[Vm-dev] RoarVM: The Manycore SqueakVM

Bert Freudenberg bert at freudenbergs.de
Sun Nov 7 14:07:40 UTC 2010


On 07.11.2010, at 10:20, Stefan Marr wrote:

> Hi:
> 
> On 07 Nov 2010, at 08:06, Michael Haupt wrote:
> 
>> today's branch predictors seem to be too advanced to be impressed by threading.
> Just as a side note: the share of machines with CPUs that have awesome branch-predictors is likely to decline. With all those low-powered systems and with manycore systems, the cost of using the die space/your power envelope for branch-prediction, out-of-order execution, and the like does not seem to be worth it when compared to just spending those transistors on additional cores.
> So, when you stay with an interpreter, these kind of optimizations will remain interesting for sequential performance.

And given the current security paranoia which leads to certain platforms preventing jitting (memory pages are writable *or* executable, but bot both) then interpreters might have to stay around for quite some time ...

- Bert -



More information about the Vm-dev mailing list