void
_nmod_poly_div_basecase_2(mp_ptr Q, mp_ptr W,
mp_srcptr A, long A_len, mp_srcptr B, long B_len,
nmod_t mod)
{
long coeff, i, len;
mp_limb_t lead_inv = n_invmod(B[B_len - 1], mod.n);
mp_ptr B2, R2;
mp_srcptr Btop;
B2 = W;
for (i = 0; i < B_len - 1; i++)
{
B2[2 * i] = B[i];
B2[2 * i + 1] = 0;
}
Btop = B2 + 2*(B_len - 1);
R2 = W + 2*(B_len - 1);
for (i = 0; i < A_len - B_len + 1; i++)
{
R2[2 * i] = A[B_len + i - 1];
R2[2 * i + 1] = 0;
}
coeff = A_len - B_len;
while (coeff >= 0)
{
mp_limb_t r_coeff;
r_coeff =
n_ll_mod_preinv(R2[2 * coeff + 1], R2[2 * coeff], mod.n, mod.ninv);
while (coeff >= 0 && r_coeff == 0L)
{
Q[coeff--] = 0L;
if (coeff >= 0)
r_coeff =
n_ll_mod_preinv(R2[2 * coeff + 1], R2[2 * coeff], mod.n,
mod.ninv);
}
if (coeff >= 0)
{
mp_limb_t c, * R_sub;
Q[coeff] =
n_mulmod2_preinv(r_coeff, lead_inv, mod.n, mod.ninv);
c = n_negmod(Q[coeff], mod.n);
len = FLINT_MIN(B_len - 1, coeff);
R_sub = R2 + 2 * (coeff - len);
if (len > 0)
mpn_addmul_1(R_sub, Btop - 2*len, 2 * len, c);
coeff--;
}
}
}
mp_limb_t
n_powmod_precomp(mp_limb_t a, mp_limb_signed_t exp, mp_limb_t n, double npre)
{
mp_limb_t x, y;
mp_limb_signed_t e;
if (n == 1UL)
return 0L;
e = (exp < 0L ? -exp : exp);
x = 1UL;
y = a;
while (e)
{
if (e & 1L)
x = n_mulmod_precomp(x, y, n, npre);
e >>= 1;
if (e)
y = n_mulmod_precomp(y, y, n, npre);
}
return (exp < 0L ? n_invmod(x, n) : x);
}
void
_nmod_poly_revert_series_lagrange(mp_ptr Qinv, mp_srcptr Q, long n, nmod_t mod)
{
long i;
mp_ptr R, S, T, tmp;
if (n >= 1) Qinv[0] = 0UL;
if (n >= 2) Qinv[1] = n_invmod(Q[1], mod.n);
if (n <= 2)
return;
R = _nmod_vec_init(n - 1);
S = _nmod_vec_init(n - 1);
T = _nmod_vec_init(n - 1);
_nmod_poly_inv_series(R, Q + 1, n - 1, mod);
_nmod_vec_set(S, R, n - 1);
for (i = 2; i < n; i++)
{
_nmod_poly_mullow(T, S, n - 1, R, n - 1, n - 1, mod);
Qinv[i] = nmod_div(T[i - 1], i, mod);
tmp = S; S = T; T = tmp;
}
_nmod_vec_clear(R);
_nmod_vec_clear(S);
_nmod_vec_clear(T);
}
void qsieve_ll_compute_B_terms(qs_t qs_inf)
{
long s = qs_inf->s;
mp_limb_t * A_ind = qs_inf->A_ind;
mp_limb_t * A_modp = qs_inf->A_modp;
mp_limb_t * B_terms = qs_inf->B_terms;
prime_t * factor_base = qs_inf->factor_base;
mp_limb_t A = qs_inf->A;
mp_limb_t B;
mp_limb_t p, temp, temp2, pinv;
long i;
for (i = 0; i < s; i++)
{
p = factor_base[A_ind[i]].p;
pinv = factor_base[A_ind[i]].pinv;
temp = A/p; /* TODO: possibly use precomputed inverse here */
A_modp[i] = (temp2 = n_mod2_preinv(temp, p, pinv));
temp2 = n_invmod(temp2, p);
temp2 = n_mulmod2_preinv(temp2, qs_inf->sqrts[A_ind[i]], p, pinv);
if (temp2 > p/2) temp2 = p - temp2;
B_terms[i] = temp*temp2;
}
B = B_terms[0];
for (i = 1; i < s; i++)
{
B += B_terms[i];
}
qs_inf->B = B;
}
void
fmpz_mat_CRT_ui(fmpz_mat_t res, const fmpz_mat_t mat1,
const fmpz_t m1, const nmod_mat_t mat2, int sign)
{
long i, j;
mp_limb_t c;
mp_limb_t m2 = mat2->mod.n;
mp_limb_t m2inv = mat2->mod.ninv;
