[concurrency-interest] forkjoin.ParallelArray and friends

Michael Hicks mwh at cs.umd.edu
Tue Aug 28 08:13:54 EDT 2007


> In fact, the differences between boxed and unboxed array
> operations on scalars are usually more extreme for parallel
> vs sequential use because of locality issues

Functional programming languages that use polymorphism have performed  
unboxing optimizations to good effect for quite a while now (e.g.,  
the simplest approach is to do a kind of in-lining or  
"monomorphization" of polymorphically-instantiated classes, which is,  
I believe, performed in the .NET runtime).  Assuming FJ becomes part  
of the assumed standard, should we not expect the VM to perform these  
optimizations for Java as well?  I know this doesn't answer your  
short-term need to provide classes that people can use now, but  
perhaps this addresses the longer-term API issue?

-Mike



More information about the Concurrency-interest mailing list