[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: goals for fc4?

>>>>> "Chip" == Chip Turner <cturner redhat com> writes:

Chip> I think at best the jury is still out on how fast gcj is
Chip> vs a hotspot-optimizing jvm for "real world" use.

Best performance depends on a lot of factors, not just your
application but also how you plan to deploy it.  For instance, using
gcj to build shared libraries may be a winner when you expect to be
reusing that library in several running processes, even if the
generated code looks marginally slower.

gcj and hotspot-like JITs have different optimization opportunities
available to them.  Another factor is how you compile your code with
gcj; the new binary compatibility ABI provides flexibility at the cost
of some performance.  The usual answer is "try your program and see".

On smaller-than-real-applications benchmarks, gcj fares pretty well
against the best JITs (especially considering how few gcj-specific
optimizations we've added to gcc, e.g. we still haven't implemented
array bounds check removal).  The 1.5 release from Sun looks like it
pulled ahead a little bit, though you can still construct tests where
gcj does better.  I would expect some bounce from gcc in the 4.1 time
frame, though, as more high-level optimizations are written.

As a practical matter touching on Fedora, gcj remains the "best" way
to deliver java applications on a free system.  In my opinion, other
free VMs have unacceptable performance, unacceptable platform
coverage, or both.


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]