mul_add between real and complex numbers

[isomer]

I'm attempting to use:

 println!("{:?}", (2.0_f32).mul_add(
      Complex::new(3.0_f32, 4.0_f32), 
      Complex::new(5.0_f32, 6.0_f32)));

which doesn't exist.

I think the equivalent would be:

fn fma<T: num::Float>(a: T, b: Complex<T>, c: Complex<T>) -> Complex<T> {
    let re = a.mul_add(b.re, c.re);
    let im = a.mul_add(b.im, c.im);
    Complex::new(re, im)
}

which does compile using two fused multiply adds which hopefully the compiler could reorder, vectorise or whatever optimisations it believes fruitful.

The code I'm writing (a convolution kernel) is attempting to be generic over real and complex numbers which relies on the num::Complex crate to provide this implementation.

My preferred solution would be for num::Complex to implement T::mul_add(Complex<T>, Complex<T>) -> Complex<T> where T: MulAdd<T, T, Output=T>

[PaulXiCao]

There is also a discussion about adding div_add at #146. It should probably also be taken into consideration for consistency purposes.

Similarly CheckedAdd/Sub/... are being added at #152 .

When deciding which arguments should be supported for Complex<T> and T we should also discuss owned vs reference arguments.