Merge pull request #1369 from IntelPython/fix-gh-1352

Fix gh-1352 dpnp.sum() for empty array crashes.

dpnp/dpnp_algo/dpnp_algo_mathematical.pyx

@@ -631,6 +631,9 @@ cpdef utils.dpnp_descriptor dpnp_sum(utils.dpnp_descriptor x1,
                                                                        usm_type=x1_obj.usm_type,
                                                                        sycl_queue=x1_obj.sycl_queue)
 
+    if x1.size == 0 and axis is None:
+        return result
+
     result_sycl_queue = result.get_array().sycl_queue
 
     cdef c_dpctl.SyclQueue q = <c_dpctl.SyclQueue> result_sycl_queue

dpnp/dpnp_iface_mathematical.py

@@ -1629,10 +1629,16 @@ def sum(x1, axis=None, dtype=None, out=None, keepdims=False, initial=None, where
         if where is not True:
             pass
         else:
+            if dpnp.isscalar(out):
+                raise TypeError("output must be an array")
             out_desc = dpnp.get_dpnp_descriptor(out, copy_when_nondefault_queue=False) if out is not None else None
             result_obj = dpnp_sum(x1_desc, axis, dtype, out_desc, keepdims, initial, where).get_pyobj()
             result = dpnp.convert_single_elem_array_to_scalar(result_obj, keepdims)
 
+            if x1_desc.size == 0 and axis is None:
+                result = dpnp.zeros_like(result)
+                if out is not None:
+                    out[...] = result
             return result
 
     return call_origin(numpy.sum, x1, axis=axis, dtype=dtype, out=out, keepdims=keepdims, initial=initial, where=where)

tests/test_mathematical.py

@@ -923,3 +923,35 @@ def test_float_to_inf(self):
         dpnp_res = dpnp.array(a) ** dpnp.array(b)
 
         assert_allclose(numpy_res, dpnp_res.asnumpy())
+
+
+@pytest.mark.parametrize("dtype", get_all_dtypes(no_complex=True, no_bool=True))
+@pytest.mark.parametrize("axis", [None, 0, 1, 2, 3])
+def test_sum_empty(dtype, axis):
+    a = numpy.empty((1, 2, 0, 4), dtype=dtype)
+    numpy_res = a.sum(axis=axis)
+    dpnp_res = dpnp.array(a).sum(axis=axis)
+    assert_array_equal(numpy_res, dpnp_res.asnumpy())
+
+
+@pytest.mark.parametrize("dtype", get_all_dtypes(no_complex=True, no_bool=True))
+def test_sum_empty_out(dtype):
+    a = dpnp.empty((1, 2, 0, 4), dtype=dtype)
+    out = dpnp.ones(())
+    res = a.sum(out=out)
+    assert_array_equal(out.asnumpy(), res.asnumpy())
+    assert_array_equal(out.asnumpy(), numpy.array(0, dtype=dtype))
+
+
+@pytest.mark.parametrize("shape", [(), (1, 2, 3), (1, 0, 2), (10), (3, 3, 3), (5, 5), (0, 6)])
+@pytest.mark.parametrize("dtype_in", get_all_dtypes(no_complex=True, no_bool=True))
+@pytest.mark.parametrize("dtype_out", get_all_dtypes(no_complex=True, no_bool=True))
+def test_sum(shape, dtype_in, dtype_out):
+    a_np = numpy.ones(shape, dtype=dtype_in)
+    a = dpnp.ones(shape, dtype=dtype_in)
+    axes = [None, 0, 1, 2]
+    for axis in axes:
+        if axis is None or axis < a.ndim:
+            numpy_res = a_np.sum(axis=axis, dtype=dtype_out)
+            dpnp_res = a.sum(axis=axis, dtype=dtype_out)
+            assert_array_equal(numpy_res, dpnp_res.asnumpy())