09:16 <rellla> anarsoul: i thought about sth like that for all the streams. standalone or integrated in the syscall tracker:

09:16 <rellla> https://pastebin.com/raw/qavykWZ3

09:17 <rellla> https://pastebin.com/raw/BCfed3C6

11:23 <MoeIcenowy> oh... is Mesa going to make NIR first-class rather than TGSI?

12:43 <melonmart> if you were not so twisted guys from your heads, implementation would had been landed longs time ago even by me, cause obviously as you are not the one who you claim to be , just violators instead of any talent in place in your actions.

12:51 <melonmart> So i am instead somewhat glad and somewhat happy lately to have been ensuring my soveranity and untouchless status for the whole life, and it is of course because all the people who became to violate me all guns blazing and all the people in a debates against me happened to make no sense at all.

13:14 <melonmart> sad because that i was not suppose to waste all my nerves and times to publish methods on IRC , i was not suppose to me in this situation, so the time loss, and in addition i care about users over the world, they still have no sane and troubless driver code.

13:16 <melonmart> happy cause in this situation where i allready was tried to be harmed many times i.e cause inevitably i was in this position after all, they were embarrassed cause of not making sense, in this position i needed warranties for my health protection.

13:27 <melonmart> but I say to others, you can not ask more money from the chip that is tweaked for performance, every chip can be, cause the lithography machine throughputs are known to be insane 50k chips it produces in 10hours.

13:28 <melonmart> you do not have to pay extra at least until low-end devices are offered by the vendor or manufacturer directly.

13:33 <melonmart> in the end it is just, the ones who know how to optimize the chip to perform real tasks, should be hired by many of growing startups or allready proven companies, and just make some sane programming, hence i just advised to listen to me there, instead of getting fired and be a complication to me during my existence, if you adjust to new paradigm fast, you get a real job instead.

13:35 <melonmart> you just have to deal with depression and be good at matrices and general computing theory. I have presonally known mathematicians who have always been upto the task, i assume if i present them the rules how to fill the matrices with correct dependencies and rules how it issues commands on chip, i'd get every month many solutions on paper which would have to be tested.

13:38 <melonmart> yeah , i have those mathematicians among my friends, who showed high end performance for athering a cum laude from university, i know many but couple of them are gurus , who solve those matrices pretty well.

14:01 <melonmart> SO i am up to the task to program per GPU the code with 1month turnaround, after i fix my laptop, I am ok with getting hired to do that as a job, it is just i do not present my curriculum vitae and job application, cause i do not fake higher education which i do not have

14:05 <melonmart> in the linux OS the most work requiring hunks are inside the page tables, which are slow and pointless for data cache accesses, scheduling itself is somewhat easier, page tables do not overly make sense even they do not make sense.

14:19 <melonmart> you have twos complent addition logic and two variables refelecting the data which was loaded, you should not traverse memory from different in cache or memory pointers

14:19 <melonmart> they say whether the data is in cache or memory

14:24 <lepurnius> paging is so fucked up feature, that you get on all OSes that use this feature around 20x added latency to data cache accesses

14:26 <lepurnius> while ddr memory should be somewhere around 1-3GB/s with paging what you get is 29MB/s lol, and it is nowhere near secure either.

14:30 <lepurnius> I head off now, so when all of that is fixed , you get all the performance in a world even on a7 cortex processor without a gpu and even cache for persistent storage can go through main memory instead, however they have second level interconnect based cache addon-chips

15:03 <MoeIcenowy> rellla: did you dump command stream for ideas?

15:03 <anarsoul> rellla: can we set +r on the channel so only registered users can talk?

15:03 <rellla> oh... no :p

15:04 <rellla> anarsoul: sure. isn't it set already?

15:04 <asriel> I wonder if it's the same guy last time

15:05 <asriel> Also putting it to "registered users only" mode should work

15:06 <anarsoul> asriel: that's +r

15:08 <anarsoul> rellla: https://pastebin.com/raw/BCfed3C6 won't work since vs cmd stream is dynamic

15:08 <anarsoul> also blob sometimes uses commands that we don't understand

15:08 <rellla> anarsoul: dynamic in size?

15:10 <anarsoul> and in commands :)

15:11 <anarsoul> see reload in lima for example, it's doesn't look like a regular draw

15:11 <anarsoul> (and it doesn't use GP at all if I remember correctly)

15:20 <rellla> anarsoul: i find that for plbu stream, but where is an example code for vs?

15:21 <rellla> afaics i just did the reverse of https://gitlab.freedesktop.org/mesa/mesa/blob/master/src/gallium/drivers/lima/lima_draw.c#L725

15:30 <anarsoul> rellla: yeah, so it'll work for lima, but probably not for the blob

15:31 <rellla> anarsoul: so as long as we don't know, how the blob creates the stream, there is no possibility to decifer it!?

15:31 <anarsoul> rellla: you have to do that dynamically, without using structs

15:33 <rellla> anarsoul: so check every uint32 pair against known schemes here https://gitlab.freedesktop.org/mesa/mesa/blob/master/src/gallium/drivers/lima/lima_draw.c#L186?

15:33 <anarsoul> yep

15:34 <rellla> and the rest stays as unknown then...

15:34 <rellla> ok, that would make code for plbu easier as well :)

15:35 <rellla> as there are some if's in cmd stream creation. dealing with declaring structs and switch is bad...

15:36 <rellla> i'll try to code that a bit smarter

15:37 <anarsoul> rellla: can you dump ideas for me first? :)

15:37 <rellla> anarsoul: the first results with hardcoded stream are promising though :)

15:38 <rellla> anarsoul: i think i have to build glmark first. and don't know if i have enough time left today. i forgot that tbh :/

15:39 <anarsoul> OK, no worries

15:39 <rellla> iirc cross building glmark2 was a mess ...

15:40 <anarsoul> I build everything natively

15:44 <rellla> doing that now. started building at least, but i have to go now.

15:53 <rellla> from a quick test: Error: Loading EGL entry points failed

15:53 <rellla> have to check that later.

15:56 <rellla> hm, do i need X for glmark?

15:57 <rellla> drm-glesv2 or x11-glesv2?

16:02 <anarsoul> drm-glesv2

19:20 <anarsoul> oh, I think I know what's wrong with my partial writeout changes

21:07 <mastermart> if all were Vasily Poopachyuks aka. analsounds , then vendors could easily scam about what hw can't do, i can author on another hand pretty easily the performance mode, i stayed on sm2.0 spec too, so it is universal or generic.

21:18 <mastermart> wayland was mainly 2009 or so started after my so called rant, where i claimed that opengl es is capable of rendering stuff to terminal without X, during that time i was largely (by set of trolls) laughed at and also violated with protest,pickets,cockblockings and general annoyance at various locations where i physically was not to mention the IRC monkey republic.