Search
14:41:25
stassats
(rational 1.5d0) about five times faster now
14:41:53
stassats
(rational 15d0) about two times
15:02:17
stassats
(= float bignum) should be trivial, first check first-bit-set and then use ldb-bignum=>fixnum
15:02:47
stassats
(< float bignum) would need to divide
17:19:11
stassats
ok, no need to divide, same thing but just use (- exp first-bit-set) to see of there's a remainder
17:19:15
pkhuong
stassats: float in the 64 bit bignum range are all integers
17:20:37
stassats
but i probably do not care about 32-bit
17:21:05
stassats
still, gotta get it right
17:25:35
stassats
but once i have (= bignum float) adding (< bignum float) for fractions is trivial
17:27:47
stassats
and we already have bignum-lower-bits-zero-p
17:40:51
stassats
bignum-float-= is about twice as fast and doesn't cons
17:41:35
stassats
even better on small floats which aren't bignums, especially if they are ratios (with the old RATIONAL it would've been even slower)
18:11:07
stassats
bignum-float-< is trickier, but seems to be twice as fast as well (and no consing, naturally)
18:48:48
stassats
i wonder if i should just punt to RATIONAL on 32-bit for <>, as it's getting quite tricky
19:12:54
stassats
can't get it right even on 64-bit
22:08:13
stassats
i guess i implemented all the stuff i can think of for cross-type comparisons
22:09:39
stassats
ok, (= complex complex) isn't using single/dobule-float specialized variants