Make the gpu version of DownsampleFactorMax and its grad don't crash on bigger inputs

theano/sandbox/cuda/tests/test_blas.py

@@ -269,6 +269,8 @@ def test_downsample():
             (1, 1, 10, 1023),
             (1, 1, 1025, 10),
             (1, 1, 1023, 10),
+            (65536, 1, 10, 10),
+            (1, 65536, 10, 10),
              ]
 
     numpy.random.RandomState(unittest_tools.fetch_seed()).shuffle(shps)
@@ -299,6 +301,14 @@ def test_downsample():
                     for node in f2.maker.env.toposort()])
                 assert numpy.allclose(f(), f2())
 
+                # The grad is too slow on GT220 GPU
+                # This cause the computer to freeze...
+                # Remove this when it get optimized enought
+                # This only bypass the last 2 checks
+                # Those tests where passing in all Mode on a GTX470
+                if shp[0] > 30000 or shp[1] > 30000:
+                    continue
+
                 g = pfunc(
                         [],
                         tensor.grad(ds_op(tensor.as_tensor_variable(a)).sum(),
@@ -314,7 +324,7 @@ def test_downsample():
                             for node in g.maker.env.toposort()])
                 assert any([isinstance(node.op, DownsampleFactorMaxGrad)
                             for node in g2.maker.env.toposort()])
-                assert numpy.allclose(g(), g2())
+                assert numpy.allclose(g(), g2()), shp
 
                 # We already check that the gpu version return
                 # the same value as the gpu version for