00:04 swee: Mesa updated to 25.1.0, Nouveau GL works, but Vulkan doesn't work, even when an app tries to use Zink (I added NOUVEAU_USE_ZINK=1 env var) it says MESA: error: ZINK: vkCreateInstance failed (VK_ERROR_INCOMPATIBLE_DRIVER)
00:08 mhenning[d]: swee: what card? what kernel?
00:12 swee: mhenning[d]: GeForce 940MX (NV118), Kernel 6.14.6-edge
00:15 mhenning[d]: Do you have nvk installed? A lot of distros package vulkan drivers separately from the rest of mesa
00:15 mhenning[d]: eg. on arch it's vulkan-nouveau
00:20 swee: I'm using alpine if thats useful
00:33 gfxstrand[d]: What's your kernel version?
00:34 swee: I already said, 6.14.6
00:48 x512[m]: gfxstrand[d]: Is it possible to efficiently do memmove on Nvidia GPU? Or src -> temp, temp -> dst is the best option? Needed for dragging windows on GUI without composition.
00:51 gfxstrand[d]: I'm not sure on that one. I would guess either the copy engine or the old 2D engine provide some sort of guarantee but it's not like we have documentation. You'd have to experiment.
00:52 x512[m]: Standard Vulkan seems have no memmove operation.
00:52 gfxstrand[d]: swee: Sorry. I missed that. The Zink thing could be a red herring. What all devices does `vulkaninfo` give you?
00:53 swee: gfxstrand[d]: https://p.swee.codes/swee/daaef86e875b4ccfbb1edd8aa2d18333
00:54 gfxstrand[d]: x512[m]: Correct. Those are often found in 2D blit entirely engines precisely because XP and earlier Windows relied on it pretty heavily. But given how old Vista is, a lot of HW companies have started deprioritizing or even straight up deleting that hardware because it's a PITA.
00:55 swee: vulkaninfo shows gpu 1 as llvmpipe (0 is Intel UHD Graphics 620), but glxinfo with DRI_PRIME=1 shows the renderer string as NV118, hmmm
00:55 gfxstrand[d]: That's because you're getting nouveau GL there.
00:56 swee: ..right
00:56 gfxstrand[d]: I'm not sure why NVK isn't enumerating. It should. But I haven't plugged in a pre-Turing card in over a month.
00:56 swee: https://pkgs.alpinelinux.org/packages?name=mesa* I already have the mesa-vulkan-* packages listed here
00:57 gfxstrand[d]: If you run something with VK_LOADER_DEBUG=all, what do you get?
00:59 swee: gfxstrand[d]: https://p.swee.codes/swee/c8093f64755a42f0aa68666afa8a5a8b
01:00 gfxstrand[d]: swee: Yes, but they seem to only have AMD (which they call ATI for some reason) and Intel. I don't see Nvidia.
01:05 gfxstrand[d]: Maybe a better question: What do you see in /use/share/vulkan/icd.d?
01:06 swee: intel_hasvk_icd.x86_64.json intel_icd.x86_64.json lvp_icd.x86_64.json radeon_icd.x86_64.json
01:07 swee: Will ask Alpine's IRC channel about mesa-vulkan-*
01:09 gfxstrand[d]: I just poked one of the Alpine devs on Fedi. We'll see what they say. But yeah, you don't have NVK on your system
01:10 swee: Good to know, nothing broken on Mesa's side :,3
01:10 gfxstrand[d]: Yup
01:13 esdrastarsis[d]: gfxstrand[d]: This is probably relevant: https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/82829
01:14 gfxstrand[d]: Relevant but depressing.
01:15 swee: as fossdd said, they could move that depend to main
01:21 orowith2os[d]: swee: can't you build mesa yourself locally, with network access and whatnot, to get NVK?
01:22 orowith2os[d]: Not sure if the Flatpak runtimes contain NVK, but I'd bet that they do. So your apps should run on it either way, too.
01:23 orowith2os[d]: Ah, yup, master has Vulkan-drivers=nouveau
01:36 gfxstrand[d]: orowith2os[d]: So they're just waiting for a rebuild or something? I don't know how Alpine works.
