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
When one uses buffering in a dummy Operator with a parallel time dimension there is a race condition due to OpenMP-izing the time loop.
Reproducer:
grid = Grid(shape=(2, 2))
time = grid.time_dim
x, y = grid.dimensions
usave = TimeFunction(name='usave', grid=grid, save=10)
v = TimeFunction(name='v', grid=grid)
eq = [Eq(v[time, x, y], usave)]
op = Operator(eq, opt='buffering')
trees = retrieve_iteration_tree(op)
assert len(trees) == 2
assert trees[1].root.dim is time
assert not trees[1].root.is_Parallel
assert trees[1].root.is_Sequential # Obv
The issue is that buffering creates a storage-related dependence and what was previously discovered to be PARALLEL should now be downgraded to SEQUENTIAL
The text was updated successfully, but these errors were encountered:
When one uses buffering in a dummy Operator with a parallel time dimension there is a race condition due to OpenMP-izing the time loop.
Reproducer:
The issue is that buffering creates a storage-related dependence and what was previously discovered to be PARALLEL should now be downgraded to SEQUENTIAL
The text was updated successfully, but these errors were encountered: