> I wonder if there's a way to make it work where all the usages are replaced by 'float_t'. Yes, in fact you don't have to change anything else. The trick is t

> I wonder if there's a way to make it work where all the usages are replaced by 'float_t'. Yes, in fact you don't have to change anything else. The trick is that on x86-64 float is promoted to double for argument passing, so the mismatch in the return type on spectral_contrast has no effect.