Hi Yaron, > I am trying to use ViennaCL with Ceemple and encountered a possible bug. > > In amg_base.hpp:1045, iter2 is decremented (code below). However, > nothing (at least locally in this function) prevents iter2 from > being pointlist.begin() and then the result is undefined, see > > http://stackoverflow.com/questions/18225651/is-stdvectorbegin-1-undefined > > Also, why are both iter and iter2 required? it seems iter alone could do > the job.
thanks for reporting. From what I can tell, you are totally right. Unfortunately, the student who contributed this code moved to the private sector and is no longer available, so I can't tell whether there are any invisible guarantees on the set. The AMG code is subject to significant refactoring, I added your message to the associated issue: https://github.com/viennacl/viennacl-dev/issues/16 Do you happen to have a test case where this particular piece of code triggers an error? A simple fix would indeed be to just remove the use of iter2, which is probably not a significant performance hit anyway. Thanks again and best regards, Karli ------------------------------------------------------------------------------ _______________________________________________ ViennaCL-devel mailing list ViennaCL-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/viennacl-devel
