Strictfp returns different results on different systems

Question

I'm using the following method on Windows 8 (Intel Atom Z3775):

public static strictfp void main(String args[])
{
    float a = 0.000000002f;
    float b = 90000300000000004f;
    float c = a * b;
    System.out.println(c);
}

which gives me: 1.80000592E8

When running

public strictfp void calculate()
{
    float a = 0.000000002f;
    float b = 90000300000000004f;
    float c = a * b;
    Toast.makeText(getApplicationContext(), c + "", Toast.LENGTH_LONG).show();
}

i get: 1.8000059E8

Why do they differ? Am I using strictfp wrong?

What happens if you print `String.format("%08x", Float.floatToIntBits(c))` instead in both cases? — Andy Turner, Apr 14 '16 at 12:25
They return exactly the same string. Using Float.toString(c) to convert to String on both plaforms doesn't help. — Geosearchef, Apr 14 '16 at 12:33
Right - so the actual numbers are bitwise-identical. There is simply a difference in the implementation of the algorithm for turning the float into a string. — Andy Turner, Apr 14 '16 at 12:34

score 1 · Accepted Answer · answered Apr 14 '16 at 12:40

You have confirmed that printing the hex representation of the floats shows that the values are the same:

String.format("%08x", Float.floatToIntBits(c))

This means that there is a difference in the implementation of how a float is converted to a String.

Notice that the OpenJDK implementation of java.lang.Float.toString(float) invokes a method from sun.misc.FloatingDecimal - i.e. it is a JDK-specific implementation.

You'd need either:

To ensure that you always run the code using the same implementation
To provide your own implementation which always produces the same result
To specify a shorter format, so that the strings are the same to that many decimal places.

score 0 · Answer 2 · answered Apr 14 '16 at 12:24

0

Well, the displaying method certainly differs...

answered Apr 14 '16 at 12:24

Adrian Roman

534
4
8

And that changes what? I now added a '+ " "' to System.out.println() and now both floats are first converted to a string before being given to the method. – Geosearchef Apr 14 '16 at 12:26
Obviously the float to string conversion also can differ. – Adrian Roman Apr 14 '16 at 12:27
2

@user2912528 you'd have to implement the float to string conversion yourself. – Andy Turner Apr 14 '16 at 12:35

Strictfp returns different results on different systems

2 Answers2