158 entries in 0.377s
a111: Logged on 2017-08-14 17:57 asciilifeform: if instead of 'mult of 64' we had 'powers of 2', we could dispense with the odd split in
karatsuba ascii_modem: nao for
karatsuba squaring case. and that's mostly it.
a111: Logged on 2017-07-15 13:00 asciilifeform: btw if you're actually doing something that doesn't need constanttime, you can simply put the obvious check-for-zero in the
karatsuba and get 2-9000x boost for mul.
mod6: (contains completed
karatsuba, but no comba)
a111: Logged on 2017-07-13 05:21 mod6: this is actually using
karatsuba, haven't even integrated the new comba code yet.
mod6: this is actually using
karatsuba, haven't even integrated the new comba code yet.
☟︎ a111: Logged on 2017-06-21 16:44 asciilifeform: ( 8192bit exponentiation: ~10min with egyptological mul; 20.5s was with first-grade mul; 17.7 with
karatsuba posted today )
mod6: this is outside the spec, but just as an experiment, what kinda timings do we get if you were to do inline asm for the W_Mul &
Karatsuba procedures? is this worth doing?
a111: Logged on 2017-06-18 23:30 asciilifeform: and realized, while doing this, that in fact you don't need 2k+2 bits for the
karatsuba intermediates, you can do instead of (x0+x1)*(y0+y1) , (x0-x1)*(y0-y1), and then you don't need to propagate carries, but only take absolutevalue and xor the borrows to see if gotta invert the resulting term
a111: Logged on 2017-06-18 23:30 asciilifeform: and realized, while doing this, that in fact you don't need 2k+2 bits for the
karatsuba intermediates, you can do instead of (x0+x1)*(y0+y1) , (x0-x1)*(y0-y1), and then you don't need to propagate carries, but only take absolutevalue and xor the borrows to see if gotta invert the resulting term
mod6: yeah, few examples i've seen of
karatsuba all use base 10..
mod6: I guess, it doesn't cover
karatsuba specifically, but gives some background perhaps for the uninitiated.
ave1: From looking at
karatsuba and how it could work bitwise, it might also be useful to have a starting offset
a111: Logged on 2017-04-28 23:50 phf: in related lulz
Karatsuba and Ofman's "Multiplication of Many-Digital Numbers by Automatic Computers" doesn't seem to be available anywhere online, but "everyone" seems to know it somehow
phf: philistine that i am, for a longest time i thought
karatsuba was japanese..
phf: in related lulz
Karatsuba and Ofman's "Multiplication of Many-Digital Numbers by Automatic Computers" doesn't seem to be available anywhere online, but "everyone" seems to know it somehow
☟︎ ascii_field: normally, acca relies on sidechannels (e.g.,
karatsuba mult. timing)
assbot: Logged on 15-07-2014 03:00:27; asciilifeform: one of my first job interviews out of uni. telephone. a fellow from one of the giant gov. contractors was really intrigued that i know x86 asm., have reversed crud for money. i ask him 'what's the job'. he: automated reversing. me: of what. he: ever hear of
karatsuba's algo? me: sure. bignum mult. him: well, we wanna find encryption softs on terrorist drives!
assbot: Logged on 15-07-2014 03:00:27; asciilifeform: one of my first job interviews out of uni. telephone. a fellow from one of the giant gov. contractors was really intrigued that i know x86 asm., have reversed crud for money. i ask him 'what's the job'. he: automated reversing. me: of what. he: ever hear of
karatsuba's algo? me: sure. bignum mult. him: well, we wanna find encryption softs on terrorist drives!