00:51wideopen: Overall the impression is very good, as to what i am did grep out from gini and triton sources, double or arbitrary int length can be encoded, unroller for sequential logic, all the needed functionality is in place, except the dma compute machine offloader, fascinating , qvm and other sources are fine too, as said every country and bigger institution have those solvers, i expect very fast performance, since i am not used to dealing with common
00:51wideopen: lisp, not much of the qvm i looked at, however soon i would say case close, and entirely completed after doing some amount of testing.
00:52wideopen: at overseas i started to build a gaming system, did a lot of building on mono, and overall impression was good there too.
01:14wideopen: It's the nvidia inference server carrying the same project name, but their ways are not memory friendly and hardly hence make any sense. Why would anyone have to install terabytes of ram to do deep learning is where i get lost in terms of sanity.
01:33wideopen: I soon get my batteries, so i could program a bit more in more restful states, however i want to look at EGL today a bit, the idea there is to have indirect buffer addresses to be identified through regex, to access those buffers to compress to, ideally since khronos rambled EGL together quite interestingly and there is an interaction with glMapBuffer among pixmaps and surfaces, no machine code of the underlying hw needs to be known to calibrate
01:33wideopen: the buffers and start to call them in zero copy ways.
01:35wideopen: if that gets done with precision lifters in compressed format, things like autopilot dekstop gl program on embedded GL ES can be achieved.
01:35wideopen: a lot of games can be automatically ported to run on ES hw
01:38wideopen: With linux there is no problem on supporting anything, but regex ways on cross platform EGL would be needed to support other OSes, me i do not use other OS systems that much, but others do.
01:50wideopen: USB displaylink kind of gud in mesa, is already very interesting, but i said it needs a bit of reprogramming , you actually should upload the dma procedure sequence from compressed state, but the rewrite is very feasible expectedly enough, it's the idea that was important and base usb hub transactions are likely there.
02:01dcbaker: zmike: \o/ Thank you!
02:09wideopen: I get back to sleep, but it very much looks like an absolute victory is near by, like Charlie Sheen would say, win here win there win win everywhere :)
04:06soreau: daniels: are you still waiting to hear back from ajax on https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24700 ?
04:08soreau: zmike: Does xwayland work for you on zink+radv?
04:08soreau: because it causes xwayland to segfault here
04:15soreau: running wayfire with zink and gdb attaching to the Xwayland pid, then starting xcalc, I get a segfault with this bt http://ix.io/4H2J
06:40habernir: Just out of curiosity what the status of mesa 23.2? why dylan don't update the release calendar or write what the status ofr mesa 23.2?
06:42HdkR: It's on its way, just running a bit late
06:43habernir: if their is a ETA people would happy to know
06:44kode54: hi
06:44habernir: because people don't like to stay in the air
06:44kode54: I saw someone on the mailing list submitting an attempt at gsc-mei for DG1/DG2 for xe.ko to the mailing list
06:45kode54: does this mean someone may use that to support the HuC firmware eventually?
07:09wideopen: well I do not have anything to spam the channel with, but everything is fully exposed by abusers anyway, they seem to surveillance my every step, that is why i went live too, cause they claim i am ill for their abuse, but my mind is ok theirs isn't. I have nothing against explaining in case of questions to someone, but spamming i would not have done otherwise. Event loop has such structure itself, it first evaluates the constant and then does
07:09wideopen: something, solvers every one have a thing called assumption for this, and you can assume based of response, this will add only one clause in triton and gini and the check would be to ask if it's variable or constant, it's how i see, in case you want to rebuild an event loop. Constraints would add all clauses of path.
07:12luc: hello, all! I encounter a _XReply stalling when I ssh my host machine which running Xorg server (no -X Forwarding). I notice that it occurs only when ssh login user is different from remote login user. is that X11 or system security mechanism something?
09:46jani: airlied: sima: there was a discussion about issue templates for the gitlab drm group, see backlog. if you agree, could you set maintainer-tools as the template repository for drm group please? instructions at https://docs.gitlab.com/ee/user/project/description_templates.html#set-group-level-description-templates
09:51airlied: jani: I might have a need for sleep, but I'm not seeing the option in settings for the drm group
09:52airlied: sima: ^ maybe you can confirm I'm not missing it
09:54jani: airlied: hmm. the docs say "Premium" and "All offerings", and the latter should include "Self-managed". I wonder if it isn't a config option in the instance level somewhere. daniels?
