
13:07 Tom^: karolherbst: https://gist.github.com/gulafaran/4ff2173d270e6caff29a141cae1952af i just put that in wat, and ran sh wat
13:08 Tom^: and no files :(
13:08 karolherbst: fix your gputest then :p
13:13 t-ask: exit
13:14 Tom^: karolherbst: oh right i was running some silly wrapper that still went into /opt and ran the binary ...
13:14 Tom^: so me moving it didnt do anything
13:15 Tom^: why didnt you say so.
13:15 karolherbst: :p
13:20 Tom^: karolherbst: blob https://gist.github.com/gulafaran/4f72226b9d9b90a8dc2adb4f1aab093e
13:21 Tom^: now for nouveau
13:25 t-ask: Tom^: so the AUR package uses v5 reclocking. I just hoped there is ne newer patch I could try
13:28 Tom^: t-ask: blame karol for not having coded a _v6 yet.
13:29 geese: salut
13:29 geese: comment allez vous
13:29 t-ask: Tom^: no need for that :)
13:29 Tom^: geese: you are better of sticking to english
13:29 Tom^: t-ask: blame him for any reason then! :D
13:30 Tom^:blames karolherbst
13:30 karolherbst: Tom^: I think after english, france is the most spoken language by nouveau devs :D
13:30 karolherbst: ..
13:30 karolherbst: *french
13:30 Tom^: perhaps, oh well mesa-git built time to reboot into nouveau.
13:32 geese: ok, so good morning everybody
13:38 Tom^: karolherbst: https://gist.github.com/gulafaran/20ff0b457f3832ac1092883d87ddd4a2
13:38 karolherbst: Tom^: awesome, thanks
13:39 karolherbst: funny that there are tests where nouveau is faster :p
13:39 Tom^: indeed
13:39 karolherbst: in all three bs tests
13:39 karolherbst: except JuliaFP32
13:51 Tom^: karolherbst: then again, at these low resolutions im probably not bottlenecking on the gpu
13:51 karolherbst: sure it does
13:51 karolherbst: well not all
13:51 karolherbst: but some
14:07 t-ask: GW2 with Nv+D9 http://www.pasteall.org/pic/show.php?id=106145
14:14 Tom^: compiling wine takes longer then the kernel ;_;
14:33 pmoreau: Good afternoon!
14:33 Tom^: t-ask: http://i.imgur.com/vbkHeLY.jpg nouveau ftw. :p
14:33 pmoreau: It’s time to pick up the fight against evil phi nodes one more time! :-)
14:35 siro__: t-ask: I'm hunting the issue right now
14:46 t-ask: siro__: you installed GW2? You most likely need to, because situation depends which map you are. I think they are runnning 3 engine versions
14:48 t-ask: interestingly the CPU is running at around 10-20% ... you've seen the related bugreport? I guess so.
14:49 siro__: it could be gpu limited and waiting a lot
14:49 siro__: is there a winehq bug report ?
14:50 t-ask: https://github.com/iXit/Mesa-3D/issues/224#issuecomment-233148368
14:53 siro__: it looks like there's a bug related to offset_units while accessing depth as texture
14:53 t-ask: it feels like ther is only a very small area around the char or props which is sharp
14:54 t-ask: not sure if the probs themselfs have their own local 'bubble'
14:59 siro__: t-ask: there seems to be a fix made on June 25th
14:59 siro__: it's not part of ixit/mesa yet
15:02 t-ask: siro__: It's a bit confusing for me which is which with all those forks :) I just hope the issue landet at the right place
15:04 siro__: t-ask: the testing and development is done on ixit/Mesa, and once it's stable and tested it'll be send to mesa
15:04 t-ask: ok then it should be ok. at least if it is mesa related
15:05 t-ask: While it is really nice to see the CPU is only runnning with 10-20%, I'm not sure if that is a good sign ,)
15:06 siro__: t-ask: this one fixes a few failing tests: https://patchwork.freedesktop.org/patch/95229/
15:10 imirkin: karolherbst: where are the crashes? in the thing that parses the text tgsi representation, or in the compiler?
