Fibonacci and performance

Anthony Green
Fri Apr 27 10:17:00 GMT 2001

Per wrote:
> What I forgot was that we do the Jv_InitClass at the start of the
> called method, not at the call site.  This saves code space, but
> it makes it harder to optimize the call away when it is unneeded.
> We might re-consider this decision if/when we get smarter about
> initialization.

Andrew Haley's suggestion from a while back was to implement two
entry-points into a method (with and without _Jv_InitClass).

I don't know if GCC's infrastructure supports this yet.


More information about the Java mailing list