10:45 pmoreau: karolherbst: How did you RE’ed that offset 0x60 thing on Kepler+? I am curious. :-)
10:48 karolherbst: uhm.. looked at traces. It was some time ago, but it also changes value when using nvaforcetherm
10:49 karolherbst: it is some time ago
10:49 karolherbst: *was
10:49 karolherbst: pmoreau: https://github.com/karolherbst/envytools/commit/d71cd71e4f0207e2ca6d33737f217010f4cc9314#diff-505e3fb1d454ce30b2e9b9f8a49e46ac
10:50 karolherbst: I think nvidia uses this reg starting with kepler or something and there were many reads from it
10:50 karolherbst: looked important
10:54 pmoreau: Ok. Shouldn’t the documentation mention that if offset == 0x60, you should substract the offset to the value?
10:54 karolherbst: and those were always read out as pairs
10:54 karolherbst: we could have been more specific with the docs, right
10:54 pmoreau: Ok
10:55 pmoreau: I”ll continue going through the patches.
10:55 karolherbst: awesome, thanks
10:56 pmoreau: I failed to get a working kernel yesterday… after the keyboard/mouse issue, I regenerated a new config file and recompiled a new kernel, but then I would get some warnings about vgaswitcheroo symbols that were not found when installing the module (despite vgaswitcheroo being selected in the config).
10:57 pmoreau: I need to try that again today.
13:49 karolherbst: pmoreau: I actually found a bug in the first patch due to your comment :)
13:49 pmoreau: \o/
13:50 pmoreau: Glad my comments were somewhat useful :-)
13:50 karolherbst: or maybe it worked anyhow.. but it was a logical bug
13:56 karolherbst: pmoreau: do you think it is a good idea to actually be able to call nvkm_therm_temp_get with val being NULL? I think we should rather keep that pain until somebody writes a function to check whether we can read it out
14:28 karolherbst: pmoreau: the call to nvkm_clk_ustate_update is gone, which was iterating over the pstates again
14:29 pmoreau: You remove the call, but you inlined the iterating directly in the function instead
14:29 pmoreau: *removed
14:30 karolherbst: ohh at the end
14:31 pmoreau: yes
14:33 pmoreau: What’s the order again: ctor (only run once), then init (run after ctor and when resuming)?
14:33 karolherbst: I think I ended up adding this, because this was actually a bug I should fix properly
14:33 karolherbst: like if you select pstate 0xd, but it isn't there, it would crash the module
14:33 karolherbst: but I think it doesn't anymore and the loop can be removed?
14:33 karolherbst: not _quite_ sure
14:33 pmoreau: That’s… slightly annoying if the module crashed
14:34 karolherbst: I think nvkm_pstate_prog is already guarded against it now
14:51 karolherbst: pmoreau: by the way, do you think you will find some reasonable amount of time during the next month?
14:52 karolherbst: then we could talk about what is all still missing for OpenCL and maybe even write it down on several trello cards or so
14:54 pmoreau: karolherbst: Hum… the SPIR-V linker got merged in SPIRV-Tools, so that is out of the way. I still need to send some fixes, and prepare my talk for the Khronos meetup on Friday… :-D
14:56 pmoreau: And I’ll be away a bit in the coming weeks, but I should still have some time to do some work on Nouveau
14:57 karolherbst: pmoreau: awesome. I was actually thinking about fixing those silly compiler bugs we know of, we should just have those collected somewhere
14:58 pmoreau: I still have a paper deadline early Dec., but I should have a decent amount of time still
14:58 pmoreau: That would be good! :-)
14:59 karolherbst: ohh, you got passe the clk rework patches, now the nice patches are coming next :D
15:00 pmoreau: I have a public Trello board with my todos, and I keep track there of some of the bugs I found, and series I haven’t sent yet to the ML, or which haven’t made it upstream yet.
15:01 karolherbst: ohh,I see
15:02 pmoreau: karolherbst: https://trello.com/b/YyQi5Wy2/nouveau for reference
15:02 karolherbst: uhh,I think I encountered that RA bug regarding loops once as well, not quite sure but I think I hit it somewhere
15:03 karolherbst: wait.. I think it was a F2015 shader
15:04 karolherbst: something looked odd there, but I didn't investigate furhter
15:04 karolherbst: but it looked a bit like RA reassigning registers it shouldn't have
16:15 karolherbst: ui, GL4.6 CTS was pushed
16:35 pmoreau: It wasn’t available on Github yet?
