On Fri, Jul 29, 2016 at 1:53 PM, David Knezevic <david.kneze...@akselos.com> wrote:
> On Fri, Jul 29, 2016 at 3:48 PM, John Peterson <jwpeter...@gmail.com> > wrote: > >> >> >> On Fri, Jul 29, 2016 at 12:54 PM, David Knezevic < >> david.kneze...@akselos.com> wrote: >> >>> Below is a short test case that causes a segfault with DenseMatrix::svd >>> with --enable-complex. I think that the issue is related to the fact that >>> LAPACK clears the "number_val" array, and then it gets cleared again when >>> it goes out of scope. I wasn't able to fix it so far, though (seems like >>> allocating the array with new would do the job, but that didn't seem to >>> work for me), so another pair of eyes on this would be appreciated. >>> >>> This is related to the PR 1042 that I just closed based on comments from >>> John and Roy (I agree that it'd be better to fix this than to require the >>> user to call the SVD themselves). >>> >> >> Thanks for the test case. Note that if you "new" memory and then Lapack >> "free's" it rather than deleting it, I think that's undefined behavior >> (?). So you might check whether malloc works any better for you. >> > > > OK, that might be the issue. If so, it should be an easy fix. I'll try > that when I get some time... > OK, I made an issue for this as well: https://github.com/libMesh/libmesh/issues/1043 -- John
------------------------------------------------------------------------------
_______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
