mirror of https://github.com/python/cpython
Issue #19171: speed some cases of 3-argument long pow().
Reduce the base by the modulus when the base is larger than the modulus. This can unboundedly speed the "startup costs" of doing modular exponentiation, particularly in cases where the base is much larger than the modulus. Original patch by Armin Rigo, inspired by https://github.com/pyca/ed25519.
This commit is contained in:
parent
7760b4eb4b
commit
81a93159d7
|
@ -3868,10 +3868,16 @@ long_pow(PyObject *v, PyObject *w, PyObject *x)
|
||||||
goto Done;
|
goto Done;
|
||||||
}
|
}
|
||||||
|
|
||||||
/* if base < 0:
|
/* Reduce base by modulus in some cases:
|
||||||
base = base % modulus
|
1. If base < 0. Forcing the base non-negative makes things easier.
|
||||||
Having the base positive just makes things easier. */
|
2. If base is obviously larger than the modulus. The "small
|
||||||
if (Py_SIZE(a) < 0) {
|
exponent" case later can multiply directly by base repeatedly,
|
||||||
|
while the "large exponent" case multiplies directly by base 31
|
||||||
|
times. It can be unboundedly faster to multiply by
|
||||||
|
base % modulus instead.
|
||||||
|
We could _always_ do this reduction, but l_divmod() isn't cheap,
|
||||||
|
so we only do it when it buys something. */
|
||||||
|
if (Py_SIZE(a) < 0 || Py_SIZE(a) > Py_SIZE(c)) {
|
||||||
if (l_divmod(a, c, NULL, &temp) < 0)
|
if (l_divmod(a, c, NULL, &temp) < 0)
|
||||||
goto Error;
|
goto Error;
|
||||||
Py_DECREF(a);
|
Py_DECREF(a);
|
||||||
|
|
Loading…
Reference in New Issue