log☇︎
20700+ entries in 0.005s
asciilifeform: the ~other~, tho less simple, speed boost, is if one were to obtain a machine with wider multiplier !
asciilifeform: ( and even presuming 100% correctness, bounds checks still give added 'cosmic ray resistance'. i ~like~ having'em )
asciilifeform: the other 'secret' speed boost is if one were to turn off the bounds checks; this gives ~2x speedup across the board. but i dun expect to do this for my personal uses for years , if ever -- it requires 100% certainty of correctness of the program under all possible input
asciilifeform: the only reason asmism even potentially invites itself, is that idjit compiler gives no primitive for add/sub-with-carry or full-word mul ☟︎
asciilifeform: bvt: correct
asciilifeform: ( and when ADD, SUB, we do it twice, where only needs once )
asciilifeform: http://www.loper-os.org/pub/ffa/hypertext/ch15/w_mul__adb.htm#95_14 << specifically here, currently we do 4 MUL instrs for where really needs only 1
asciilifeform: the remaining virgin land for speed revvup, is asmism for base cases. ( and ~possibly~ in combination with unrolled comba )
asciilifeform: as i noted previously -- i do not expect to find any moar ~asymptotic~ speedups for ffa algos , such that are relevant to the sizes of numbers typically used in public key crypto ☟︎
asciilifeform: bvt: comba itself ended up on the list of things that made it in by very small margin ( it wins perhaps 10%, on most pc iron , vs straight word*word mul as base case, try it yourself by turning the threshhold knob )
asciilifeform: it allows reader to see what is bought by adding the moving parts.
asciilifeform: sorta why, when i first started preparing the thing for publication, wound it back to the simplest known ('egyptian') variant, and walked from there.
asciilifeform: bvt: there's a long list of things that asciilifeform considered and (for time being) rejected from ffa, on acct of costing substantial complexity for very small saving of cpu cost. e.g. unrolled comba.
asciilifeform: there was another fella (since wandered off) who posted last yr an attempt at de-recursivizing karatsuba ( it's definitely in the logs, can't find it just yet tho )
asciilifeform: http://btcbase.org/log/2019-01-20#1888467 << gotta nitpick here: it aint allocations (which in ffa planet are always done by stack frame, in O(1) ) that leads to slow, but cache eviction ( as well as linear overhead from doing moar ops in general ) ☝︎
asciilifeform: 'i built atomic dirigible but fughet why!111'
asciilifeform: i assumed bvt knew of a use, given that he dug out the ru materials on subj
asciilifeform: bvt: i don't, which is why never bothered with fft
asciilifeform: ( not particularly useful for ffa, but potentially elsewhere.. )
asciilifeform: i saw. and imho would be interesting to have a constant-spacetime, no-floats fft
asciilifeform: same effect applies to karatsuba, and errywhere else.
asciilifeform: despite both being in same order
asciilifeform: 'locality' effect is actually why e.g. comba wins over ordinary 'kindergarten' o(n^2) mult
asciilifeform: of course 2x slower.
asciilifeform: bvt: you have more than twice the # of cache evictions
asciilifeform: mircea_popescu: i'm not surprised, considering that bounds check overhead (in ada with all safeties switched on) magnifies the 'losing' of a losing (for particular width) algo
asciilifeform: bvt: ty for reading, signing, and publishing experiment -- i will include your seals in ch16 article
asciilifeform: bvt, mircea_popescu : http://bvt-trace.net/2019/01/experiment-n-way-split-karatsuba/comment-page-1/#comment-6
asciilifeform: BingoBoingo: it's a crying shame they didn't send ~all~ the 'java gurlz' to syria, to use for sandbags.
asciilifeform: when he ran outta juice for that, i quit reading, there was literally 0 else worth even passing look
asciilifeform: kinda like in late '90s asciilifeform read sci.physics just to watch al schwartz trash idjits
asciilifeform: imho that was the 1st and last readable item there
asciilifeform admits : was never able to stomach reading tardstalk with any regularity, cuz it ALWAYS read like this, from birth
asciilifeform: if not today then tomorrow, verily
asciilifeform: 'pleez sir buy this very fine islamic death ray , only 0.1 btc'
asciilifeform: ^ see also
asciilifeform: !#s death ray
asciilifeform: dunno, cuz collects 100% garbage ? ( tho admittedly this never stopped'em before.. )
asciilifeform: lol i had nfi tardstalk were still around
asciilifeform: stands to reason there would be 'fractal posers', on 9000 levels of magnification.
asciilifeform: whoknows. but given the existence of the masochist folx, not much of a leap to picture the existence also of 'double agent' masochist, who takes delight in seekritly jimmying the cock seal, etc
asciilifeform: little plastic padlock, comes in crates of 9000 , with consecutive stamped #s. ( not only quite simple to forge, but jimmies with ordinary hairpin, like 19th c. handcuffs , and snaps closed again... )
asciilifeform: ^ see also re subj.
asciilifeform: !#s szabo seals
asciilifeform believes
asciilifeform: in ye olde sovok they used pb seals on iron wire, stamped with signet; but in usa pb is envirowhiner-verboten, ergo plastic
asciilifeform: holding the cock gasket in place
asciilifeform: the item on left hand of photo.
asciilifeform: lol that's a , didjaknow, standard usa mains current meter seal.
asciilifeform: wtf's that mean in context of (what is evidently a) trans-whatever ?
asciilifeform: 'lock my clit' ?!