fmpz_t m1m2;
c = fmpz_fdiv_ui(m1, m2);
c = n_invmod(c, m2);
if (c == 0)
{
printf("Exception in fmpz_mat_CRT_ui: m1 not invertible modulo m2!\n");
abort();
}
fmpz_init(m1m2);
fmpz_mul_ui(m1m2, m1, m2);
for (i = 0; i < mat1->r; i++)
{
for (j = 0; j < mat1->c; j++)
_fmpz_CRT_ui_precomp(fmpz_mat_entry(res, i, j),
fmpz_mat_entry(mat1, i, j), m1,
nmod_mat_entry(mat2, i, j), m2, m2inv, m1m2, c, sign);
}
fmpz_clear(m1m2);
}
void _nmod_poly_rem_basecase_1(mp_ptr R, mp_ptr W,
mp_srcptr A, long lenA, mp_srcptr B, long lenB,
nmod_t mod)
{
if (lenB > 1)
{
const mp_limb_t invL = n_invmod(B[lenB - 1], mod.n);
long iR;
mp_ptr R1 = W;
mpn_copyi(R1, A, lenA);
for (iR = lenA - 1; iR >= lenB - 1; iR--)
{
if (R1[iR] != 0)
{
const mp_limb_t q = n_mulmod2_preinv(R1[iR], invL, mod.n, mod.ninv);
const mp_limb_t c = n_negmod(q, mod.n);
mpn_addmul_1(R1 + iR - lenB + 1, B, lenB - 1, c);
}
}
_nmod_vec_reduce(R, R1, lenB - 1, mod);
}
}
void
_nmod_poly_divrem_basecase_1(mp_ptr Q, mp_ptr R, mp_ptr W,
mp_srcptr A, slong lenA, mp_srcptr B, slong lenB,
nmod_t mod)
{
const mp_limb_t invL = n_invmod(B[lenB - 1], mod.n);
slong iR;
mp_ptr ptrQ = Q - lenB + 1;
mp_ptr R1 = W;
flint_mpn_copyi(R1, A, lenA);
for (iR = lenA - 1; iR >= lenB - 1; iR--)
{
if (R1[iR] == 0)
{
ptrQ[iR] = WORD(0);
}
else
{
ptrQ[iR] = n_mulmod2_preinv(R1[iR], invL, mod.n, mod.ninv);
if (lenB > 1)
{
const mp_limb_t c = n_negmod(ptrQ[iR], mod.n);
mpn_addmul_1(R1 + iR - lenB + 1, B, lenB - 1, c);
}
}
}
if (lenB > 1)
_nmod_vec_reduce(R, R1, lenB - 1, mod);
}
void
_nmod_poly_taylor_shift_convolution(mp_ptr p, mp_limb_t c, slong len, nmod_t mod)
{
slong i, n = len - 1;
mp_limb_t f, d;
mp_ptr t, u;
if (c == 0 || len <= 1)
return;
t = _nmod_vec_init(len);
u = _nmod_vec_init(len);
f = 1;
for (i = 2; i <= n; i++)
{
f = n_mulmod2_preinv(f, i, mod.n, mod.ninv);
p[i] = n_mulmod2_preinv(p[i], f, mod.n, mod.ninv);
}
_nmod_poly_reverse(p, p, len, len);
t[n] = 1;
for (i = n; i > 0; i--)
t[i - 1] = n_mulmod2_preinv(t[i], i, mod.n, mod.ninv);
if (c == mod.n - 1)
{
for (i = 1; i <= n; i += 2)
t[i] = nmod_neg(t[i], mod);
}
else if (c != 1)
{
d = c;
for (i = 1; i <= n; i++)
{
t[i] = n_mulmod2_preinv(t[i], d, mod.n, mod.ninv);
d = n_mulmod2_preinv(d, c, mod.n, mod.ninv);
}
}
_nmod_poly_mullow(u, p, len, t, len, len, mod);
f = n_mulmod2_preinv(f, f, mod.n, mod.ninv);
f = n_invmod(f, mod.n);
for (i = n; i >= 0; i--)
{
p[i] = n_mulmod2_preinv(u[n - i], f, mod.n, mod.ninv);
f = n_mulmod2_preinv(f, (i == 0) ? 1 : i, mod.n, mod.ninv);
}
_nmod_vec_clear(t);
_nmod_vec_clear(u);
}
static __inline__ mp_limb_t
invert_det_divisor_modulo_pk(mpz_t dd,p_k_pk_t const* pp,nmod_t const* mod)
// take dd modulo p_deg_k, then invert it
{
mp_limb_t r,m=pp->p_deg_k;
r=mpz_fdiv_ui(dd, m);
if(pp->k==1)
return n_invmod(r,m);
else
return inv_mod_pk_3arg(r,pp[0],mod[0]);
}
mp_limb_t
n_powmod2_preinv(mp_limb_t a, mp_limb_signed_t exp, mp_limb_t n, mp_limb_t ninv)