01:36 orowith2os[d]: I was talking about fd-sdk, for flatpak
01:36 orowith2os[d]: Flatpak apps on Alpine will have nvk
01:37 gfxstrand[d]: Ah
11:24 asdqueerfromeu[d]: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35037
11:51 nandedahli91: Who is the idol for you who said, it's worth to mess around me? Do i pick sides, yes i pick sides, but not in conventional sense, since i have the strongest hand i pick sides (repeated words) from all of the best sides to accumulate my strongest hand , for example i get the best of United State, Germany, Austria, Russia and such and accumulate my strongest hand to the conflict, because me
11:51 nandedahli91: myself am Estonian , very powerful but little nation, i am their only GOD!
11:57 nandedahli91: I build you up, and you apply to my side, not that i ever apply onto yours.
11:57 nandedahli91: Because i have so magical people who's real man i am, I am holding so much better grip compared to you.
11:57 nandedahli91: They never want me die from my lines.
11:58 nandedahli91: I am their biggest brain.
11:58 nandedahli91: so i also have some privileges at hand.
11:59 nandedahli91: my senses are though extraordinary well coordinated, but it's just genetical permutations more like mutation in medicine, it's we are all aliens, I am not more than you an alien.
12:00 nandedahli91: When you consider, that monkey was just a first realease of an alien.
12:01 nandedahli91: That's why i do not care when you dump me, because i am the one with biggest brain anyways.
12:01 nandedahli91: Holding such position for very long time, it provides some benefits.
12:02 nandedahli91: People mostly start to love you by helping them out with their issues in life.
12:02 nandedahli91: If you did not feel helped, we can talk about it , but i am not going to beg my own.
12:03 nandedahli91: I am not the murderer i do not want you to die, i wanna train you to better version of you, cause i know how to do that.
12:03 nandedahli91: It's that i am so smart, i know how to do those things.
12:04 nandedahli91: i am not your enemy , but i am not gonna beg you to work for AMD; Intel, Nvidia etc.
12:04 nandedahli91: I am so much better than any of their biggest name is.
12:06 nandedahli91: So are you ready for the last flowless computation of all elements in the sequence?
12:07 nandedahli91: it's that only one logics cross i.e some sneaker in this method, where control flow is totally not needed anymore.
12:07 nandedahli91: than over eather human inputs the correct sequence if the equipment has no memory storage at them, like pocket calculator
12:07 nandedahli91: *either
12:08 nandedahli91: or you just switch the parameters in the arithmetid, those are called operands
12:08 nandedahli91: that is why i like the WASM best, it's the most closest to the hardware.
12:08 nandedahli91: arithmetic has two source operands and one result operand
12:09 nandedahli91: so before we had a deviation at the calculation
12:10 nandedahli91: now we do 224(aka twice 112)-288-144+48-280+144+72
12:11 nandedahli91: you notice that the answer is -224
12:12 nandedahli91: but there is a minor difference, it's now for everyone exactly the same -224
12:13 nandedahli91: cause now we do 112+112=224-172-136+4-130-136-68
12:14 nandedahli91: oops -130/-260/s'
12:16 nandedahli91: 224-272-136+4-260+136+68
12:20 nandedahli91: -236 should be the answer, and just 224 removed from it is -12
12:21 nandedahli91: so 116+6=122
12:27 nandedahli91: 224−193−193
12:27 nandedahli91: is 162
12:28 nandedahli91: 162-193-193=-224
12:29 nandedahli91: 224-162=62
12:29 nandedahli91: now you see we are not on 35 like before we are now at 31 times two
12:30 nandedahli91: so now only the last sigma proof, that deviation is totally removed comes from say 132 index
12:30 nandedahli91: that corresponds to indexes subindex or same index 52
12:31 nandedahli91: so now the calculation is 224-208-72-240+104+52
12:33 nandedahli91: but we remove from there 256-132-132
12:33 nandedahli91: *2 of that so it's gonna be -16
12:34 nandedahli91: so altogether last index is 224-208-72-16-240+104+52
12:35 nandedahli91: that results into -112-112-68
12:37 nandedahli91: and now the reference point being 31 for first and second, there is no flow needed anymore, cause the scale is aligned to four4 gap alignment
12:42 nandedahli91: so we now have perfectly accurate formula already
12:44 nandedahli91: 34-31=3 31+3=34 where as we use a reference that is 31
12:45 nandedahli91: and it was now 31 for both first elements, because one of them would lie , cause of odds and evens
12:53 nandedahli91: the algorithm is now perfect but it needs once current delta as first parameter and on arithmetic subraction first, then at second instance of the calculation as follow up, you switch reference to be first and the algorithm is now flowless.
