🚀 go-pugleaf

gcc version: 2.7.2
os version:  linux 2.0.18

-------------------------------------------------------------------------------
float fooFloat( float, float );
 
void main()
{
  float a, b, f;
 
  a = 3.14156;
  b = 0.123456789;
  f = 123126.266560;
 
  printf( "%f\n", fooFloat( a, b+123123 ) );
  f -= fooFloat( a, b+123123 );
  printf( "%10.15f\n", f );
}
 
float fooFloat( float a, float b )
{
#ifdef NO_BUG
  float f = a+b;
  return( f );
#else
  return( a+b );
#endif
}
-------------------------------------------------------------------------------

On my linux box, compiling like:

$ gcc foo.c
$ a.out

Gives a non zero result at the second printout

Compiling like:

$ gcc -DNO_BUG foo.c
$ a.out

Gives a zero result at the second printout.

Playing around with that and having a look at the generated assembly
language, it seems to me that GCC is, when NO_BUG is not defined,
returning a double (or rather keeping a double, thanks to the Pentium
FPU architecture) as the result of the evaluation of a+b. Explicitly
storing the result into a float makes the function return a float. It
doesn't work though, if you force the cast in the a+b return
statement.

I'd be grateful to receive some feedback about this submission.