MPI-CUDA backend is broken
naoyam opened this issue · comments
Branch develop has fixed almost all regressions except for user-defined types with array members. See #16 for the remaining issue.
The current state of the develop branch is not optimized for performance. For example, all communications are performed synchronously without no overlapping. Still, it would be fine for small-scale runs. Performance optimizations will be addressed in the next iteration of development.