diff options
Diffstat (limited to 'numpy')
| -rw-r--r-- | numpy/core/src/umath/ufunc_object.c | 12 |
1 files changed, 9 insertions, 3 deletions
diff --git a/numpy/core/src/umath/ufunc_object.c b/numpy/core/src/umath/ufunc_object.c index bf5a4ead3..6dd597b3a 100644 --- a/numpy/core/src/umath/ufunc_object.c +++ b/numpy/core/src/umath/ufunc_object.c @@ -3215,9 +3215,15 @@ PyUFunc_Accumulate(PyUFuncObject *ufunc, PyArrayObject *arr, PyArrayObject *out, */ ndim_iter = ndim; flags |= NPY_ITER_MULTI_INDEX; - /* Add some more flags */ - op_flags[0] |= NPY_ITER_UPDATEIFCOPY|NPY_ITER_ALIGNED; - op_flags[1] |= NPY_ITER_COPY|NPY_ITER_ALIGNED; + /* + * Add some more flags. + * + * The accumulation outer loop is 'elementwise' over the array, so turn + * on NPY_ITER_OVERLAP_ASSUME_ELEMENTWISE. That is, in-place + * accumulate(x, out=x) is safe to do without temporary copies. + */ + op_flags[0] |= NPY_ITER_UPDATEIFCOPY|NPY_ITER_ALIGNED|NPY_ITER_OVERLAP_ASSUME_ELEMENTWISE; + op_flags[1] |= NPY_ITER_COPY|NPY_ITER_ALIGNED|NPY_ITER_OVERLAP_ASSUME_ELEMENTWISE; op_dtypes_param = op_dtypes; op_dtypes[1] = op_dtypes[0]; NPY_UF_DBG_PRINT("Allocating outer iterator\n"); |