{
if (exp < WORD(0))
{
a = n_invmod(a, n);
exp = -exp;
}
return n_powmod2_ui_preinv(a, exp, n, ninv);
}
void _nmod_poly_integral(mp_ptr x_int, mp_srcptr x, slong len, nmod_t mod)
{
mp_limb_t r;
slong k = len - 1;
while (k > 0)
{
if (k > 3 && k < PROD_TAKE4)
{
r = n_invmod(k*(k-1)*(k-2)*(k-3), mod.n);
x_int[k] = MUL3(x[k-1], r, (k-1)*(k-2)*(k-3));
x_int[k-1] = MUL3(x[k-2], r, k*(k-2)*(k-3));
x_int[k-2] = MUL3(x[k-3], r, k*(k-1)*(k-3));
x_int[k-3] = MUL3(x[k-4], r, k*(k-1)*(k-2));
k -= 4;
}
else if (k > 2 && k < PROD_TAKE3)
{
r = n_invmod(k*(k-1)*(k-2), mod.n);
x_int[k] = MUL3(x[k-1], r, (k-1)*(k-2));
x_int[k-1] = MUL3(x[k-2], r, k*(k-2));
x_int[k-2] = MUL3(x[k-3], r, k*(k-1));
k -= 3;
}
else if (k > 1 && k < PROD_TAKE2)
{
r = n_invmod(k*(k-1), mod.n);
x_int[k] = MUL3(x[k-1], r, k-1);
x_int[k-1] = MUL3(x[k-2], r, k);
k -= 2;
}
else
{
r = n_invmod(k, mod.n);
x_int[k] = n_mulmod2_preinv(x[k-1], r, mod.n, mod.ninv);
k -= 1;
}
}
x_int[0] = UWORD(0);
}
void
_nmod_poly_divrem_basecase_3(mp_ptr Q, mp_ptr R, mp_ptr W,
mp_srcptr A, slong lenA, mp_srcptr B, slong lenB,
nmod_t mod)
{
const mp_limb_t invL = n_invmod(B[lenB - 1], mod.n);
slong iR, i;
mp_ptr B3 = W, R3 = W + 3*(lenB - 1), ptrQ = Q - lenB + 1;
for (i = 0; i < lenB - 1; i++)
{
B3[3 * i] = B[i];
B3[3 * i + 1] = 0;
B3[3 * i + 2] = 0;
}
for (i = 0; i < lenA; i++)
{
R3[3 * i] = A[i];
R3[3 * i + 1] = 0;
R3[3 * i + 2] = 0;
}
for (iR = lenA - 1; iR >= lenB - 1; )
{
mp_limb_t r =
n_lll_mod_preinv(R3[3 * iR + 2], R3[3 * iR + 1],
R3[3 * iR], mod.n, mod.ninv);
while ((iR + 1 >= lenB) && (r == WORD(0)))
{
ptrQ[iR--] = WORD(0);
if (iR + 1 >= lenB)
r = n_lll_mod_preinv(R3[3 * iR + 2], R3[3 * iR + 1],
R3[3 * iR], mod.n, mod.ninv);
}
if (iR + 1 >= lenB)
{
ptrQ[iR] = n_mulmod2_preinv(r, invL, mod.n, mod.ninv);
if (lenB > 1)
{
const mp_limb_t c = n_negmod(ptrQ[iR], mod.n);
mpn_addmul_1(R3 + 3 * (iR - lenB + 1), B3, 3 * lenB - 3, c);
}
iR--;
}
}
for (iR = 0; iR < lenB - 1; iR++)
R[iR] = n_lll_mod_preinv(R3[3 * iR + 2], R3[3 * iR + 1],
R3[3 * iR], mod.n, mod.ninv);
}
void qsieve_ll_compute_A_factor_offsets(qs_t qs_inf)
{
long s = qs_inf->s;
mp_limb_t * A_ind = qs_inf->A_ind;
mp_limb_t * A_modp = qs_inf->A_modp;
mp_limb_t * soln1 = qs_inf->soln1;
mp_limb_t * soln2 = qs_inf->soln2;
mp_limb_t p, D;
mp_limb_t hi = qs_inf->hi;
mp_limb_t lo = qs_inf->lo;
mp_limb_t B = qs_inf->B;
mp_limb_t temp, temp2, B_modp2, index, p2;
prime_t * factor_base = qs_inf->factor_base;
mp_limb_t * inv_p2 = qs_inf->inv_p2;
mp_limb_t pinv;
long j;
for (j = 0; j < s; j++)
{
index = A_ind[j];
p = factor_base[index].p;
p2 = p*p;
pinv = factor_base[index].pinv;
D = n_ll_mod_preinv(hi, lo, p*p, inv_p2[j]);
if ((mp_limb_signed_t) B < 0)
{
B_modp2 = n_mod2_preinv(-B, p2, inv_p2[j]);
B_modp2 = p2 - B_modp2;
if (B_modp2 == p2) B_modp2 = 0;
} else
B_modp2 = n_mod2_preinv(B, p2, inv_p2[j]);
