Not dropped, but given we could not find a way to root out that issue we took the simple approach for 1.0, but the thought was that your solution should be better. We have three folks who can configure things such that they see problems, possibly Petri who has a new many core setup can see if he sees the issue.
What would also be good is to add a test to test/performance that highlights the performance penalty of the locks especially if you can ramp up to many cores in your set up with/without the patch. I will try it again and get back to you with what I currently see. On 2 March 2015 at 15:35, Geoffrey Blake <[email protected]> wrote: > Hi all, > > > > I see that Bill reverted the global buffer allocator to using normal > spin/ticketlocks in the run-up to releasing v1.0 of ODP, is the tagged > pointer patch going to be dropped for the time being? I’ll keep it in my > tree and resubmit if I see a performance case for it in the future on > larger multi-core machines if the locks start to break down. For Anders > and Mike, was there ever a root-cause to the segmentation faults seen? > > > > Thanks > > Geoff Blake > > _______________________________________________ > lng-odp mailing list > [email protected] > http://lists.linaro.org/mailman/listinfo/lng-odp > > -- *Mike Holmes* Linaro Sr Technical Manager LNG - ODP
_______________________________________________ lng-odp mailing list [email protected] http://lists.linaro.org/mailman/listinfo/lng-odp
