Fixes for gpuarray in DebugMode

theano/sandbox/gpuarray/basic_ops.py

@@ -996,6 +996,8 @@ class GpuJoin(HideC, Join):
     def perform(self, node, axis_and_tensors, out_, ctx):
         out, = out_
         axis = int(axis_and_tensors[0])
+        if axis < 0:
+            axis += axis_and_tensors[1].ndim
         tensors = axis_and_tensors[1:]
         out[0] = pygpu.concatenate(tensors, axis=axis, context=ctx).astype(
             node.outputs[0].dtype)

theano/sandbox/gpuarray/blas.py

@@ -287,7 +287,8 @@ class GpuDot22(BlasOp):
     def perform(self, node, inputs, outputs):
         x, y = inputs
 
-        out = pygpu.empty((x.shape[0], y.shape[1]), dtype=x.dtype)
+        out = pygpu.empty((x.shape[0], y.shape[1]), dtype=x.dtype,
+                          context=x.context)
         outputs[0][0] = blas.gemm(1., x, y, 0., out,
                                   overwrite_c=True)
 

theano/sandbox/gpuarray/neighbours.py

@@ -462,6 +462,10 @@ class GpuImages2Neibs(GpuKernelBase, Images2Neibs, Op):
         } // END NESTED SCOPE
         """ % locals()
 
+    def perform(self, node, inp, out, ctx):
+        # Disable the perform method from the CPU version
+        Op.perform(self, node, inp, out, ctx)
+
 
 @op_lifter([Images2Neibs])
 def use_gpu_images2neibs(node, context_name):

theano/tensor/tests/test_basic.py

@@ -3884,11 +3884,7 @@ class T_Join_and_Split(unittest.TestCase):
         got = f(-2)
         assert numpy.allclose(got, want)
 
-        try:
-            got = f(-3)
-            assert False
-        except IndexError:
-            pass
+        self.assertRaises((IndexError, OverflowError), f, -3)
 
     def test_join_matrixC_negative_axis(self):
         """constant join negative axis"""
@@ -3920,11 +3916,7 @@ class T_Join_and_Split(unittest.TestCase):
         got = f()
         assert numpy.allclose(got, want)
 
-        try:
-            s = join(-3, a, b)
-            assert False
-        except IndexError:
-            pass
+        self.assertRaises((IndexError, OverflowError), join, -3, a, b)
 
         utt.verify_grad(lambda a, b: join(-1, a, b), [v, 2 * v],
                         mode=self.mode)