exp with floats and doubles

[riched158]

hi

i am running a program in C++ that calls (amongst other things) the exp function several hundred times. When i use doubles the program runs approx 10% faster than with floats. Can anyone explain why this might be?

cheers
richard

[VictorN]

Which one from a lot of exp types are you using?

//edit: ... and how did you measure and for what build?

[riched158]

hi

i am using c++ under Gcc. As i understand it, in this system the compiler automatically selects the appropiate exp for example expf for a float. One cannot specify which type of exp to use.

http://www.cplusplus.com/reference/clibrary/cmath/exp/

[VictorN]

i am using c++ under Gcc.
...
http://www.cplusplus.com/reference/clibrary/cmath/exp/

Well, I don't know how Gcc implements this function but assuming your link:

Portability
In C, only the double version of this function exists with this name.

I'd guess that Gcc may use casting/converting operations for floats to double and back to use only this variant:

  double exp (      double x );

BTW, you haven't answer how you measured...