Rounding off floating point values

Question

While rounding off the floating point values I observed some discrepancy in values. I have extracted following part of code. Here if variable var_d is assigned value> 5.3 to then I am getting proper values for variable var_d, but for values like 5.01 and 5.02 I am getting 500 and 501 respectively.

#include<stdio.h>
int main()
{
double var_d=5.02;
long var_l;
var_l = (double)(var_d*100);
printf("var_d : %f  var_l= %ld\n ",var_d,var_l);
printf("result : %ld\n",var_l);
return 0;
}

Rounding up floating point values is bound to a certain degree of unpredictability. Fixed point arithmetics should be used instead. — Alex, May 16 '13 at 11:59
possible duplicate of [Floating point inaccuracy examples](http://stackoverflow.com/questions/2100490/floating-point-inaccuracy-examples) — AndersK, May 16 '13 at 12:01
@Alex This has much more to do with the choice of a base than with the choice of fixed/floating point. Binary floating-point fails at these examples. With reasonable parameters, decimal floating-point gets these examples right. Binary fixed-point fails at these or similar examples. With reasonable parameters, decimal fixed-point gets these examples right. — Pascal Cuoq, May 22 '13 at 09:43
Indeed it would be nice if C had BCD (binary coded decimal) float types, even if they had to be done in software. — Adrian Ratnapala, Jun 13 '13 at 14:59

Adrian Ratnapala · Accepted Answer · 2013-05-16T12:09:15.773

2

Use

double var_d=5.02;
long var_l = rint(var_d*100);

Since 100 * 5.02 is not not exactly equal to 502, you are getting a rounded down.

To be clearer: 5.02 has no exact representation in binary floating point. Thus var_l*100 == 5.02*100 is not exactly 502. In fact it is closer to 501.99999999999994. When you cast it to an integer, this is rounded down to 501.

edited May 16 '13 at 12:09

answered May 16 '13 at 12:01

Adrian Ratnapala

5,485
2
29
39

score -1 · Answer 2 · answered May 16 '13 at 12:22

-1

float var_d=5.02;
long var_l;
var_l = (long)(float)(var_d*100);

This should works properly

answered May 16 '13 at 12:22

I tried it but it does not work, Instead I used following var_l = (float)(var_d*100) + 1e-9 ; which is working for now. – aashoo May 17 '13 at 05:35

Rounding off floating point values

2 Answers2