view release on metacpan or  search on metacpan

libsecp256k1/src/modinv32_impl.h  view on Meta::CPAN

     * down by 30 bits). */
    for (i = 1; i < 9; ++i) {
        fi = f->v[i];
        gi = g->v[i];
        cf += (int64_t)u * fi + (int64_t)v * gi;
        cg += (int64_t)q * fi + (int64_t)r * gi;
        f->v[i - 1] = (int32_t)cf & M30; cf >>= 30;
        g->v[i - 1] = (int32_t)cg & M30; cg >>= 30;
    }
    /* What remains is limb 9 of t*[f,g]; store it as output limb 8. */
    f->v[8] = (int32_t)cf;
    g->v[8] = (int32_t)cg;
}

/* Compute (t/2^30) * [f, g], where t is a transition matrix for 30 divsteps.
 *
 * Version that operates on a variable number of limbs in f and g.
 *
 * This implements the update_fg function from the explanation in modinv64_impl.h.
 */
static void secp256k1_modinv32_update_fg_30_var(int len, secp256k1_modinv32_signed30 *f, secp256k1_modinv32_signed30 *g, const secp256k1_modinv32_trans2x2 *t) {
    const int32_t M30 = (int32_t)(UINT32_MAX >> 2);
    const int32_t u = t->u, v = t->v, q = t->q, r = t->r;
    int32_t fi, gi;
    int64_t cf, cg;
    int i;
    VERIFY_CHECK(len > 0);
    /* Start computing t*[f,g]. */
    fi = f->v[0];
    gi = g->v[0];
    cf = (int64_t)u * fi + (int64_t)v * gi;
    cg = (int64_t)q * fi + (int64_t)r * gi;
    /* Verify that the bottom 62 bits of the result are zero, and then throw them away. */
    VERIFY_CHECK(((int32_t)cf & M30) == 0); cf >>= 30;
    VERIFY_CHECK(((int32_t)cg & M30) == 0); cg >>= 30;
    /* Now iteratively compute limb i=1..len of t*[f,g], and store them in output limb i-1 (shifting
     * down by 30 bits). */
    for (i = 1; i < len; ++i) {
        fi = f->v[i];
        gi = g->v[i];
        cf += (int64_t)u * fi + (int64_t)v * gi;
        cg += (int64_t)q * fi + (int64_t)r * gi;
        f->v[i - 1] = (int32_t)cf & M30; cf >>= 30;
        g->v[i - 1] = (int32_t)cg & M30; cg >>= 30;
    }
    /* What remains is limb (len) of t*[f,g]; store it as output limb (len-1). */
    f->v[len - 1] = (int32_t)cf;
    g->v[len - 1] = (int32_t)cg;
}

/* Compute the inverse of x modulo modinfo->modulus, and replace x with it (constant time in x). */
static void secp256k1_modinv32(secp256k1_modinv32_signed30 *x, const secp256k1_modinv32_modinfo *modinfo) {
    /* Start with d=0, e=1, f=modulus, g=x, zeta=-1. */
    secp256k1_modinv32_signed30 d = {{0}};
    secp256k1_modinv32_signed30 e = {{1}};
    secp256k1_modinv32_signed30 f = modinfo->modulus;
    secp256k1_modinv32_signed30 g = *x;
    int i;
    int32_t zeta = -1; /* zeta = -(delta+1/2); delta is initially 1/2. */

    /* Do 20 iterations of 30 divsteps each = 600 divsteps. 590 suffices for 256-bit inputs. */
    for (i = 0; i < 20; ++i) {
        /* Compute transition matrix and new zeta after 30 divsteps. */
        secp256k1_modinv32_trans2x2 t;
        zeta = secp256k1_modinv32_divsteps_30(zeta, f.v[0], g.v[0], &t);
        /* Update d,e using that transition matrix. */
        secp256k1_modinv32_update_de_30(&d, &e, &t, modinfo);
        /* Update f,g using that transition matrix. */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, 9, &modinfo->modulus, -1) > 0); /* f > -modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, 9, &modinfo->modulus, 1) <= 0); /* f <= modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, 9, &modinfo->modulus, -1) > 0); /* g > -modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, 9, &modinfo->modulus, 1) < 0);  /* g <  modulus */

        secp256k1_modinv32_update_fg_30(&f, &g, &t);

        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, 9, &modinfo->modulus, -1) > 0); /* f > -modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, 9, &modinfo->modulus, 1) <= 0); /* f <= modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, 9, &modinfo->modulus, -1) > 0); /* g > -modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, 9, &modinfo->modulus, 1) < 0);  /* g <  modulus */
    }

