02:35 mareko: why is debian-arm64-asan timing out? https://gitlab.freedesktop.org/mesa/mesa/-/jobs/69952104
08:58 daniels: mareko: I have a fix for this in flight
14:37 kisak: mareko: Note that mesa!33211 will force Ubuntu 20.04 and 22.04 to be discontinued in my PPA. There's no clean way around it because https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113258 didn't make its way to those Ubuntu releases.
14:40 mareko: ok
17:23 mareko: robclark: what do you think of enabling glthread by default for all drivers and removing the option to disable it? (to optimize GL dispatch) do we really have many single-core CPUs that we would care about?
17:24 robclark: do we have _any_ single core cpu's that we care about? That said, I don't have any particular concerns atm but haven't looked too much into perf of glthread
17:25 mareko: robclark: it's enabled by default on radeonsi and zink
17:51 DemiMarie: robclark: Single-core CPUs seem unlikely outside *deeply* embedded systems, and I think those would be microcontrollers.
17:52 robclark: right.. or really old things that are on amber perhaps
19:30 DemiMarie: I think there may be a lot of single-core CPUs in deeply embedded systems even today, but they aren't things one would put a GPU on.
20:08 mareko: I think that if we enabled glthread by default for everybody, I could nuke most of the glapi spaghetti: GetProcAddress can just return _mesa_marshal_* and internal dispatch tables can just be direct pointers to _mesa_*
20:11 mareko: oh and we would need to nuke indirect GL as well
20:48 swisslandlord: You will find that this formula works well. Which was bit surprising to me, but it descends into some kind of linear algebra formula likely, that i was not so strong to recap/recall without excessive testing. 29+3-52=-20 and -20+29+3=12 -20+29+29=38 38-12-12=14 and 38-52=-14 , so -20+29+3=12 that is needed to be done for all elements, so that is single 12 that i tried to tell. but
20:48 swisslandlord: instead of doing this again for all elements, now you selectively do -20+3+3=-14, so you have elements on 24 and 12-14=-2 , hence when you add 14 as per 52-38 and -12 as per 26-38 that is 2 selectively, you now have 24 at some elements and and zero for others. and answer according to real algebra not mentioned is 26-24=2+26=28*2=56 and 128-56=72. filtered answe, i am optimizing this for
20:48 swisslandlord: r300, but i do not see shorter routines atm. so i hammered/bombarded the best one to the channel already. But real motivation is never r300, but 945gme, that china manufactures in stacks for atom processors that of intel. r300 is rare setup and already out of any manufacturing habbit.
21:01 swisslandlord: now this is for data and tbh. computation on those formats is lot easier, which i had already before, but did not print this to the channel, it's very logical and short , it's simpler for me to understand cause no linear algebra but calculus is needed only.
21:06 mareko: swisslandlord: you should come to XDC sometime
21:15 swisslandlord: i think i mentioned yes my brainpower is on top but it has nothing to do with my brainpower that backed this method out, since the computer is developed through the powers of two's and intentionally they allow such methods to be hacked in. The real lifting happened in the hardware and presentation of bits.
21:42 swisslandlord_: i love X and linux and such things yeah, but formal answer is , i've been assaulted too many times in life and too many times in foreign countries, and i am unsure if this is greatest idea, however i tested the things on those hashmaps, and i would infact have many slides to offer.
21:51 swisslandlord_: So as to how much the upstream expects modernized hashes i do not know, but i seem to have better methods than java and scale sets, which i could not read well, but chatgpt said to me, as to how they work in words, but actually the filesystem i talked about puts the memory above 2 in power of 22 or so, can be more but with little more latency involved . Definitely amazing things i have
21:51 swisslandlord_: talked about in public, are the people going to do amazing things if we add such code, is not known.
21:55 swisslandlord_: so and the paper from jvesely and his collegues were trying the first steps to run kernel on gpu , i am trying the same non-the less, account with his ideas when i do code, i always account with gpus one day doing such things as in the paper.
21:57 swisslandlord_: but actually i looked at nasa coding guidelines video, where it made all sense as to how code can be debugged the best, and avoid very creapy bugs that could explode anything.
22:35 DemiMarie: mareko: please ditch indirect GL so that Xorg can also ditch it.
23:02 swisslandlord_: I am also very busy on security related things :( for past couple of days, some prankster here attacks my wifi with deauth packages. Typical stuff, i am not being saved even at my homeland, but when troubles are ahead at least i can reach some of my friends here, i.e not entirely alone among wolves.
23:05 swisslandlord_: I suspect russian related people, who likely know the same code, but don't want ukranians to update their systems.
23:07 swisslandlord_: ukranian military so far only depends on support of united states, and some other countries, classical format as to how things should not exclusively be. They are in deep trouble in fact.
23:27 karolherbst: dcbaker: how does b_sanitize interact with rust targets? I _think_ I figured ou the weird issue I was hitting with "rust.test" and the tldr is that if the address sanitizer is used, and rust.test uses a crate using "rust_abi: c" it fails to link not finding the address sanitizer stuff
23:34 karolherbst: mhh maybe it's not in fact the rust_abi, but more linking against C/C++ libraries
23:51 dcbaker: karolherbst: I think I have patches to make rust work with the sanitizers that I never landed… I wouldn’t be surprised if there are corners like that