Fix the case where there is a single array to index by.

theano/gpuarray/subtensor.py

@@ -502,6 +502,7 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor):
 
         x = x.reshape(nshp)
 
+        narrays = 0
         transp = list(range(x.ndim))
         p = 0
         for k, i in enumerate(list(nidx)):
@@ -512,6 +513,7 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor):
                 i = nidx.pop(k)
                 nidx.insert(p, i)
                 p += 1
+                narrays += 1
         x = x.transpose(*transp)
 
         idx_ = ([slice(None)] * p + nidx[p:])
@@ -535,7 +537,17 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor):
 
         # finish up
         out_flat_shp = take_idx.shape + x.shape[p:]
-        out[0] = out_flat.reshape(out_flat_shp)
+        o = out_flat.reshape(out_flat_shp)
+
+        # If there was only one array we need to move the indexed
+        # dimension back
+        if narrays == 1:
+            k = transp[0]
+            ntransp = list(range(1, o.ndim))
+            ntransp.insert(k, 0)
+            o = o.transpose(*ntransp)
+
+        out[0] = o
 
 
 class GpuAdvancedIncSubtensor1(Op):

theano/tensor/tests/test_subtensor.py

@@ -55,6 +55,7 @@ class T_subtensor(unittest.TestCase, utt.TestOptimizationMixin):
                  inc_sub=tensor.IncSubtensor,
                  adv_sub1=tensor.AdvancedSubtensor1,
                  adv_incsub1=tensor.AdvancedIncSubtensor1,
+                 adv_sub=tensor.AdvancedSubtensor,
                  mode=None,
                  dtype=theano.config.floatX,
                  type=tensor.TensorType,
@@ -65,6 +66,7 @@ class T_subtensor(unittest.TestCase, utt.TestOptimizationMixin):
         self.inc_sub = inc_sub
         self.adv_sub1 = adv_sub1
         self.adv_incsub1 = adv_incsub1
+        self.adv_sub = adv_sub
         self.dimshuffle = dimshuffle
         if mode is None:
             mode = theano.compile.mode.get_default_mode()
@@ -359,7 +361,7 @@ class T_subtensor(unittest.TestCase, utt.TestOptimizationMixin):
         numpy_version = [int(v) for v in numpy.version.version.split('.')[0:2]]
         if numpy_version >= [1, 9]:
             test_cases.append(
-                (1, AdvancedSubtensor, AdvancedSubtensor,
+                (1, AdvancedSubtensor, self.adv_sub,
                  numpy.index_exp[..., numpy.newaxis, [1, 2]]))
 
         for length, op_type, op_type_opt, slice_ in test_cases: