Remove usage of pygpu_fromhostdata and simplify code.

theano/gpuarray/basic_ops.py

@@ -702,39 +702,35 @@ class GpuFromHost(Op):
         if (%(name)s_tmp == NULL)
           %(fail)s
 
-        if (%(out)s != NULL && GpuArray_IS_C_CONTIGUOUS(&%(out)s->ga) &&
+        if (%(out)s == NULL || GpuArray_IS_C_CONTIGUOUS(&%(out)s->ga) ||
             theano_size_check(%(out)s, PyArray_NDIM(%(name)s_tmp),
                               (size_t *)PyArray_DIMS(%(name)s_tmp),
                               get_typecode((PyObject *)PyArray_DESCR(%(name)s_tmp)))) {
-          Py_BEGIN_ALLOW_THREADS
-          err = GpuArray_write(&%(out)s->ga, PyArray_DATA(%(name)s_tmp),
-                               PyArray_NBYTES(%(name)s_tmp));
-          Py_END_ALLOW_THREADS
-          Py_DECREF(%(name)s_tmp);
-          if (err != GA_NO_ERROR) {
-            PyErr_Format(PyExc_RuntimeError, "Could not write data to gpu");
-            %(fail)s;
-          }
-        } else {
           Py_XDECREF(%(out)s);
-          // This method will release the GIL when needed.
-          %(out)s = pygpu_fromhostdata(PyArray_DATA(%(name)s_tmp),
-                                       get_typecode((PyObject *)PyArray_DESCR(%(name)s_tmp)),
-                                       PyArray_NDIM(%(name)s_tmp),
-                                       (size_t *)PyArray_DIMS(%(name)s_tmp),
-                                       (ssize_t *)PyArray_STRIDES(%(name)s_tmp),
-                                       %(ctx)s,
-                                       Py_None);
-          Py_DECREF(%(name)s_tmp);
+          %(out)s = pygpu_empty(PyArray_NDIM(%(name)s_tmp),
+                                (size_t *)PyArray_DIMS(%(name)s_tmp),
+                                get_typecode((PyObject *)PyArray_DESCR(%(name)s_tmp)),
+                                GA_C_ORDER, %(ctx)s, Py_None);
           if (%(out)s == NULL) {
-              %(fail)s
+            Py_DECREF(%(name)s_tmp);
+            %(fail)s;
           }
         }
+
+        Py_BEGIN_ALLOW_THREADS
+        err = GpuArray_write(&%(out)s->ga, PyArray_DATA(%(name)s_tmp),
+                             PyArray_NBYTES(%(name)s_tmp));
+        Py_END_ALLOW_THREADS
+        Py_DECREF(%(name)s_tmp);
+        if (err != GA_NO_ERROR) {
+          PyErr_Format(PyExc_RuntimeError, "Could not write data to gpu");
+          %(fail)s;
+        }
         """ % {'name': name, 'inp': inputs[0], 'ctx': sub['params'],
                'out': outputs[0], 'fail': sub['fail']}
 
     def c_code_cache_version(self):
-        return (9,)
+        return (10,)
 
 
 class GpuToGpu(Op):