06:58 sunnyday001: I also added the code to edaplayground how it looks in the vcd files, i removed now the hardcoded signals and epwave window, just download the files and open the vcd file with gtkwave or other vcd viewer
06:59 sunnyday001: however slight knowledge of the HDL is needed to read it properly, which is why i added the verilator trace also, cause edaplayground does not have verilator.
07:00 sunnyday001: https://www.edaplayground.com/x/5Bvr#_=_
08:15 sunnyday001: no way with in HIGH level code to do it, needs machine code, cause LSU recursion is not allowed by default in conventional drivers, it breaks it by separating the registers spaces. I my own do not know nouveaus internal assembler well, and nvidia does not offer sass to machine code assembler.
08:17 sunnyday001: i know that for some stuff, except semantic mappings and all the glsl support which is not included, there is also asfermi, maxas and kepler or whatever is sm4.0 nomenclature for nvidia multi2sim or gpgpu-sim offers the sass to machine code assembler there
08:18 sunnyday001: and there is something called SASSi which allows direct machine code editing in hexadecimal form, not easy for beginners, and later trigger that into the code
08:21 sunnyday001: it'd be lot wasier, if i offer the guidelines and you implement the bits in nouveaus code, and also try to get rid of the bugs continously, rather than me needing to deal with the internal codegen, which i just do not have time for
08:30 sunnyday001: jvesely: i glanced over your paper from 2018, i did not entirely read it, but it almost like seemed you are trying to run an os with using graphic acceleration
08:31 sunnyday001: in concept, cause those cards are too powerful mostly and also for mobile platforms , this could be quite ok idea
08:33 sunnyday001: http://www.cs.yale.edu/homes/abhishek/jvesely-isca18.pdf
08:42 sunnyday001: in theory when you call tomasulo algorithm from CPU side, it also has some ROB/RATs to accelerate CPU code, but resource wise CPU register files are normally quite small, and hence those queues are also tiny, cause cpu performs also io controlling and stuff like that which uses lots of resources
08:45 sunnyday001: Linus has thought the way i understand that accelerator should be integrated on CPUs core , which is also possible, depends on the cache sizes since regfile is small in the spec, correct code on those would perform also quite well though
08:52 sunnyday001: for instance llvmpipe would become also a lot faster if you again would call the procedures and loops correctly on x86 cpus, context switch though would be slow inherently on cpus but other code is so fast if called right, that it would not matter
08:54 sunnyday001: tomasulo algorithm executes alus also in picoseconds , i have a paper again. offered it long time ago though zwiss or and norwgegian ones
09:00 sunnyday001: this requires a large cache , not large but reasonable at least, cause SIMD on cpus is otherwise difficult
09:05 sunnyday001: https://arxiv.org/pdf/1401.1084.pdf
09:07 sunnyday001: https://arxiv.org/abs/1401.1084
09:14 sunnyday001: https://github.com/unison-code/unison also very interesting, and openmp has gpu backends too
09:26 sunnyday001: jvesely: https://divine.fi.muni.cz/2017/divm/ in several asbects very quite interesting stuff
14:23 sunnyday001: but.. i do not want to insult people who work on the games and this field, i worked abit my own too, and threw some theories at you, however mainly i every day try to find some nonfurious moment, not egoist entirely, but i think mostly why my personal life has gone the way it has, i do not want any help either, I am just doing analysing there daily basis, cause something has gone entirely wrong
14:24 sunnyday001: it is done by me, just for the sake that when i was younger i was capable of feeling well, and i want to do that today too when i mostly do not feel well
14:29 sunnyday001: which is to ssay again, i never cared that much about the shape of the drivers above probably, and i consider my own problems higher ranked than the code, what you have spotted also probably, which is to say i actually do not care what you do with my ideas even if you manage to understand
14:31 sunnyday001: for me the humanity is higher ranked then the quality of the code
14:43 sunnyday001: as earlier life all was cracked up and full of shit, it boiled down to me having smile on the face, this shit does not affect me, i want badly to feel it again, to say that life is just wonderful
14:45 sunnyday001: i think now days it requires more mental effort, on some organs failing just a little, but can be still done
15:10 sunnyday001: I can break it down to vague lines over what is considered as violation or just human psychology which is sometimes too raw. Such a painful thing at times, but humans in some testordome maybe for others, have grown so smart, that they have their own science how to deal among others, it is strategy
15:10 sunnyday001: it feels like, if some is too raw to be compassionate about others feelings, than it does not work out, and others claim that one to be too raw
15:10 sunnyday001: to recognize the psychology as science
15:18 sunnyday001: i feel a strategy as advanced player behind a table, i recognize that science, but need to yet mature abit in the bigger games as everyday explanations why someone behaved the way he did etc. sometimes it remains unkown normally there should explanation of some type, which i am bothered enough with
18:52 diogenes_: Hello guys, doest it happen that nouveau after a period of time, drops support for older cards?
18:55 karolherbst: diogenes_: no
18:59 diogenes_: karolherbst, wow, so even if nvidia no longer supports some cards in the newest drivers, nouveau will keep on supporting them? also nvidia is known to slow down older cards with driver updates, i'm sure nouveau does't follow this policy?
19:01 karolherbst: no idea if they really slow it down on purpose, but I guess they just don't care and the compiler might make a decision which is bad for older gens
19:01 HdkR:toggles MAKE_FERMI_SLOW option
19:01 karolherbst: which is something which could happen for nouveau as well though, because we aren't able to benchmark everywhere
19:02 diogenes_: karolherbst, thank you!
19:02 karolherbst: HdkR: ohh, I mean, there are a few things people were able to bnechmark and notice slowdowns with newer drivers, just I am not convinced it's on purpose. More like "it makes newer cards faster", but nobody bothered checking if it makes older cards slower ;)
19:04 HdkR: Yea. I don't need to put on a tin foil hat. That is exactly what happens
19:04 orbea: marketing doesn't ever suggest misfeatures?
19:05 HdkR: ?
19:05 orbea: idk about nvidia, but there are documented cases of people designing misfeatures for marketing reasons
19:06 HdkR: :shruggie:
19:06 diogenes_: it's called planned obsolescence.
19:06 orbea: simply not caring is very plausible though
19:06 orbea: DRM is another example :)
20:28 SciresM: Hi -- is there anybody around who might be able to help me? I'm looking for information on how to poke at the falcon processor for a desktop GPU.
21:18 RSpliet: ScoresM: "the falcon processor"? There's a dozen on every graphics card, which one are you interested in?
21:18 RSpliet: *SciresM^
22:41 mardination: the way i have understood things, i had fans among two groups, females and sports enthusiasts, now i lost both of the qualities, since those were taken from me with rough work, so i am neither attractive nor active in sports, now i had also third quality which is going to end up as a loss too, i am a sane programming visioner, since you are cockblockers and violators, i am thinking that i lost this one too allready
23:11 mardination: maybe you enlighten me about something else for a while by not stalking me by not violating females and by not saying me who does not like me, cause this cockblocking subject is fairly old to me, man kind have terrorised me with their group attacks ever since i was born, i am aware person , and i have seen this is as unwinnable battle