00:08 kisak: generally speaking, mesa doesn't implement D3D12, but projects like vkd3d and vkd3d-proton are around which translate that to Vulkan.
00:18 kitzune: ohh, thanks. i'll check that
12:33 takenout1: smallbitset is even too rough rust crate for me to grasp immediately. But i think the smallest set is the key operation there, for example, you place all combinations as a query of set x, that maps the addition coefs. so 4+6 is 10, so it orders a set as all small numbers, say you had 1to122, but there is a gap you also has 512to634, so altogether you have 122*2 numbers, you query hence all combinations of 122+10 i.e 132 numbers, and the
12:33 takenout1: smallest subset/combination is passed forward. and you bitwise out the unneeded bases like whatever to land to a correct arithmetic result. in that case the mapping between answers and their source operands is always the same.
12:55 takenout1: overall it is very interesting crate still, such a small code but reasonable effort needed, but biggest question comes the way or along the lines of those iteration unrolling and checkpointing friendliness.
12:56 takenout1: and for this documentation is missing, but next weekend i need to test, it took so long to read it.
13:02 takenout1: it's very small crate/lib, but it does so many things, and much is not yet documented, in this regard yes, it looks very interesting.
13:32 takenout1: The data flow graph is possible to be built, not the end of the world hard, but biggest problem was raised, parallelization and loops merged and unrolled. was it called fission or fusion, w3c stuff is already so large in all implementations like signal collect, rdf graph implementations, very difficult to read.
15:34 karolherbst: mareko: what's the timeline on using ACO by default in radeonsi or who would be the one I should talk about that?
15:53 kisak: ^ I'm the wrong person to anwser, but yuq825 is the one who's been working on that.
15:54 kisak:frowns at the typo
16:20 karolherbst: but they aren't on IRC are they?
17:13 mareko: karolherbst: no timeline
19:42 takenout1: There one of the most primitive but efficient ways is radix tree, patricia tree, but more powerful ones as modified ven de moas and fusion structures, but fun but serious bosses there in science come from center of the europe, dutch, belgians, lux, but germans austraians and swizz and such are beyond of them all
19:43 takenout1: those guys are so fun it's pleasure to read what they do
19:44 takenout1: maxwell and all this english stuff, besides faraday is overhyped.
19:44 takenout1: all the real bosses were in central axis
19:52 takenout1: he was way before, they had another good guy too, but he was way earlier, thermodynamics laws, Newton.
19:52 takenout1: Newton was their big gun
19:52 takenout1: later Edison was just a businessman
19:59 takenout1: spanish , portuguese , french and italians and rumanians more roman related, but also very impressive
20:01 takenout1: roumanians , me, arian races guy myself, but this line is also the best in science
20:01 takenout1: my father is a jew and his mother, but my mother is more like german blond
20:07 takenout1: germans document things very formatted and strictly, cause they do not put prank in their work
20:07 takenout1: they leave very few open questions
20:08 takenout1: and smallbitset is another example, though dark headed belgian more to the roman side, having such vibes as signal collect guys, just brilliant
20:08 takenout1: signal connect guy is also shultz
20:09 takenout1: is dark color hair guy, but still pretty good
20:09 takenout1: he wrote the files as swizz zurich scientist
20:10 takenout1: those had markings from 2013-2014 is part of the world wide web initiative solution stuff
20:10 takenout1: but smallbitset is also quite striking , a genius stuff
20:11 takenout1: karolherbst started to strike as well in rust, appears to be good too, and really is i think too
20:11 takenout1: me i have not yet started , maybe don't want to show out either
20:12 takenout1: computer technology is amazing powers are from central europe
20:13 takenout1: i see those make a lot of sense
20:14 takenout1: karolherbst and zgillard those are over my skills for an example
20:14 takenout1: many from slavic countries too
20:14 takenout1: this guy already has found a way to express things in rust language so well
20:15 takenout1: all that i have and had he already has and lot more
21:25 KariMunne0XF: so what in place algorithm does is use some memory to read from and return things in registers of cpu, by using alus, so it only uses one memory of external type to base on
21:26 KariMunne0XF: the further you think the better it is, so now you come across a fiction where you create several instances at the same memory and daisy chain things
21:26 KariMunne0XF: and they document this in code
21:26 KariMunne0XF: up from that point you can have several sets in that memory and you can fuse the loops
21:27 KariMunne0XF: and unroll them, which is funny side
21:27 KariMunne0XF: this library looks good to me
21:31 KariMunne0XF: it's often to the web land cause people demand huge databases and are dependent on computers telling them every step they need to do, but programmers need to do a huge amount of work to store and feed so big amounts of info
21:32 KariMunne0XF: so they need the best methods there are to compress their data structures
21:43 KariMunne0XF: You should now understand the polarity of the arithmetic, it eliminates all first bigger ones in a set to leave the smallest into the next phases
21:44 KariMunne0XF: if it was to use indexes like i told, it could access the biggest value first too, but fragmenting that address space this way leaving the smallest values to the programmer is way smarter
21:46 KariMunne0XF: and it is much simpler, cause all of this is natural
22:05 kisak: *rolls dice* Is it human?
22:15 airlied: definitely
22:57 bl4ckb0ne: hi joss
23:00 stepasideplease: The way i understand is the scoping stuff, it's about self &mut, this is returning to itself, unless some other scope that does this very same thing (returning to self &mut) is also open, in that case it belongs to another instance or scope, cause it's automatic variable, those get closed when scope closes, but it looks like feature alloc is used by another instance that has constructor destructor type of thing, not sure how that gets
23:00 stepasideplease: deallocated