15:11 imirkin: karolherbst: also the compiler isn't resilient against highly-illegal tgsi
15:11 imirkin: karolherbst: imo fuzzing it isn't a useful endeavour
15:25 karolherbst: imirkin: yeah, I noticed that I hit some asserts
15:25 karolherbst: imirkin: but the fuzzer found 19 crashes in glsl_compiler
15:26 karolherbst: so far
15:27 karolherbst: hihi, glsl compiler crashes with a line like this: " q.yz = ma|2(0.9,-0.43,0.43,0.9)*q.yz;"
15:27 karolherbst: instead of " q.yz = mat2(0.9,-0.43,0.43,0.9)*q.yz;"
15:27 imirkin: TGSI isn't meant to be resilient to fuzzing
15:27 imirkin: however GLSL parsing is
15:27 karolherbst: right
15:28 karolherbst: that's why I run it on glsl_compiler now
15:28 imirkin: you don't want to be able to crash the browser when using webgl
15:28 karolherbst: yup
15:28 karolherbst: my thought exactly
15:28 karolherbst: silly me is using the pixmark_piano shader, so execution time between runs is like 100ms :/
15:28 karolherbst: wait
15:28 karolherbst: <100 / sec
15:28 karolherbst: so 10ms+
15:29 karolherbst: 1961 different paths taking after 1h 22min
15:42 karolherbst: uhh, that one crash inside the hash table lookup thing
15:42 karolherbst: nasty
15:42 siro__: t-ask: fixed one of the issue
15:42 karolherbst: find_symbol (table=<optimized out>, name=name@entry=0xffffffffffffffff <error: Cannot access memory at address 0xffffffffffffffff>)
15:42 imirkin: karolherbst: make a small repro, file a bug
15:43 karolherbst: I was thinking about fixin stuff myself as far as I am able to
15:43 karolherbst: :p
15:43 imirkin: oh, ok. that works too
15:43 karolherbst: I don't really plan on opening 26 bugs or so
15:43 karolherbst: except I don't get what is wrong
15:44 imirkin: that's how bugs go
15:44 imirkin: everything is right, and yet it messes up somehow :)
15:45 t-ask: siro__: nice!
15:45 karolherbst: ... I just removed my results... how silly am i
15:45 karolherbst: :D
15:46 imirkin: siro__: fyi, i just fixed some issues with setting depth near/far planes. i *think* it only affects depth clamping (which i don't think existed in DX9), but i'm not 100% sure.
15:46 t-ask: Tom^: siro__: an example who GW2 can look like: https://www.youtube.com/watch?v=0Wgg1UrHSeQ ;)
15:48 imirkin: [the issue was that the near/far calculation didn't take clip_halfz into account]
15:49 siro__: imirkin: I found the rendering is corrupted when there's no zbuffer bound, but pipe_rasterizer_state offset_tri and offset_units is set
15:50 siro__: llvmpipe and r600 are fine with this, but nouveau isn't
15:51 imirkin: hmmmm.... well i made a bit of a guess as to the units that offset_units should be in when no zbuffer is bound...
15:51 imirkin: https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/nvc0/nvc0_state_validate.c#n702
15:51 imirkin: i just assume it's 24-bit
15:52 imirkin: feel free to play with it
15:52 imirkin: and/or make a repro that i can play with
15:52 imirkin: iirc mannerov had something for the offset_units_unscaled thing that checked out when i ran it
15:53 imirkin: that said, i wonder if any of this stuff is even legal to set when there's no depth buffer bound. dunno.
15:53 imirkin: it should be - it just offsets the z value passed into the PS...
