Nan behaves differently in #pragma cpp

September 30, 2022

I am learning about the NaN datatype, so, I ran a code to understand it and it goes well, the code worked as expected, but when I add a line #pragma GCC optimize("Ofast") in my code then, it behaves unexpected. I am curious why??

#pragma GCC optimize("Ofast")
// C++ code to check for NaN exception
// using "==" operator
#include <cmath>
#include <iostream>
using namespace std;

// Driver Code
int main()
{
  float a = sqrt(2);
  float b = sqrt(-2);

  // Returns true, a is real number
  // prints "Its a real number"
  a == a ? cout << "Its a real number" << endl
    : cout << "Its NaN" << endl;

  // Returns false, b is complex number
  // prints "Its nan"
  b == b ? cout << "Its a real number" << endl
    : cout << "Its NaN" << endl;

  return 0;
}

The ouput without pragma is

Its a real number
Its NaN

But after using pragma it gives

Its a real number
Its a real number

>Solution :

-Ofast turns on -ffast-math, which turns on -ffinite-math-only, which does the following (from the gcc documentation here: https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html)

-ffinite-math-only

Allow optimizations for floating-point arithmetic that assume that
arguments and results are not NaNs or +-Infs.

This option is not turned on by any -O option since it can result in
incorrect output for programs that depend on an exact implementation
of IEEE or ISO rules/specifications for math functions. It may,
however, yield faster code for programs that do not require the
guarantees of these specifications.

I think the "-O option" part is wrong since the -Ofast documentation says it "disregards strict standards compliance". That, or by "any -O option" they mean -O1, -O2, -O3.