temp = B_modp2*A_modp[j];
temp = n_mod2_preinv(temp, p, pinv);
temp2 = n_invmod(temp, p);
D -= (B_modp2*B_modp2);
if ((mp_limb_signed_t) D < 0)
temp = -(-D/p); /* TODO consider using precomputed inverse */
else
temp = (D/p); /* TODO consider using precomputed inverse */
temp *= temp2;
temp += qs_inf->sieve_size/2;
if ((mp_limb_signed_t) temp < 0)
{
temp = p - n_mod2_preinv(-temp, p, pinv);
if (temp == p) temp = 0;
}
else temp = n_mod2_preinv(temp, p, pinv);
soln1[index] = temp;
soln2[index] = -1;
}
}
long
hmod_mat_lu_classical(long * P, hmod_mat_t A, int rank_check)
{
hlimb_t d, e, **a;
nmod_t mod;
long i, m, n, rank, length, row, col;
m = A->r;
n = A->c;
a = A->rows;
mod = A->mod;
rank = row = col = 0;
for (i = 0; i < m; i++)
P[i] = i;
while (row < m && col < n)
{
if (hmod_mat_pivot(A, P, row, col) == 0)
{
if (rank_check)
return 0;
col++;
continue;
}
rank++;
d = a[row][col];
d = n_invmod(d, mod.n);
length = n - col - 1;
for (i = row + 1; i < m; i++)
{
e = n_mulmod2_preinv(a[i][col], d, mod.n, mod.ninv);
if (length != 0)
_hmod_vec_scalar_addmul_hmod(a[i] + col + 1,
a[row] + col + 1, length, nmod_neg(e, mod), mod);
a[i][col] = 0;
a[i][rank - 1] = e;
}
row++;
col++;
}
return rank;
}
void
nmod_mat_solve_tril_classical(nmod_mat_t X, const nmod_mat_t L,
const nmod_mat_t B, int unit)
{
int nlimbs;
long i, j, n, m;
nmod_t mod;
mp_ptr inv, tmp;
n = L->r;
m = B->c;
mod = L->mod;
if (!unit)
{
inv = _nmod_vec_init(n);
for (i = 0; i < n; i++)
inv[i] = n_invmod(nmod_mat_entry(L, i, i), mod.n);
}
else
inv = NULL;
nlimbs = _nmod_vec_dot_bound_limbs(n, mod);
tmp = _nmod_vec_init(n);
for (i = 0; i < m; i++)
{
for (j = 0; j < n; j++)
tmp[j] = nmod_mat_entry(X, j, i);
for (j = 0; j < n; j++)
{
mp_limb_t s;
s = _nmod_vec_dot(L->rows[j], tmp, j, mod, nlimbs);
s = nmod_sub(nmod_mat_entry(B, j, i), s, mod);
if (!unit)
s = n_mulmod2_preinv(s, inv[j], mod.n, mod.ninv);
tmp[j] = s;
}
for (j = 0; j < n; j++)
nmod_mat_entry(X, j, i) = tmp[j];
}
_nmod_vec_clear(tmp);
if (!unit)
_nmod_vec_clear(inv);
}
void qsieve_ll_compute_off_adj(qs_t qs_inf)
{
long num_primes = qs_inf->num_primes;
mp_limb_t A = qs_inf->A;
mp_limb_t B = qs_inf->B;
mp_limb_t * A_inv = qs_inf->A_inv;
mp_limb_t ** A_inv2B = qs_inf->A_inv2B;
mp_limb_t * B_terms = qs_inf->B_terms;
mp_limb_t * soln1 = qs_inf->soln1;
mp_limb_t * soln2 = qs_inf->soln2;
int * sqrts = qs_inf->sqrts;
prime_t * factor_base = qs_inf->factor_base;
long s = qs_inf->s;
mp_limb_t p, temp, pinv;
long i, j;
for (i = 2; i < num_primes; i++) /* skip k and 2 */
{
p = factor_base[i].p;
pinv = factor_base[i].pinv;
A_inv[i] = n_invmod(n_mod2_preinv(A, p, pinv), p);
for (j = 0; j < s; j++)
{
temp = n_mod2_preinv(B_terms[j], p, pinv);
temp = n_mulmod2_preinv(temp, A_inv[i], p, pinv);
temp *= 2;
if (temp >= p) temp -= p;
A_inv2B[j][i] = temp;
}
temp = n_mod2_preinv(B, p, pinv);
temp = sqrts[i] + p - temp;
temp *= A_inv[i];
