asciilifeform: arm cpu is incidentally ~egregious~ offender
asciilifeform: of course it is even easier if idiot luser uses 'library' or 'os api' crypto. hence why the insistent and incessant schneiering, 'don't write own crypto!'
asciilifeform: 'i know! the number of bits equalling 1! i'ma broadcast'
asciilifeform: cpu, designed by enemy, goes 'aha! he wants to mul! what can i leak about his operands..!'
asciilifeform: but having a 'mul' instruction is in fact a thing on the same continuum!
asciilifeform: the multiplication problem is incidentally a textbook instance of 'specificity of diddling'. if someone offered any of you a cpu that has a 'rsa instruction', you'd barf. because very clearly paints a target for the enemy, giving him info he has no business having re what you intend to eat, when, with what spoon
asciilifeform: but at least sees that it is a problem..
asciilifeform: witness, incidentally, the lack of a 'constant-MUL' instruction on any extant cpu.
asciilifeform: (what's it gonna do ? not shift? not add ? how'd it know when.)
asciilifeform: shift-and-add ('egyptian') mult. has advantage that cpu CANNOT attempt to optimize
asciilifeform: in fact, it is well-known not to be constant time on recent intels.
asciilifeform: that there is NO guarantee that MUL/IMUL (or equiv. on other cpu) is constant time !!
asciilifeform: a 'schoolbook' (word-by-word, using * (which will compile to IMUL instr on x86) multiplier will be faster than the 'egyptian' one. the catch is,
asciilifeform: possibly i explained this next bit, but can't seem to find it in the l0gz, so will post here ftr :
asciilifeform: in other news, a 4096-bit A**B takes approx 14 seconds (3GHz) .☟︎☟︎
asciilifeform: as for 'ar is too far', d00d was (from his pov) poor -- a small wad of btc, but the damocle's sword of having to do honest work one day, loomed; then he 'sold the motherland' and now rich, problemsolved.
