Interface nfloat and nfloat_complex

[fredrik-johansson]

These are available in flint-3.2(-rc1) and often much faster than arf/acf. Plus, they should be possible to wrap as GC-free Julia structs (using metaprogramming to generate one for each limb size).

[Joel-Dahne]

Yes, wrapping these would be great! A few weeks ago I started to experiment a little bit with a potential interface, but didn't get so far.

I'm not sure if it is possible to generate them in a fully GC-free way. They have a fixed size, but they still need to be mutable. From what I recall, mutable structs in Julia are generally heap allocated and handled by the GC. In principle I think the compiler can sometimes make optimization to be able to put them on the stack, but probably not if they are passed to opaque C-functions. The GC should however be significantly faster than for afc/arb since there is no need to have a custom finalizer, which has a fairly large performance impact (though it might be possible to improve upon this).

There might be some way to get around the heap allocation. I was hoping to discuss this a bit with people in the Zulip chat during the workshop next week (since unfortunately I will not be able to attend in person).

[Joel-Dahne]

I started making an attempt at wrapping these functions, see flintlib/flint#2184 for some comments.

[Joel-Dahne]

See #202