#trilema | 2018-12-28

meanwhile: $ wc -l libffa/* >> 4559 total ; $ wc -l ffacalc/* >> 1226 total .

10:05

http://www.loper-os.org/?p=2892#selection-4511.0-4595.6 << rsa timings.

~ 24 minutes ~

10:29

not even bad.

10:42

there's a surprise waiting, gimme a coupla min

10:46

in other random not-news, 37.187.150.139 is "ddosing" trilema by loading random pages ; but with a few select "lurkmore.to" pages as referrer.

10:47

i hope i didn't say something to offend teh russki patriot, did i ? http://btcbase.org/log/2018-12-20#1882391 is just a neutral observation, not in any sense ~my fault~ they're morons.

10:47

Logged on 2018-12-20 15:46 mircea_popescu: asciilifeform on meditation, that piece makes it plainly evident redditardation is ~generational~, not ideological. there sit the russki kids, who'd be in a fine position to understand the problem. god knows they have much better access to much better priors that readily illuminate it.

10:47

mircea_popescu: i got these too !

10:47

9000 just today.

10:48

possibly you shouldn't have agreed!

10:53

feedbot

http://www.loper-os.org/?p=2906 << Loper OS -- Finite Field Arithmetic vs MPI.

10:53

^ newyear present for mircea_popescu & diana_coman

10:53

*

asciilifeform brb,teatime

10:55

Turns out, Kochs pile of shit, despite eschewing constant time arithmetic, and being implemented in Overflowandcrashlang loses the footrace, when given a full-width modular exponentiation (i.e. one where it cannot cheat by skipping over leading zeroes.)

10:55

this was never controversial was it.

~ 34 minutes ~

11:30

that it would lose on ~speed~ ?!

11:31

until hr ago i was deadly certain that it couldn't

11:31

( considering that, e.g. http://www.loper-os.org/?p=2798#selection-5181.0-5211.75 )

11:32

and that he still gets to skip , theoretically , ~half~ of the work of a modexp with random inputs (half the bits of exp are 0, and in kochian modexp dun result in a mul/square cycle)

11:33

but what we find is that -- even on decade-old irons -- he loses moar to the branch predictor, than he wins from skipping

11:34

incidentally, it's a 'fair fight', i.e. both ch14 ffa and mpi-koch lack asmism

11:34

but i expect ffa-cum-asmism will still beat shit out of koch-cum-asmism

11:35

( and incidentally if you disable the bounds checks, having been satisfied with the proofs, ffa speeds up by another 2x 'for phree' )

11:40

*

asciilifeform fixed formatting barf in post

11:46

turns out, the entire justification for use of branching in rsa at all, is bogus. ( and 'short exps' aint a justification -- if you know the max length of your e in advance, you can use fixed exponentiator , same as i do )

11:47

i.e. you win moar from not upending the cache, than you lose by doing erry single bit of modexp and muxing.

11:50

mircea_popescu: phunphakt, since last night , massive uptick in heathens visiting ffa www, even curl'ing patches. but so far none have the balls to speak up.

11:51

asciilifeform in my mind the idea was that if you force non-republican software to use republican exponents it will either crash or take forever.

11:52

hence why original choice for full sized exponents, "why let enemy enjoy the fruits of his lies"

11:52

moreover you know, http://btcbase.org/log/2016-11-26#1573488

11:52

Logged on 2016-11-26 16:31 asciilifeform: koch's shitball per se is written in such a way that the cruft is glued on with broken glass (all the ciphers are modularized in very gnarly multilayered way, whole thing relies through and through on his weirdo streams thing, 1,001 idiocies)

11:52

mircea_popescu: this is tru re gpg per se

11:52

test was re koch's arithm engine (which does take arbitrary exps etc, and a patched ver is used in e.g. phuctor)

11:53

yeah but i mean...

11:53

WHY was it true re gpg per se ?

11:53

he hardcoded a short buffer, iirc, lol

11:53

gpg itself is substantially moar crippled than koch's mpi lib

11:53

( which i suspect is largely cribbed, but i dun specifically know from where )

11:53

still, gpg itself uses mpi to do the mathing

11:54

aha

11:54

same place rms got emacs im sure.

11:54

teh goddess.

11:54

the copy on my www ( and later cleaned further by diana_coman ) is from gpg 1.4.10 ( itself from signed tar that mircea_popescu dug in in '14 )

11:54

aha.

11:55

anyway, sorry if i inadvertently rained. tis indeed great news, and it being expected great news dun detract from greatness.

11:55

possibly mircea_popescu is simply less thick than asciilifeform and expected the find

11:55

( i.e. that branch predictor nukes ~all~ win from skipping zeros )

11:55

possibly more thick, expected the find :D

11:55

lol

11:56

anyway, yes, we're here ending "computing", and good for you.

