Derby benchmark inside SPECjvm2007 has some hot code path like below:

private int getTimeIndex(long time) {
            int res1 = 0;
            long tmp = time;
            for (int i = 0; i < 20; i ++) {
                res1 += tmp % 10;
                tmp = tmp / 10;
            }
            return (int)(res1 % 15);
        }

collector and analyzer show about it takes  4.54% of user CPU time on Solaris x86. The current c2 compiler only optimizes the integer division, but not for the long division case according to Vladimir. This partially explains why we have 3 times division instruction than some of our competitors.
Posted Date : 2007-09-10 23:42:46.0

N/A

Rework the div by constant code in divnode.cpp to support longs as well as
to consistent across longs and ints.
Posted Date : 2008-03-14 15:46:56.0