15:53 imirkin: even if that never gets stored anywhere
15:53 siro__: imirkin: the patch seems fine as it fixes failing wine tests
15:54 imirkin: right, i just mean how it handles the !fb->zsbuf case
15:56 siro__: radeonsi doesn't set poly offset in case no zbuffer is bound
16:03 karolherbst: this shader file crashes the compiler for example: https://gist.github.com/karolherbst/f677e0c26835e874f5b36a8aaf0e0a1e
16:03 karolherbst: :)
16:07 pmoreau: So much rubber paste to keep everything holding together for the parsing… :-/ I’ll need to (almost) completely rewrite everything to clean it up. But at least, I am able to concentrate on the OoSSA pass now.
16:17 karolherbst: k
16:20 t-ask: siro__: If it helps, I can try to test those patches. I just need an introduction what I have to do. I'm on Arch. Maybe I can customize mesa-git...
16:23 siro__: t-ask: it'll fix reading depth as texture. character names will be displayed correct
16:23 siro__: give me 5 minutes
16:23 karolherbst: imirkin: can I set the severity to "major" for glsl shader crashes? due to potential remote code execution?
16:25 karolherbst: well I will set them to normal and let the devs decide
16:30 karolherbst: nice, every other crash is due to the same issue
16:30 karolherbst: seems like the glsl compilation is pretty solid
16:31 t-ask: karolherbst: maybe it is ok to set it to major, because then it gets attention and the devs decide ;)
16:34 karolherbst: :D
16:46 kloofy: i am in my own bubble, science i think provides opportunities even to me
16:48 t-ask: kloofy: as long as you like your bubble and no one wants to move you somewhere else ... everything is ok :)
16:52 kloofy: t-ask: yeah it's ok, i've done lot's of theories and tried develop my thinking sort of mechanically, i am really fine about how things are finally, i have not been moved for long time, people also see my thinking is stable almost currently
16:59 kloofy: stays in mental instutition are left in the past allready, aside from really big handicaps no one had anything against me tested in piles, key persons still mostly say i am a normal guy
17:01 kloofy: when you have potential you always progress in the end, there is nothing but time lost before, but in the end right persons progress through mostly
17:02 kloofy: key in psychology how you handle that you lost lot of time due to this and that, when you are able to handle it, then in end stronger ones still win
17:05 imirkin: karolherbst: you can do whatever you want, nobody pays attention to those things anyways
17:06 karolherbst: k
17:06 karolherbst: you mean such bugs or the bug details?
17:06 kloofy: for instance in the first half of life seeing not nothing comes together, cause being violated in greater extend it's hard to play the tactics that you write all the youth to the garbage bin, and play for a very strong second half it's only little concilation to see all those idiots cracking down
17:07 imirkin: siro__: hm, that's inconvenient, i'd have to make rast state dependent on zsbuf state. although i guess i already have a vehicle for doing so. i thought the offset things were supposed to not matter when there is no depth buffer bound... i'll have a glance.
17:07 imirkin: karolherbst: such bug details
17:07 imirkin: i.e. the severity/etc
17:07 karolherbst: k
17:07 kloofy: because of something that was always clear that they never had internal content to go through in legit ways
17:10 karolherbst: and the next one
17:10 imirkin: siro__: but it's unclear to me whether that's correct. i'd have to look at the polygon offset spec... afaik it should still be applied even if no zsbuf
17:11 imirkin: although since it mainly exists to avoid zfighting issues... yeah, dunno.
17:11 siro__: imirkin: is the spec public available ?
17:12 imirkin: the opengl spec? yeah
17:13 kloofy: final comments before i start my work career, we say sometimes eagle wings touch the ground, but chicken never fly, it's a hard fact that some no matter what they try to do never have the content to compete, so all the time they are tried to be pushed up is lost
17:14 siro__: there's another strange issue
17:15 siro__: wine takes about 2 seconds to start GW2, but on Gallium nine it takes about 20 seconds to compile all the shaders
17:15 karolherbst: siro__: cause for d3d stuff is mostly pre compiled already
17:15 imirkin: my guess is that the application precompiles all its shaders
17:16 imirkin: but wine doesn't generate the glsl immediately
17:16 kloofy: so in my life i have hated that fact that weaker are tried to be helped and encouraged to violate stronger persons and trying to to life someting in sense of missions which they can't never managed while stronger ones are put to threadmill, because it ends up with a big loss
17:17 imirkin: karolherbst: i thought you were going to try to fix all those
17:17 kloofy: but as said, people here on the channels are pretty intelligent
17:17 kloofy: i find myself being better on the channels
17:17 karolherbst: imirkin: I thought so too
17:18 karolherbst: but seems like I don't understand enough of the compiler now to actually fix those
17:18 karolherbst: and I am sure if a dev knowing the code sees those, he has a fix in like 2 minutes
17:19 imirkin: aren't you a dev? :p
17:19 karolherbst: yeah, but I don't know the code
17:19 karolherbst: mhhh, I actually did some chances on the compiler for include shaders....
