Josh Stone

Hi again -- I don't know why I didn't merge this before, but let's try now... bors r+

AFAICT, those intrinsics are now supported by `rustc_codegen_cranelift`, `rustc_codegen_gcc`, and miri (and of course LLVM), and I've still found them to be faster than what the standard library is using....

On further reflection, the *measured* gains should outweigh theoretical optimization losses. Maybe someday we could try [`is_val_statically_known`](https://doc.rust-lang.org/std/intrinsics/fn.is_val_statically_known.html) to detect divisors that might be optimizable.

Both `iter_u32_digits` and `iter_u64_digits` are explicitly documented as being "ordered least significant digit first." That's roughly little-endian, although we're not talking about _byte_ order -- the `u32`/`u64` digits that come...

Merged in #56 without the inline change.

Given the large amount of existing uses, I think even deprecation would be too much churn. https://github.com/search?q=lang%3Arust+%28Complex32+OR+Complex64%29&type=code

Unfortunately, this is a breaking change for external implementors, especially since this trait isn't sealed. It's tempting to add a default to `8 * mem::size_of::()`, but nothing says that the...

@aj-r interesting, thanks! I'm open to "fixing" unsigned with that kind of implementation, as long as we're careful to document the difference -- and ideally how to reconstruct what the...

Maybe we could then drop `Signed` from `extended_gcd_lcm` -- but I think that's technically a breaking change for anyone that overrode that method, if they depended on the conditional `Signed`...

Closed by #53