Floating point comparisons with NaN values produce wrong results

[Olli Saarela]

Stephen L Moshier wrote:
> 
> > static const char double_q_nan_bytes_be[8] = {0x7f, 0xf8, 0, 0, 0, 0,
> 0, 0};
> 
> > #define double_q_nan  (*(const double *)double_q_nan_bytes_be)
> 
> This is not valid C language.  You need to fix your test case.

A fixed version is included below. This doesn't affect the results, the
inequality comparisons (< > <= >=) still give wrong results.

#include <stdio.h>
#include <math.h>

union DoubleInit {
  char   bytes[8];
  double f;
};

/* Quiet NaN : depends on byte order */

const union DoubleInit double_q_nan_bytes_le = {{0, 0, 0, 0, 0, 0, 0xf8,
0x7f}};
const union DoubleInit double_q_nan_bytes_be = {{0x7f, 0xf8, 0, 0, 0, 0,
0, 0}};

#define double_q_nan  double_q_nan_bytes_be.f

void compare(double x, double y) {
  printf("%10f <  %10f %s\n", x, y, (x <  y) ? "TRUE" : "FALSE");
  printf("%10f <= %10f %s\n", x, y, (x <= y) ? "TRUE" : "FALSE");
  printf("%10f >  %10f %s\n", x, y, (x >  y) ? "TRUE" : "FALSE");
  printf("%10f >= %10f %s\n", x, y, (x >= y) ? "TRUE" : "FALSE");
  printf("%10f == %10f %s\n", x, y, (x == y) ? "TRUE" : "FALSE");
  printf("%10f != %10f %s\n", x, y, (x != y) ? "TRUE" : "FALSE");
  printf("%10f == %10f %s\n", x, 0.0, (x == 0.0) ? "TRUE" : "FALSE");
  printf("%10f != %10f %s\n", x, 0.0, (x != 0.0) ? "TRUE" : "FALSE");
  printf("\n");
}

int main(int argc, char *argv[]) {
  compare(1.0, double_q_nan);
  compare(double_q_nan, 1.0);
  compare(double_q_nan, double_q_nan);
  return 0;
}

-- 
Olli Saarela, KCL - The Finnish Pulp and Paper Research Institute
Olli.Saarela@kcl.fi                tel. +358-9-4371538 (office)
Tekniikantie 2, Espoo-Otaniemi     fax. +358-9-464305
P.O. Box 70, FIN-02151 Espoo, Finland