17:19 karolherbst: *changes
17:20 karolherbst: seems like I am out of excuses already
17:20 imirkin: learn? :p
17:21 imirkin: i mean .. obviously your call what you do, but chances are these issues are moderately simple to fix
17:21 karolherbst: right and I still have that mmiotrace bug to fix
17:21 imirkin: btw, i assume that your second issue has nothing to do with |
17:21 imirkin: i assume just a shader with 0(); is enough to trigger the issue
17:21 imirkin: or maybe even just ();
17:22 karolherbst: mhh I doubt that
17:22 karolherbst: I will try
17:22 karolherbst: just (): 0:3(3): error: syntax error, unexpected ')'
17:22 imirkin: what about 0() ?
17:22 karolherbst: 0:2(1): error: function `bug' has non-void return type float, but no return statement
17:22 imirkin: huh ok
17:23 karolherbst: O|(); 0:3(12): error: syntax error, unexpected ')
17:23 imirkin: surprising.
17:23 karolherbst: I think it is due being an operator or so
17:23 karolherbst: because + and * also triggers it
17:24 karolherbst: allthough 2+2(); does indeed look somewhat legit
17:24 imirkin: somewhere in parser-land
17:24 imirkin: probably
17:24 karolherbst: mhh
17:24 karolherbst: a+a(); doesn't crash
17:24 karolherbst: but a+2();
17:25 kloofy: http://www.diva-portal.org/smash/get/diva2:302536/fulltext01 i contribute one more link here, based of what i did register count/amounts calculations
17:25 karolherbst: but my goal with the fuzzer was more like spotting bugs in the nouveau compiler... :D
17:26 kloofy: relevant information is 49-50 pages and before that couple of pages of LUT schematics
17:26 karolherbst: but somewhat it doesn't work with the piglit shader_runner
17:26 karolherbst: no idea why yet
17:26 imirkin: karolherbst: don't need a fuzzer for that ... just look at the bug list :)
17:26 kloofy: it really has shift register amounts in kb's those chips
17:26 karolherbst: imirkin: :D right
17:27 imirkin: i fixed a good one recently
17:27 karolherbst: which one?
17:27 imirkin: where the CFG iterator just kinda stops sometimes
17:27 karolherbst: uhhh
17:27 imirkin: "oops"
17:27 imirkin: [which in turn leads to all manners of other problems down the line]
17:27 karolherbst: I imagine
17:27 imirkin: https://patchwork.freedesktop.org/patch/106232/
17:28 karolherbst: :D
17:28 karolherbst: I bet this took a while to spot?
17:28 imirkin: hehe, yeah
17:28 kloofy: so every LUT has one flip-flop and every clb (that composes of 8 6input LUTS on some, and 16 6input luts on ultrascale and newer) has 128bit shift register
17:29 imirkin: i thought it was the logic that handled atomic ops on shared mem on kepler
17:29 imirkin: but that logic just happened to futz with the CFG enough to make the issue appear
17:29 imirkin: probably took me like 2-4h of investigation total.
17:30 karolherbst: that isn't that much though
17:31 karolherbst: ohh wait, I try that #version 0 shader in webgl now :D
17:31 karolherbst: mhh
17:31 karolherbst: seems okay
17:31 karolherbst: odd
17:33 karolherbst: ohh, might be compiled on a server...
