Avoid function call with JVMPI enabled

Bryce McKinlay mckinlay@redhat.com
Fri Aug 27 21:10:00 GMT 2004

Tom Tromey wrote:

>Hans> Presumably you can still move the size computation out of
>Hans> loops if you pass it as a parameter?  Similarly for the finalizer
>Hans> lookup?
>One option would be to have an atable slot representing `new of class
>X'.  We could then point this at AllocObjectNoFinalizer when resolving
>the slot.
Thats an interesting idea. That would allow us to swap in a direct call 
to the debug allocation routines, too.

However, one thing I worry about with using indirect dispatch for 
everything, though, is that it makes the compiler's asm output a whole 
lot less readable. A solution for that might be to allow the name of the 
target decl of the call to persist through gimplification somehow, so 
that -fdebug-asm can print the name of the function it thinks its 
calling alongside the call instruction.


More information about the Java mailing list