Skip to main content

Python/ceval.c (part 52)

Source:

cpython 3.14 @ ab2d84fe1023/Python/ceval.c

This annotation covers BINARY_OP specializations for integer and string arithmetic. See python_ceval51_detail for LOAD_GLOBAL specializations.

Map​

LinesSymbolRole
1-80BINARY_OP_ADD_INTInteger addition specialization
81-160BINARY_OP_ADD_UNICODEString concatenation specialization
161-240BINARY_OP_MULTIPLY_INTInteger multiplication
241-320BINARY_OP_SUBTRACT_INTInteger subtraction
321-500BINARY_OP (generic)Fallback with slot dispatch

Reading​

BINARY_OP_ADD_INT​

// CPython: Python/ceval.c:4620 BINARY_OP_ADD_INT
inst(BINARY_OP_ADD_INT, (unused/1, left, right -- res)) {
    DEOPT_IF(!PyLong_CheckExact(left) || !PyLong_CheckExact(right));
    STAT_INC(BINARY_OP, hit);
    res = _PyLong_Add((PyLongObject *)left, (PyLongObject *)right);
    DECREF_INPUTS();
    ERROR_IF(res == NULL, error);
    DISPATCH();
}

_PyLong_Add has a fast path for small integers (those that fit in a single digit). It checks for the compact int representation and does a direct C addition, falling back to multi-digit arithmetic only when necessary.

BINARY_OP_ADD_UNICODE​

// CPython: Python/ceval.c:4680 BINARY_OP_ADD_UNICODE
inst(BINARY_OP_ADD_UNICODE, (unused/1, left, right -- res)) {
    DEOPT_IF(!PyUnicode_CheckExact(left) || !PyUnicode_CheckExact(right));
    STAT_INC(BINARY_OP, hit);
    res = PyUnicode_Concat(left, right);
    DECREF_INPUTS();
    ERROR_IF(res == NULL, error);
    DISPATCH();
}

s1 + s2 for exact unicode strings. PyUnicode_Concat allocates a new string of the combined length. Note: repeated + in a loop is O(n²); use ''.join(...) instead.

BINARY_OP_MULTIPLY_INT​

// CPython: Python/ceval.c:4740 BINARY_OP_MULTIPLY_INT
inst(BINARY_OP_MULTIPLY_INT, (unused/1, left, right -- res)) {
    DEOPT_IF(!PyLong_CheckExact(left) || !PyLong_CheckExact(right));
    STAT_INC(BINARY_OP, hit);
    res = _PyLong_Multiply((PyLongObject *)left, (PyLongObject *)right);
    DECREF_INPUTS();
    ERROR_IF(res == NULL, error);
    DISPATCH();
}

_PyLong_Multiply also has a single-digit fast path. For larger integers it uses Karatsuba multiplication (O(n^1.585)) above a threshold.

gopy notes​

BINARY_OP_ADD_INT is in vm/eval_simple.go. It calls objects.IntAdd which uses Go's int64 fast path for small integers and falls back to big.Int.Add for large ones. BINARY_OP_ADD_UNICODE calls objects.UnicodeConcat. BINARY_OP_MULTIPLY_INT calls objects.IntMul.