09:57jani: the gitlab tiers are fairly annoying
10:11sima: jani, I think that's just all offerings of the ee edition
10:13jani: sima: get your terminology right :p https://about.gitlab.com/handbook/marketing/brand-and-product-marketing/product-and-solution-marketing/tiers/
10:13sima: hm
10:13sima: maybe some knob I need to hit to enable this first?
10:14jani: I'm not saying you're wrong, it could very well be that this is only available for "customers"
10:14sima: jani, hm isn't this two dimensions?
10:15sima: like there's the tier
10:15sima: and then there's the deployment type, which can be "gitlab.com" "self-managed" or "all offerings"
10:15sima: like the instance-level templates obviously is only a thing on self-managed, not on gitlab.com
10:15daniels: ‘all offerings’ means that it works on gitlab.com and self-hosted, but ‘premium’ indeed means only the highest $$ tier (no matter how it’s hosted)
10:17sima: jani, yeah so not a thing for us :-/
10:18jani: daniels: so should I be reading that as you need to have Premium AND All offerings, not Premium OR All offerings
10:18jani: I hate gitlab more every day
10:18daniels: correct
10:20jani: daniels: I'll read that as an answer to both my comments
10:21jani: robclark: can't have nice things ^
10:32jani: sima: I don't seem to be able to do anything with the git repo of https://gitlab.freedesktop.org/drm/intel, as it's empty. does that need owner privileges?
11:29wideopen: a seeming mental illness goes with all those cases around programming, especially since human beings are valued and evaluated by quick response capabilities, that are all prepared more over a robotic mode, which makes few sense, but composing algorithms and seeing things through as of being some kind of inventor, that is backed by stamina and brain power combination. Which by this fraudulent todays world standard both
11:29wideopen: might seem a bit ill, as well as gets classified as ill being too.
11:30wideopen: that means that you are often more creative when your quick response instincts they value so much and reflexes are rather down.
11:37wideopen: i have created many algorithms they use today in the best kit's though i recreated them, hence i am understanding what the paper says about gini compression, cause i was rambling it the same way, you apply division three times, then modulus and then shift, and each index is a continuous bound taken from each separate bit length
11:38wideopen: but to compose that algorithm, you really need to think hard, and show some accuracy, and that combination of stamina and brain power and work mode, does not feel very much mental well being.
11:41wideopen: mental well being is that you have a girl who is agreeing with you , or supporting you, and you have time to commit to your family etc, you get some at times, not that you work on things that demand very high end accuracy and brain power.
11:54wideopen: but division itself, with power of two operand , is a bit shift anyway, so you see that it divides with 32 three times, so this indeed is a power of two arithmetic which can be done with bitshift or even look up tables itself.
12:05wideopen: So yeah it looks correct, i had previously lately read the same results on the web, so you can store over million entries with one round of such minification, the exact number is 32*32*32*32 which is 1048576
12:08wideopen: web page suggests over 800k entries, but that is within the range, since they hardcode some MSB bits in elias fano sometimes, to reflect the mode of operation.
12:48sima: jani, owner is only for groups, gives you permissions to great projects/groups in that group
12:48sima: if you're maintainer, you should be maxed out
12:48sima: jani, I guess plan B is a gitlab-config branch and make that the HEAD?
12:49jani: sima: you and airlied are still owners for drm/intel, and I'm just a maintainer... and I can't figure out how to create a repo
12:49sima: at least with tools like dim the confusion should be minimal ...
12:49sima: jani, yeah but for a repo owner==maintainer
12:50sima: hm I can make you owner, but I thought that does nothing for repos/projects
12:51sima: jani, you should be owner now
12:51pq: there is a difference between owner and maintainer for a project too, IIRC, like the permission to delete it.
12:51sima: but looking at https://docs.gitlab.com/ee/user/permissions.html that shouldn't help
12:51sima: pq, yeah you can do some namespace operations if you're owner
12:51jani: sima: thanks... I guess PEBKAC then :(
12:51sima: jani, does it work now?
12:52sima: if owner fixed it, we need to make the other maintainers owners too I guess
12:53sima: jani, oh do you want to create a repo within intel? like drm/intel/kernel?
12:55jani: sima: no, not like that, just top level like what https://gitlab.freedesktop.org/drm/amd has, but I don't know how :/
12:55sima: ah
12:55jani: on the left side, the Code menu item doesn't even exist
12:55sima: settings -> general -> visibility, project features, permissions
12:56sima: you need to enable the repo first
12:56sima: we disabled that so people don't try to clone the wrong thing :-)
12:56jani: sima: yay. I've looked at that a dozen times, but I don't know how I failed to see that
12:57jani: sima: thanks!
