Equation 0.05 - 0.04 = 0.01 is not correct?! Help!!!

[ernestasju]

My code (Visual C++ Express 2008, CRT Console Application):

#include "iostream"

void main()
{
float rez = 0.05f - 0.04f;

printf_s("%f\n%i\n%i\n",
0.05f - 0.04f, // 0.010000, so that is 0.01
0.05f - 0.04f == 0.01f, // should be true, but...
0.05f - 0.04f == rez); // true
}

and I get this:

0.010000
0
1
Press any key to continue . . .

As you can see, my problem is 0.05f - 0.04f == 0.01f...
What project settings should I change to get correct results?

[newb16]

Because 0.05 - 0.04 is not exactly 0.01 due to rounding - internal representation is binary and not decimal you can see in source code. And if you need to check if float/double equals exactly to some other number, you are doing it wrong.

float rez = 0.05f - 0.04f;
float d1 = 0.01f;
int *pp;
pp = (int*)&rez;
printf("%x\n", *pp);
pp = (int*)&d1;
printf("%x\n", *pp);
 

if int is 32 bit, you can see that rez and d1 differ in some lowermost bit, although displayed with the same string.

[ernestasju]

Thank you.

Only one thing bad 1% more complex code...

You said that string representations will be equal... I'll use that...

[Banfa]

If you are using a fixed number of decimal places then most programmers use integers.

That is if I am holding values in the range -120.000 to 120.000 instead of using a float use an 32 bit integer and times the value by 1000 to hold values in the range -120000 to 120000. Do all you calculations in integers, being careful to take account of integer wrap round, just put in the decimal point for display to the user.

[donbock]

You should read What Every Computer Scientist Should Know About Floating-Point Arithmetic.

This and other helpful links are in the PLEASE READ FIRST thread.