Am 21.05.2008 19:33, Xueming Shen schrieb:
It is a surprise to see that the String.charAt is way faster than the
char[] index access...I had
the assumption that the array access is faster when considering to
move the mappable table
from the String storage to a simple 256 length char[] for both
singlebyte and doublebyte charsets.
Yes, it's surprising. I assume that there are some native machine code
bypasses in the VM for String processing, whereas char[] index access is
treated by the standard hotspot optimisation.
I'm also wondering how array initialization is compiled to byte code.
Compile
char[] chars = {'\u0000', '\u0001', '\u0002', ..., '\u00FF' };
... and compare it with
String charString = "\u0000+\u0001+\u0002+ ...+\u00FF";
by help of javap.
Ulf, Martin and I were (will continue to, hopely) working on a project
to improve the performance
of the charset implementation, your contribution will definitely
appreciated.
Thanks very much. Where can I find your project?
-Ulf
