custom bitwise_and.reduce function using identity value -1

theano/tensor/tests/test_elemwise.py

@@ -330,6 +330,22 @@ class test_Broadcast(unittest.TestCase):
             assert (f(xv) == zv).all()
 
 
+def reduce_bitwise_and(x, axis=-1, dtype='int8'):
+    identity = numpy.array((-1,), dtype=dtype)[0]
+
+    if 0 in x.shape and x.shape[axis] != 0:
+        new_shape = tuple([s for i, s in enumerate(x.shape) if i != axis])
+        return numpy.empty(shape=new_shape, dtype=x.dtype)
+
+    def custom_reduce(a):
+        out = identity
+        for i in range(a.size):
+            out = numpy.bitwise_and(a[i], out)
+        return out
+
+    return numpy.apply_along_axis(custom_reduce, axis, x)
+
+
 class test_CAReduce(unittest_tools.InferShapeTester):
     op = CAReduce
     cases = [((5, 6), None),
@@ -430,7 +446,7 @@ class test_CAReduce(unittest_tools.InferShapeTester):
                     zv = numpy.bitwise_or.reduce(zv, axis)
             elif scalar_op == scalar.and_:
                 for axis in reversed(sorted(tosum)):
-                    zv = numpy.bitwise_and.reduce(zv, axis)
+                    zv = reduce_bitwise_and(zv, axis, dtype=dtype)
             elif scalar_op == scalar.xor:
                 # There is no identity value for the xor function
                 # So we can't support shape of dimensions 0.