register c code for Shape op with GpuArrayVariable and doc the register fct().

doc/extending/ctype.txt

@@ -553,3 +553,20 @@ respectively. See an example for the type ``CudaNdarrayType`` (GPU
 array) in the file `theano/sandbox/cuda/type.py`. The version
 parameter is what is returned by ViewOp.c_code_cache_version(). By
 default, it will recompile the c code for each process.
+
+
+Shape and Shape_i
+=================
+
+We have 2 generic Ops Shape and Shape_i that return the shape of any
+Theano Variable that have a shape attribute and Shape_i return only of
+the element of the shape.
+
+
+.. code-block:: python
+
+   theano.compile.ops.register_shape_c_code(YOUR_TYPE_CLASS, THE_C_CODE, version=())
+   theano.compile.ops.register_shape_i_c_code(YOUR_TYPE_CLASS, THE_C_CODE, version=())
+
+The c code work as the ViewOp. Shape_i have the additional i parameter
+that you can use with %(i)s.

theano/sandbox/gpuarray/type.py

@@ -298,6 +298,23 @@ theano.compile.register_view_op_c_code(GpuArrayType, """
     Py_XINCREF(%(oname)s);
 """, version=(0,))
 
+# Register TensorType C code for ViewOp.
+theano.compile.register_shape_c_code(
+    GpuArrayType,
+    """
+    npy_intp shape[] = {%(iname)s->ga.nd};
+    if(%(oname)s == NULL || (PyArray_DIMS(%(oname)s)[0] != shape[0]))
+    {
+        Py_XDECREF(%(oname)s);
+        %(oname)s = (PyArrayObject*) PyArray_SimpleNew(1, shape, NPY_INT64);
+    }
+    for(int i=0;i<shape[0];i++)
+    {
+        ((npy_int64*)PyArray_GETPTR1(%(oname)s, i))[0] = %(iname)s->ga.dimensions[i];
+    }
+    """,
+    version=1)
+
 theano.compile.register_shape_i_c_code(GpuArrayType, """
     if(!%(oname)s)
         %(oname)s=(PyArrayObject*)PyArray_ZEROS(0, NULL, NPY_INT64, 0);