    /* At this point sufficient iterations have been performed that g must have reached 0
     * and (if g was not originally 0) f must now equal +/- GCD of the initial f, g
     * values i.e. +/- 1, and d now contains +/- the modular inverse. */

    /* g == 0 */
    VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, 9, &SECP256K1_SIGNED30_ONE, 0) == 0);
    /* |f| == 1, or (x == 0 and d == 0 and f == modulus) */
    VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, 9, &SECP256K1_SIGNED30_ONE, -1) == 0 ||
                 secp256k1_modinv32_mul_cmp_30(&f, 9, &SECP256K1_SIGNED30_ONE, 1) == 0 ||
                 (secp256k1_modinv32_mul_cmp_30(x, 9, &SECP256K1_SIGNED30_ONE, 0) == 0 &&
                  secp256k1_modinv32_mul_cmp_30(&d, 9, &SECP256K1_SIGNED30_ONE, 0) == 0 &&
                  secp256k1_modinv32_mul_cmp_30(&f, 9, &modinfo->modulus, 1) == 0));

    /* Optionally negate d, normalize to [0,modulus), and return it. */
    secp256k1_modinv32_normalize_30(&d, f.v[8], modinfo);
    *x = d;
}

/* Compute the inverse of x modulo modinfo->modulus, and replace x with it (variable time). */
static void secp256k1_modinv32_var(secp256k1_modinv32_signed30 *x, const secp256k1_modinv32_modinfo *modinfo) {
    /* Start with d=0, e=1, f=modulus, g=x, eta=-1. */
    secp256k1_modinv32_signed30 d = {{0, 0, 0, 0, 0, 0, 0, 0, 0}};
    secp256k1_modinv32_signed30 e = {{1, 0, 0, 0, 0, 0, 0, 0, 0}};
    secp256k1_modinv32_signed30 f = modinfo->modulus;
    secp256k1_modinv32_signed30 g = *x;
#ifdef VERIFY
    int i = 0;
#endif
    int j, len = 9;
    int32_t eta = -1; /* eta = -delta; delta is initially 1 (faster for the variable-time code) */
    int32_t cond, fn, gn;

    /* Do iterations of 30 divsteps each until g=0. */
    while (1) {
        /* Compute transition matrix and new eta after 30 divsteps. */
        secp256k1_modinv32_trans2x2 t;
        eta = secp256k1_modinv32_divsteps_30_var(eta, f.v[0], g.v[0], &t);
        /* Update d,e using that transition matrix. */
        secp256k1_modinv32_update_de_30(&d, &e, &t, modinfo);
        /* Update f,g using that transition matrix. */

        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, -1) > 0); /* f > -modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, 1) <= 0); /* f <= modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &modinfo->modulus, -1) > 0); /* g > -modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &modinfo->modulus, 1) < 0);  /* g <  modulus */

        secp256k1_modinv32_update_fg_30_var(len, &f, &g, &t);
        /* If the bottom limb of g is 0, there is a chance g=0. */
        if (g.v[0] == 0) {
            cond = 0;
            /* Check if all other limbs are also 0. */
            for (j = 1; j < len; ++j) {
                cond |= g.v[j];
            }
            /* If so, we're done. */
            if (cond == 0) break;
        }

        /* Determine if len>1 and limb (len-1) of both f and g is 0 or -1. */
        fn = f.v[len - 1];
        gn = g.v[len - 1];
        cond = ((int32_t)len - 2) >> 31;
        cond |= fn ^ (fn >> 31);
        cond |= gn ^ (gn >> 31);
        /* If so, reduce length, propagating the sign of f and g's top limb into the one below. */
        if (cond == 0) {
            f.v[len - 2] |= (uint32_t)fn << 30;
            g.v[len - 2] |= (uint32_t)gn << 30;
            --len;
        }

        VERIFY_CHECK(++i < 25); /* We should never need more than 25*30 = 750 divsteps */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, -1) > 0); /* f > -modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, 1) <= 0); /* f <= modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &modinfo->modulus, -1) > 0); /* g > -modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &modinfo->modulus, 1) < 0);  /* g <  modulus */
    }

    /* At this point g is 0 and (if g was not originally 0) f must now equal +/- GCD of
     * the initial f, g values i.e. +/- 1, and d now contains +/- the modular inverse. */

