Improve performance of ilog and checked_ilog for primitive integer types

[FedericoStra]

Currently checked_ilog is implemented with iterated divisions:

rust/library/core/src/num/int_macros.rs

Lines 2475 to 2495 in e7f9f48

	pub const fn checked_ilog(self, base: Self) -> Option<u32> {
	if self <= 0 || base <= 1 {
	None
	} else {
	let mut n = 0;
	let mut r = self;
	// Optimization for 128 bit wide integers.
	if Self::BITS == 128 {
	let b = Self::ilog2(self) / (Self::ilog2(base) + 1);
	n += b;
	r /= base.pow(b as u32);
	}
	while r >= base {
	r /= base;
	n += 1;
	}
	Some(n)
	}
	}

It's pretty straightforward to convert it to iterated multiplications:

fn checked_int_log(self, base:Self) -> Option<u32> {
    if self <= 0 || base <= 1 {
        None
    } else {
        let mut n = 0;
        let mut r = 1;

        // Optimization for 128 bit wide integers.
        if Self::BITS == 128 {
            let b = Self::ilog2(self) / (Self::ilog2(base) + 1);
            n += b;
            r *= base.pow(b);
        }

        while r <= self / base {
            n += 1;
            r *= base;
        }
        Some(n)
    }
}

and this results in some decent speedups (1.5x - 7x).

[ChaiTRex]

@FedericoStra In case you didn't know, it seems that you closed the related pull request.

[FedericoStra]

@FedericoStra In case you didn't know, it seems that you closed the related pull request.

Ooops... Yes I had some troubles forking and cloning the repository (possibly slow connection) and I did something wrong. Thanks for telling me!

[FedericoStra]

New PR: #115913.

I'll make sure not to accidentally close the new one this time 😬