12:57sima: glad to be of service :-)
12:57jani: sima: perhaps you can just demote me back to maintainer (or since I seem to suck so bad with the interface, even that's too much :p)
12:57sima: yeah done, we want to make sure you don't click the wrong button and accidentally delete it after all :-P
12:58sima: jani, I guess when we enable the repo might be good excuse to also drive the migration from legacy cgit to gitlab a bit forward?
12:59sima: since "why is there nothing in the intel drm repo" confusion was why we disabled it in the first place
12:59jani: sima: dang it!
12:59sima: jani, oh maybe disable merge requests?
12:59sima: and forks and all that
13:00sima: at least until there actually is something else there than gitlab config templates
13:00jani: sima: for starters, I was not planning on pushing kernel code
13:01sima: yeah hence disable all that stuff
13:01jani: how do you propose the config templates get changed then?
13:01jani: *issue templates
13:02sima: git push or so
13:02sima: but yeah if you want the web ui then you need mr and forks
13:02jani: I mean sure I can update them, but I'd rather delegate that to other people ;)
13:02sima: developer should be enough to push
13:03sima: creating an MR iirc only needs you to be able to see the project (i.e. public or you're a member if your private)
13:03sima: like creating an issue
13:03sima: roughly it's "project visible to you" = you can create issues/mr
13:04sima: "reporter" = you can wrangle issues
13:04sima: "developer" = you can also push code and wrangle MR
13:04sima: "maintainer" = you can touch settings on top of the above
13:04sima: "owner" = you can also delete it
13:19jani: sima: so I added the most basic thing for starters, to be continued. https://gitlab.freedesktop.org/drm/intel/-/issues/new
13:25sima: nice
13:25sima: w/e now I guess for you?
15:05jani: sima: yes. I've got a bit of an old amiga 500 revival project going on. :)
15:09pq: jani, haha, enjoy! :-)
15:10jani: pq: thanks :)
15:26cmarcelo: folks, we are baking a solution to handle the "misleading indentation" note from GCC that involves disabling the particular warning for libnir and possibly enabling it as a -Werror in clang CI: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25315#note_2094272 if someone has feedback on either of those, please reply on the linked thread.
15:29wideopen: For what i see, we are equal or level so to speak, i did consume your code, and act as research backing up department, you wanted to code the base yourself, and it indeed worked out, but you wanted to do this by yourself, and that make sense, since most of you get paid to do this, however i do not owe you, since all the research i went over, i reported back.
15:38gfxstrand: We really should get rid of INTERP_MODE_NONE outside of GLSL....
15:45DavidHeidelberg[m]: with new kernel 6.5, iris-apl-egl timeouting, but I'm not sure what is fatal issue and what just warning, can someone point me: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/49344107 in right direction?
15:46wideopen: Forced adjustments really sometimes open up some chakras , all the evil came down to me, cause i was extremely fortunate which so to speak i had every shape going the way needed and on the route to have no single issue , none could compete with such ways , however there is the slim bits of truth, that being so fortunate did not get my brain or thinking going at full potential, cause outside of conspiracy i did not have too much issues at all, but
15:46wideopen: it's like very very long time ago i understood, that i have to simulate the issues but not get hit, so computers is only one field where to simulate and stimulate forced thinking and crisis resolution, and gain more experience.
15:47wideopen: confidence when doing well etc.
15:51wideopen: and only recently in table tennis, i had played before hundreds of tournaments, and was invited to participate in table tennis again being injured, and i found that lowering the pressure point to muscles lower the knee, you can time the attack from half bounce , and attack very consistently and strong. I did not have that back up system in the game before, so such things changes open up some new methods during crisis.
16:13gfxstrand: I really want INTERP_MODE_NONE to die...
16:13gfxstrand: I might just make spirv_to_nir not generate it
16:14gfxstrand: I think it's needed for GL for $REASONS
16:14gfxstrand: Like you san set flat shading through the API on ancient versions.
16:16gfxstrand: Hrm... The GLSL spec says "When no interpolation qualifier is present, smooth interpolation is used."
16:16karolherbst: ahh yeah.. in gallium we live patch the shader depending on what the API set...
16:19karolherbst: ohh, anybody with proper nir knowledge have any opinions on my vec8/vec16 source lowering I'm doing in https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25330 ?
16:20gfxstrand: Oh, wait... is ShadeModel only a fixed-function thing?
16:21gfxstrand: Hrm... No. lower_flatshade is a thing...
16:22gfxstrand: Does anyone NOT do flatshade as a variant?
16:23gfxstrand: I'm guessing nvidia theoretically can but I doubt we care
16:24gfxstrand: Oh, shader_has_one_variant...
