00:14samimullikulli: There is bit of so far unknown information that i never wrote to you, 42 from hitchikers road to galaxy is not special in any way as a prime or even number. however the synthesis never mentioned 219 i suppose, or i can not remember. where 112 was mirror if 232 same as 192 being mirror of 81 where 232 also mirror of 61 etc. so 51 comes in three random powers selected 37 41 and 57, you
00:14samimullikulli: see we do not have mirror for 41 since we have unusual odd number of powers where 32 and 64 bit is saner, but this was presented for sake of clearity and simplicity only (quite random selection) so 64 of them or 32of them would work exactly alike. so then 193 is 51+61+81 hence where 84 was 4 44 and 36, hence 51 is just 135-84 aka where 37+57+41=135, so i am afraid you go to jail, cause
00:14samimullikulli: i personally never lacked brainpower nor skills and i had been victim, however i have no victims mentality i will kill abusers with my soldiers. And the solutions have been all authored by me as to what you have in apis. Just you are debiliated scammers. But both ends of the spectrum can be handled with various compensation values like constant 51 until it is enough middle of the
00:14samimullikulli: range, where 42 in that thee power paradigm can be forced to suite as well. There is no special number without their relational ecosystem i.e a function on sequences or around number algebra. But i developed batched no iteration fastest in-place quantum replacement paradigm which works anyhow. You know that Russia ain't gonna win the war against ukranians and estonians like me, our
00:14samimullikulli: allies are way stronger, but it's more of a question what india and china brazil iran and all all the other brics countries do. 100 dollars around hundred bucks is a cost of truck trailor worth full of 486 chips which can be tuned immensely powerful for defense attack or surveillance drones, and ukranians are not retarded it seems either in addition to being also one of the best sports
00:14samimullikulli: attenders in the world. Once again the destiny of the world is in the hands of china mostly, if they go fighting alike we can not stop that issue anymore i am afraid, it's the end of the civilization frankly extinction, they as you however were never smarter than me or my friends however but again enough smart to cause trouble like you do days for days. I have material to put you
00:14samimullikulli: lifetime behind bars whenever i choose to do that, i hold you in choke position for the rest of your pathetic abortion leftovers lives.
02:22cmarcelo: alyssa: sent you a link for a CTS change
09:37bonzini: hello - i created my first MR at https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36284
09:38bonzini: but i am a bit confused by all the manual jobs in CI
09:38bonzini: i started the ones that seem appropriate to not consume too many resources
09:44bonzini: but i can't find a job that covers all the Rust code
10:23daniels: bonzini: the build jobs cover ~all of it
10:28glehmann: why is test-d3d12-quick_shader using a different piglit compared to everyone else?
10:33daniels: just to fit them into the execution time
10:44alyssa: mareko: Hmm, alright
10:44alyssa: I wonder if this is a spec bug or a CTS one then
10:47valentine: glehmann: plus it's windows while everything else is debian
10:58arnd: airlied, sima: Are you planning another drm-fixes pull request for 6.16? I think we need to revert e5478166dffb ("drm/nouveau: check ioctl command codes better") because of the regressions https://lore.kernel.org/all/aH9n_QGMFx2ZbKlw@debian.local/
11:06glehmann: does anyone understand what's going on with the build jobs here? https://gitlab.freedesktop.org/mesa/mesa/-/pipelines/1475926
11:27sima: arnd, yeah we'll do a drm-fixes every week usually
11:28sima: dakr, ^^
11:29daniels: glehmann: timeout: sending signal TERM to command 'bash'
11:29daniels: the jobs are timing out because they take too long
11:30glehmann: are these jobs running some tests or are they just building?
11:30daniels: I assume this means that your changes now make clc run longer
11:31dakr: arnd, it seems I wasn't Cc'd on the regression report..
11:31daniels: glehmann: well, for example, if you click on the debian-release job, that takes you to https://gitlab.freedesktop.org/mesa/mesa/-/jobs/80989087
11:31daniels: you can see at the bottom that it's still running a meson build
11:31daniels: if you scroll up to the top, you can see that it doesn't run any tests before it does the build
11:32daniels: if you click on the other jobs with the red X, you can see the logs of those jobs, and what they're doing / have done
11:33glehmann: those logs tell me what was successfully compiled, but not what's causing the timeout
11:33glehmann: or am I reading them wrong
11:34daniels: what causes the timeout is that the build takes longer than the timeout
11:35glehmann: mhm, okay
11:36glehmann: are these jobs building more than the other build jobs?
11:37daniels: all the jobs suffixed with -release do a release build, i.e. -Dbuildtype=release
11:37daniels: it's a little bit of a scattershot, but you can see what each job builds by looking at the meson configure summary at the top of the job, or by reading .gitlab-ci/build/gitlab-ci.yml
11:39glehmann: I guess maybe there could be a clc shader that's infinitely unrolled now
11:40glehmann: this MR is cursed, first the broken piglit test, and now that it's fixed (well, except for d3d12), it's blocked by this
11:42daniels: it's a canary to let us know when build jobs are getting too long, because throughput is an issue: assuming we have 15min of hardware testing per MR, then if build jobs take 5min, then we can turn around 72 MRs per day. if build jobs go up to 15min, then we can only turn around 48 MRs per day.
