03:04 Shallow193: In general a remainder bank is something that has a dependency on the flow control or io or operand load or interrupt, so such would push into certain range. It looks as if nothing more needs settling now, as i have a golden break on how to do those things. So there is nothing to discuss , this year i author the code , your prankster Ermine managed to handle one subject at #nouveau which is relevant for event loop of binaries too, legal
03:04 Shallow193: reverse engineering issue, cpu has that , however accelerator drivers not, so one is not supposed to recompile proprietary programs that has eula restricting it.
11:12 Shallow193: I was hurt many times very long time ago, they have continuously searched for more success in similar way, so i can not possibly accet such people or solutions. Treatment to such scammers has to brutal as the way it has been, but i have not yet participated in treating them, they have gotten treated by others, not sure if my fans, due the abuse scams they arranged.
11:12 Shallow193: *accept
11:18 Shallow193: *has to be
11:21 Shallow193: I know this , but that is somewhat orthogonal to compute powering and routines that i developed, so i have tried so even though as my sports career took early entire drawback and crack up i touch this subject not too much
11:22 Shallow193: Those who push code and hw heroics are similar professionals like athletes but active more in the educational and mental domain, where i feel i am not that much worse either.
11:24 Shallow193: academics so to speak, lots of interesting things have been delivered by them too.
11:35 Shallow193: If you were to see me in physical space, those half baked scammers went over the world to scam with their nonsense terror content, without computing i would not be able to keep my temper there at all, they were such a trash in the era i was active.
12:28 Shallow193: It's the latest stanford research, but all know that i already have it, since i demonstrated those routines here, so it's good, it's banking theory, the accesses are done based of what banks the dep lands to, i.e from which bank the bound that got combined to was removed.
12:29 Shallow193: the code is not so hard, but alone it is a bit, but i am a fighter and hope to get it done this year.
12:33 Shallow193: The order itself is the order how things were added, so if we need to make an io write, certain bank is filledd in, and if that io write modified the flow control a certain bank is pushed to, and that is how things work in the end. So one has operand loading, io and branch banks.
12:59 Shallow193: Yes, i am already imagining and working on the compiler routines indexes deps offsets and compressed answer sets and banks to generate, but i am not ready before couple of months more spent at this all
12:59 Shallow193: runtime routines are easier, but would need also more testing to commit, compiler routine testing is easier.
13:34 Shallow193: Generally it is indexes that determine where in the banks the data goes concept of shared indexes, banks share indexes, but one subtracts the index and one adds, so when the data comes it can only be eliminated from one bank, which it effectively belongs to
13:47 Shallow193: they all push to the remainder, so as an example for 512-500-1 is 11 524-11 is 13..say that is branch bank 499, if that value is needed to land in branch you add +1 to io bank the elimination buffer has no indexes, hence it gets eliminated from the branching bank cause io added +1
17:35 Shallow193: words is not the most expressive way to describe code.
18:56 Shallow193: so how can branches be handled in parallel? conditional branches in pc relative sequence. So we have these answer sets which are inside the branch i.e alus, we would want to discard the branch that is not suppose to execute. So branch condition is if x==5 jmp to one/next else jmp to other PC. So one way of doing it is always raise the bound of the condition, so if answer is 5 it will push say 1232 from 2048 if it's below it pushes only in
18:56 Shallow193: hundreds
18:57 Shallow193: now that should work okay, but branch targets are asymmetric
19:17 Shallow193: but how we add such conditional without external pass that would end up being sequential in nature, it still has to determine whether it was two hundreds or 1k and two hundreds it pushed, so that is the distance theory
19:20 Shallow193: once the branch target is added new bound it would end up being times one or times two the value
19:29 Shallow193: so that says we must annotate the answer 5 to bigger bound, while 0....4 as lower bound
19:34 Shallow193: say 4 is the outcome of comparison
20:02 Shallow193: whatever , i give up on wording, well technically when you raise da stuff, you take some number away from 2048 which is 4 so 2044 to 2048 for branch 1 and 2048+1... for branch 2
23:42 biker_rat: https://archive.mesa3d.org/ is down ftp is up