TotalEq -0 in fabs test

[Rudxain]

assert_eq! uses PartialEq, which doesn't discern between -0 and +0. total_cmp does distinguish them. So this patch makes the test spec-compliant

[Rudxain]

I recommend reviewing any other code that uses == and/or assert_eq!, to check if it should be replaced by total_cmp. total_cmp also distinguishes between signed NANs, which may be useful somewhere else in the codebase

[Amanieu]

It would be better to use .to_bits() to ensure an exact match.

[Rudxain]

.to_bits()

I agree, but that would only apply to other cases (such as performance-sensitive code), not here. The spec imposes no requirements about fabs return bit pattern

[Amanieu]

There is only a single bit pattern for 0 or -0.

[Rudxain]

There is only a single bit pattern for 0 or -0.

Yes, but some platforms use "flush-to-zero" for subnormal numbers. I'm not entirely sure how that mechanism works 🤔

[Amanieu]

That doesn't have any effect here. It's only done during arithmetic operations, not comparisons. In any case the Rust language doesn't allow flush-to-zero behavior, so such modes must be turned off in the CPU when running Rust code.