[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

comparison between native gcj and bytecode



>>>> Sun seem to have aquired the leading thinkers behind the Self VM, and
have only just begun
applying these techniques to their JVM to get the hotspot JVM.

In as much as these techniques do not necessarily require JIT, they can be
applied to AOT as well, and would not change the balance.

But then again, some information can only be acquired while the application
is running, favouring JIT optimization, while quite a few optimization
techniques require large amounts of processing power, such as escape
analysis, favouring AOT optimization. Further, there is only limited scope
for collecting and processing JIT runtime information, before the processing
power required by these activities start swamping any performance gains from
them.

The question remains whether virtually unlimited amounts of processing time,
ahead of time, can be beaten by having this additional information at run
time with actually few opportunities to draw conclusions from it?

This problem has the touch and feel of the Heisenberg principle of
uncertainty. Maybe the person who manages to equate this problem
theoretically, is ready for a page in the history books :)