02:02 miinus2: penguin kind of punisher on the channel, does not know how to count, but punishes.
09:31 miinus2: yeah infact there appears to be a way to unblock the texunit, by writing into the address reg from another unit or with indirection , but i am yet unsure what is going to happen then.
09:42 miinus2: to me it seems it will try to run current wfid on the arbiter then out of the blue, unless its blocked by scbd it will immediately run.
09:50 miinus2: imo it does not unblock itself in miaow, unless tmu can deal with blocked units, or there is some hw clamping or discard logic missing
09:51 miinus2: what it does it gets stuck in address calculator state in the fsm
09:51 miinus2: and it is brought out when only a new value a proper value is posted to address registers of the lsu load
09:55 miinus2: if one does not want that LSU load to schedule something it writes to, it needs to also redirect the destination with indirection
09:55 miinus2: but r300 is not allowing this without modifications etc.
10:10 miinus2: so when the alu operand wants to schedule cause of freeing the blocked lsu, it needs source operands to be redirected against something that blocks
10:16 miinus2: https://community.amd.com/thread/159507
10:16 miinus2: look at fibonacci to get further ideas
10:25 miinus2: so to sum up, once the lsu stalls at addr_calc stage in fsm, lsu to all it's operands are unblocked in scbd, alu operands to destination are blocked but source operands unblocked
10:32 miinus2: i will leave you alone now...
13:03 miinus2: whole thing is a mess in mesa, elaboration needed why dest operands are not indirectable! The operand that is dest in some instruction happens to be source in another , LOL!
13:30 miinus2: I can not fully precisely comment this stuff, cause i do not understand :( i mean what are the symptomps of redirecting dest on intel hw that has incompatible format or radeon hw which is incapable to do that etc.
13:51 miinus2: Yeah i read all documentation of course, but it there is still something unknown.
19:11 leidurleo: I am expecting that it hits you once, how large of gulf there is in between you an me, when you start picking up all the debates i purely beat you in and start to think how wrong you have been, then you also understand why this big brain was assaulted and attacked during youth ages allready!
19:41 karolherbst: or not really
19:41 karolherbst: he stopps after a few replies
19:52 juri_: please, more of that. i don't care if you ban me a few times, by accident. :)
19:56 Lyude: skeggsb: hm, I think we might be able to make nouveau use bridges instead of encoders for msto's
20:29 kreyren: how do i parse nvclock (or other variables that makes nouveau work better then nvidia on everything that is not vulkan) without rebooting to parse kernel cmdline?
20:32 kreyren: .. and that is not maxwell and higher >.>
20:35 kreyren: expecting nouveau.config=NvClkMode=15vBoost=1 to translate in runtime instructions
20:47 kreyren: can i change licence of nouveau fork on gpl-2 ? i got impulsively triggered by nvidia issue on my system and openned git and text editor
21:59 karolherbst: juri_: it's a lost cause though :/ Don't want to get into details in public, but let me phrase it like that: he is causing issues for years already :(
21:59 karolherbst: and if we block IP ranges or make the channel for registered accounts only, he grabs a different VPN and creates a new account
22:00 karolherbst:cleaning up the ban list *sigh*
22:14 karolherbst: *sigh*
22:15 karolherbst: there is more, but now I got tired