Floor vs int cast difference

Question

I have a weird problem. Here's part of my code:

int temp=1100;
int foo=floor(0.03*temp);
int foo1=0.03*temp;
if(foo-foo1){
        cout<<foo<<endl;
        cout<<foo1<<endl;
}

If 3% of temp = integer then foo differs from foo1 by 1.
For example:
1100*0.03=33.
foo=33
foo1=32.
In addition if i write it like that :

int foo=floor(0.03*1100);
int foo1=0.03*1100;

There is no such problem.
Why?

Casting to `int` rounds toward zero, whereas `floor()` always rounds down (toward negative infinity). — Leon, Dec 25 '16 at 17:32
Which compiler are you using? Can't reproduce it here: http://coliru.stacked-crooked.com/a/2df015f881f60122 — alain, Dec 25 '16 at 17:37
Code blocks is not a compiler, it's an IDE. Please look in Settings->Compiler... — alain, Dec 25 '16 at 17:44
Now, instead of multiplying by .03, try dividing by 100, then multiply by three. You will now get same results. — Sam Varshavchik, Dec 25 '16 at 17:46
True. Same when i try different build configuration(release). — Mr Patience, Dec 25 '16 at 17:49
Could you please type `g++ --version` into a terminal, I think you must have a really old version. — alain, Dec 25 '16 at 17:49
Your problem has same with topic http://stackoverflow.com/questions/588004/is-floating-point-math-broken — nartoan, Dec 25 '16 at 18:07
Expecting binary floating point to work the same as textbook computations is bounds for automatic closure, as @SamVarshavchik pointed out. — PaulMcKenzie, Dec 25 '16 at 18:07

score 2 · Accepted Answer · answered Dec 25 '16 at 18:01

Floating point numbers have trouble rendering decimal places. It is more of an approximation.

2**-6 + 2**-7 + 2**-8 + 2**-9 + 2**-11 + 2**-13 + 2**-14 + 2**-15 + 2**-20 + 2**-22 + 2**-26 = 0.0299999863

Using binary, I drove the accuracy to 26 binary. I got close to 0.03 but not quite. In my example I chose to be under 0.03 but I could have gone a little above (i.e. 0.03000001. I don't think it is possible to represent 0.03 perfectly in floating point notation.

Multiplying 0.03 by any number produces yet another approximation. Casting to type int will cut out everything after the decimal place. I assume the implementation of floor is more elegant. Your compiler probably choose a floating point value of 32.99999 so and int would make 32.

Floor vs int cast difference

1 Answers1