11:56

btw i expect that on moar recent irons, with gigantic caches, diff will be even moar palpable

11:56

imo putting an end to "software development" much greater achievement for human race than "decrypting bolix"

11:56

( anyone have an 'i7' handy ? )

11:57

bolix is a sort of rommel's afrika front to this.

11:57

( which reminds me, i still have a bolix post queued on conveyor )

12:00

re 'end to software' : http://www.loper-os.org/?p=2892#selection-985.1-1041.25 .

12:01

*

asciilifeform loox at scrolling wwwbox log, waves to the academitards

12:02

i'm not even sure how that kelvin thing's supposed to work ; notably bothj knuth and mod6 attempted it, but the problem with starting with a number is that you don't know what to pick.

12:03

nature found 273.15 i suppose the same way it computed ocean level ; but we're not quite equal to nature.

12:03

it requires knowing what yer doing.

12:03

so how did you pick 255 and he 10k and knuth whatever it was, 100

12:03

with guts.

↖ ↖

12:04

i.e. 'if this thing needs >255 vpatches after today, then asciilifeform is a tard and somebody else ought to do the whole fucking job over'

12:04

lmao, your guts know what you're doing ?

12:05

looky, management is a rational activity. there's nothing wrong with hanging kids for having failed to master spanish in two weeks -- just as long as you can answer as to how you came up wih the two weeks.

12:05

it also has gut component.

12:05

just like shoemakin' and any other art.

12:05

this is like saying language has innate poetry.

12:05

only because someone had a fucking clue ~before~.

12:06

at some pt you gotta make a cut where you dun have exact logical support for precisely what cut.

12:06

whether it's 'attack is at quarter past dawn' or 'this needs <=255 patches or fuck it'

12:07

incidentally i expected mircea_popescu to say '255 is laughably generous'

12:07

there's great cause to attack at dawn : morale superiority. there's also great cause to attack at night, note eg dracula's not famous for attacking the turkish camp at dawn and nearly making sultan run off.

12:08

ya but why quarter past, and not 33min.

12:08

gut.

12:08

asciilifeform the discussion spawned from me going "wtf is the POINT of even counting to 255, nobody can fit such in head" that then merged into "how do you know ?" which resulted in the predictable "i wonder how the fuck does ~he~ know"

12:08

fwiw i gave orig vtron, 'v.py' , 100. it's on 97 (phf's revision) and when i get around to massaging it, i suppose will be 96, and i dunsee why to touch it again after that..

12:09

turns out he doesn't know ; in context probably dunbar's number is the saner approach.

12:09

mircea_popescu: how does bridge constructor decide 'it should hold 50 tanks time margin of 3' and not 55 and not 45.

12:09

correct.

12:10

the larger problem is "what is meant by a patch". i doubt the item can survive ~even one~ major alterations. meanwhile "hey, you spelled theyre with an apostrophe" needn't be counted (but need be fixed)

12:11

a patch in the context is a vpatch.

12:11

in short, putting numbers on the human mind's http://btcbase.org/log/2018-12-19#1881724 process is gonna be painful.

12:11

Logged on 2018-12-19 17:09 mircea_popescu: natural language, however, ablates the trees for "convenience" so to speak, ie, uses commonly what's known in computing as sparse trees.

12:11

asciilifeform i know, but not all vpatches are created equal.

12:11

which can carry some qty of change ( i hold to the practice of keeping'em reasonably compact )

12:11

the declaration i want to make with 'kelvins' is that there is such a thing as a correct program.

12:11

to which one can converge.

12:12

the vpatch that adds a "except if number is 5577, math doesn't work for 5577" is a VERY different patch from the vpatch that adds "here's a sentence ended by a comma turned ended by a dot instead"

12:12

( where it dun make sense to make anymoar changes without simply baking new item out of the dismembered corpse of old )

12:12

asciilifeform it's more than that : a) that there's such a thing as a correct program ; b) to which one can converge ; c) in a finite number of itemized steps ; d) the count therewhich can be known in advance.

12:13

mircea_popescu: if i v-versioned the ~docs~, we'd be at 50-ish, nao, i suppose

12:13

yet our ideal is literate code, as in, docs in the program.

12:13

mircea_popescu: upper bounds can be known in advance . ( in ffadom, at least. ) in larger meta-heuristics, naturally cannot be ~known~, but can be estimated.

12:13

ok, how.

12:14

mircea_popescu: if someone knows how to make ffa moar 'docs is the program'-y, oughta write in and tell asciilifeform , cuz i came as close to it as i knew how.