16:24gfxstrand: ugh
16:27mattst88: wideopen: we made an agreement. do you remember?
16:28gfxstrand: Iris shader-keys it internally
16:28gfxstrand: radeonsi does as well
16:29gfxstrand: mareko: Is there a good reason why radeonsi does flatshade as an internal key instead of just letting st/mesa lower it?
16:29zmike: because st/mesa lowering breaks precompile
16:30gfxstrand: *sigh*
16:30zmike: smh you thought
16:31gfxstrand: Looking at it all, though, it seems like flatshade only affects COL0, COL1, BFC0 and BFC1, none of which are ever declared in shaders.
16:32zmike: "ever" is a strong word
16:32zmike: piglit tests would probably argue otherwise
16:34gfxstrand: Right...
16:35gfxstrand: And... getting chewed out by joss in the DMs. :rolling_eyes:
16:35gfxstrand: It's been a while. :sweat_smile:
16:36karolherbst: you live in the worst of timelines
16:36zmike: flatshading is still a thing and so is xfb
16:37gfxstrand: https://www.youtube.com/watch?v=POT3plx0vBs
16:39zmike: and it has french subtitles
16:39zmike: truly the worst timeline
16:39mareko: gfxstrand: the internal flatshade key field is not used for compiling the user shader, it's used for compiling the prolog, whose binary is placed before the user shader binary
16:39gfxstrand: mareko: Ah
16:39mareko: so we can interpolate colors outside the user shader without interpolating colors outside the hw shader
16:41mareko: and that's why colors are sysvals
16:44gfxstrand: Yeah, that makes sense.
16:45karolherbst: lowering of i2f32@8 is kinda a pain....
16:45gfxstrand: Intel has lowering for this. On NV, I think we can use InterpFreq::State
16:50gfxstrand: I wonder what happens if you IPA.STATE on gl_FragCoord.w....
16:50gfxstrand: Does it give you 1.0? That would be very convenient.
16:54mareko: I don't know IPA.STATE
16:55karolherbst: IPA.STATE just interpolates according to how it's set up in the shaders attributes
16:55karolherbst: and passes the value as is if it's instructed to not interpolate
16:55mareko: and what does gl_FragCoord.w have to do with interpolation?
16:58mareko: our FS gets values of 3 vertices for each shader input component, and (i,j) barycentric coordinates to interpolate between them, which are different for center, centroid, sample, perspective, noperspective, at_offset, and at_sample
16:59gfxstrand: mareko: On NV, we have to multiply by gl_FragCoord.w in the shader.
17:00mareko: I see, we just get (i,j) and do: v0 + i*(v1 - v0) + j*(v2 - v0)
17:00gfxstrand: If IPA.STATE does what I think it does and if IPA.STATE on gl_FragCoord.w gives you 1, then we can just use IPA.STATE for all INTERP_MODE_NONE things and smash the SHADE_MODEL HW bit and everything happens by magic.
17:01gfxstrand: if IPA.STATE on gl_FragCoord.w doesn't give you 1, then we'd have to predicate the FMUL by 1/w.
17:02gfxstrand: NVIDIA loves to make ancient GL go brrrr
17:02mareko: we also have the shader model for explicit interpolation where we get (i/w, j/w, 1/w) as a sysval, but we don't use it
17:04mareko: and I'm surprised you interpolate gl_FragCoord, we have it as a sysval
17:05gfxstrand: Yeah, IDK. That's the way the HW works.
17:31kchibisov: Is binding texture could be a relatively fast operation? As in, there's no reason to `if` when you use 2 textures and swap between each other from time to time.
17:36gfxstrand: Depends on how often you're switching
17:36gfxstrand: But if you're doing that, I recommend an array binding and indexing it instead of an if
17:37kchibisov: Ah, we bind once a frame basically.
17:38kchibisov: We just have this weird issue when someone decided to inject their code to do a texture based screencapture, and they fail to restore our texture so the state messes up...
17:38kchibisov: We basically have one texture and could end up rebinding when we started adding more textures, but most of the time we run with exactly one texture so we never bother rebinding it.
17:56gfxstrand: Oh, once a frame is fine
17:56gfxstrand: a few thousand times a frame is probably fine
17:57gfxstrand: 10s or 100s of thousands of times per frame and I'd get worried
17:58kchibisov: Ah, ok, thx.
18:51mareko: oh don't you like open-coded TES interpolation using an array of patch varyings instead of 1 per-vertex varying
18:59mareko: oh yeah and let's use patch outputs are temporaries because why not
21:22alyssa: mareko: HLSL->GLSL translation gunk?