12:53 nandedahli91: in other words, it is self-accurate but i already optimized it too
12:55 nandedahli91: I just do not like your attitude that i do a lot of hard work to you and You say i am mentally ill, when i am the one who is doing it the hardest.
12:55 nandedahli91: Cause i am the best what estonia has, I can not be wrong!
12:56 nandedahli91: I earn so much more money by not working for Nvidia Intel and AMD.
12:57 nandedahli91: ancient maths does not have such idioticy like Markov Chain.
12:58 nandedahli91: it has sequences, probability theory and derivatives and logarithmic ones too pre derivative
12:58 nandedahli91: factorial based algorithms and such, i gave you the highest grade
12:58 nandedahli91: which is the best of them all
13:01 nandedahli91: without my contribution to sceience of embedded and desktop systems, you'd be in a slum and in bears ass with your things.
13:01 nandedahli91: I did all the hard work.
13:01 nandedahli91: So cause i am so much more powerful and accurate in what i do, i care nothing about you!
13:14 nandedahli91: You think you will have a good life showing disrespect to person who has not earned any of that?
13:15 nandedahli91: Me i take none of such decisions but my reputation is so much higher
13:15 nandedahli91: so let's continue with what he have to guard are science with new modern computing facilitites
13:16 nandedahli91: so what you do is accomdate a format based of my findings, so that format is very powerful format 5 9 13 17 21 is the ideal sequence to represent powers of two's
13:17 nandedahli91: to accommodate multiple times of such number limited to it's max powers in
13:18 nandedahli91: you are getting any number of banks, and the calculations there are easy, but i tested only the base , so you have on max 72 as value possible 34 banks when base is 256
13:19 nandedahli91: so it's not exact power, but reduced by two elements it's exact power 32, but it never matters
13:25 nandedahli91: The decision we take is going to carry success, because we have the most skilled people at my side.
13:27 nandedahli91: If you have powers you do not want to die or anything , it's war about nationalism which does not exist, it's based of the other more narrower criterias , i.e per the certain human
13:27 nandedahli91: every nation has the top that is very good and skilled
13:28 nandedahli91: actually nationalism does not exist, but boy are ukranians and Russians proving wrong, i dunno!
13:28 nandedahli91: I'm more into thinking that Putin is in defence of his own nation to get rid of the discrimination and rusophobia
13:29 nandedahli91: but i do not have it, i like russians anyways, but it's too big nation
13:29 nandedahli91: in europe the nation never counts
13:29 nandedahli91: Like Swedish think they are better than Estonians, and they have their own top events
13:30 nandedahli91: but the rest cares none of their home nation stuff, me am like that too
13:30 nandedahli91: I doubt all swedish are better than me
13:30 nandedahli91: so in that case i can communicate with one from sweden who shares not the dogmas that they are better than our men
13:31 nandedahli91: and there is always one and ridiculously more from every nation
13:31 nandedahli91: But putin has a job to do, Russian nation is so moral, so beautiful that if he fails, he has hard time
13:31 nandedahli91: So after him, maybe no one tries
13:32 nandedahli91: Me i do not have that card to play that we have in majority anyways
13:32 nandedahli91: Estonia so little
13:33 nandedahli91: So this war is in the land of different matrix, Ukranians are proud as Russians are, they are one of the bigsmarty[d]
13:33 nandedahli91: but we are one of the littles
13:34 nandedahli91: Ukranians are bigger nut to crack compared to Estonians
13:36 nandedahli91: Me i know when Estonia is being attacked, all my allies come and only my allies , the country has less than i have
13:36 nandedahli91: In human force, so we fight for disntinction and littles are loved too
13:37 nandedahli91: My lines come from everywhere in the world to fight evil not nations
13:38 nandedahli91: I can not take decisions publicly for the other Estonians , i get jailed for it, they think they are better than me perhaps, so i can risk my ass and fight evil
13:39 nandedahli91: cause evil is worth a risk to be eliminated we need none of this
14:08 bigsmarty[d]: nandedahli91: Bro
14:08 bigsmarty[d]: What does any of this have to do with me
14:09 orowith2os[d]: Ignore him. You'll get used to it.
