gcj performance on Solaris 2.6

[Kresten Krab Thorup]

Andrew Haley wrote:

>Jeff Sturm writes:
> > On Wed, 25 Jul 2001, Cedric Berger wrote:
> > > Are you running hotspot (the default in 1.3)?> 
> > I don't think so.  1.2.2 is what was known as the "Solaris
> > Production VM".  It has a JIT and generational GC, however it is
> > unrelated to Hotspot AFAIK.
>
>I thought hotspot only optimized if you called th same function many
>times.  That's not happening here.
>
The "Solaris production VM" is actually rather advanced, with dynamic 
compilation and exact garbage collection (it was also known under the 
name, ExactVM).  At some point however, it was decided to go with 
HotSpot, because it was easier to maintain from what I hear.  For 
several cases it was more efficient that hotspot, and there was a heavy 
debate inside Sun on which VM to go forward with.  Ole Agesen (now with 
VM Ware) was in charge of the ExactVM effort, while Lars Bak was in 
charge of HotSpot.  Several of the optimizations now used in hotspot was 
invented by the ExactVM team, e.g. techniques for on-stack code 
replacement; and efficient synchronization.

HotSpot can do such things as in-place loop unrolling, while the loop is 
being run.  

Kresten