Make GpuOp(old back-end) and GpuElemwise(new back-end) use prepare_node

theano/gpuarray/elemwise.py

@@ -2640,11 +2640,9 @@ class GpuCAReduceCPY(GpuKernelBase, HideC, CAReduceDtype):
     def get_params(self, node):
         return node.outputs[0].type.context
 
-    def make_thunk(self, node, storage_map, compute_map, no_recycling):
+    def prepare_node(self, node, storage_map, compute_map):
         # cache the kernel object
         self.get_kernel_cache(node)
-        return super(GpuCAReduceCPY, self).make_thunk(
-            node, storage_map, compute_map, no_recycling)
 
     def get_kernel_cache(self, node):
         attr = '@cache_reduction_k'

theano/sandbox/cuda/__init__.py

@@ -246,18 +246,14 @@ class GpuOp(theano.gof.Op):
 
     """
 
-    def make_thunk(self, node, storage_map, compute_map, no_recycling):
+    def prepare_node(self, node, storage_map, compute_map):
         if use.device_number is None:
             use("gpu",
                 force=True,
                 default_to_move_computation_to_gpu=False,
                 move_shared_float32_to_gpu=False,
                 enable_cuda=False)
-        return super(GpuOp, self).make_thunk(node, storage_map,
-                                             compute_map, no_recycling)
 
-theano.compile.debugmode.default_make_thunk.append(
-    get_unbound_function(GpuOp.make_thunk))
 
 # We must do those import to be able to create the full doc when
 # nvcc is not available

theano/sandbox/cuda/blas.py

@@ -2120,6 +2120,8 @@ class GpuConv(GpuOp):
         return flops
 
     def prepare_node(self, node, storage_map, compute_map):
+        super(GpuConv, self).prepare_node(node, storage_map, compute_map, impl)
+
         if node.op.max_threads_dim0 is None:
             cuda = theano.sandbox.cuda
             device_id = cuda.use.device_number

theano/sandbox/cuda/dnn.py

@@ -1513,6 +1513,8 @@ class GpuDnnPool(DnnBase):
         self.mode = mode
 
     def prepare_node(self, node, storage_map, compute_map):
+        super(GpuDnnPool, self).prepare_node(node, storage_map, compute_map)
+
         if len(node.inputs) == 2:
             warnings.warn("Theano GPUDnnPoolGrad internal changed.", stacklevel=3)
             # Old interface