compiler: fix halo placement for non out dimm exchange

Author: mloubout

devito/ir/stree/algorithms.py

@@ -197,8 +197,18 @@ def preprocess(clusters, options=None, **kwargs):
                 diff = dims - distributed_aindices
                 intersection = dims & distributed_aindices
 
+                # Does the cluster writes into the halo scheme
+                halo_write = False
+                for f, w in c.scope.writes.items():
+                    loc_vals = c1.halo_scheme.loc_values
+                    if f in c1.halo_scheme.fmapper:
+                        for a in [wi.access for wi in w]:
+                            if set(a.indices) & loc_vals:
+                                halo_write = True
+                                break
+
                 if all(c1.guards.get(d) == c.guards.get(d) for d in diff) and \
-                   len(intersection) > 0:
+                   len(intersection) > 0 and not halo_write:
                     found.append(c1)
                     queue.remove(c1)
 

tests/test_mpi.py

@@ -2203,6 +2203,24 @@ def test_lift_halo_update_outside_distributed(self, mode):
         halo_update = tloop.nodes[0].body[0].body[0].body[0]
         assert isinstance(halo_update, HaloUpdateList)
 
+    @pytest.mark.parallel(mode=4)
+    def test_halo_inner_dim(self, mode):
+        grid = Grid((11, 11, 11))
+
+        np.random.seed(0)
+        v = TimeFunction(name="v", grid=grid, space_order=4,
+                         time_order=1, save=Buffer(1))
+        v.data[:] = np.random.randn(*grid.shape)
+        e = TimeFunction(name="dummy", grid=grid, space_order=4, time_order=0)
+
+        eq = [Eq(v.forward, v + 1), Eq(e, v.forward.dydz)]
+
+        op = Operator(eq)
+        assert_structure(op, ['txyz', 't', 'txyz', 'txyz'], 'txyzxyzz')
+        op(time=100)
+
+        assert np.isclose(norm(e), 23484.863, rtol=0, atol=1e-1)
+
 
 class TestOperatorAdvanced: