merge

theano/sandbox/cuda/cuda_ndarray.cu

@@ -572,9 +572,9 @@ static PyMethodDef CudaNdarray_methods[] =
     {"__array__", 
         (PyCFunction)CudaNdarray_CreateArrayObj, METH_NOARGS,
         "Copy from the device to a numpy ndarray"},
-    {"__copy__", 
-        (PyCFunction)CudaNdarray_Copy, METH_NOARGS,
-        "Create a copy of this object"},
+    {"__copy__",
+        (PyCFunction)CudaNdarray_View, METH_NOARGS,
+        "Create a shallow copy of this object. used by module copy"},
     {"__deepcopy__", 
         (PyCFunction)CudaNdarray_DeepCopy, METH_O,
         "Create a copy of this object"},

theano/sandbox/cuda/opt.py

@@ -70,6 +70,9 @@ gpu_cut_copies.register('cut_gpu_host_transfers', local_cut_gpu_host_gpu,
         'fast_run', 'inplace', 'gpu')
 gpu_cut_copies.register('cut_gpu_constant_transfers', tensor.opt.constant_folding, 
         'fast_run', 'gpu')
+#register it into canonicalize to allow other optimization to work without
+#botering with this useless pattern.
+compile.optdb['canonicalize'].register('local_cut_gpu_host_gpu', local_cut_gpu_host_gpu, 'fast_run')
 
 @register_opt()
 @local_optimizer([])

theano/sandbox/cuda/tests/test_cuda_ndarray.py

@@ -90,6 +90,11 @@ def test_copy():
     assert numpy.allclose(a, numpy.asarray(b))
     assert numpy.allclose(a, numpy.asarray(c))
     assert numpy.allclose(a, numpy.asarray(d))
+    b+=b
+    assert numpy.allclose(a+a, numpy.asarray(b))
+    assert numpy.allclose(a+a, numpy.asarray(c))
+    assert numpy.allclose(a, numpy.asarray(d))
+    
 
 def test_dot():
     print >>sys.stdout, 'starting test_dot'