temp += qs_inf->sieve_size/2;
soln1[i] = n_mod2_preinv(temp, p, pinv);
temp = p - sqrts[i];
if (temp == p) temp -= p;
temp = n_mulmod2_preinv(temp, A_inv[i], p, pinv);
temp *= 2;
if (temp >= p) temp -= p;
soln2[i] = temp + soln1[i];
if (soln2[i] >= p) soln2[i] -= p;
}
}
ulong
dlog_crt_init(dlog_crt_t t, ulong a, ulong mod, ulong n, ulong num)
{
int k;
n_factor_t fac;
ulong * M, * u;
ulong cost = 0;
n_factor_init(&fac);
n_factor(&fac, n, 1);
t->num = fac.num;
nmod_init(&t->mod,mod);
nmod_init(&t->n, n);
M = t->expo = flint_malloc(t->num * sizeof(ulong));
u = t->crt_coeffs = flint_malloc(t->num * sizeof(ulong));
t->pre = flint_malloc(t->num * sizeof(dlog_precomp_struct));
for (k = 0; k < t->num; k++)
{
ulong p, e, mk;
p = fac.p[k];
e = fac.exp[k];
if (0 && mod % p == 0)
{
flint_printf("dlog_crt_init: modulus must be prime to order.\n");
abort();
}
mk = n_pow(p, e);
M[k] = n / mk;
u[k] = nmod_mul(M[k], n_invmod(M[k] % mk, mk), t->n);
/* depends on the power */
#if 0
flint_printf("[sub-crt -- init for size %wu mod %wu]\n", mk, mod);
#endif
dlog_precomp_pe_init(t->pre + k, nmod_pow_ui(a, M[k], t->mod), mod, p, e, mk, num);
cost += t->pre[k].cost;
}
#if 0
if (cost > 500)
flint_printf("[crt init for size %wu mod %wu -> cost %wu]\n", n,mod,cost);
#endif
return cost;
}
void
_nmod_poly_exp_series_monomial_ui(mp_ptr res, mp_limb_t coeff, ulong power,
slong n, nmod_t mod)
{
slong k, r;
mp_limb_t rfac;
mp_limb_t a;
r = (n - 1) / power;
rfac = n_factorial_mod2_preinv(r, mod.n, mod.ninv);
rfac = n_invmod(rfac, mod.n);
if (power > 1)
_nmod_vec_zero(res, n);
res[0] = UWORD(1);
if (coeff == UWORD(1))
{
a = rfac;
for (k = r; k >= 1; k--)
{
res[k * power] = a;
a = n_mulmod2_preinv(a, k, mod.n, mod.ninv);
}
}
else
{
a = coeff;
for (k = power; k < n; k += power)
{
res[k] = a;
a = n_mulmod2_preinv(a, coeff, mod.n, mod.ninv);
}
a = rfac;
for (k = r; k >= 1; k--)
{
res[k * power] = n_mulmod2_preinv(res[k * power],
a, mod.n, mod.ninv);
a = n_mulmod2_preinv(a, k, mod.n, mod.ninv);
}
}
}
void
_nmod_poly_interpolate_nmod_vec_barycentric(mp_ptr poly,
mp_srcptr xs, mp_srcptr ys, slong n, nmod_t mod)
{
mp_ptr P, Q, w;
slong i, j;
if (n == 1)
{
poly[0] = ys[0];
return;
}
P = _nmod_vec_init(n + 1);
Q = _nmod_vec_init(n);
w = _nmod_vec_init(n);
_nmod_poly_product_roots_nmod_vec(P, xs, n, mod);
for (i = 0; i < n; i++)
{
w[i] = UWORD(1);
for (j = 0; j < n; j++)
{
if (i != j)
w[i] = nmod_mul(w[i], nmod_sub(xs[i], xs[j], mod), mod);
}
w[i] = n_invmod(w[i], mod.n);
}
_nmod_vec_zero(poly, n);
for (i = 0; i < n; i++)
{
_nmod_poly_div_root(Q, P, n + 1, xs[i], mod);
_nmod_vec_scalar_addmul_nmod(poly, Q, n,
nmod_mul(w[i], ys[i], mod), mod);
}
_nmod_vec_clear(P);
_nmod_vec_clear(Q);
_nmod_vec_clear(w);
}
void
_nmod_poly_div_basecase_1(mp_ptr Q, mp_ptr W,
mp_srcptr A, long A_len, mp_srcptr B, long B_len,
nmod_t mod)
{
mp_limb_t lead_inv = n_invmod(B[B_len - 1], mod.n);
