This adds some documentation to the complex divide kernel about the potential pitfall in some architectures of obtaining inf or nan with a small denominator (even if the numerator is zero) due to reduced range because of -fcx-limited-range and -freciprocal-math.
This adds some documentation to the complex divide kernel about the potential pitfall in some architectures of obtaining inf or nan with a small denominator (even if the numerator is zero) due to reduced range because of -fcx-limited-range and -freciprocal-math.
This closes #608.