05:14 <mastermart> making new slides is just the matter of opening the pdf/ppt and editing abit, it is highly easy for me to see how the modified flow works, i allready but the indices in place, so no scandals are needed!

05:14 <mastermart> but/put

05:19 <mastermart> Yes , SSA can be interpreted, SSA form from twos complement variables, brings benefits, that one can issue multiple instructions pending on writeback, and reisse the ones who after the non-graduated writebacks at that point.

05:22 <mastermart> Even if you have no SSA in the code, in some mali IR, it is reasonably easy to make an abstract machine based of DIVINE code to lift maliIR to LLVM IR back and forth, bunch instrinsics needs to be added only.

05:24 <mastermart> DIVINE abstract machine is nice code, cause it has SSA in LLVM IR layer, not the machininstr layer. It can also come out of it

05:34 <mastermart> algorithms-by-blocks is one of the easiest part of the implementation to generate the needed dependencies, on 8issue or more pipeline

05:38 <mastermart> The harder part is to checksum the stream.

05:54 <mastermart> anyhow, i will go now, I have the code ready when i will program a bit there, and it is almost piece of cake.

05:56 <mastermart> So as i said, if you skip from 0 this is fast to anywhere on the line, two computed in sequence is also fast when first 5were skipped, and 6th and 7 was selected is faster than 6th and 9nth selected

05:56 <mastermart> so if we gonna do this my way no extra power nor latency by the arbiter is added

05:57 <mastermart> so all 500instuctions on nvidia 8SM card can be scheduled when they are basic MULs in 0.33 ns

05:58 <mastermart> cause fermi the reference benchmark card in blum-blum shub cpu perf countered code is 40nm process/tech/transistor size based card

05:59 <mastermart> if you are not doing it my way, you lose FLOPS

06:00 <mastermart> then only for instance 50or100 out of 500 can be computed in the timeframe

06:07 <mastermart> I simulated the arbiter.v code which is responsible of replaying the holes, and the rules are from id1 skipped to two consequent issues is fast, 8 wrapped around back to anywhere is also fast

06:08 <mastermart> 8 non wrapped around to say 10 or 30 is adds delays

06:11 <mastermart> power loss and delays even if you do not seem to agree nor spot them doing that type of mistake, are actually quite serious there and should not be ignored like this.

06:12 <mastermart> doing things right you get almost full expected single precision band.

06:18 <mastermart> barrelled arbiter should behave the same in case of queues, at least in sense of power, and latency, or even worse, if you never intend to run the 32barrels slots, you gotta skip them in order

06:23 <mastermart> I am a little silent genius my own, however if things have been done before and fleshed out just fine, even under huge pressure stuff seems easy, however if they would not been done before and designers were to make little mistakes and stuff needs to be done from scratch i would not be so confident under this kind of pressure.

06:27 <mastermart> as longs as i have been violated, this hasn't made much difference when i have had access to internet, cause planet earth has so manycitizens who are not being violated, and do their tasks so i can cling onto their articles even if i am violated by closer circuise of people.

06:28 <mastermart> it takes me an hour to prove all my theories that i present and reference them so you can not scam with me.

06:36 <mastermart> And of course my little invention based of the hardware inherent methods is the selector, when all slots issue this is going to be fast as well, but graduates should be in program order for the rest of the code in duplets it is the most sane as i presented.

06:37 <mastermart> bye, and i come with next time either with pseudo or real code that does it, or slides or something more specific, since prolly you yet do not understand.

14:49 <yann> is it possible to use lima with Xorg ? I only find references to plain drm and wayland...

14:53 <enunes> yann: Xorg itself works, specific desktop environments on top of Xorg may vary

14:53 <yann> using the generic modesetting driver ?

14:53 <enunes> yes

14:55 <enunes> there are some performance issues and I think it affects most desktop environments though

14:55 <enunes> but as of "Xorg" like if you want to launch a raw Xorg and run applications on it, that works

19:45 <yann> I guess I'm not far from a working setup but something's still wrong. I can't see why Xorg tries to load fbdev (which is not built) - it gives no reason: https://pastebin.com/rBCK9E5p (this is an Orange Pi PC (H3) for which I'm trying to get a yocto build to work with lima

20:04 <anarsoul> lima should be card1 and your display card should be card0

20:04 <anarsoul> make sure you load modules in correct order

21:04 <anarsoul> enunes: any luck with debugging BO leak?

21:11 <yann> hm, I only have card0, which other module do I miss ?

21:13 <anarsoul> whatever drives display on your board? I guess sun4i-drm? and probably related hdmi modules

21:27 <enunes> anarsoul: didnt have much time for it today, probably tomorrow

21:28 <anarsoul> I see

21:31 <anarsoul> I'll try to reproduce it tonight

21:46 <yann> enunes: that's better with sun4i-drm, thx. Nowit even enumeratees video modes, but still fails with "AddScreen/ScreenInit failed for driver 0"

22:08 <anarsoul> yann: make sure that sun4i-drm is card0 and lima is card1

22:22 <yann> looks like they are already so: I see /sys/devices/platform/display-engine/drm/card0/ and /sys/devices/platform/soc/1c40000.gpu/drm/card1/

22:26 <anarsoul> well, then check the log to see why it fails

22:27 <yann> if only it gava any info :) - it is much very like eg. the last log in https://forum.armbian.com/topic/6071-next-not-installing-on-orange-pi-plus-2e/ - seems to start up but suddenly fails (and I don't like the last post in that thread...)

22:32 <yann> my full log is https://pastebin.com/yDv8KA1J - no EE tag (or meaningful WW) before that final "AddScreen/ScreenInit failed for driver 0"

22:33 <yann> "Xorg -configure" OTOH says "No devices to configure. Configuration failed."

22:45 <anarsoul> maybe you don't have enough CMA to allocate FB for 3840x2160

22:45 <anarsoul> and btw 5.2 is too old, you need at least 5.3 to use lima

22:45 <anarsoul> also make sure you're using mesa master

22:45 <anarsoul> *git master

22:48 <anarsoul> but it's unlikely to be related to your current issues with X

22:48 <anarsoul> get it working without lima first?