Feature or enhancement

Proposal:

Currently BINARY_SLICE looks like this:

        op(_BINARY_SLICE, (container, start, stop -- res)) {
            PyObject *slice = _PyBuildSlice_ConsumeRefs(PyStackRef_AsPyObjectSteal(start),
                                                        PyStackRef_AsPyObjectSteal(stop));
            PyObject *res_o;
            if (slice == NULL) {
                res_o = NULL;
            }
            else {
                res_o = PyObject_GetItem(PyStackRef_AsPyObjectBorrow(container), slice);
                Py_DECREF(slice);
            }
            PyStackRef_CLOSE(container);
            ERROR_IF(res_o == NULL);
            res = PyStackRef_FromPyObjectSteal(res_o);
        }

Note that we build a slice just for getitem, only to throw it away immediately after.

We can scalar replace the slice in BINARY_SLICE after recording the types. If we record the container type, we can call the slice dispatcher underneath directly without boxing the slice object. See for example BINARY_OP_SUBSCR_LIST_SLICE

We'd need to guard on the type recorded.

I propose to do this in the current optimizer pass rather than the partial evaluation pass, as this is just a simple optimization within a single op, while the PE pass can handle inter-instruction optimization.

Has this already been discussed elsewhere?

No response given

Links to previous discussion of this feature:

No response

Linked PRs

• gh-144569: Avoid creating temporary objects in BINARY_SLICE for list, tuple, and unicode #144590