12:14

( and it still needed conventional docs, illustrations, etc )

12:15

seems to me the unspoken heuristic is, "large enough so it's not meaningful [and therefore large enough to not bother] and small enough so it's not larger than some other number i thought about".

↖ ↖

12:15

this doesn't make for much.

12:15

imho the haskellism where 'program carries proof' is generally a sad thing and results in soup

12:15

not proof ; comments. like knuth wrote them.

12:15

thing's about 50-60% comment by mass.

12:16

right. and such a thing as typos in those comments is possible. this entire line is the result of your ineffectual sideswiping.

12:16

there's at least 1 still in'ere.

12:17

http://btcbase.org/log/2018-12-28#1883605 << it is impossible to defend against this criticism, as formulated, so i won't try

12:17

Logged on 2018-12-28 17:15 mircea_popescu: seems to me the unspoken heuristic is, "large enough so it's not meaningful [and therefore large enough to not bother] and small enough so it's not larger than some other number i thought about".

12:18

in short : "why not say the magic number 3 ?" "because maybe i need to fix a typo, and i dun wanna lose 1 out of 3 for this, but i do wanna fix it" "so how do you know 1 in 255 is ok to lose for this ?" "um http://btcbase.org/log/2017-07-18#1686299 "

12:18

Logged on 2017-07-18 22:58 asciilifeform: whaack is quite likely thinking of the bulk of the b00k, which consists of blockcipher liquishit which is complicated for no reason at all other than the religion where 'it is confusing to ME, author, and therefore Must Be Hard To Break'

12:18

http://btcbase.org/log/2014-04-29#650478 << see also.

12:18

Logged on 2014-04-29 22:56 asciilifeform: 'Dijkstra said angrily. (Hed been visibly shaking his head through out the talk even before this outburst.) How many bugs are we going to tolerate? he demanded. Seven, Teitelman shot back.'

12:18

!Qcalc 255/7

12:18

lobbesbot

mircea_popescu: 36.4285714286

12:18

seems we've improved.

12:18

lol

12:20

i'll be surprised if i haven't nailed the last comment typo by the time done adding chapter-sized knobs, tbh

12:20

the fundamental promise involved in the concept of number, as a scalar measure of something

12:20

is thgat it'll be applied to like things

12:20

my objection inter alia is that this is not satisfied here ; makes us look like a bunch of retards.

12:21

let's hear mircea_popescu's logical end of the stick tho. mircea_popescu finds himself partizan commander again, in the field. has 300 men. how he decides how many to send to ambush nazi convoy ?

12:22

any details re this convoy ?

12:23

let's say scout reported a 1k-man garrison in the point of origin. and nobody knows precisely how many hauptmann fuckface will be sending.

12:23

im not sending anyone.

12:24

lol

12:24

well what's the incentive ?

12:25

e.g. you gotta keep'em from linking up with neighbouring pile o'jerries, they'll encircle you and yer done

12:25

im sending everyone.

12:25

see ? there's ~reasoning~.

12:25

good, yer erryone register is 8bit long

12:25

you nao sent 255, congrats

12:26

so your reasoning was that if some bits get left out the koch is gonna eat them ?

12:26

not as such

12:26

my reasoning is 'must pick a number, distinguish item from the endless moron march of firefox etc'

12:29

not even ff reached 255 yet. but anyway, i think this is exhausted

12:29

"gotta pick something" and all that.

~ 1 hours 21 minutes ~

13:50

meanwhile, pehbot updated

13:50

!!up pehbot

13:50

deedbot

pehbot voiced for 30 minutes.

13:51

!A V_[FFACalc Ver: ]#

13:51

pehbot

asciilifeform: FFACalc Ver: 00000000000000000000000000000000000000000000000000000000000000FF

13:51

!A V[FFA Ver: ]#

13:51

pehbot

asciilifeform: FFA Ver: 00000000000000000000000000000000000000000000000000000000000000FF

13:51

nuffin to write home about; but there, current.

13:53

( if not for irc line char # cap, could even make it do 2048bit widths, rsa etc. will still fire faster than ping lag on typical day..)

~ 37 minutes ~

14:31

meanwhile, in ancient l0gz, http://btcbase.org/log/2016-11-26#1573520 << still, interestingly, live.

14:31

Logged on 2016-11-26 19:16 mircea_popescu: and in other news, a... bar. http://68.media.tumblr.com/95befa3bb9c2e4fac4f3dfb4a94b23e6/tumblr_mrxj9415lX1sfcru6o1_1280.jpg

14:31

despite 'ban'

14:31

http://btcbase.org/log/2016-11-26#1573536 << ditto. and, come to think of it, ~all of'em.

