I support reluctance in fixing performance. AFAIK, no performance change should be done without demonstrating incerase on real loads.
2009/11/13 Tim Ellison <t.p.elli...@gmail.com>: > On 12/Nov/2009 19:20, Vijay Menon wrote: >> Right, writes - even unsynchronized ones - to the same cache line are >> problematic. Here's a good quick read on this: >> http://en.wikipedia.org/wiki/MESI_protocol. Recent Intel architectures >> adapt this protocol. Here's the interesting bit: >> >> A write may only be performed if the cache line is in the Modified or >> Exclusive state. If it is in the Shared state, all other cached copies must >> be invalidated first. This is typically done by a broadcast operation known >> as *Read For Ownership (RFO)*. >> >> >> If two cores continually write to the same cache line, they'll repeatedly >> invalidate each other. > > Thanks for the pointers. > > Given that the current code has been there since time immemorial and is > not broken, I'm inclined to leave it for now and pick this thread up > again after M12. That'll give us time to see what the difference is > between the local and static versions (though if I were a betting person > I'd say skip is used infrequently enough that nobody's app will notice). > > Regards, > Tim > -- With best regards / с наилучшими пожеланиями, Alexei Fedotov / Алексей Федотов, http://www.telecom-express.ru/ http://harmony.apache.org/ http://www.expressaas.com/ http://openmeetings.googlecode.com/