11:46dakr: arnd, caught it on the list and replied.
11:58alyssa: running NIR passes as part of build jobs was definitely one of the ideas of all time (-:
11:58arnd: dakr, sima: ok, revert sent.
11:59alyssa: (at least from my POV though still, benefit >> cost)
11:59dakr: arnd, thanks!
16:22glehmann: https://gitlab.freedesktop.org/mesa/mesa/-/commit/7b70b419b5282fb01b7202e5ee1d391fcba15ef8 effectively means that just printing a shader invalidates nir_metadata_live_defs
16:23glehmann: I guess it's fine because nir_index_ssa_defs actually clears the meta data bit
16:23glehmann: but it's still weird
16:41glehmann: jenatali: can someone at microsoft deal with their part of https://gitlab.freedesktop.org/mesa/mesa/-/issues/10409 ?
16:42glehmann: it's one of the last remaining parts, together with lima/powervr
16:42jenatali: Yeah I'll try to find some time
16:42jenatali: Thanks for the ping
17:03anholt: jnoorman: VK_STRUCTURE_TYPE_WSI_IMAGE_CREATE_INFO_MESA / wsi_image_create_info is an example of where we define an internal extension struct. Though if you wanted to use such an ext with an out-of-tree tool like vkrunner / amber, we'd end up wanting a real spec.
17:20jnoorman: anholt: thanks! I'll try to get something quick and dirty working before thinking about specs :)
17:20anholt: exactly :)
17:25anholt: for anyone else wondering, our idea is "instead of having freedreno's computerator for running some assembly as a compute shader, wouldn't it be neat for GPU experiments to have vkrunner or whatever be able to pass your backend's assembly text to be parsed as the shader instead of spirv. then, in an ideal world, we have renderdoc be able to pull assembly from the normal spirv-based pipeline, you can edit it in the ui, and see the
17:25anholt: difference live.
17:30cwabbott: that would indeed be very neat
17:32cwabbott: I think that if it's a MESA extension we could probably skip most of the debate and skip it as a "vendor extension" of sorts
17:37anholt: yeah, if this was a generic "MESA driver backend assembly ext, and the driver gets to be fussy and reject your assembly and maybe give you an error string about it", that would be pretty neat to be able to share. I think pipeline linking for non-compute is going to be the tricky part.
17:43mattst88: jfyi, intel has an `executor` tool that takes .lua/pseudo-asm files as input (see src/intel/executor/examples/bfi.lua)
17:45anholt: it was extending a similar tool to support emitting another kind of GPU state, and wishing we didn't have to (we already *wrote* a driver!), that triggered this experiment.
17:56robclark: in-place shader asm editing in trace replay would indeed be neat
18:12zmike: has anyone ever seen an issue with ralloc usage and asan?
18:13zmike: trying to figure out if what I'm seeing is my problem or asan
19:01xeyler: dianders: hi, i’m brigham! i’ve been communicating with you on the mailing list about updating the jdi and novatek drivers to remove deprecated mipi dsi macros. thanks for all your feedback!
19:24xeyler: i’m a CS undergraduate at utah state university. i graduate next spring and i’m considering pursuing kernel development as a career. do you mind if i message
19:28xeyler: you directly to ask you a couple questions?
20:00alyssa: jenatali: can you ack this trivial patch https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36147/diffs?commit_id=586f8316daad463720f8a18601f06364c0553d28 ?
20:00alyssa: I think this should be ok for all gcc/clang/msvc but I don't know computer
20:01alyssa: https://learn.microsoft.com/en-us/cpp/c-language/typeof-c?view=msvc-170 suggests it should be ok
20:01alyssa: but needs to VS 17.9 which hopefully CI has
20:02alyssa: nope CI has 16.0
20:02alyssa: good grief
20:04alyssa: #ifdef it is
20:08alyssa: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36147/diffs?commit_id=ac53c39e037d07c7236bf7bf7b56aed0041916d0
20:08alyssa: jenatali: slightly less trivial ^
20:17jenatali: Looking
20:17jenatali: I thought we had up-to-date CI
20:24jenatali: alyssa: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/81037774 says that the actual reassociate pass was fine
20:25alyssa: jenatali: there was another issue with reassociate, which I fixed
20:25jenatali: Right just pointing out that typeof was fine?
