You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Right now, fusion in the CEESD branches is only based on loop length, but this can be insufficient. In the tensor product setting, the number of quadrature DOFs and interpolatory DOFs can be the same, but vectors from those two element groups should never be involved in the same expression, hence there won't ever be horizontal fusion.
Right now, fusion in the CEESD branches is only based on loop length, but this can be insufficient. In the tensor product setting, the number of quadrature DOFs and interpolatory DOFs can be the same, but vectors from those two element groups should never be involved in the same expression, hence there won't ever be horizontal fusion.
At the same time, the current fusion algorithm is quite eager to discard fusion candidates, which can wreck fusion if a non-fusible pair of inames is found.
Thus IMO it makes sense to distinguish axes more finely.
cc @majosm
The text was updated successfully, but these errors were encountered: