Added some tests for the new optimizations.

theano/tensor/tests/test_opt.py

@@ -3924,6 +3924,38 @@ class T_useless_elemwise(unittest.TestCase):
         assert len(topo) == 1
         assert topo[0].op == deep_copy_op
 
+    def assert_eqs_const(self, topo, val):
+        elem = topo[0]
+        assert len(topo) == 1
+        assert elem.op == deep_copy_op
+        assert len(elem.inputs) == 1
+        assert isinstance(elem.inputs[0], T.TensorConstant)
+        assert T.extract_constant(elem.inputs[0]) == val
+
+    def test_and(self):
+        x = T.scalar('x', dtype='int64')
+
+        func = theano.function([x], T.and_(x, 0), mode=self.mode)
+        self.assert_eqs_const(func.maker.fgraph.toposort(), 0)
+
+        func = theano.function([x], T.and_(0, x), mode=self.mode)
+        self.assert_eqs_const(func.maker.fgraph.toposort(), 0)
+
+    def test_or(self):
+        x = T.scalar('x', dtype='int64')
+
+        func = theano.function([x], T.or_(x, 1), mode=self.mode)
+        self.assert_eqs_const(func.maker.fgraph.toposort(), 1)
+
+        func = theano.function([x], T.or_(1, x), mode=self.mode)
+        self.assert_eqs_const(func.maker.fgraph.toposort(), 1)
+
+    def test_xor(self):
+        x = T.scalar('x', dtype='int64')
+
+        func = theano.function([x], T.xor(x, x), mode=self.mode)
+        self.assert_eqs_const(func.maker.fgraph.toposort(), 0)
+
 
 class T_cast_cast(unittest.TestCase):
     def setUp(self):