ok. it's slower, maybe less accurate as well. but code size should be smaller(?). I'm thinking for the the future, when i want to run on real hardware. I avoided adding sw implementations because of all these constants.
(1) a browser / (2) a c compiler / (3) a go compiler. this is indirect. I have extracted soft float for sin/cos/atan2/hypot from the go stdlib as a source of inspiration. not too much code, i'll send you by mail.
go is bsd-like i guess. golang.org/LICENSE. the source code for these math algorithm also mentions original licences. browse here, e.g.: golang.org/src/math/sin.go (and one directory upwards)
@ngn there is also an arg reduction branch for very large arguments, that i removed because of it's complexity. see the original if you are interested.
@ktye for ngn/k there are bigger questions i must answer first
what's the point of floating point? (yeah, i know.. it's "floating") i hardly ever use it. maybe it belongs to a library that should be written by someone who does use it.
also: what's the point of carrying on with ngn/k? golfing is dead. k5-6 is dead, and the community have moved on to k9, following the leader rather than trying to develop some kind of reasonable alternative.