lower the number of GPU op compiled

theano/sandbox/cuda/basic_ops.py

@@ -298,6 +298,8 @@ class GpuDimShuffle(GpuOp):
     """
     Implement DimShuffle on the gpu.
     """
+    check_broadcast = False
+
     def __init__(self, input_broadcastable, new_order):
         input_broadcastable = tuple(input_broadcastable)
         self.input_broadcastable = input_broadcastable
@@ -2355,6 +2357,8 @@ class GpuSubtensor(GpuOp, tensor.Subtensor):
     """
     Implement subtensor on the gpu.
     """
+    check_broadcast = False
+
     # __hash__, __eq__, __str__ come from tensor.Subtensor
     def make_node(self, x, *inputs):
         assert isinstance(x.type, CudaNdarrayType)

theano/sandbox/cuda/blas.py

@@ -515,6 +515,7 @@ class BaseGpuCorrMM(GpuOp):
     :param pad: *deprecated*, now you should always use border_mode
 
     """
+    check_broadcast = False
 
     def __init__(self, border_mode="valid", subsample=(1, 1), pad=(0, 0)):
         if pad != (0, 0):
@@ -1498,6 +1499,8 @@ class GpuConv(GpuOp):
     """
     Implement the batched and stacked 2d convolution on the gpu.
     """
+    check_broadcast = False
+
     @staticmethod
     def logical_output_shape_2d(imshp, kshp, mode):
         if mode == 'valid':