How do you force a computed float value to be stored in a variable?

[DWilli3501]

I have two short programs that use floating point numbers. The first gives an incorrect result. By simply placing a "cout" (or any executable line) in the correct location of the program I get a correct result. The first program is:

/* floor example */
#include <iostream>
#include <math.h>
 
int main ()
{
    float a=10.0f;
    float b=0.9f;
 
    float c=a*b;
    float d=floor(c);
    std::cout << "floor of " << c << " is " << d << std::endl;
 
    return 0;
}
 
 

which outputs:
floor of 9 is 8

The second program is

/* floor example */
#include <iostream>
#include <math.h>
 
int main ()
{
    float a=10.0f;
    float b=0.9f;
 
    float c=a*b;
    std::cout << "";  //inserted line of executable code
    float d=floor(c);
    std::cout << "floor of " << c << " is " << d << std::endl;
 
    return 0;
}
 
 

which outputs:
floor of 9 is 9

I am using MSVC++ 6.0 on an XP workstation. (SP2 for XP, and SP6 for VC++) I understand how floating point numbers are represented in the computer. I also know that the representation of c in the first program is 8.999... and in the second program the representation is 9.000. I think that due to compiler optimizations, the value of c in the first program is not being rounded correctly and stored, but the register containg the result of a*b is being used directly without rounding. In the second program the inserted line of executable code seems to cause the value stored in c to be correctly rounded.

Is this a bug in the compiler, and if so is there a fix?

[Banfa]

Is this a bug in the compiler, and if so is there a fix?

Upgrade to VS 2005 my VS 2005 Express Edition does not seem to exhibit this behaviour the result is always

floor of 9 is 9

This is probably classifiable as a feature rather than a bug. All bets are kind of off if you are going to perform floating point arithmetic.

[weaknessforcats]

Keep in mind that in C floor() takes a double argument and returns a double. Conversion back to float can cause truncation and possible loss of data. Like maybe what you are seeing.

Do not mix float an double. Unless you have a technical reason that you can write down on paper, use double throughout. float is an obsolete holdover ffrom the days of computers with small memories.

[JosAH]

What every computer scientist should know about floating-point arithmetic.

kind regards,

Jos