16:35 karolherbst: no clue when, I just fetched and saw it got pushed
16:48 pmoreau: Ok
16:53 karolherbst: oh crap. trello started to change :O damn you atlassian
16:53 karolherbst: the first obvious change it is already worse
16:59 karolherbst: imirkin: and idea if you find time to finish those enhanced_layouts patches this year? If you say it won't be this year, then I will at least try to work on those.
16:59 karolherbst: *any
17:17 pmoreau: Which changes are you talking about, Karol?
17:17 karolherbst: pmoreau: https://patchwork.freedesktop.org/patch/175959/
17:18 karolherbst: ohhh
17:18 karolherbst: you meant trello
17:18 karolherbst: the right menu
17:18 pmoreau: Yes :-)
17:19 pmoreau: Hum… what did they change there? It seems to me it has been like that for ages.
17:19 karolherbst: force reload
17:20 karolherbst: it is different and it annoys me how it is now
17:22 pmoreau: Hum… might have gotten used to it then I guess.
17:23 karolherbst: mhh, it is quite new
17:23 karolherbst: it wasn't like this 1 month ago
17:26 karolherbst: pmoreau: it looks like this for you, right? https://i.imgur.com/gb5Hyc5.png
17:27 pmoreau: Huh, absolutely not
17:27 karolherbst: ;)
17:27 karolherbst: guess why I said force reload
17:27 pmoreau: I did try the force reload, but still…
17:27 karolherbst: or maybe this is like testing for 1% of users
17:27 karolherbst: anyhow, it is on all boards for me like this
17:28 pmoreau: Ok
17:28 karolherbst: okay, even in their newest blog post there is the old design
17:29 karolherbst: so maybe they are currently deploying it?
17:29 pmoreau: Could be
17:30 karolherbst: anyhow, it looks worse than before
17:31 karolherbst: I wouldn't be surprised if trello ends up looking just like Jira and gets abondened in a year or two
17:45 pmoreau: karolherbst: I won’t be able to review 10 & 11, I have no idea if CLK could have some dependencies on THERM, nor how all that pm_runtime works.
17:47 karolherbst: pmoreau: it doesn't, but therm has some on clk
17:50 karolherbst: but yeah, no worries for those. skeggsb actually suggested those changes
17:51 karolherbst: ohh wow, and a bunch of new tests were added to the 4.5 CTS
17:51 karolherbst: we moved from 98.6% to 98.7% though
17:53 karolherbst: new fails: KHR-GL45.gpu_shader_fp64.builtin.mod_dvec{2,3,4}
17:53 karolherbst: but I think imirkin knew about those already
18:08 pmoreau: karolherbst: I’m a bit surprised you are using u8 for `temp` in 13, after patch 1 which introduced using ints cause Pascal had more precision and Kepler+ could report negative values.
18:10 karolherbst: pmoreau: uhhh, I wrote that patch before writing that hwmon rework
18:10 pmoreau: Ah, ok :-) Might want to update 13 then
18:42 karolherbst: uiui running the CTS now
18:47 docmax: hello
18:48 pmoreau: Hello docmax
18:49 docmax: are you guys working on nouveau?
18:49 docmax: for free? how is the progress?
18:52 docmax: i have the feeling the linux community is stagnating
18:53 pmoreau: yes, yes, and slow for my part
18:53 karolherbst: docmax: there might be good news next month ;) also I will focus more on getting students to do an EVoC project for nouveau
18:54 karolherbst: docmax: are you a student? :D
18:55 docmax: im working at IT support at a telecommunications company
18:55 karolherbst: sad :(
18:55 karolherbst: do you know students who might be interested?
18:55 docmax: what skills do they need?
18:56 karolherbst: uhm, self organization, self learning, motivation
18:56 karolherbst: maybe the first thing isn't as important
18:56 karolherbst: but time is
18:57 karolherbst: 3 months free time basically, because EVoC as GSoC is a full time job for 3 months
19:03 pmoreau: Phew, almost reviewed 2/3 of the patches. Only took me the whole day (almost). :-D
19:03 karolherbst: :) many thanks from my side though!
19:04 karolherbst: I still have some unrelated things mixed together, but I will clean those up after we discussed the stuff I wrote about in the cover letter
19:04 pmoreau: Ok
19:05 karolherbst: pmoreau: and for the CTS, we can ignore everything which isn't a Fail
19:06 pmoreau: Really?
