SIMDRegister <std::complex<float>> or dsp::SIMDRegister <std::complex<double>>?

Hi. Does anyone have any experience with using complex SIMD registers?
I’ve always done complex multiplies by hand, and I wanted to start using the SIMDRegisters to visually simplify the code, and hopefully gain a speedup.

Is it best to use the double version, or the float version? I feel the double version is more compatible, but I don’t know how the float version is set up. Are there two complex numbers in the float version?
And does it work well with the Arm version of the SIMD complex code?