long len, coeff = A_len - B_len;
mp_ptr R1 = W;
mp_srcptr Btop = B + B_len - 1;
mpn_copyi(R1, A + B_len - 1, A_len - B_len + 1);
while (coeff >= 0)
{
R1[coeff] = n_mod2_preinv(R1[coeff], mod.n, mod.ninv);
while (coeff >= 0 && R1[coeff] == 0L)
{
Q[coeff--] = 0L;
if (coeff >= 0)
R1[coeff] = n_mod2_preinv(R1[coeff], mod.n, mod.ninv);
}
if (coeff >= 0)
{
mp_limb_t c, * R_sub;
Q[coeff] =
n_mulmod2_preinv(R1[coeff], lead_inv, mod.n, mod.ninv);
c = n_negmod(Q[coeff], mod.n);
len = FLINT_MIN(B_len - 1, coeff);
R_sub = R1 + coeff - len;
if (len > 0)
mpn_addmul_1(R_sub, Btop - len, len, c);
coeff--;
}
}
}
void
_nmod_poly_cosh_series(mp_ptr f, mp_srcptr h, long n, nmod_t mod)
{
mp_ptr g, T, U, hprime;
g = _nmod_vec_init(n);
T = _nmod_vec_init(n);
U = _nmod_vec_init(n);
hprime = _nmod_vec_init(n);
_nmod_poly_derivative(hprime, h, n, mod); hprime[n-1] = 0UL;
__nmod_poly_exp_series_prealloc(f, g, h, hprime, T, U, n, mod, 1);
_nmod_vec_add(f, f, g, n, mod);
_nmod_vec_scalar_mul_nmod(f, f, n, n_invmod(2UL, mod.n), mod);
_nmod_vec_free(hprime);
_nmod_vec_free(g);
_nmod_vec_free(T);
_nmod_vec_free(U);
}
void _nmod_poly_rem_basecase_3(mp_ptr R, mp_ptr W,
mp_srcptr A, long lenA, mp_srcptr B, long lenB,
nmod_t mod)
{
if (lenB > 1)
{
const mp_limb_t invL = n_invmod(B[lenB - 1], mod.n);
long iR, i;
mp_ptr B3 = W, R3 = W + 3*(lenB - 1);
for (i = 0; i < lenB - 1; i++)
{
B3[3 * i] = B[i];
B3[3 * i + 1] = 0;
B3[3 * i + 2] = 0;
}
for (i = 0; i < lenA; i++)
{
R3[3 * i] = A[i];
R3[3 * i + 1] = 0;
R3[3 * i + 2] = 0;
}
for (iR = lenA - 1; iR >= lenB - 1; iR--)
{
const mp_limb_t r = n_lll_mod_preinv(R3[3*iR + 2], R3[3*iR + 1],
R3[3*iR], mod.n, mod.ninv);
if (r != 0)
{
const mp_limb_t q = n_mulmod2_preinv(r, invL, mod.n, mod.ninv);
const mp_limb_t c = n_negmod(q, mod.n);
mpn_addmul_1(R3 + 3 * (iR - lenB + 1), B3, 3 * lenB - 3, c);
}
}
for (iR = 0; iR < lenB - 1; iR++)
R[iR] = n_lll_mod_preinv(R3[3 * iR + 2], R3[3 * iR + 1],
R3[3 * iR], mod.n, mod.ninv);
}
}
int nmod_mat_inv(nmod_mat_t B, const nmod_mat_t A)
{
nmod_mat_t I;
long i, dim;
int result;
dim = A->r;
switch (dim)
{
case 0:
result = 1;
break;
case 1:
if (nmod_mat_entry(A, 0, 0) == 0UL)
{
result = 0;
}
else
{
nmod_mat_entry(B, 0, 0) =
n_invmod(nmod_mat_entry(A, 0, 0), B->mod.n);
result = 1;
}
break;
default:
nmod_mat_init(I, dim, dim, B->mod.n);
for (i = 0; i < dim; i++)
nmod_mat_entry(I, i, i) = 1UL;
result = nmod_mat_solve(B, A, I);
nmod_mat_clear(I);
}
return result;
}
void _nmod_poly_divrem_q1(mp_ptr Q, mp_ptr R,
mp_srcptr A, long lenA, mp_srcptr B, long lenB,
nmod_t mod)
{
const mp_limb_t invL = (B[lenB-1] == 1) ? 1 : n_invmod(B[lenB-1], mod.n);
if (lenB == 1)
{
_nmod_vec_scalar_mul_nmod(Q, A, lenA, invL, mod);
}
else
{
mp_limb_t t;
Q[1] = n_mulmod2_preinv(A[lenA-1], invL, mod.n, mod.ninv);
t = n_mulmod2_preinv(Q[1], B[lenB-2], mod.n, mod.ninv);
t = n_submod(A[lenA-2], t, mod.n);
