Operations on float64 are more expensive than on float32. The library should be refactored so that float64 is as limited as possible. Things to consider:
cost of converting from float64 to float32 to perform further operations. It might be cheaper to continue with float64 in between areas where float64 is required.
If the return type of the function is not float64 the conversion cost might be considered free as it would need to happen at some point anyway.
Operations on
float64are more expensive than onfloat32. The library should be refactored so thatfloat64is as limited as possible. Things to consider:float64tofloat32to perform further operations. It might be cheaper to continue withfloat64in between areas wherefloat64is required.float64the conversion cost might be considered free as it would need to happen at some point anyway.