class initialization check overhead

Andrew Haley aph@redhat.com
Fri Oct 25 06:10:00 GMT 2002


Jeff Sturm writes:
 > On 24 Oct 2002, Adam Megacz wrote:
 > > I assume this means inlining across .java files when they are all
 > > compiled in a single invocation of jc1?
 > 
 > Yes.  That, together with -fprofile-arcs/-fbranch-probabilities yields a
 > very favorable improvement on my code.  I also tuned down the inliner
 > params a little.
 > 
 > > Care to share the patch? =)
 > 
 > I have many patches lumped together at the moment.   Soon I'll have to see
 > which are suitable for submission.
 > 
 > (The inlining patch is not, it has serious bugs, like omitting
 > `synchronized' 

There's other problems too.

Scope and variable info for debugging is generated during Java
expand_expr.  That means that it's impossible to get correct debugging
info until a method has been expanded -- and it can't be inlined until
then either.

 > and the class init test.)

That should be fine now.

Andrew.



More information about the Java mailing list