jsantala at tml.hut.fi
Mon May 27 05:36:15 PDT 2002
On Sun, 26 May 2002, Archie Cobbs wrote:
> Jukka Santala writes:
> > How does this compare to a native method call? I've been under the
> > impression that JNI calls carry significant overhead, so would not
> > neccessarily be efficient for simple functions. On the other hand we have
> There is also the "kaffe native" method call which is much faster
> than JNI but kaffe-specific.
I thought of that, but seeing as the "kaffe native" is undocumented and
nonstandard and there aren't even all that many examples of it being used
in Kaffe, either somebody more familiar with the internals would have to
do it, or then I'll need to learn how to do it, which isn't worth holding
your breath for, considering I'm not paid to develop Kaffe :)
On the other hand, I found out that according to the "-prof" flat profile
generated by Kaffe, the patch I posted changes the average runtime of the
find(Object) method on X-Smiles browser from 7.3 microseconds to 5.3
microseconds, on this weird 800Mhz Duron running RedHat. This would seem
to indicate about one microsecond per in-class calls for thsi very
unofficial test. Unfortunately the Xprofiling output was less conclusive,
giving some weird artifacts.
At any case 25% speedup is enough for me for now (Altough for the public
get() method it will be less), there are bigger problems to deal with.
More information about the kaffe