Added docstrings and fixed two errors

theano/gpuarray/extra_ops.py

@@ -2,14 +2,13 @@ from __future__ import absolute_import, print_function, division
 import os
 from theano import Apply, Op
 from theano.tensor.extra_ops import CumsumOp
-
+from .type import GpuArrayType
 try:
     from pygpu import gpuarray
 except ImportError:
     pass
 
-from .basic_ops import (as_gpuarray_variable, GpuKernelBase, Kernel,
-                        infer_context_name)
+from .basic_ops import (as_gpuarray_variable, GpuKernelBase, Kernel)
 from .opt import register_opt, op_lifter, register_opt2
 
 
@@ -40,7 +39,6 @@ class GpuCumsum(GpuKernelBase, Op):
 
     def make_node(self, x):
         assert x.type.dtype == 'float32', "Only float32 supported for GpuCumSum"
-        x = as_gpuarray_variable(x, infer_context_name(x))
 
         if x.ndim > GpuCumsum.SUPPORTED_NDIMS:
             raise NotImplementedError('Only cumsum on 1D, 2D and\
@@ -456,6 +454,8 @@ class GpuCumsum(GpuKernelBase, Op):
 @register_opt2([CumsumOp], 'fast_compile')
 def use_gpu_cumsumop(op, ctx_name, inputs, outputs):
     if inputs[0].dtype == 'float32':
+        if isinstance(inputs[0].type, GpuArrayType):
+            return
         axis = op.axis
         x = inputs[0]
 

theano/gpuarray/opt.py

@@ -259,12 +259,19 @@ gpu_seqopt.register('InputToGpuArrayOptimizer', InputToGpuOptimizer(),
 class GraphToGPU(NavigatorOptimizer):
     """
     Transfer the graph as a whole to GPU instead of transfering node by node.
+
+    Parameters
+    ----------
+    local_optimizers_all : List or Set
+        The local optimizations to apply to a node.
+    local_optimizers_map : Dict
+        Dictionary object containing the mapping of Op to list of
+        LocalOptimizers.
     """
 
     def __init__(self, local_optimizers_all, local_optimizers_map):
         self.local_optimizers_all = local_optimizers_all
         self.local_optimizers_map = local_optimizers_map
-        self.failure_callback = None
 
     def add_requirements(self, fgraph):
         fgraph.attach_feature(toolbox.ReplaceValidate())
@@ -1082,8 +1089,8 @@ def local_gpua_careduce(op, context_name, inputs, outputs):
                     for i in range(x.ndim):
                         if i not in op.axis:
                             out_shp.append(shape_i(x, i))
-                    unreshaped_reduce = GpuReshape(len(out_shp))(
-                        tensor.stack(out_shp))
+                    unreshaped_reduce = GpuReshape(len(out_shp))(reduce_reshaped_x,
+                                                                 tensor.stack(out_shp))
                 else:
                     unreshaped_reduce = reduce_reshaped_x
                 return [unreshaped_reduce]