19:06 karolherbst: so, current status with the master tree: " Failed: 35/7391 (0.5%)"
19:06 karolherbst: yes
19:06 pmoreau: So a Crash isn’t bad?
19:06 karolherbst: well
19:06 karolherbst: there are no crashes ;)
19:06 pmoreau: Doh, not fun!
19:06 karolherbst: if there is a crash, the entire run fails
19:07 pmoreau: Ok
19:07 karolherbst: it ain't piglit
19:07 pmoreau: I need to fix my IADD patch for GM107+, and see if it helps with any piglit/CTS tests.
19:08 karolherbst: ohh right, still wanted to run the CTS on more modern GPUs chipsets
19:08 karolherbst: but for now I will finish it for kepler1
19:09 pmoreau: Sounds good
19:10 karolherbst: I see a few fails with the current CTS run I do, but they can be bugs within the CTS as well... this will be fun to figure out
19:10 karolherbst: especially those bs errors: "got GL_ERROR_X1, but expected GL_ERROR_X2"...
19:40 feaneron: Is https://nouveau.freedesktop.org/wiki/FeatureMatrix/ updated? I'm thinking specificaly of the NV110 column (Maxwell)
19:46 pmoreau: feaneron: I don’t think it gets updated that often. I think EXA and Xrender are done.
19:47 feaneron: yeah. I'd be glad to help here, but looks like I need to authenticate to be able to edit
19:48 feaneron: by the way, did you check for EXA/Xrender somewhere else?
19:48 feaneron: (sorry, I'm completely lost in Mesa website)
19:51 pmoreau: No, from memory
19:52 karolherbst: mhh
19:52 karolherbst: 6 fails which aren't most likely CTS bugs
19:54 feaneron: karolherbst: what a nice surprise. I was just reading your patches :)
19:55 pmoreau: feaneron: GM10x has EXA support for sure, and GM20x has acceleration support through the Nouveau DDX (not sure if that means EXA support though). imirkin would know for sure since he’s the one who did it.
19:55 karolherbst: airlied: is there a way to get the additional tests for the CTS integrated into the current public master branch?
19:56 karolherbst: pmoreau: if we disable pipeline_statistics_query, we have one fail in the CTS as far as I can tell. Would need to investigate all other failures, but those seems to be actual bugs withing the CTS
19:57 airlied: karolherbst: in theory sending pull requests on github should start a process
19:57 airlied: karolherbst: if you are crazy enough to write CTS tests :-)
19:57 karolherbst: I meant the confidential tests
19:57 pmoreau: airlied: Congrats on radv being conformant by the way!! :-)
19:58 karolherbst: airlied: doing a run right now, but a lot of fails look like things imirkin and others fixed already
19:58 karolherbst: in the cTS
19:58 airlied: karolherbst: oh doh, not sure, I should see if X.org adopter gets access
19:58 airlied: karolherbst: oh you are doing a run on the adopter cts 4.5?
19:58 karolherbst: yeah
19:58 karolherbst: 15 fails in the first seed :)
19:58 airlied: there are set of patches from intel/radneom
19:58 airlied: radeon
19:58 airlied: pmoreau: thx!
19:58 karolherbst: and 9 of those should be fixed in the master
19:58 airlied: karolherbst: on top of the 4.5 branch
19:58 karolherbst: yeah
19:58 karolherbst: I have the intel ones
19:59 karolherbst: but I don't want to start mixing those with the radeon ones now...
19:59 karolherbst: maybe I have to in the end
19:59 airlied: the radeon ones were mostly the same
19:59 pmoreau: karolherbst: Don’t you have access to the internal CTS through RH? I had access to it through Unity.
19:59 karolherbst: pmoreau: duh... ;) no, through xorg
20:00 pmoreau: You got a @xorg address?
20:00 airlied: karolherbst: so does x.org adopter let you download the git repo?
20:00 pmoreau:is jealous
20:00 karolherbst: yeah
20:01 karolherbst: airlied: not quite sure, maybe? They wanted my github account name, but I didn't find a link except to their gitlab repository, which I can't access
20:01 airlied: there's a script in cts
20:01 karolherbst: ohh, true
20:01 karolherbst: let me check
20:01 pmoreau: Link to the gitlab should be available from the member page IIRC
20:02 karolherbst: I have that link, but I need an account
20:02 karolherbst: there as well
20:02 pmoreau: karolherbst: What’s the protocol for asking for such an address? Last time I asked mupuf, he told me they were still trying to figure out how to distribute those.