14:10 bigsmarty[d]: 👍
14:27 nandedahli91: Me i cared about you only to show you the best code, cause i am the inventer of it , i have no problems , cause i am so much more capable head than you, what is your question about predicates, well it's that you have flow stack as deep as you have threads per batch, since gpus are different than cpus, they have wider registering register files per work-item
14:28 nandedahli91: it's the most powerful paradigm, but as we learned it does not matter wether you have internal registers or external as number, or no registers at all, cause at the end of the day it's straight line of code what i produced
14:29 nandedahli91: it can be read from memory fifo just alike
14:29 nandedahli91: so what MILL cpu came up with is belt, which is lot less powerful than my algorithms, but is classified hance as registerless stackless architecture
14:30 nandedahli91: meaning you do not have external stack of registers in whatever memory at all, it's anyways that the memory read in with opcodes is contiguous memory or repeating fifo
14:31 nandedahli91: orowith2os[d]: hit the thumbs up lgbt enthusiast, it's not like i was not there for HdkR i like him in some sense
14:32 nandedahli91: but he was seeming to be on a road where he would had grown delusions on his reporetoir and arsenal, that things actually work so
14:32 nandedahli91: like he and llvm guys see.....
14:32 nandedahli91: so final time i came out to save you, you are on your own if wished now
14:34 nandedahli91: I also looked that something that translates from x86-64 to arm-64 and vice versa is indeed needed in the end, the solution is awful what HdkR has , but even this was one of my dreams, since llvm is strong in what it quickly offers
14:34 nandedahli91: memory that they use is insane, but this can change, the API is mature
14:35 nandedahli91: but FIRRTL being the one chisel the austrian and sziss hw project jointly uses is accurate, but it actually comes from YOSYS
14:36 nandedahli91: i would do it way better but IR based semantics also work, or heuristics or idiomatic expressions or intrinsics and such
14:36 nandedahli91: they offer very wide flexibility and some of it can be misleading to others, cause people are not so capable always to understand everything like me
14:37 nandedahli91: so in the worst case what HdkR achieved was binary translation during runtime
14:38 nandedahli91: which is totally unneeded but in terms of experience that is how giants grow into field of work
14:38 nandedahli91: so it's learning curve that makes them big and ability to learn
14:40 nandedahli91: I could use this fex translator but it has a big dep in the core to llvm which is very roundtripish
14:40 nandedahli91: for example wasmati is way better skeleton, and if skeleton was missing than i'd say solvers can generate it based of whatever complilation template
14:41 nandedahli91: so you configure a project than build it and make cmake template
14:41 nandedahli91: nowdays you deal with this dcbakers python make system
14:41 nandedahli91: but non the less, as long as it's cross platform which meson is, it matters few
14:43 nandedahli91: the algorithm which i offered is one percent accurate on it's range, so it hardly matters what you throw at it
14:45 nandedahli91: since the deviation came from, remember what we logically constructed, we constructed that every number divided by one and than two can be subtracted from distance so that remainder is going to be left behind
14:46 nandedahli91: but this remainder when reduced to single can only be even on single, on double distance it can be odd or even now
14:47 nandedahli91: so if you stretch out to the derivatives specification where you have 4 sections per full
14:47 nandedahli91: you actually get no error at all on the calculation, so there is no deviation fluctuating error in it
14:49 nandedahli91: since the remainder is now equally so to speak divided between the participants
15:49 gfxstrand[d]: swee: If you're interesting in helping with testing, it looks like the author of that MR trying to pull NVK into Alpine is going to give it another go.
16:36 HdkR: Nice nice nice.
16:53 jannau: the missing HoneyKrisp (asahi vulkan) driver is also enabled in the meantime as result of the post
20:15 gfxstrand[d]: I told the author just to get them enabled and not to worry about runtime testing. Hopefully musl doesn't bite us in the ass in untested ways.
20:17 HdkR: musl bit FEX recently in that some of their `long double` cmath transcendental operations were implemented at 64-bit precision instead of the 128-bit that AArch64 desires. That's such a wacky edge case that no one real cares about.
20:22 gfxstrand[d]: We got bit a few months ago by differences between their and glibc's memstream implementations. But now that's nicely abstracted and unit tested in CI (which runs on Alpine).
20:23 HdkR: Interesting
21:36 gfxstrand[d]: Most of the driver doesn't have very interesting interactions with libc. A qsort here, a printf there.
21:40 HdkR: Which is a good thing :D
22:06 gfxstrand[d]: Yeah....