FULL PRODUCT VERSION :


A DESCRIPTION OF THE PROBLEM :
The Math.exp() function is extremely slow for some exponents.

STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
Run the applied test case. Change the base from 500 to 800 and see that the code is 1000x slower.

EXPECTED VERSUS ACTUAL BEHAVIOR :
EXPECTED -
The time should be roughly the same. It is the same when running a similar C program using the libm library.

REPRODUCIBILITY :
This bug can be reproduced always.

---------- BEGIN SOURCE ----------
import java.util.concurrent.TimeUnit;

public class MathExpTest {
    private static final int BASE = 500; // change to 800 and 1000x slower

    public static void main(String[] args){
        for(int i=0;i<=10;i++) {
            long start = System.nanoTime();
            long total = 0;
            for(int j=0;j<10000000;j++) {
                total+=testExp(i+BASE);
            }
            long diff= TimeUnit.NANOSECONDS.toMicros(System.nanoTime()-start);
            System.out.printf("time is %f\n",diff/10000000.0);
            System.out.println("total is "+total);
        }
    }

   static int testExp(int i) {
        return (int) (Math.sqrt(Math.exp(i)));
    }
}

---------- END SOURCE ----------

CUSTOMER SUBMITTED WORKAROUND :
write our own exp function, which is slower in most cases