20:03 airlied: karolherbst: oh I'm not sure if adopters can get gitlab access
20:03 karolherbst: pmoreau: https://www.x.org/wiki/Khronos/
20:03 karolherbst: airlied: yeah, that's why I just wanted to get those additional tests for the public repository
20:03 karolherbst: should be sufficient already
20:03 airlied: karolherbst: if those tests were able to be made public, they'd be in there
20:04 airlied: they aren't hiding them in gitlab because they want to be difficult
20:04 karolherbst: well I can download the tarballs
20:04 pmoreau: If I read correctly, it won’t be possible to submit conformance tests for hardware that hasn’t been declared conformant by the hardware vendor. So, unless NVIDIA does submit OpenCL>=2.0 conformance tests for its hardware, we won’t be able to submit those either. :-(
20:04 airlied: yeah only via adopter status, but yeah I don't think adopter has any ability to get anything other than the offical CTS
20:05 airlied: pmoreau: start with OpenCL 1.2 :-P
20:05 karolherbst: airlied: okay
20:05 pmoreau: airlied: I am not even targeting 1.2, 1.0 is good enough for me atm. :-D
20:05 karolherbst: at least I know now that it looks like I expected it
20:05 karolherbst: pretty good
20:05 airlied: karolherbst: but the offical GL4.5 + Intel patches should be pretty much complete
20:06 airlied: I'm sure AMD has one or two patches for gallium related things
20:06 karolherbst: mhh, well I know some of those tests were fixed in the CTS
20:07 karolherbst: uhh, I can actually check if those fail on intel as well
20:08 airlied: karolherbst: okay amd added 4 patches on top
20:10 karolherbst: okay
20:10 karolherbst: good to know
20:40 karolherbst: 10 fails to go :)
20:42 airlied: karolherbst: will mesa 17.3 be submittable then? :-)
20:43 karolherbst: airlied: ;) depends when 17.3 will be released
20:43 airlied: still have to convince imirkin on some of the fixes? :-)
20:43 karolherbst: most likely in December I guess?
20:44 karolherbst: ohh imirkin won't have time afaik
20:44 karolherbst: but... I guess I will be able to make a good progress next month? dunno
20:44 karolherbst: actually planned to work more on compiler bugs
20:45 karolherbst: that 3d image fail is really something though
20:45 airlied:should go for cts fixes on r600 someday
20:45 RSpliet: should we go for cts fixes on vieux?
20:45 karolherbst: RSpliet: do you like pain?
20:46 RSpliet: only in the French sense...
20:46 karolherbst: :D
20:46 airlied: I doubt public cts has much vieux coverage
20:46 airlied: and cts before GL4.4 is pointless
20:47 RSpliet: Un petit-pain n'a jamais blessé personne...
20:48 RSpliet: sorry, I'll try and keep it serious again.... although airlied: why do you say cts before 4.4 is pointless? Bad coverage? No coverage?
20:48 karolherbst: it doesn't run
20:48 karolherbst: 4.4 doesn't even run if you don't support 4.5
20:49 RSpliet: Oh that's a bit daft...
20:49 karolherbst: allthough that is quite fixable
20:49 karolherbst: but still
20:49 karolherbst: there aren't many tests for 4.3 and prior
20:49 airlied: RSpliet: Khronos only enforce adopter on 4.4/4.5
20:49 airlied: and lots of tests rely on extensions without testing for them
20:49 airlied: and the gl wrapper they use is crap
20:50 karolherbst: well we could try to run the mustpass lists from the public one
20:50 karolherbst: that should somehow work
20:50 RSpliet: airlied: right. So near-zero coverage and worse than piglit from the sounds of it
20:51 karolherbst: RSpliet: what exposes vieux at most? 2.1?
20:51 karolherbst: or are there some with 3.x support?
20:51 pmoreau: RSpliet: :-D
20:51 RSpliet: Heh... not sure. Doesn't 2.0 require shaders? I'm betting 1.3 or 1.4
20:51 karolherbst: ;)
20:52 karolherbst: soo pretty pointless I would assume
20:52 RSpliet: Yeah... investing in nvfx is one thing, but vieux is... well, if mooch wants to have a shot after her emulator works :-P
20:55 mooch: RSpliet, well, i'd need some help getting it to work anyway
20:56 mooch: the stupid timers are too slow and they barely work, so
20:57 RSpliet: mooch: I feared/suspected you might run into trouble long before the point of investing in something that you didn't sign up for :-P You might just be the biggest vieux expert right now though!
