Merge pull request #5903 from lamblin/fix_5898

theano/gpuarray/subtensor.py

@@ -544,6 +544,13 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor):
         idx_ = ([slice(None)] * p + nidx[p:])
         x = x.__getitem__(idx_)
 
+        if p == 0:
+            # The only indexing was through slices and indices.
+            # This can happen with symbolic slices for instance.
+            # Since no view_map is set, we need to copy the returned value
+            out[0] = x.copy()
+            return
+
         # flatten the array-indexed dimensions
         shape = ((np.prod(x.shape[0: p]),) +
                  x.shape[p:])

theano/tensor/subtensor.py

@@ -2169,9 +2169,14 @@ class AdvancedSubtensor(Op):
 
     def perform(self, node, inputs, out_):
         out, = out_
-        # TODO: in general, we need to re-pack the inputs into a valid
-        # index, just like subtensor
-        out[0] = inputs[0].__getitem__(inputs[1:])
+        rval = inputs[0].__getitem__(inputs[1:])
+        # When there are no arrays, we are not actually doing advanced
+        # indexing, so __getitem__ will not return a copy.
+        # Since no view_map is set, we need to copy the returned value
+        if not any(isinstance(v.type, TensorType) and v.ndim > 0
+                   for v in node.inputs[1:]):
+            rval = rval.copy()
+        out[0] = rval
 
     def connection_pattern(self, node):
         rval = [[True]]

theano/tensor/tests/test_subtensor.py

@@ -1623,6 +1623,14 @@ class TestAdvancedSubtensor(unittest.TestCase):
         out = X[b_idx, r_idx, c_idx].eval({X: xx})
         utt.assert_allclose(out, xx[b_idx, r_idx, c_idx])
 
+    def test_adv_sub_slice(self):
+        # Reported in https://github.com/Theano/Theano/issues/5898
+        var = self.shared(np.zeros([3, 3], dtype=config.floatX))
+        slc = tensor.slicetype()
+        f = theano.function([slc], var[slc])
+        s = slice(1, 3)
+        f(s)
+
     def test_grad(self):
         ones = np.ones((1, 3), dtype=self.dtype)
         n = self.shared(ones * 5, broadcastable=(True, False))