Simple Math Error in gfortran?

Question

I am a beginner with regards to programming in Fortran. I am currently using cygwin and the gfortran compiler running on my Windows XP PC. I'm having difficulty with some simple math - the program I wrote simply won't do the math. The code is:

program convert
real t

t=0
t=8320671.25 - 8000000.00

write(*,*) t
end

The program should give me the answer "320671.25" but gives me 320671.00 instead! What am I doing wrong?

Answer 1

You're running into the single precision limit. The result of the subtraction is a real*4, and can be safely stored in t. The values used in the subtraction, however, are outside the real*4 range. Outside in the range of precision, with the result that the numbers are rounded to fit into a real*4 before calculating the subtraction.

Try this for example:

program convert
real t

t=0
t=8320671.25_8 - 8000000.00_8

write(*,*) t
end

The appended _8 ensures the two numbers are double precision; the result is then converted to real before assigned to t, but the calculation is now in double precision, and the .25 is "saved" in the subtraction.