17:35 imirkin: there's a translation layer between what you feed into webgl and what gets sent to libGL
17:35 karolherbst: ahh oaky
17:35 karolherbst: *okay
17:36 karolherbst: ohh right, chromium has it's own layer
17:36 karolherbst: totally forgot about that
17:36 imirkin: among other things, webgl glsl != glsl es
17:37 imirkin: [despite the obvious enormous similarities]
17:37 karolherbst: I think chromium even compiles to desktop gl
17:37 karolherbst: or rather translates
18:02 pmoreau: imirkin: If you are writing a new pass for NV50 IR (by subclassing the Pass class), how do you get to iterate over every insn, phi ones included? I said that I don’t want to skip phi nodes when calling `run()`, but iterating over the insn of a BB which contains a phi insn, does not return that insn.
18:03 pmoreau: Is it due to `bb->getEntry()` returning the first insn of a BB by skipping the phi insns found at the top?
18:03 pmoreau: *returning the first insn which is not a phi insn?
18:04 karolherbst: pmoreau: right
18:04 pmoreau: :-(
18:05 karolherbst: getFirst() I think is the right name
18:05 karolherbst: or
18:05 karolherbst: getPhi
18:05 pmoreau: I’ll try both, thanks
18:05 karolherbst: both actually
18:05 karolherbst: but getPhi does only return phi nodes
18:05 karolherbst: where getFirst returns the first insn ;)
18:06 karolherbst: check the "src/gallium/drivers/nouveau/codegen/nv50_ir.h" file when in doubt
18:06 karolherbst: " Instruction *getEntry() const { return entry; } // first non-phi instruction" ;)
18:06 pmoreau: Hum, can a BB contain multiple phi insns, or only a single one?
18:06 imirkin: pmoreau: start with bb->getPhi() and iterate until you hit bb->getEntry()
18:06 imirkin: multiple
18:06 pmoreau: Not bad
18:07 pmoreau: Thanks!
18:08 pmoreau: It does indeed work way better at getting the insns I want. :-D
18:21 imirkin: pmoreau: may i ask what you're trying to do?
18:22 pmoreau: The famous pass which will allow me to have working conditionals and loops! :-)
18:23 pmoreau: Getting out-of-SSA, for the SPIR-V -> NV50 IR code
18:24 pmoreau: (Well, "working", if the remaining of the code works as well :-D )
18:24 imirkin: so ... i haven't thought a *ton* about how to get that to work, but i think it should be possible to do it as you're doing the conversion
18:24 imirkin: with a pre-scan over the spir-v code
18:24 Tom^: yea karolherbst still freezes randomly even tho i only boost to the next highest state
18:24 imirkin: the nice thing is that you don't have to generate clean code, since the opt passes will clean it all up for you down the line
18:25 imirkin: so you can add all the moves your heart desires
18:25 pmoreau: The SPIR-V phi node stores pairs of <var,BB> already, so I was simply planning to add a copy at the end of each of those blocks, as you have suggested a couple of months ago already (IIRC).
18:25 imirkin: so ... that's tricky
18:26 imirkin: unfortunately i don't think that works in the generic case
18:26 imirkin: since you could have something like
18:26 pmoreau: At least to get a first working version; and later, once I understand things a bit better, I’ll implement a proper pass.
18:26 imirkin: well, it's a pain to write it out, but you could end up with a funny swap situation
18:26 imirkin: where you can't stick just a single mov in
18:26 imirkin: and have to create a separate bb
18:27 imirkin: that's what the current RA logic does
18:27 pmoreau: :-/
18:27 imirkin: for critical edges and whatnot
18:27 imirkin: iirc there's a moderately simple thing one can do if one doesn't care about extra mov's
18:27 glennk: swap of values is probably the most common tricky case
18:28 imirkin: i can never remember what the proper way to deal with all this is :( i always figure it out, and then promptly forget.
