Closed psakievich closed 7 months ago
After getting into the issues with Ilker and Wyatt, this change does get at the problem. However, to fix the iblank array, it need to be "parallel_min" instead of "parallel_max", because the value of -1 needs to trump the other values. Though this fixes the iblank array for nalu-wind, the data pathways still don't exist for the nodes whose iblank values have been corrected. This leaves them uninitialized, preventing nalu-wind from going forward (the code crashes).
We concluded that the underlying issue isn't this part of nalu-wind, but it is in TIOGA itself (it should get the same answer for a node that is shared by two different chunks). Because of that, it makes the most sense to address this bug through TIOGA directly, and Ilker has some ideas for doing that.
I am not sure update shared is the right operation. It seems like if a cell is detected as fringe anywhere it should be fringe even if it is not the owner. In this case
parallel_max
makes more sense to me. I have not tested this.This could explain the behavior https://github.com/Exawind/exawind-driver/issues/65