Q[0] = n_mulmod2_preinv(t, invL, mod.n, mod.ninv);
if (FLINT_BITS + 2 <= 2 * mod.norm)
{
mpn_mul_1(R, B, lenB - 1, Q[0]);
if (lenB > 2)
mpn_addmul_1(R + 1, B, lenB - 2, Q[1]);
_nmod_vec_reduce(R, R, lenB - 1, mod);
}
else
{
_nmod_vec_scalar_mul_nmod(R, B, lenB - 1, Q[0], mod);
if (lenB > 2)
_nmod_vec_scalar_addmul_nmod(R + 1, B, lenB - 2, Q[1], mod);
}
_nmod_vec_sub(R, A, R, lenB - 1, mod);
}
}
void
nmod_poly_xgcd_hgcd(nmod_poly_t G, nmod_poly_t S, nmod_poly_t T,
const nmod_poly_t A, const nmod_poly_t B)
{
if (A->length < B->length)
{
nmod_poly_xgcd_hgcd(G, T, S, B, A);
}
else /* lenA >= lenB >= 0 */
{
const slong lenA = A->length, lenB = B->length;
mp_limb_t inv;
if (lenA == 0) /* lenA = lenB = 0 */
{
nmod_poly_zero(G);
nmod_poly_zero(S);
nmod_poly_zero(T);
}
else if (lenB == 0) /* lenA > lenB = 0 */
{
inv = n_invmod(A->coeffs[lenA - 1], A->mod.n);
nmod_poly_scalar_mul_nmod(G, A, inv);
nmod_poly_zero(T);
nmod_poly_set_coeff_ui(S, 0, inv);
S->length = 1;
}
else if (lenB == 1) /* lenA >= lenB = 1 */
{
nmod_poly_fit_length(T, 1);
T->length = 1;
T->coeffs[0] = n_invmod(B->coeffs[0], A->mod.n);
nmod_poly_one(G);
nmod_poly_zero(S);
}
else /* lenA >= lenB >= 2 */
{
mp_ptr g, s, t;
slong lenG;
if (G == A || G == B)
{
g = _nmod_vec_init(FLINT_MIN(lenA, lenB));
}
else
{
nmod_poly_fit_length(G, FLINT_MIN(lenA, lenB));
g = G->coeffs;
}
if (S == A || S == B)
{
s = _nmod_vec_init(FLINT_MAX(lenB - 1, 2));
}
else
{
nmod_poly_fit_length(S, FLINT_MAX(lenB - 1, 2));
s = S->coeffs;
}
if (T == A || T == B)
{
t = _nmod_vec_init(FLINT_MAX(lenA - 1, 2));
}
else
{
nmod_poly_fit_length(T, FLINT_MAX(lenA - 1, 2));
t = T->coeffs;
}
if (lenA >= lenB)
lenG = _nmod_poly_xgcd_hgcd(g, s, t, A->coeffs, lenA,
B->coeffs, lenB, A->mod);
else
lenG = _nmod_poly_xgcd_hgcd(g, t, s, B->coeffs, lenB,
A->coeffs, lenA, A->mod);
if (G == A || G == B)
{
flint_free(G->coeffs);
G->coeffs = g;
G->alloc = FLINT_MIN(lenA, lenB);
}
if (S == A || S == B)
{
flint_free(S->coeffs);
S->coeffs = s;
S->alloc = FLINT_MAX(lenB - 1, 2);
}
if (T == A || T == B)
{
flint_free(T->coeffs);
T->coeffs = t;
T->alloc = FLINT_MAX(lenA - 1, 2);
}
G->length = lenG;
S->length = FLINT_MAX(lenB - lenG, 1);
T->length = FLINT_MAX(lenA - lenG, 1);
MPN_NORM(S->coeffs, S->length);
MPN_NORM(T->coeffs, T->length);
if (G->coeffs[lenG - 1] != 1)
{
inv = n_invmod(G->coeffs[lenG - 1], A->mod.n);
nmod_poly_scalar_mul_nmod(G, G, inv);
nmod_poly_scalar_mul_nmod(S, S, inv);
nmod_poly_scalar_mul_nmod(T, T, inv);
}
}
}
}
void
nmod_poly_xgcd(nmod_poly_t G, nmod_poly_t S, nmod_poly_t T,
const nmod_poly_t A, const nmod_poly_t B)
{
const long lenA = A->length, lenB = B->length;
mp_limb_t inv;
if (lenA == 0)
{
if (lenB == 0)
{
nmod_poly_zero(G);
nmod_poly_zero(S);
nmod_poly_zero(T);
}
else
{
inv = n_invmod(B->coeffs[lenB - 1], B->mod.n);
nmod_poly_scalar_mul_nmod(G, B, inv);
nmod_poly_zero(S);
nmod_poly_set_coeff_ui(T, 0, inv);
