https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121818
--- Comment #3 from Benjamin Schulz <schulz.benjamin at googlemail dot com> --- Created attachment 62323 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62323&action=edit mathfunctions_mpi.h mathfunctions_mpi.h with omp atomics on shared variables of teams distribute parallel for simd loops on target from line 1610 ff that does at least not lead to nans but still wrong values, and strangely, different values than with the simple reduction statements instead of variable sharing and atomic updates, but that should yield the same results and it does not... For compilation, just replace the mathfunctions_mpi.h from this attachment with the file of the same name in the tar.gz archive that i attached....
