asciilifeform: ( http://btcbase.org/log/2015-01-10#971993 << thread. possibly elsewhere also. ) ☝︎

asciilifeform: see the classic, https://lwn.net/Articles/629259/ , and the rms thread, re why this is not available for c/cpp gcc

asciilifeform: ikr?

asciilifeform: mod6: http://wotpaste.cascadianhacker.com/pastes/Ibqes/?raw=true << example with w_mul

asciilifeform: and /me considered pasting the example, but turns out it potentially fires in every single line, lol

asciilifeform: so for instance i throw in pragma Restrictions (No_Exception_Propagation); and see "Constraint_Error may result in unhandled exception" and then wonder where,

asciilifeform: pretty great for debugging failures of, e.g., pragma Restrictions(No_Implicit_Conditionals);

asciilifeform: barfs raw ast out of gnat

asciilifeform: in other 'news', '-gnatG' flag is very very spiffy ☟︎☟︎

asciilifeform: http://210.86.180.160/ << lol, readable dirs

asciilifeform: ( they get set afresh each time it muls )

asciilifeform: ( also posted earlier, at some point )

asciilifeform: mod6: they dun need initializing, they are the low and high words of word*word mul

asciilifeform: mod6: see in https://archive.is/DGDu1

asciilifeform: note that our L does not depend on the results of the arithmetic at any point

asciilifeform: same reason i didn't implement fft

asciilifeform: ( only begins to +ev a considerable way above 8192b )

asciilifeform: generalization of karatsuba, but pretty useless for fitsinhead rsatron imho

asciilifeform: mod6: toom-cook ?

asciilifeform: ( and -- damningly -- i can't think of why. )

asciilifeform: http://btcbase.org/log/2017-07-10#1681458 << massive ☝︎

asciilifeform bbl, meat

asciilifeform: if you want ( and we do ) an ABSOLUTE, iron grasp.

asciilifeform: the tightest, in practice, space of all, however, is.. the head

asciilifeform: ( lessay you had to put one in a very... tight space. e.g. 'uci' payload... )

asciilifeform: part of what asciilifeform was even doing re 'p' is answering the q of 'what is the minimal practical rsatron'

asciilifeform: ( esp. if you use muller's delay-insensitive logic )

asciilifeform: or even... mechanical

asciilifeform: aha

asciilifeform: mircea_popescu: aka 'path dependent'

asciilifeform: probably an ada build and a handrolled asm, per cpu make, is best that can be had.

asciilifeform: i don't expect to see many reimplementations, sadly, however, because one of the items of said spec is 'the following ops are in constant spacetime...'

asciilifeform: p ain't a 'code is the spec' crock-o-shit tho. up asciilifeform's sleeve, is an actual spec

asciilifeform: but try and get back lost clarity!!

asciilifeform: we can always get the speed BACK later

asciilifeform: upstack, ffa wants to double as a 'how to rsa' 'b00k' , i.e. at every point tradeoff was made for clarity rather than speed.

asciilifeform: the way folx used to stitch shifters together.

asciilifeform: and if made correctly they oughta plug in for longer arity.

asciilifeform: same principle

asciilifeform: in my mind's eye thing would be a 32kBit-arity alu...lol

asciilifeform: tru!

asciilifeform: but unrolled is moar cacheable.

asciilifeform: constant-arity loops - equally fixed

asciilifeform: ( i dun know of any compelling practical reason to do this, outside of something like a hand-sewn gossipd packet filter in asm on bare iron, for max horsepower )

asciilifeform: but iirc this came up in old thread.

asciilifeform: pretty lulzy, jmpless rsatron then

asciilifeform: e.g. mircea_popescu's 4096

asciilifeform: phunphakt -- if you nail down, to a fixed bitness, the N-arity of the arithmetic, you can unroll all of the loops.

asciilifeform: ( ffa in asm is - somewhat surprisingly - easier than ada or even c )

asciilifeform: in the interest of bootstrapology.

asciilifeform: relatedly, after adatronic 'p' is fully nailed down and flying, i'ma bake one in x64 asm.

asciilifeform: my gnat is mine, but what's in yours - only beelzebub knows

asciilifeform: ftr asciilifeform doing this chore does not excuse other folx from same

asciilifeform: seems like it works - almost astonishingly - as printed on the box...

asciilifeform: mod6: aha. i wanted to see how the bounds checks are done, and whether gcc4.9 takes any liberties with'em

asciilifeform: on -O2 it's surprisingly notbad

asciilifeform: the asm

asciilifeform: of gnat, i mean

asciilifeform: i've been reading, btw, the output

asciilifeform: lol

asciilifeform: ( and -- fits-in-head !! )

asciilifeform: in some critical applications (airplane, rsatron) this is still unacceptable and hence spark

asciilifeform: most you'll get is a stop.

asciilifeform: mod6: you won't physically ever see a segfault in ada proggy built with default 'fascism' level

asciilifeform: mod6: note that 'blow up' still dun mean 'segfault'

asciilifeform: every single time it was 'guess wat, this is How It Works in the adult hut. for reasons that Make Sense'...

asciilifeform: quite possibly i'dve given up long ago if not for this repeated experience

asciilifeform: 'oooh hey turns out, Right Thing'

asciilifeform: every time, to date, when i bashed head against wall and went 'WAI DOES IT DOOO THAT!1!!1' i ended up repenting

asciilifeform: ada is a thing for a reason, i found. hence this entire thread.

asciilifeform: slice has own bounds, the correct ones.

asciilifeform: nope.

asciilifeform: aha.

asciilifeform: every. time.

asciilifeform: they get checked against the bounds

asciilifeform: wrong

asciilifeform: about which you can prove particulars.

asciilifeform: in that they are replaced with a far more constrained item

asciilifeform: rid.

asciilifeform: in much the same way as 'in' parameters (in ada procedures)

asciilifeform: upstack, array slices ( which commonlisp also has ) are important to being rid of having to use c-style pointers

asciilifeform: (no_implicit_loops pragma iirc)

asciilifeform: ( behind the curtain )

asciilifeform: seekrit : in the 'most fascist' restriction mode, ada... copies slices

asciilifeform: worst of all worlds.

asciilifeform: and bloats the algo.

asciilifeform: nope. blows the cache.

asciilifeform: that's a copy, we do this also sometimes.

asciilifeform: *slice

asciilifeform: that ain't called a sloce

asciilifeform: no!!!!

asciilifeform: so you can get it back from a slice

asciilifeform: and ergo retain the original's indexing.

asciilifeform: slices are not arrays in their own right, they are offsets into another array

asciilifeform: though i also found it at first difficult to understand why.

asciilifeform: because not doing it would lead to increased ugly.

asciilifeform: mircea_popescu: it ain't tho.

asciilifeform: rot13 spoiler: Nqn neenl fyvprf xrrc gur 'Svefg naq 'Ynfg bs jurerire va gur cnerag gurl pnzr sebz! sbe fbzr ernfba guvf vf abg zragvbarq naljurer ohg va gur fgnaqneq. Xnengfhon bs pbhefr raqf hc vaibxvat gur onfrpnfr zhygvcyvre jvgu fyvprf gung qba'g ortva jvgu mreb, naq gb qrfgvangvba neenl gung yvxrjvfr qbrfa'g. Fb jr tbggn abeznyvmr.

asciilifeform: incidentally, (imaginary) prize to the first d00d who understands why this had to be rewritten with the loops indexing from 0 .. L-1, rather than, e.g., for i in X'Range ... as formerly

asciilifeform: so may as well get head start.