07:51 Reagan_rain: I take it as you are not interested, extortion you should not do near my territory and we can call it quit, so to speak underground who comes at me, so implement me some bitcoin code i am not interested in, i saw those so called bosses and gangsters who invaded into my personal life, these are nasty fools. Other deluded people dealt with such sexually (thought it was enough to come to me placing charges), and they have no money to pay for
07:51 Reagan_rain: the job either, they are just fools who did not pick their victim well.
08:08 Reagan_rain: over the long run you get the code from university X anyhow, it's Stanford latest for an example log minor w logu access time they formalize seems about right (maybe code was not offered yet, sure they do offer it at some point) , cause there is no logloglogu
08:38 Reagan_rain: And if you ever wonder why would that work, well hardware can process the added operands in sequence inside one machine word very fast, if you map the information based of added operands, it technically goes sequential where the order is the one you inputed, but as the engineers optimized the hardware to perform at this task it can do many in sequence which appears as very low latency and almost as parallel set of operations. You can access
08:38 Reagan_rain: any amount of entries throughout any banks, it's likely to just happen indeed this year where they expose hardware methods in software.
10:16 posselbox: I figure you suggested that it already has happened, things like qvm,gini,varisat,miden-vm,divine , none of them i tested, but each having some modules to suggest that things alike has been done. And what matters for the flow graph is exactly what i said, the thing that hw does it in serial or parallel matters not, what matters it that you sample the answers for alus in each adder sample of two operands meeting each other to formulate the
10:16 posselbox: result, that's possible for doingit parallel just cause through accessing several elements at once, so it is a bit like ffm, cause it takes many samples (each sample is indexed at intermediate result of an alu after doing index as a sum of operands at that pc and access from the hash as multiple accesses at the same time) , but ffm has different formula as the main function, some euler number inside of it.
21:54 DottorLeo: hi!
21:55 DottorLeo: is it possible to donate some hardware to the mesa project? Like a Sapphire 7950 3Gb? :)
22:01 emersion: maybe mupuf would be intersted in hw donations, but not sure
22:02 DottorLeo: for CI? :)
22:02 emersion: yea
22:02 DottorLeo: also for development if someone want it, i know it's GFX6 but still punching :D
22:03 emersion: oh, GFX6, that's quite old
22:03 DottorLeo: maybe RADV or radeonsi
22:19 DottorLeo: well, i don't usually connect here so much but if a developer is interested contact me at [REMOVE]kinetocore86[REMOVE]AT[REMOVE]gmail[REMOVE]DOT[REMOVE]com ;)
23:33 posselbox: it's been a long weekend off from other work and i suspect i have done a golden break already by now. I think to propagate/transmit forward a remainder from alu banks it's best to mix the routine i gave with value receiver banks, that are inserted after every lookup i.e intermediate result, it needs a little linear algebra but it probably propagates and contributes to the increasing index to look another result from. And for this there are
23:33 posselbox: data access routines generated by compiler, as alu permutes are static with their masks. this tandem has elimination middle pass. Because decreasing intermediates can not be inserted otherwise.
23:37 posselbox: It's parallel indeed, so only one lookup table is needed to make multiplies needed for asymmetric accesses, so this bank tells 1 or 2 or 3 times 512, is how much