Help with performance issues.

[Andrew Haley]

Tom Tromey writes:
 > >>>>> "Andrew" == Andrew Haley <aph@redhat.com> writes:
 > 
 > >> I think we could consider doing this for any final method with fixed
 > >> semantics.  For instance, anything in java.lang.Math (taking strictfp
 > >> concerns into account, of course).  If other final methods are
 > >> important for performance, we could inline those too.  For instance,
 > >> there might be situations where inlining System.arraycopy makes sense.
 > 
 > Andrew> All this falls down, though, when we need to inline a *native*
 > Andrew> method.  That would be a good trick...
 > 
 > In the above case (final methods with fixed semantics), we can do
 > this.  That's because we aren't "inlining" -- we are recognizing the
 > name of the function and replacing the call with a builtin

Yeah.  I was having a flight of fantasy...

 > Inlining Java into C++ can be done much more easily.  We could replace
 > the trivial decompilation code in gcjh with a real decompiler.

Cool...  :-)

Andrew.