18:29 imirkin: hopefully each successive figure-it-out stage takes less and less time
18:30 pmoreau: It will do for now, I can pick up my sample codes to avoid swapping values :-D
18:30 karolherbst: Tom^: happens, I think the high engine clocks is at fault here
18:30 Tom^: :(
18:30 pmoreau: Just to at least process simple conditionals to check for boundaries when dealing with textures some time later in the future
18:31 imirkin: pmoreau: yeah, that's fine
18:31 imirkin: pmoreau: also remember that in nouveau, the order of phi node arguments has to implicitly match the order of incoming bb edges
18:31 imirkin: it's been on my todo list to fix that for ... literally years
18:31 pmoreau: Nevertheless, I’ll write down your comments to remember about it. :-)
18:32 imirkin: but there have always been other things to do, and "if it ain't broke don't fix it" and all that
18:32 pmoreau: Oh right, that rings a bell…
18:34 karolherbst: it is indeed a little messy to modify the CFG ;)
18:49 t-ask: compiling mesa may take some time :)
19:00 t-ask: wow, nice gfx crash with vertical pattern :P
19:15 kloofy: of course there is one big optimisation, which i haven't been intenting to ever implement, it is somewhat doable, and i linked to pdf where some dutch hackers did with cuda, blob does not do that
19:16 kloofy: mannerov said he might be intending to do such things on top of llvm
19:16 kloofy: it's called register reuse
19:16 kloofy: for me it is slightly too complex
20:20 t-ask: well, if I did compile and install mesa patched right, then the character names are still visible
20:21 pmoreau: imirkin, karolherbst: If I understand the Graph API, to iterate over the incoming edges, I simply need to get the EdgeIterator by using `bb->cfg.incident()`, and then use `get()->getOrigin()` and `next()` to iterate over those. Is that correct? And how do I link back from a node/edge to a BB?
20:21 imirkin: pmoreau: have a look at EdgeIterator uses
20:22 imirkin: basically you do BasicBlock::get(ei.getNode())
20:22 imirkin: or something along those lines
20:22 t-ask: What's interesting, the oveall system performance is weak . I can barely write, while htop shows only 20% CPU usage
20:22 karolherbst: pmoreau: part of my empty branch elim pass: https://github.com/karolherbst/mesa/blob/d9c247e643a1a65a23fe59626db136e928bc3218/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp#L3422-L3436
20:23 imirkin: pmoreau: https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/codegen/nv50_ir_ra.cpp#n465
20:23 pmoreau: imirkin: Silly me! I had an example of that, like right under my eyes, but didn’t saw it… --"
20:25 pmoreau: Thanks again for the pointers
20:26 karolherbst: mhh, odd, nvidia thinks that there is a fan on my system :/
20:26 karolherbst: well, on my gpu nvidia might be able to control
20:26 karolherbst: I think
20:28 karolherbst: k, I try to port this patch to nvbios: https://github.com/skeggsb/nouveau/commit/b66d8d56fac0508ff37f2d51f0d83bfc9b46d818
20:28 karolherbst: we need it for pascal vbios'
20:28 karolherbst: git
20:29 karolherbst: ...
21:12 t-ask: Does it matter if I disable Vsync and the graphic is suddenly brighter?
21:13 t-ask: I experienced that before, but never thought of reporting it
21:20 t-ask: anyways I report it then, who knows maybe it helps with another issue
21:25 karolherbst: the gp104 vbios I've got makes no friggin sense, even with the adjustments :/
21:25 karolherbst: I see a lot of fun coming towards us the next years *sigh*
21:46 t-ask: You mean, we book a flight to NV HQ and camp there until they hand over the docs? *g
21:47 pmoreau: WTF! I have one sample code using 2 phi nodes which now runs correctly, but the initial one I used, which only has 1 phi nodes still fails… --" (and by fail, it does compile, but the hardware isn’t happy about the code)
21:52 imirkin: pmoreau: feel free to paste the source + compiled version
21:52 imirkin:has had a lot of experience debugging this bs
22:13 karolherbst: mupuf: how did you figure out to force a temperature? :D
22:14 karolherbst: ohh wait, just messing with the reg, got it