Merge pull request #2379 from devitocodes/sparse-gather

mpi: Fix data_gather  for sparse functions

devito/mpi/distributed.py

@@ -481,6 +481,20 @@ def decompose(cls, npoint, distributor):
             raise TypeError('Need `npoint` int or tuple argument')
         return tuple(glb_npoint)
 
+    @cached_property
+    def all_ranges(self):
+        """The global ranges of all MPI ranks."""
+        ret = []
+        for i in self.decomposition[0]:
+            # i might be empty if there is less receivers than rank such as for a
+            # point source
+            try:
+                ret.append(EnrichedTuple(range(min(i), max(i) + 1),
+                                         getters=self.dimensions))
+            except ValueError:
+                ret.append(EnrichedTuple(range(0, 0), getters=self.dimensions))
+        return tuple(ret)
+
     @property
     def distributor(self):
         return self._distributor

tests/test_data.py

@@ -3,7 +3,8 @@
 
 from devito import (Grid, Function, TimeFunction, SparseTimeFunction, Dimension, # noqa
                     Eq, Operator, ALLOC_GUARD, ALLOC_ALIGNED, configuration,
-                    switchconfig)
+                    switchconfig, SparseFunction, PrecomputedSparseFunction,
+                    PrecomputedSparseTimeFunction)
 from devito.data import LEFT, RIGHT, Decomposition, loc_data_idx, convert_index
 from devito.tools import as_tuple
 from devito.types import Scalar
@@ -1485,6 +1486,32 @@ def test_gather_time_function(self, mode):
         else:
             assert ans == np.array(None)
 
+    @pytest.mark.parallel(mode=[4, 6])
+    @pytest.mark.parametrize('sfunc', [SparseFunction,
+                                       SparseTimeFunction,
+                                       PrecomputedSparseFunction,
+                                       PrecomputedSparseTimeFunction])
+    @pytest.mark.parametrize('target_rank', [0, 2])
+    def test_gather_sparse(self, mode, sfunc, target_rank):
+        grid = Grid((11, 11))
+        myrank = grid._distributor.comm.Get_rank()
+        nt = 10
+        coords = [[0, 0], [0, .25], [0, .75], [0, 1]]
+        s = sfunc(name='s', grid=grid, npoint=4, r=4, nt=nt, coordinates=coords)
+
+        np.random.seed(1234)
+        try:
+            a = np.random.rand(s.nt, s.npoint_global)
+        except AttributeError:
+            a = np.random.rand(s.npoint_global,)
+
+        s.data[:] = a
+        out = s.data_gather(rank=target_rank)
+        if myrank == target_rank:
+            assert np.allclose(out, a)
+        else:
+            assert not out
+
 
 def test_scalar_arg_substitution():
     """