BUG: ifort 12.0.4 Strange behavior for constans and double precision

[Kamil_Kie]

[fortran]program prec double precision :: a,b,x,y a=1./3. x=1. y=3. b=x/y print *,"kind=",kind(a), " val=",a print *,"kind=",kind(b), " val=",b if(a==b) then print *,"OK!" else print *,"Why 'a' and 'b' are different?!" end if end program[/fortran]

We compile this by:

ifort prec.f90

Result:

kind= 8 val= 0.333333343267441

kind= 8 val= 0.333333333333333

Why 'a' and 'b' are different?!

Why such a misleading behavior?

[mecej4]

Please read a few Fortran/programming-related text books and manuals. There is nothing unusual or buggy here at all.

The right-hand sides of lines 5, 6 and 7 are single-precision expressions. In particular, the expression 1./3. is single precision, which implies that it is correct to about seven decimal digits. Assigning such values to double precision variables cannot restore lost precision.

The values assigned to x and y, however, happen to have values (small integers) that are exactly representable in single or double precision. Therefore, the expression x/y is calculated in double precision and the quotient is the best that double precision is capable of delivering.