20:25alyssa: (enum bitfields not sizing right)
20:25alyssa: oh, hmm, ok
20:26alyssa: we can try merging with __typeof__ I guess
20:26jenatali: Yeah fine by me at least
20:26alyssa: i dont understand MSVC's versions
20:26alyssa: and no please don't explain it ;)
20:26jenatali: IIRC I went through a bunch of pain to get it in for Faith a while ago
20:26jenatali: Psh I'd have to understand it to explain
20:32alyssa: :clown:
20:40jenatali: glehmann: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36299. Probably needs iteration with CI but it's a start :)
21:19jenatali: Man my metadata handling was bad
21:19alyssa: jenatali: don't worry, everyone's was
21:19alyssa: which is half the reason I pushed for nir_pass
21:20alyssa: (over nir_pass_v I mean)
21:20jenatali: Yeah
21:21alyssa: (the other half is that it runs nir_validate a lot less often with minimal loss of coverage, which speeds up CTS)
21:22jenatali: Yeah it totally makes sense
21:22alyssa: jenatali: btw, `nir_shader_lower_instructions` is usually not what you want
21:22alyssa: we should really rename it tbh
21:22jenatali: Hm? Why not?
21:23alyssa: That's specifically designed when you need to recursively lower the output of your lowering pass
21:23alyssa: nir_lower_int64 is the prime example
21:23alyssa: In 95% of cases, nir_shader_*_pass is all you need (smaller hammer)
21:24alyssa: usually ends up being less code anyway, especially now that we have the alu/intrinsics/etc aliases
21:24alyssa: we should probably rename to nir_shader_lower_instructions_recursive or something but it's not something I care to drive forward right now
21:24alyssa: this is more of a "helpful effective NIR tip" then me telling someone in code review, haha
21:25alyssa: I've had "write a doc on how to write effective NIR passes" on my todo list since 2023 so uhh
21:26jenatali: Yeah fair enough
21:26jenatali: I just personally like the ergonomics of returning the nir_def *
21:27jenatali: If there was a non-recursive variant of this which didn't split filtering from lowering that would generally be my preference
21:27jenatali: Especially if it automatically filtered based on instruction type with alu/intrinsic variations
21:29alyssa: we could probably do that...
21:31jenatali: Wow that's a lot fewer failures than I expected...
21:32alyssa: it is fascinating to me that other people don't surround themselves with test boxes for CTS runs :P
21:33jenatali: I'm not sure I've ever actually built the GL CTS... and I've gotten a new dev PC and haven't set up the Vk CTS on it yet
21:33alyssa: i have a 3 letter alias to push whatever is in my ~/mesa/ to a dedicated asahi test box and trigger a full GL+VK CTS, i cannot imagine going without loll
21:33alyssa: no contention with other people's changes, it's on my LAN :P
21:34alyssa: (n.b. this is not a suitable replacement for CI of course. but it's fascinating to me that for some people CI is a replacement for this.)
21:35jenatali: Well my last dev machine was... not great. Pushing to CI was significantly faster than trying to run things locally
21:35alyssa: o_o
21:35jenatali: Probably not true after my upgrade
21:36jenatali: Managed to convince leadership I needed an 18-core Sapphire Rapids chip for WARP dev ;)
21:36alyssa: seems good to me
21:36alyssa: i do like when things are sapph and rapid..
21:40airlied: alyssa: my office is mostly CTS test machines
21:41airlied: I think I spent a day a release cycle on dnf upgrades
21:43alyssa: airlied: now that I understand =D
22:05alyssa: jenatali: https://gitlab.freedesktop.org/mesa/mesa/-/jobs/81049611
22:05alyssa: I.. what?
22:05alyssa: I guess it just wants me to add a cast?
22:05alyssa: why is this Werror?!
22:06alyssa: back into the queue it go
22:11zmike: alyssa: zink+hk testing when
22:12alyssa: when there's a reason for it
22:12zmike: brutal
22:47alyssa: gfxstrand: are you planning to look at https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36265 ? should I merge with Mary's rb and Lionel's ack?
22:47alyssa: (we want this even for NVK. Mary has a patch and the stats are really good.)
22:52alyssa: running fossil stats on Mary's nvk patch now
22:53jeffreysand: Your achievements during 30years of development is kind of pinchartl and matts88 fecalist type of total stirred saliva, you reverse engineered X11 replaced this with wayland without single compression techinque nor performance enhancement, let me do all the hardlifting and you are brave enough to call inventors and engineers of real logic as mentally ill schizoophrenics. You are deluded
22:53jeffreysand: mistakes in life. I already wrote all the opencl video input drivers, also cracking the RSA, did all the networking setups. You are undoudedly very ill people extorting money with your anal actors backing of development with deluded garbage which comes from your fingers and mouth. You have forgatten to write even single solid AI solution or filesystem that makes sense, in other words your
22:53jeffreysand: code is huge mindfart and mistake, yet you terror other people. Performance code has nothing to do with say bayesian probablility not even ffm no markov chains, those formulas can be presented in maths with integral computation, which is part of higher math course using inetgreable variables and subterms, it composes of calculus like derivatives that show rate of change and rate of
22:53jeffreysand: change of a change, aka in physics velocity and acceleration and linear algebra, where you derive another value or variable or formula or subterm by first subtistuting the variables possibly with values , your video was kit was generated from tensor products that i offered, your opencl by mentally ill people was 10years of garbage that came out, until i got involved. You are all entirely
22:53jeffreysand: mentally ill people and should forget spawning terror at me, until it is too late for you.