"Bruce Momjian" <pgman@candle.pha.pa.us> wrote
>
> Are you willing to say that we should always prefer pgport over glibc's
> qsort()?
>
At least for Linux and windows. My test is performed on a dataset ranges
from 10 to 15000000 elements. Each elements contains a 64 bytes garbage
character area and an integer key, which is uniformly distributed from 1 to
RANGE. RANGE takes values from 2 to 2^31. In all cases, our qsort absolutely
wins. Maybe skewed distribution should be tested?
Another interesting thing is that the qsort on RANGE=2 or other small number
in windows is terriblly slow - our version does not have this problem.
The test code could be found here (Note: it mixed with some other
experiements I am doing but might be a good start point to construct your
own tests):
http://www.cs.toronto.edu/~zhouqq/sort.c
Regards,
Qingqing