'implemented CudaNdarray.i{add,div} for ndim==0 and test it.'

theano/sandbox/cuda/cuda_ndarray.cu

@@ -1048,6 +1048,31 @@ CudaNdarray_inplace_add_div(PyObject* py_self, PyObject * py_other, int fct_nb)
 
     switch(self->nd)
     {
+        case 0:
+            {
+                dim3 n_blocks(1, 1, 1);
+                dim3 n_threads(1);
+                k_iop_3<<<n_blocks, n_threads>>>(1,
+                        1, //CudaNdarray_HOST_DIMS(self)[0],
+			1, //CudaNdarray_HOST_DIMS(self)[0],
+                        CudaNdarray_DEV_DATA(self),
+                        1,
+                        1, //CudaNdarray_HOST_STRIDES(self)[0],
+                        CudaNdarray_HOST_STRIDES(self)[0],
+                        CudaNdarray_DEV_DATA(other),
+                        1,
+                        1, //CudaNdarray_HOST_STRIDES(other)[0],
+                        CudaNdarray_HOST_STRIDES(other)[0]);
+                CNDA_THREAD_SYNC;
+                cudaError_t err = cudaGetLastError();
+                if( cudaSuccess != err)
+                {
+                    PyErr_Format(PyExc_RuntimeError, "Cuda error: %s: %s.\n", "k_iop_3", cudaGetErrorString(err));
+                    return NULL;
+                }
+                Py_INCREF(py_self);
+                return py_self;
+            }
         case 1:
             {
                 dim3 n_blocks(1, 1, 1);

theano/sandbox/cuda/tests/test_cuda_ndarray.py

@@ -59,17 +59,16 @@ def test_add_iadd_idiv():
         print shape, 'adding ', a0.size, 'cpu', cpu_dt, 'advantage', advantage(cpu_dt, gpu_dt)
         assert numpy.allclose(asum,  numpy.asarray(bsum))
 
-        if len(shape)>0:
-            #test inplace version, not implemented with 0 dims
-            b0 += b1
-            a0 += a1
-            assert numpy.allclose(a0, numpy.asarray(b0))
-            assert numpy.allclose(a0,a1*2)
-
-            b0 /= b1
-            a0 /= a1
-            assert numpy.allclose(a0, numpy.asarray(b0))
-            assert numpy.allclose(a0,numpy.ones(a0.shape)*2)
+        # test inplace version
+        b0 += b1
+        a0 += a1
+        assert numpy.allclose(a0, numpy.asarray(b0))
+        assert numpy.allclose(a0,a1*2)
+
+        b0 /= b1
+        a0 /= a1
+        assert numpy.allclose(a0, numpy.asarray(b0))
+        assert numpy.allclose(a0,numpy.ones(a0.shape)*2)
 
         if len(shape)==2:
             #test not contiguous version.