20:57 RSpliet: (and potentially the only potential user left...)
20:58 mooch: lmao
20:58 mooch: yeah
20:59 mooch: mwk is another expert
21:00 mooch: RSpliet, also, i only emulate nv3, nv4, and nv5 right now so
21:00 mooch: yeah
21:00 RSpliet: Yeah... and NV3 is pre-vieux even iirc
21:00 RSpliet: Riva 128 right?
21:00 mooch: yeah
21:01 RSpliet: From the times where your screen resolution was limited to the amount of RAM on the board...
21:01 mooch: yep
21:01 mooch: dos quake does work in 1024x768x256 tho so
21:01 mooch: then again, that's worked since forever
21:05 karolherbst: imirkin: are you aware of oddness with GL_DEPTH32F_STENCIL8?
21:10 karolherbst: seems that something is broken with samples >2
21:53 imirkin: karolherbst: some piglits fail
21:54 imirkin: when in doubt as to what GL versions are supported where -- https://people.freedesktop.org/~imirkin/glxinfo/#v=Mesa%2017.1.0&p=compat
21:54 karolherbst: imirkin: ahh, that makes sense
21:55 karolherbst: guess I will try to fix those piglits tests first, because those should be _much_ easier to debug
21:56 imirkin: there's currently an issue in nouveau too regarding msaa and glReadPixels/glDrawPixels
21:56 imirkin: and glCopyPixels
21:56 imirkin: we're supposed to do a resolve and we don't
21:56 karolherbst: sounds most likely like the issue those tests hit
21:57 imirkin: dunno - those are ancient-gl calls
21:57 imirkin: oh, and this only affects winsys buffers with msaa > 1
21:57 karolherbst: not > 2?
21:58 imirkin: no
21:58 karolherbst: mhh
21:58 imirkin: nothing special with 2 vs 4
21:58 karolherbst: well
21:58 karolherbst: the test only fails for >2
21:59 karolherbst: and the error offset is the same for 3-8
22:00 karolherbst: 0x7a00 in all cases
22:02 karolherbst: it works for UINT_24_8, DEPTH24_STENCIL8, but not for F32_UINT_24_8_REV, DEPTH32_STENCIL8
22:03 karolherbst: and in the latter case only if the dest is sampled and passes if the source is sampled
22:08 karolherbst: imirkin: ohh by the way, it's a 3.0 test which fails
22:09 karolherbst: and the tests uses glDrawArrays/glReadPixels
22:10 imirkin: yeah dunno
22:12 karolherbst: basically this is what happens: create FBO/RBO, Render, blit, downsample blit, verify
22:13 karolherbst: downsample blit is a blit from tested samples to 0 samples
22:14 karolherbst: ohh
22:15 karolherbst: and there is a format to RGBA8 conversion somewhere
22:17 karolherbst: well, will try to fix those related piglit tests first
22:17 karolherbst: easier to talk about the code
23:03 imirkin: airlied: some of the current fixes are a bit more questionable.
23:04 imirkin: i think it's doubtful we'll be ready to submit in time for 17.3 given our current time commitment, which is quite minimal
23:05 imirkin: airlied: i haven't looked at the official tests yet, but from the public repo most of the remaining fail is common though. however for the official ones, i'm not sure what we'll do as i suspect that the spec has changed since they were written.
23:26 airlied: imirkin: the 4.5 fails arent common
23:26 imirkin: airlied: they are for the public repo
23:26 airlied: since amd/intel already pass all those tests
23:26 imirkin: with patches, i suspect.
23:27 imirkin: i'll have to find out what it is they precisely pass, and whether they still do
23:27 imirkin: with current tree
23:27 airlied: you habe those patches
23:27 imirkin: i haven't done the khr access thing yet -- no time
23:27 airlied: at least x.org adoptersdo
23:28 airlied: the remaining gtf fails are likely nouveau specific
23:28 airlied: gtf hasnt changed.in years
23:28 imirkin: like i said, my only info is based off the public tree
23:28 imirkin: oh, and i guess the tests you ran a while back
23:29 imirkin: although i think there have been a number of fixes to mesa since then