flint3-sys 3.6.0

Rust bindings to the FLINT C library
Documentation
/*
    Copyright (C) 2026 Fredrik Johansson

    This file is part of FLINT.

    FLINT is free software: you can redistribute it and/or modify it under
    the terms of the GNU Lesser General Public License (LGPL) as published
    by the Free Software Foundation; either version 3 of the License, or
    (at your option) any later version.  See <https://www.gnu.org/licenses/>.
*/

#include "radix.h"

/*
    Karp-Markstein Hensel division.

    Develops n limbs of the p-adic (Hensel) quotient q with

        q * b == a   (mod B^n),

    using one Karp-Markstein refinement step on top of a half-precision Hensel
    inverse of b. This is the Hensel analogue of radix_div_approx; it is
    asymptotically fast (cost dominated by an inverse and a few multiplications,
    O(M(n)) when bn is comparable to n), and so is preferable to the schoolbook
    radix_divmod_bn_classical when the divisor b is long.

    With m = ceil(n/2):

        y  = b^(-1) mod B^m                 (half-precision inverse)
        q0 = a * y mod B^m                  (low m limbs of q)
        d  = a - b*q0                       (== 0 mod B^m, since b*q0 == a mod B^m)
        qh = y * (d / B^m) mod B^{n-m}      (high n-m limbs of q)
        q  = q0 + B^m * qh

    Correctness: writing b*y = 1 + B^m*s, with q = q0 + y*d and d = a - b*q0,

        b*q = b*q0 + b*y*d = b*q0 + (1 + B^m s) d = a + B^m s d == a (mod B^{2m}),

    and 2m >= n, so q*b == a (mod B^n). The standard "compute b^(-1) to full
    precision then multiply by a" would instead need the inverse to n limbs; the
    Karp-Markstein trick halves that.

    Two of the products only need their high limbs, whose low limbs are known in
    advance, so they use _radix_mulhigh_known_low (an approximate high product):
      - b*q0: only limbs [m, n) are used and (b*q0) == a (mod B^m);
      - q*b (for the remainder): only limbs [n, n+bn) are used and q*b == a
        (mod B^n).

    q receives n quotient limbs (caller provides room for n) and may alias a.
    Returns 1 on success and 0 if b[0] is not invertible modulo B (nothing is
    written in that case).

    If rem != NULL it receives the resume (Henselian) remainder (a - q*b)/B^n
    reduced to its low bn limbs (room for bn required), unnormalised; then
    a == q*b + B^n*rem (mod B^{n+bn}). Exactness can be tested by the caller as
    in radix_divmod_bn_classical.
*/
int
radix_divmod_bn_karp_markstein(nn_ptr q, nn_ptr rem, nn_srcptr a, slong an,
    nn_srcptr b, slong bn, slong n, const radix_t radix)
{
    slong m, nm;
    nn_ptr y, qf;
    TMP_INIT;

    FLINT_ASSERT(an >= 1);
    FLINT_ASSERT(bn >= 1);
    FLINT_ASSERT(n >= 1);

    TMP_START;

    m = (n + 1) / 2;            /* ceil(n/2) */
    nm = n - m;                 /* floor(n/2), 0 <= nm <= m */

    /* y = b^(-1) mod B^m; 0 iff b[0] is a non-unit. */
    y = TMP_ALLOC(m * sizeof(ulong));
    if (!radix_invmod_bn(y, b, bn, m, radix))
    {
        TMP_END;
        return 0;
    }

    /* Build the quotient in scratch so that q may alias a. */
    qf = TMP_ALLOC(n * sizeof(ulong));

    /* q0 = a * y mod B^m -> low m limbs (a low product) */
    radix_mulmid(qf, a, FLINT_MIN(an, m), y, m, 0, m, radix);

    if (nm > 0)
    {
        nn_ptr bq0h, dh, scratch;
        slong ah;
        ulong bo;

        scratch = TMP_ALLOC(n * sizeof(ulong));

        /* bq0h = (b*q0)[m, n); the low m limbs of b*q0 equal a mod B^m. */
        bq0h = TMP_ALLOC(nm * sizeof(ulong));
        _radix_mulhigh_known_low(bq0h, b, bn, qf, m, a, an, m, n, scratch, radix);

        /* dh = (a - b*q0)[m, n) = a[m, n) - bq0h, where a is zero above limb an.
           No incoming borrow (the low m limbs of a - b*q0 vanish). Built without
           a zero-padded operand: subtract where a has limbs, negate the rest,
           fold in the borrow from the low subtraction. */
        dh = TMP_ALLOC(nm * sizeof(ulong));
        ah = (an > m) ? FLINT_MIN(an - m, nm) : 0;
        bo = 0;
        if (ah > 0)
            bo = radix_sub(dh, a + m, ah, bq0h, ah, radix);
        if (ah < nm)
        {
            radix_neg(dh + ah, bq0h + ah, nm - ah, radix);
            if (bo)
                radix_sub(dh + ah, dh + ah, nm - ah, &bo, 1, radix);
        }

        /* qh = y * dh mod B^{n-m} -> high n-m limbs (a low product) */
        radix_mulmid(qf + m, y, m, dh, nm, 0, nm, radix);
    }

    if (rem != NULL)
    {
        nn_ptr prodh, scratch;
        slong ac;
        ulong bo;

        scratch = TMP_ALLOC((n + bn) * sizeof(ulong));

        /* prodh = (q*b)[n, n+bn); the low n limbs of q*b equal a mod B^n. */
        prodh = TMP_ALLOC(bn * sizeof(ulong));
        _radix_mulhigh_known_low(prodh, qf, n, b, bn, a, an, n, n + bn, scratch, radix);

        /* rem = (a - q*b)/B^n mod B^bn = a[n, n+bn) - prodh, a zero above an.
           No incoming borrow; built without a zero-padded operand. */
        ac = (an > n) ? FLINT_MIN(an - n, bn) : 0;
        bo = 0;
        if (ac > 0)
            bo = radix_sub(rem, a + n, ac, prodh, ac, radix);
        if (ac < bn)
        {
            radix_neg(rem + ac, prodh + ac, bn - ac, radix);
            if (bo)
                radix_sub(rem + ac, rem + ac, bn - ac, &bo, 1, radix);
        }
    }

    flint_mpn_copyi(q, qf, n);

    TMP_END;
    return 1;
}