Different precision in C++ and Fortran

问题:

For a project I'm working on I've coded in C++ a very simple function :

Fne(x) = 0.124*x*x, the problem is when i compute the value of the function

for x = 3.8938458092314270 with both Fortran 77 and C++ languages , i got different precison.

For Fortran I got Fne(x) = 1.8800923323458316 and for C++i got Fne(x) = 1.8800923630725743. For both languages, the Fne function is coded for double precision values, and return also double precision values.

C++ code:

double FNe(double X) {
    double FNe_out;
    FNe_out = 0.124*pow(X,2.0);
    return FNe_out;
}

Fortran code:

  real*8 function FNe(X)
  implicit real*8 (a-h,o-z)
  FNe = 0.124*X*X
  return
  end

Can you please help me to find where this difference is from?

回答1:

One source of difference is the default treatment, by C++ and by Fortran, of literal constants such as your 0.124. By default Fortran will regard this as a single-precision floating-point number (on almost any computer and compiler combination that you are likely to use), while C++ will regard it as a double-precision f-p number.

In Fortran you can specify the kind of a f-p number (or any other intrinsic numeric constant for that matter and absent any compiler options to change the most-likely default behaviour) by suffixing the kind-selector like this

0.124_8

Try that, see what results.

Oh, and while I'm writing, why are you writing Fortran like it was 1977 ? And to all the other Fortran experts hereabouts, yes, I know that *8 and _8 are not best practice, but I haven't the time at the moment to expand on all that.

回答2:

As High Performance Mark pointed out, the default precision of literals is the issue. Using

double xx = 3.8938458092314270;
std::cout << std::setprecision(16);
std::cout << " (float) * x*x: " << 0.124f*xx*xx << std::endl;
std::cout << "(double) * x*x: " << 0.124*xx*xx << std::endl;

We get

 (float) * x*x: 1.880092332345832
(double) * x*x: 1.880092363072574

which is the same difference you noticed.