14:31

Logged on 2016-11-26 22:20 mircea_popescu: meanwhile at the slut party, http://68.media.tumblr.com/tumblr_lard124fV41qzw9y2o1_1280.jpg

14:44

oh wow asciilifeform !

14:44

*

diana_coman catches up

14:49

feedbot

http://qntra.net/2018/12/stanislav-datskovskiy-publishes-fully-constant-time-code-for-barretts-modular-reduction-as-part-of-ffa-library/ << Qntra -- Stanislav Datskovskiy Publishes Fully Constant Time Code For Barrett's Modular Reduction As Part Of FFA Library

14:52

BingoBoingo

Meanwhile in local lulz, confiscated drugs liberated from the police in motorcycle snatch and grab https://www.teledoce.com/telemundo/policiales/en-san-jose-robaron-145-envoltorios-de-droga-que-habian-incautado-a-un-hombre-minutos-antes/

15:02

lollers BingoBoingo

15:06

diana_coman: lulzy, neh. we do 4 MUL instrs for each word x word , and we do each ADD & SUB ~twice~ ( http://www.loper-os.org/?p=1913#selection-3319.14-3343.1 ) ( plus buncha ands, ors, shifts ) , and ~still~ win...

15:09

on one hand there is what mircea_popescu says that yes, by now and given the type of rot it's actually to be expected; on the other hand it's of the sort that still sounds...unbelievable, yes

15:10

btw there's a 'seekrit' 30% linear speed boost if one uses unrolled combas.

15:10

( not in any patch yet, but pasted last summer )

15:10

and incidentally it wins for same reason -- branch predictors suck

15:11

for some particular applications (e.g. gossipd) it (and asmism etc) justified.

15:12

( http://btcbase.org/log/2018-10-12#1860844 << see also oldthread where the same effect )

15:12

Logged on 2018-10-12 13:52 asciilifeform: ( bonus is that the closed form is not only constant time, but substantially faster on pc, nomoar branch prediction misses )

15:13

the other thing we get, is of course http://btcbase.org/log/2017-10-08#1722435 .

15:13

Logged on 2017-10-08 00:20 asciilifeform: http://btcbase.org/log/2017-10-07#1722411 << 1 ) ffa is closed form. i.e. it CAN be written as a number of nand gates, with a 'funnel' at the top, to which you present a,b,c, e.g. 4096bit, numbers, and at the bottom in a little cup you get a^b mod c , and with NO UPWARDS FEEDBACK FLOW of information , i.e. answer comes after same interval of time always, and with strictly downwards signals.

15:13

( i.e. thing is siliconizable, in principle. )

15:26

diana_coman -- will be interesting to see benchmarks from your irons, if you get the chance.

15:27

asciilifeform, I'll try to get to it but more probably in January really

15:27

no particular hurry

15:27

( the key work for diana_coman et al, from ffaology pov, is to eat the proofs )

15:28

that's precisely why the slow pace - I need a bit of fresher mind to get back and read through the rest of ffa

15:29

this is the place where slow an' careful is the Right Thing.

15:29

that being said, I dug up and yes, I even have an i7 around here

15:29

you may find it interesting that asciilifeform's 1st shot at barrett managed to have a subtly fatal mistake in the proof, and correspondingly a proggy that failed on <0.1% of input space.

15:30

sadly though no amount of slow is any guarantee of perfection still

15:30

( hence the somewhat pedantic treatment of the subj, 2 separate proofolade articles. was going for 'measure 7777 times, cut 1ce' there. )

15:31

fwiw I like very much that "pedantic" there

15:33

btw it is possible to dispense with FZ_LoMul , and simply discard upper chunk of ordinary karatsuba product. this costs ~10% cpuwise, and may be justified for some proggies.

15:33

( where yer tryin' to save coad space )

15:34

( http://btcbase.org/log/2018-12-18#1881560 << see also. technically applies to ~errything after ch6. )

15:34

Logged on 2018-12-18 20:14 asciilifeform: barrett is 'optional' (technically errything after ch6 is!)

15:35

diana_coman: sadly i do not know how to 'guarantee perfection'. all i know how to do is to bake maximally 'fits in head' and bank that the folx here will find mistake if it turns out that i made one.

↖

15:37

( and it is fortunately much easier to find mistake in 5k loc , with ~one~ execution path, than in 150kloc with 900000 paths... )

15:38

certainly

15:43

the other thing, if one is writing a proggy where only 1 ffawidth is used, it is possible to unroll all of the loops, derecursivize karatsuba, etc. and get straight line of instructions

15:44

( i.e. sequence where only ADD, SUB, MUL, SHR/SHL, MOV, all the way from head to toe , a,b,c in, a^b mod c comes out )

15:47