(it is not clear which number is first and which is second)

It is specified in the description:

sol[0]: the real part of the first number

sol[1]: the coefficient of the imaginary part of the first number (expressed in terms of √-1, more commonly known as i)

sol[2]: the real part of the second number, must not be greater than sol[0]

sol[3]: the coefficient of the imaginary part of the second number, must not be greater than sol[1]

relative errors

Sorry, I'm not quite sure - does that mean percentage difference?

Sure, will fix this issue and the other one later today

Does it work now? I've done that type-cast, and reduced the range of sum and product in the random tests.

"The first number" is the first solution (complex) whose real part is sol[0] and imaginary part is sol[1]. The second number is the first number subtracted from the sum parameter of the function (Re = sol[2], Im = sol[3]).

They are relevant because they need to be returned to the caller as the only set of complex numbers (here, a double[4]) such that the sum and product of the first and second numbers are very close to the passed parameters. I hope this helps!

The more you know about obscure inline ASM hacks... :D

Oh okay, thanks! Is there anything else that needs to be fixed?

Fixed both - is it safe to use memcpy() like this, though?

Sure, you can approve it, it looks good to me!

Oh, OK - thanks a lot for explaining, I'll do my best to remember that!
I've unpublished my fork, I agree it wasn't anything unique...

Hi,

I do want to make a NASM translation - sorry if I caused a mess by publishing my fork, will be more careful about this next time!

So if I understood correctly, I can approve the version that I think is better without the author having to review it?

NASM translation, please review :)