gh-143732: Specialize __setitem__ dunder method for STORE_SUBSCR

[johng]

This is precursuor for JIT improvements but also results in standard interpreter improvements due to specializing the bytecode.

This improves performance by about 5% on linux and macos micro benchmark

import pyperf


class Grid:
    __slots__ = ("_cell",)

    def __setitem__(self, key, value):
        self._cell = value


def bench_store_subscr(loops):
    obj = Grid()
    range_it = range(loops)
    t0 = pyperf.perf_counter()
    for _ in range_it:
        # 20 stores per loop; keep in sync with inner_loops below
        obj[0] = 0; obj[0] = 0; obj[0] = 0; obj[0] = 0; obj[0] = 0
        obj[0] = 0; obj[0] = 0; obj[0] = 0; obj[0] = 0; obj[0] = 0
        obj[0] = 0; obj[0] = 0; obj[0] = 0; obj[0] = 0; obj[0] = 0
        obj[0] = 0; obj[0] = 0; obj[0] = 0; obj[0] = 0; obj[0] = 0
    return pyperf.perf_counter() - t0

if __name__ == "__main__":
    runner = pyperf.Runner()
    runner.metadata["description"] = "STORE_SUBSCR on an object with a Python __setitem__"
    runner.bench_time_func("store_subscr_py_dunder", bench_store_subscr, inner_loops=20)

• Issue: Broader specialization in the Specializing Adaptive Interpreter for better JIT performance #143732

[markshannon]

This does not implement the specialization of STORE_SUBSCR for __setitem__ implemented in Python, as described in the issue. If the description in the issue is not clear, please ask for clarification (on the issue, not here)