C251: ERRORS IN DOUBLE ARITHMETIC

Information in this article applies to:

• C251 Version 3.50 or before

SYMPTOMS

The following C251 program does not produce correct double precision results.

#pragma FLOAT64

double v1 = 4.0;
double v2 = 6.0;
double x1, x2;

void double_error (void) {
  x1 = v1/v2;          // result: 0.666666666666666666
  x2 = x1 + x1 + x1;   // wrong result: 1.0 should be 2.0
  x2 = x1 * v2;        // wrong result: 2.0 should be 4.0
}

CAUSE

The C251 Compiler V3.50 and earlier contains a potential problem in the run-time library which may cause this effect. It only occurs when the end result is exactly a power of 2 (i.e. 0.25. 0.5, 1.0, 2.0, 4.0, 8.0). It occurs while rounding-up the final result and causes the result to be off by a factor of 2.

The problem does not exist when you use constants that are calculated at the compiler level by the constant folding optimization.

RESOLUTION

This problem is resolved in C251 Compiler V3.51 and available in C251 V3.53a.

SEE ALSO

• GENERAL: PRECISION OF PRINTF %F FORMAT STRING

Last Reviewed: Friday, May 20, 2005