T->length = 1;
}
}
else if (lenB == 0)
{
inv = n_invmod(A->coeffs[lenA - 1], A->mod.n);
nmod_poly_scalar_mul_nmod(G, A, inv);
nmod_poly_zero(T);
nmod_poly_set_coeff_ui(S, 0, inv);
S->length = 1;
}
else
{
nmod_poly_t tG, tS, tT;
mp_ptr g, s, t;
long lenG;
if (G == A || G == B)
{
nmod_poly_init2(tG, A->mod.n, FLINT_MIN(lenA, lenB));
g = tG->coeffs;
}
else
{
nmod_poly_fit_length(G, FLINT_MIN(lenA, lenB));
g = G->coeffs;
}
if (S == A || S == B)
{
nmod_poly_init2(tS, A->mod.n, lenB - 1);
s = tS->coeffs;
}
else
{
nmod_poly_fit_length(S, lenB - 1);
s = S->coeffs;
}
if (T == A || T == B)
{
nmod_poly_init2(tT, A->mod.n, lenA - 1);
t = tT->coeffs;
}
else
{
nmod_poly_fit_length(T, lenA - 1);
t = T->coeffs;
}
if (lenA >= lenB)
lenG = _nmod_poly_xgcd(g, s, t, A->coeffs, lenA,
B->coeffs, lenB, A->mod);
else
lenG = _nmod_poly_xgcd(g, t, s, B->coeffs, lenB,
A->coeffs, lenA, A->mod);
if (G == A || G == B)
{
nmod_poly_swap(tG, G);
nmod_poly_clear(tG);
}
if (S == A || S == B)
{
nmod_poly_swap(tS, S);
nmod_poly_clear(tS);
}
if (T == A || T == B)
{
nmod_poly_swap(tT, T);
nmod_poly_clear(tT);
}
G->length = lenG;
S->length = lenB - lenG;
T->length = lenA - lenG;
MPN_NORM(S->coeffs, S->length);
MPN_NORM(T->coeffs, T->length);
if (G->coeffs[lenG - 1] != 1)
{
inv = n_invmod(G->coeffs[lenG - 1], A->mod.n);
nmod_poly_scalar_mul_nmod(G, G, inv);
nmod_poly_scalar_mul_nmod(S, S, inv);
nmod_poly_scalar_mul_nmod(T, T, inv);
}
}
}
int main()
{
flint_rand_t state;
slong nmax, n, bound, count;
mp_limb_t p, pinv, m1, m2;
nmod_poly_t A;
flint_printf("rev....");
fflush(stdout);
flint_randinit(state);
bound = 100000;
p = n_nextprime(UWORD(1) << (FLINT_BITS - 1), 0);
pinv = n_preinvert_limb(p);
nmod_poly_init(A, p);
nmod_poly_set_coeff_ui(A, 1, 1);
nmod_poly_exp_series(A, A, bound);
nmod_poly_shift_right(A, A, 1);
nmod_poly_inv_series(A, A, bound);
m1 = 1;
for (n = 0; n < A->length; n++)
{
A->coeffs[n] = n_mulmod2_preinv(A->coeffs[n], m1, p, pinv);
m1 = n_mulmod2_preinv(m1, n + 1, p, pinv);
}
for (nmax = 0; nmax < bound; nmax = 1.5 * nmax + 2)
{
fmpz_t numer, denom;
bernoulli_rev_t iter;
fmpz_init(numer);
fmpz_init(denom);
nmax += (nmax % 2);
bernoulli_rev_init(iter, nmax);
if (nmax < 8000)
count = 4000;
else
count = 100;
/* flint_printf("nmax = %wd, count = %wd\n", nmax, count); */
for (n = nmax; n >= 0 && count > 0; n -= 2, count--)
{
bernoulli_rev_next(numer, denom, iter);
m1 = fmpz_fdiv_ui(numer, p);
m2 = fmpz_fdiv_ui(denom, p);
m2 = n_invmod(m2, p);
m1 = n_mulmod2_preinv(m1, m2, p, pinv);
m2 = nmod_poly_get_coeff_ui(A, n);
if (m1 != m2)
{
flint_printf("FAIL:\n");
flint_printf("nmax = %wd, n = %wd\n", nmax, n);
flint_printf("m1 = %wu mod %wu\n", m1, p);
flint_printf("m2 = %wu mod %wu\n", m2, p);
abort();
}
}
bernoulli_rev_clear(iter);
fmpz_clear(numer);
fmpz_clear(denom);
}
flint_randclear(state);
flint_cleanup();
flint_printf("PASS\n");
return EXIT_SUCCESS;
}