Post #1782148
2026-04-28 06:52 UTC
Replies (2)
-
@simon_brooke@mastodon.scot 2026-04-28 06:57
@screwlisp @vnikolov Look, are we talking 'should I', or, 'will I have time in my lifetime to'? Certainly I think I should; practically, my next goal after the compiler is exploring ways of getting computations to spread across the hypercube that actually deliver performance. I mean, a massively parallel computation which is actually slower than a single threaded one is pointless. A computation in which a thousand processors are cooperating isn't going to be a thousand times faster...
-
@vnikolov@ieji.de 2026-04-28 19:53
P.S. Just to note this; people must have been doing it for decades. Consider the following well-known identities for rational p and q: (sqrt x) ≡ (expt x 1/2) (* (expt x p) (expt x q)) ≡ (expt x (+ p q)) With an evaluator that applies the respective transformations, perhaps with a bit of lazy evaluation thrown in, we "automatically" get (* (sqrt 2) (sqrt 2)) => 2 (exactly). @screwlisp @simon_brooke