Add fibonacci benchmark, for fairly low-level eval performance

[devyn]

This is a benchmark that I created while testing IR. It's meant to basically do only very simple operations but be running 100% Nushell code most of the time, without any heavy lifting being done by commands, and without any closure calls or anything like that. It seemed useful to keep around so I'm adding it to nu_scripts.

[fdncred]

Thanks

[amtoine]

just for fun, because generate is cool and helps annihilate mut 😏

# compute the Fibonacci numbers until $F_n$
#
# # Example
# ```nushell
# # compute $F_42$
# fib 42 | last
# ```
def fib [n: int]: [ nothing -> list<int> ] {
    generate [0, 1] { |fib| {
        out: $fib.0,
        next: [$fib.1, ($fib.0 + $fib.1)],
    } }
    | first ($n + 1)
}

[devyn]

Yeah we can add that too, it's actually a good comparison / goal for optimization.

It is much slower though. It would be really nice to be able to get things to the point where it has similar performance, but the mutable version really is so much simpler to execute unfortunately

[fdncred]

it would be interesting to compare the ir of these two methods through debug profile

[devyn]

One of the things you won't see in debug profile is the extra work being done to make a closure call. That's something I'd like to make more efficient

[amtoine]

aah yeah, that's a bummer if the generate version is so much slower... i find it much much better than using mut, more Nushelly 😌

[devyn]

Certainly. I think the main thing holding it back really is closure eval performance, though I should try profiling it

There's also a free/alloc on each loop there because of the list. Mimalloc is efficient, but I think the mutable version involves no extra heap allocations for each loop