How numerical constants are handled in DIRAC¶
In Dirac numerical constants are handled in a few different ways.
- For example the code pieces ::
- double precision :: x,y y = 2*x
- with an integer “two”, ::
- double precision :: x,y y = 2.0D0*x
- with a double precision “two” and ::
- double precision :: x,y double precision, parameter :: DTWO = 2 y = DTWO*x
all results in the same value being assigned to the variable y. The two first examples often result in exactly the same machine code, since the compiler know how to convert the integer or double precision number “two” to the most efficient form. In the third example the number is first stored in a (parameter) variable, and can then be used. Besides obvious uses for constants such as pi, this is perhaps most important when the constant is passed as an argument to a subroutine.
Since Fortran does not know what type of argument the subroutine expects it cannot convert the argument given by the programmer to the correct type. Therefore it is very important that the argument is passed with the correct type by the programmer. By making, for example, a double precision number DTWO the type is guaranteed to have this representation when passed to the subroutine. The same effect can be achieved by writing i.e. 2.0D0 (if a double is expected), or 2 (if an integer is expected). This does not mean that it is somehow “wrong” to write y = 2*x in a line that is not a subroutine call, even if x and y are floating point numbers. In this case the compiler knows what to do, and you don’t have to write y = 2.0D0*x. In the same way you can write y = x/3, and get correct results. On certain older cpus it was slightly faster to execute
onethird = 1.0D0/3.0D0 y = onethird*x
than a simple divide by 3 (if repeated maybe times). On modern cpus the difference is most likely minimal. Using this style might cost one bit of precision, but in most cases it is best to write code in the most readable way possible and how to do this has to be decided on a case by case basis.
Finally most compilers are now smart enough to evaluate constants such as sqrt(2.0D0) at compile time, so if you only use this constant once there is no point in declaring this value as a parameter. However, here you have to be careful to pass the correct type to sqrt(). The call sqrt(2.0) will return a single precision approximation to the square root of two.
See [http://www.ibiblio.org/pub/languages/fortran/ch2-2.html] for more information.