On Fri, 25 Mar 2016, Salazar De Troya, Miguel wrote: > Is there any reason why PatchRecoveryError calls a threaded loop > over the elements while JumpErrorEstimator does not?
I don't recall, but my guess would be: Patch recovery is more expensive, so I bothered to thread it. I'm lazy, so I stopped there. > If I were to create a FEMContext for PatchRecoveryError and > initialize it before the threaded loop in > PatchRecoveryError::estimate_error(), could I have problems with > race conditions? Definitely. You want a FEMContext per-thread. > I cannot find a way to ensure that the patches do not contain > elements in other threads, and if we were to reuse the patches, that > would be assigning different error_estimators for the same element. We set the error_per_cell values, we don't increment them, and we wrap that setting in a scoped_lock to be doubly safe, so you should get a valid error estimate on each element. It won't be deterministic, though; two different runs of the same problem could pull an element's error estimate from two different patches depending on which thread "got there first". --- Roy ------------------------------------------------------------------------------ Transform Data into Opportunity. Accelerate data analysis in your applications with Intel Data Analytics Acceleration Library. Click to learn more. http://pubads.g.doubleclick.net/gampad/clk?id=278785351&iu=/4140 _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
