Native compilation article

[Bryce McKinlay]

Nic Ferrier wrote:

>>Hi, 
>> 
>>perhaps you are interested in an article I just found: 
>> 
>>  http://www-106.ibm.com/developerworks/java/library/j-native.html?loc=j 
>>
>
>Ooohhh... what a suprise. The IBM jre wins every test.
>
>Who'd have thought it?
>

They used GCC 3.0 for their testing. The current cvs GCJ beats IBM and 
Sun on Scimark, and is in fact much faster if we cheat with 
--no-bounds-check.

The primes.java test used in the article is a bit tricky because they 
used "long" for all the calculation instead of "int", and GCC doesn't 
produce particularly efficient code for the 64-bit mod operation. After 
changing "long" to "int", gcj beats both IBM and Sun Linux JDKs. With 
"long" we are slightly faster than Sun, but a lot slower than IBM - so 
obviously their JIT must have better optimized "long" operations.

regards

Bryce.