    /* g == 0 */
    VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &SECP256K1_SIGNED30_ONE, 0) == 0);
    /* |f| == 1, or (x == 0 and d == 0 and f == modulus) */
    VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &SECP256K1_SIGNED30_ONE, -1) == 0 ||
                 secp256k1_modinv32_mul_cmp_30(&f, len, &SECP256K1_SIGNED30_ONE, 1) == 0 ||
                 (secp256k1_modinv32_mul_cmp_30(x, 9, &SECP256K1_SIGNED30_ONE, 0) == 0 &&
                  secp256k1_modinv32_mul_cmp_30(&d, 9, &SECP256K1_SIGNED30_ONE, 0) == 0 &&
                  secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, 1) == 0));

    /* Optionally negate d, normalize to [0,modulus), and return it. */
    secp256k1_modinv32_normalize_30(&d, f.v[len - 1], modinfo);
    *x = d;
}

/* Do up to 50 iterations of 30 posdivsteps (up to 1500 steps; more is extremely rare) each until f=1.
 * In VERIFY mode use a lower number of iterations (750, close to the median 756), so failure actually occurs. */
#ifdef VERIFY
#define JACOBI32_ITERATIONS 25
#else
#define JACOBI32_ITERATIONS 50
#endif

/* Compute the Jacobi symbol of x modulo modinfo->modulus (variable time). gcd(x,modulus) must be 1. */
static int secp256k1_jacobi32_maybe_var(const secp256k1_modinv32_signed30 *x, const secp256k1_modinv32_modinfo *modinfo) {
    /* Start with f=modulus, g=x, eta=-1. */
    secp256k1_modinv32_signed30 f = modinfo->modulus;
    secp256k1_modinv32_signed30 g = *x;
    int j, len = 9;
    int32_t eta = -1; /* eta = -delta; delta is initially 1 */
    int32_t cond, fn, gn;
    int jac = 0;
    int count;

    /* The input limbs must all be non-negative. */
    VERIFY_CHECK(g.v[0] >= 0 && g.v[1] >= 0 && g.v[2] >= 0 && g.v[3] >= 0 && g.v[4] >= 0 && g.v[5] >= 0 && g.v[6] >= 0 && g.v[7] >= 0 && g.v[8] >= 0);

    /* If x > 0, then if the loop below converges, it converges to f=g=gcd(x,modulus). Since we
     * require that gcd(x,modulus)=1 and modulus>=3, x cannot be 0. Thus, we must reach f=1 (or
     * time out). */
    VERIFY_CHECK((g.v[0] | g.v[1] | g.v[2] | g.v[3] | g.v[4] | g.v[5] | g.v[6] | g.v[7] | g.v[8]) != 0);

    for (count = 0; count < JACOBI32_ITERATIONS; ++count) {
        /* Compute transition matrix and new eta after 30 posdivsteps. */
        secp256k1_modinv32_trans2x2 t;
        eta = secp256k1_modinv32_posdivsteps_30_var(eta, f.v[0] | ((uint32_t)f.v[1] << 30), g.v[0] | ((uint32_t)g.v[1] << 30), &t, &jac);
        /* Update f,g using that transition matrix. */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, 0) > 0); /* f > 0 */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, 1) <= 0); /* f <= modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &modinfo->modulus, 0) > 0); /* g > 0 */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &modinfo->modulus, 1) < 0);  /* g < modulus */

        secp256k1_modinv32_update_fg_30_var(len, &f, &g, &t);
        /* If the bottom limb of f is 1, there is a chance that f=1. */
        if (f.v[0] == 1) {
            cond = 0;
            /* Check if the other limbs are also 0. */
            for (j = 1; j < len; ++j) {
                cond |= f.v[j];
            }
            /* If so, we're done. If f=1, the Jacobi symbol (g | f)=1. */
            if (cond == 0) return 1 - 2*(jac & 1);
        }

        /* Determine if len>1 and limb (len-1) of both f and g is 0. */
        fn = f.v[len - 1];
        gn = g.v[len - 1];
        cond = ((int32_t)len - 2) >> 31;
        cond |= fn;
        cond |= gn;
        /* If so, reduce length. */
        if (cond == 0) --len;

        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, 0) > 0); /* f > 0 */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&f, len, &modinfo->modulus, 1) <= 0); /* f <= modulus */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &modinfo->modulus, 0) > 0); /* g > 0 */
        VERIFY_CHECK(secp256k1_modinv32_mul_cmp_30(&g, len, &modinfo->modulus, 1) < 0);  /* g < modulus */
    }

    /* The loop failed to converge to f=g after 1500 iterations. Return 0, indicating unknown result. */
    return 0;
}

#endif /* SECP256K1_MODINV32_IMPL_H */