inlining methods

Tue Oct 26 15:06:00 GMT 1999

Under what circumstances, if any, should gcj be inlining Java methods? 
I'd expect some `private' and/or `final' methods to be inlined given the
appropriate optimization level, but that doesn't seem to be happening
(using gcj 2.96).
Right now I'm testing a JDBC driver (the FreeTDS driver, compiled native
under gcj) and hitting a performance wall. It runs large queries about
half as fast as the IBM JDK.
One issue I'm aware of is that libgcj's object allocation runs about 50%
slower than the JDK (using boehm-gc). Another is that the IBM JIT
inlines methods aggressively (or so they claim). Both of these are
likely to affect my code. I use a lot of private methods in lieu of C++
inlines or C macros.
-- 
Jeff Sturm
jsturm@sigma6.com