Corrected few import statements in scan module and fixed error with elemwise, fft and nerv

theano/gpuarray/fft.py

@@ -374,11 +374,11 @@ if scikits_cuda_available:
     @register_opt('fast_compile')
     @op_lifter([theano.tensor.fft.RFFTOp])
     @register_opt2([theano.tensor.fft.RFFTOp], 'fast_compile')
-    def local_gpua_curfft_op(node, context_name):
+    def local_gpua_curfft_op(op, ctx_name, inputs, outputs):
         return curfft_op
 
     @register_opt('fast_compile')
     @op_lifter([theano.tensor.fft.IRFFTOp])
     @register_opt2([theano.tensor.fft.IRFFTOp], 'fast_compile')
-    def local_gpua_cuirfft_op(node, context_name):
+    def local_gpua_cuirfft_op(op, ctx_name, inputs, outputs):
         return cuirfft_op

theano/gpuarray/neighbours.py

@@ -471,6 +471,6 @@ class GpuImages2Neibs(GpuKernelBase, Images2Neibs, Op):
 @register_opt('fast_compile')
 @op_lifter([Images2Neibs])
 @register_opt2([Images2Neibs], 'fast_compile')
-def local_gpua_images2neibs(op, context_name, inputs):
+def local_gpua_images2neibs(op, context_name, inputs, outputs):
     if op.mode in ['valid', 'ignore_borders', 'wrap_centered']:
         return GpuImages2Neibs(op.mode)

theano/gpuarray/tests/test_opt.py

@@ -451,7 +451,7 @@ def test_local_assert_no_cpu_op():
     out = theano.tensor.tanh(ms).dot(ms.T)
 
     mode_local_assert = mode_with_gpu.including("assert_no_cpu_op")
-    mode_local_assert = mode_local_assert.excluding("local_gpu_elemwise")
+    mode_local_assert = mode_local_assert.excluding("local_gpua_elemwise")
 
     old = theano.config.assert_no_cpu_op
     old2 = theano.config.on_opt_error

theano/scan_module/scan_utils.py

@@ -152,12 +152,14 @@ def traverse(out, x, x_copy, d, visited=None):
         return d
     visited.add(out)
     from theano.sandbox import cuda
-    from theano.gpuarray.basic_ops import gpu_from_host
+    from theano.gpuarray.basic_ops import gpu_from_host, host_from_gpu
+    from theano.gpuarray import pygpu_activated
+    from theano.gpuarray.type import GpuArrayType
     if out == x:
         if isinstance(x.type, cuda.CudaNdarrayType):
             d[out] = cuda.gpu_from_host(x_copy)
         else:
-            assert isinstance(x.type, gpuarray.GpuArrayType)
+            assert isinstance(x.type, GpuArrayType)
             d[out] = gpu_from_host(x.type.context_name)(x_copy)
         return d
     elif out.owner is None:
@@ -167,8 +169,8 @@ def traverse(out, x, x_copy, d, visited=None):
           out.owner.inputs == [x]):
         d[out] = tensor.as_tensor_variable(x_copy)
         return d
-    elif (gpuarray.pygpu_activated and
-          out.owner.op == gpuarray.host_from_gpu and
+    elif (pygpu_activated and
+          out.owner.op == host_from_gpu and
           out.owner.inputs == [x]):
         d[out] = tensor.as_tensor_variable(x_copy)
         return d