asciilifeform: quite likely they have an eventual capitulation to llvm ( and 'unification of the churches'(tm) ) planned as well, but i have nfi what's the holdup there
asciilifeform: the gcc5+ gnomes, occupy selves with cranking out 'mandatory' kludges for intelism; removing backend support for vintage, marginally-sane archs (alpha, hitachi, etc); gluing-with-broken-glass various incompatibilities to prevent coad developed under 5+ from building under 4.x; inserting 'optimizations' that snake around naive cprogrammer attempts at bounds-constraint; and so forth.
asciilifeform: as for gcc, the mushroom let it go in mid-2000s, moar or less.
asciilifeform: nao it's a google et al. kernel.
asciilifeform: mircea_popescu: his last , near as i can tell, act as a human, was to try & keep this liquishit out of his signed kernel tree. for which he was 'retired'.
asciilifeform: Official gcc is , didja know, up to version ~8~ nao. and packed fulla 'beautiful surprises', this is by no means the only 1.
asciilifeform: on ~that~ end, the human mushroom looong ago opened the gates.
asciilifeform: aa
asciilifeform: mircea_popescu: it's a google-submitted patch, and it got eaten by the new, 'improved' torvalds, near as i can tell
asciilifeform: whole thrust is to 'nobus-maintain' the scenario where you buy 5 cpu-minutes on lullazon's cluster and get root. rather than to keep neighbouring lullazon-hruscheba tenants from fucking one another
asciilifeform: mircea_popescu: observe that all of the various Official 'speculative leak' kludges, ~even supposing they worked 100% of time~, only apply to ~kernel~ -- they do 0 to prevent 'ring3' (luserland) processes from leaking
asciilifeform: the retardation of intel was quite clear even when 386 was being pushed.
asciilifeform: mircea_popescu: it's been high-time for many many yrs.
asciilifeform: it appears 'fast' cuz fulla various 'and now we can cut THIS corner' crapola.
asciilifeform: naggum's http://www.loper-os.org/?p=165#selection-85.0-85.510 thing dun merely apply to c. it applies just same to the 'progress' of intel shitirons.
asciilifeform: errything from tlb cache to branch diddling to 'hyperthread' to 'register-renaming' to... etc
asciilifeform: (i.e. kludges introduced so that winblowz will run in slightly less geological time, 1990s-current)
asciilifeform: rright, but to get any actual effect would have to also offer offswitch for 9000 other idjit 'optimizations'
asciilifeform: it dun have an off switch tho
asciilifeform: yea i get
asciilifeform: it 'works' in the same way as other 'nobus-maintenance' kludges (e.g. koch's 2016 patch) 'work' -- raise bar so that nobus
asciilifeform: mircea_popescu: what would it mean for the kludge to 'work' ? it aint as if x86 actually offers process isolation.
asciilifeform takes the occasion to observe that ffa is 'speculation'-proof, not only in theory but as empirically tested on 7 ( & counting ) intel boxen of various vintages
asciilifeform: mircea_popescu: i dug, and , unsurprisingly, a google production
asciilifeform: ( and before anyone loses appetite, this experiment was on a test box in torture room, not in pizarro )
asciilifeform: briefly revisiting the http://btcbase.org/log/2019-01-18#1888182 find : 'CONFIG_RETPOLINE=n' yields a working kernel (with gcc 4.x), for nao. ☝︎☟︎
asciilifeform: in other noose, all temps on uy1 in-range.
asciilifeform: mircea_popescu: sounds potentially interesting. ( if you publish outputs, plox to not forget to say what pressing that trb was made from )
asciilifeform: BingoBoingo: i was doing the annoying but occasionally inescapable chore of sweeping out a spamola-encrusted mailbox, and ~100% of the spams were for something nominally smokable/pillable/etc
asciilifeform: ( and near as i can tell, ~100% of usg 'dope policy' is simply own competition for dope konsoomers. but i'm not particularly qualified to opine, nao if only gabriel_laddel were around... )
asciilifeform: moar narrow-minded orcs back on asciilifeform's home planet, drank paint thinner, laquer, etc when couldnt afford liquor, for same
asciilifeform: BingoBoingo: i find it interesting how usaschwitz inmates will do literally ~any~ dope, even the expensive obummercare-'prescription' crapola, just to take a few hour's 'vacation' from 'how world works'(tm)(r)
asciilifeform: BingoBoingo: http://btcbase.org/log/2014-08-13#794524 ☝︎
asciilifeform: !#s jekem
asciilifeform: wtf is kratom
asciilifeform: con-sangre-sudor-y-lagrimas << lol, did he also add that 'it will be long, and it will be hard, and there shall be no withdrawal!'(tm)(r) ?
asciilifeform: ty BingoBoingo . i'd rather not waste precious cargo mass on items we already have, etc
asciilifeform: BingoBoingo: btw, plox to re-inventory the bilge at yer earliest convenience
asciilifeform: BingoBoingo: LSIiBBU07 .
asciilifeform: BingoBoingo: 1st time since instrumenting, but happens regularly on my local boxen, generally not fatal ( controller will cut it off before it catches fire. ) contributes to overall thermal climate tho
asciilifeform: BingoBoingo: i've never seen'em sold anywhere other than china.
asciilifeform: ( it hovers in high 40s when 0 traffic on box, and 50s when there is )
asciilifeform: cpu at 54.
asciilifeform: nope
asciilifeform: it powers the write cache on the card. but with fast ssd, that cache gets flushed pretty quickly. ( and battery only comes into play if mains fail )
asciilifeform: and incidentally the cards run fine without'em, it's a 'parachute' component
asciilifeform: BingoBoingo: i'ma include new ones in the cargo, it's a 20 $ part
asciilifeform: >60
asciilifeform: BingoBoingo: meanwhile, on uy1, http://p.bvulpes.com/pastes/7QcPb/?raw=true