01:36 kisak: Now that the literal week of failing llvm builds is over with launchpad, the kisak-mesa PPA is now on mesa 24.1.0. karolherbst, no new snafus with rusticl, same oddities as 24.0.x for Jammy and Mantic. I dropped the hacks with the fresh Noble build.
01:39 kisak: There's a _really_ high fail rate with llvm builds on launchpad's ARM runners whenever the build farm is saturated with a distro rebuild. Not a problem for anyone here to deal with.
03:09 DavidHeidelberg: Venemo: hope it helps! :)
15:50 lumag: dianders, anything remaining for your 'mipi_dsi_*_multi()' series?
16:35 lumag: dianders, ah, excuse me. everything is in drm-misc-next, but the tree is not a part of linux-next yet.
17:03 dianders: lumag: you had me worried. I was pretty sure it had all landed.
18:21 heapsniffer: I had realized that it's possible to encode 64bit datapath on 16bit hardware it's possible to do upto 1024bit in theory, I was getting sceptical about performance it's pretty complex set of ASIP for ASIC dicts to do that extraordinary fast for 64bit grouped alus. So I am dealing with it but summer is full of other jobs, expecting perhaps to release an alpha towards the end of the year
18:21 heapsniffer: finally, it uses customized encoders, that's a body blow after inspective number testing I see a way. so even avx can be emulated but this could on 1024bit avx do one inst at time on 16bit hardware. on 64bit arch roughly max number divided with 6144 can be grouped, yeah so freaking many, it's a rocket science, I am unsure if I could make it while I live, in theory yes it seems, all the
18:21 heapsniffer: base hardware would suite, those numbers come with latest testings, last two weeks work, very tough compiler arch task though for injured men like me to complete, but you are way to arrogant and Ill to even get close to sanity. sw work should be done differently and more aggressive than you imagined.
18:36 heapsniffer: 64bit instructions on 64bit hardware can be grouped as many as max number divided with 6144, that's how it would perform with 10 to 20 base arithmetic, and compilation is something similar it accesses preassembled dicts only in theory so it's highly fast compilation times too, those dict combinations can be cached too, so I am going crazy, it's rough but possible, it's like I wish it
18:36 heapsniffer: wasn't, cause I am enough lunatic already, maximalist, idealist etc.
18:37 heapsniffer: but OS as Linux base is more complex to write to run only single arch even.
18:40 heapsniffer: so those people who did it like Gary Killdall and Linus Torvalds were thinking even better, they had to think ahead of trends and time.
18:44 heapsniffer: it's a hardware and software histories heroes that make such computation possible today, it's not by coincidence, they should know all what they did if by chance it was not a fluke they made so capable sw and hw.
19:03 Ermine: dwfreed: requesting banhammer
19:38 heapsniffer: it is actually seeming to be so, the magic lives in that formula 2xdistfromconst+offset+index+value but you doubly encoded the distance and at runtime the incoming data, so those in formula between each other share 1 common bit I figured, that needs attention, so 2x distance buffer needs a custom encoder alike thingy, I had success in my head with it, slightly controlled over with bitcalc.
19:39 heapsniffer: so yeah those numbers are roughly correct, that I offered as to how capable Asics actually are, FPGA goes even better.
19:48 heapsniffer: your abuse and terror seems enough deep so that so far any code has not materialized, but only numbers of true real hashing expected throughput or performance which yeah seems to hold, not sure if I can follow through the code too, your terror has been noted.
19:54 heapsniffer: according to my tests encoder + add or subtract behaves for that formula like some custom gate
19:55 heapsniffer: it's like bitwise and/or and add sub mixture
19:56 heapsniffer: so I had one more discovery that some mutant gates alike can do things effectively.
20:16 heapsniffer: technically Unix predated BSD minix and everything dos and whatever, Dennis passed away but Thomson may be even living this time around.
20:17 heapsniffer: but this was very cool work they did.
20:17 heapsniffer: I am fine as to how unixes and oses work
20:23 heapsniffer: graphics API wise directx is the simplest and best, then I like opengl es, slightly tougher paradigm but still ok actually I like es more than DX from user perspective
20:24 heapsniffer: but vulkan I do not like.
20:25 heapsniffer: so many to choose from and there's no actual problem, Intel hardware has some new very stable driver it never fails on my MacBook
20:27 heapsniffer: but why they did opengl was that one day Microsoft could drag people to courts and possibly win
20:28 heapsniffer: it's like you had to have or get opengl to avoid and rely on something in case this happens
20:29 heapsniffer: whether you had to have vulkan too is by me not, but I can live with other opinions
20:30 heapsniffer: I.e different opinions that say vulkan is good, my opinion it's trouble but whatever
20:33 heapsniffer: it's a way to feed developers cause users encounter many locking and freezing issues, but whatever developers have to be paid anyways some how
20:44 heapsniffer: it's actually meant so that some violent group can take over the development as new people are born who have no clue what real men did and even look alike, they think that modern ways are better and that's all they know about anyhow, cause history was erased.
20:52 heapsniffer: in the world where everyone wants to shine and many love to abuse, even memory shows or games tend to think that invention of mice or input devices mice are so aged, so are touchpads and all modern is better to not give credit for geniuses like Butler Lampson.
21:05 heapsniffer: the world was not meant to become so sick as it is, but I do not know why it's so sick and how to regulate sick people better, cause I was not born as one.
21:06 heapsniffer: it's rather open subject as to how to avoid wars alike and abusers getting a worldwide choke hold, as to over engineer and constraint things more or do it less, somehow it's possible to sentence narcissist crews.
21:08 heapsniffer: but for me issues are stereotypical I get more abuse and danger than the average person but I get more kindness too, I am myself in charge what to store as more important.
21:39 riteo: why are there so many synthetic spambots here
21:42 Sachiel: there are none
21:51 heapsniffer: what is significant about this composure or this mutant gate that takes couple or more instructions in the doubly packed format according to the formula, it seems to work on vectors of arbitrary length.
21:59 karolherbst: dwfreed: fyi ^^
22:09 iive: riteo, there is only one and he is not synthetic
22:17 DavidHeidelberg: karolherbst: my pipeline won't finish in time (rootfs rebuilt took too long), and ask you to assign mine again in circa 15 minutes and then yours? Mine will be done in 5 I'll not have to re-run everything again
22:17 DavidHeidelberg: *can I ask
22:17 karolherbst: yeah, you can remove mine from the queue
22:18 DavidHeidelberg: thx! :) btw. I realized with current rootfs I'll be able to test rusticl on iris too
22:52 kode54: what is suggested for getting debug logs of GPU hangs that kill the system so totally, there's no hang in the journal?
22:53 dwfreed: try netconsole?
22:57 riteo: iive: what do you mean they aren't synthetic
22:57 riteo: is that all handwritten, like, from a real person?
22:57 dwfreed: yes
22:57 dwfreed: but there's no need to comment on the spam
22:57 riteo: yeah right, sorry