Elliotte Harold wrote:
> derek denny-brown wrote:
>
>> Except that we are talking about the performance of XML parsers, and
>> XML is all about string processing and Java string processing is slow.
>> Java XML parsing will never be faster than a good XML parser written
>> in C.  There is just too much overhead, and C benefits from 'struct's,
>> the lack of which hinders ones ability to write certain constructs
>> efficiently in Java.   
>
> My god! Are we moving back on topic? Has this ever happened before? `-)
Until Java gets some functionality (whether some WORA based annotations 
or smarter compilers) to give or generate the equivalent of C++'s 
"intrinsic functions", it is necessarily slower than optimized C++. It 
simply will not use the pipelining capabilities of modern CPUs well.

I had an interesting email from Rob Cameron just this week:
-----------------------------------------------------------------------------------------

You wrote an interesting article on using SIMD intrinsics
for improving UTF-8 to UTF-16 transcoding performance.

Using C++ Intrinsic Functions for Pipelined Text Processing
November 7, 2005

We have been developing a much more extensive application
of SIMD techniques to XML/Unicode/character processing and
have considered the UTF-8 to UTF-16 problem in depth.   Our
methods allow us to speed up the handling of non-ASCII code
units as well as the ASCII ones.    Our performance improvements
on Intel are typically 4X (no ASCII optimization) to 25X (all ASCII),
while Power PC speed ups using the more capable Altivec
SIMD facilities are at least 8X.

Open source software applying these techniques is available
from our development website u8u16.costar.sfu.ca.

-----------------------------------------------------------------------------------------

For some reason, our LAN cannot access any systems at Simon Fraser 
University, so I have not been able to look further into the u8u16 
software.

I welcome the advent of "Efficient" XML, because some competition may 
stimulate the XML parser market, and give more impetus for research 
techniques (such as the IBM parser experiment that Noah has written 
about) to get folded into production code.

Cheers
Rick Jelliffe