static int
ref_modinv (mp_limb_t *rp, const mp_limb_t *ap, const mp_limb_t *mp, mp_size_t mn)
{
mp_limb_t tp[4*(mn+1)];
mp_limb_t *up = tp;
mp_limb_t *vp = tp + mn+1;
mp_limb_t *gp = tp + 2*(mn+1);
mp_limb_t *sp = tp + 3*(mn+1);
mp_size_t gn, sn;
mpn_copyi (up, ap, mn);
mpn_copyi (vp, mp, mn);
gn = mpn_gcdext (gp, sp, &sn, up, mn, vp, mn);
if (gn != 1 || gp[0] != 1)
return 0;
if (sn < 0)
mpn_sub (sp, mp, mn, sp, -sn);
else if (sn < mn)
/* Zero-pad. */
mpn_zero (sp + sn, mn - sn);
mpn_copyi (rp, sp, mn);
return 1;
}
void _fmpz_poly_sqrlow(fmpz * res, const fmpz * poly, long len, long n)
{
mp_size_t limbs;
if (n < 7)
{
_fmpz_poly_sqrlow_classical(res, poly, len, n);
return;
}
limbs = _fmpz_vec_max_limbs(poly, len);
if (n < 16 && limbs > 12)
{
int i;
fmpz *copy;
copy = flint_malloc(n * sizeof(fmpz));
for (i = 0; i < len; i++)
copy[i] = poly[i];
mpn_zero((mp_ptr) copy + len, n - len);
_fmpz_poly_sqrlow_karatsuba_n(res, copy, n);
flint_free(copy);
}
else if (limbs <= 4)
_fmpz_poly_sqrlow_KS(res, poly, len, n);
else if (limbs/2048 > len)
_fmpz_poly_sqrlow_KS(res, poly, len, n);
else if (limbs*FLINT_BITS*4 < len)
_fmpz_poly_sqrlow_KS(res, poly, len, n);
else
_fmpz_poly_mullow_SS(res, poly, len, poly, len, n);
}
void fmpq_poly_set_coeff_mpz(fmpq_poly_t poly, long n, const mpz_t x)
{
long len = poly->length;
const int replace = (n < len && !fmpz_is_zero(poly->coeffs + n));
if (!replace && mpz_sgn(x) == 0)
return;
if (n + 1 > len)
{
fmpq_poly_fit_length(poly, n + 1);
_fmpq_poly_set_length(poly, n + 1);
mpn_zero((mp_ptr) poly->coeffs + len, (n + 1) - len);
}
if (*poly->den == 1L)
{
fmpz_set_mpz(poly->coeffs + n, x);
if (replace)
_fmpq_poly_normalise(poly);
}
else
{
fmpz_set_mpz(poly->coeffs + n, x);
fmpz_mul(poly->coeffs + n, poly->coeffs + n, poly->den);
if (replace)
fmpq_poly_canonicalise(poly);
}
}
static int
modinv_gcd (const struct ecc_curve *ecc,
mp_limb_t *rp, mp_limb_t *ap, mp_limb_t *tp)
{
mp_size_t size = ecc->p.size;
mp_limb_t *up = tp;
mp_limb_t *vp = tp + size+1;
mp_limb_t *gp = tp + 2*(size+1);
mp_limb_t *sp = tp + 3*(size+1);
mp_size_t gn, sn;
mpn_copyi (up, ap, size);
mpn_copyi (vp, ecc->p.m, size);
gn = mpn_gcdext (gp, sp, &sn, up, size, vp, size);
if (gn != 1 || gp[0] != 1)
return 0;
if (sn < 0)
mpn_sub (sp, ecc->p.m, size, sp, -sn);
else if (sn < size)
/* Zero-pad. */
mpn_zero (sp + sn, size - sn);
mpn_copyi (rp, sp, size);
return 1;
}
void
fmpz_poly_realloc(fmpz_poly_t poly, long alloc)
{
if (alloc == 0) /* Clear up, reinitialise */
{
fmpz_poly_clear(poly);
fmpz_poly_init(poly);
return;
}
if (poly->alloc) /* Realloc */
{
fmpz_poly_truncate(poly, alloc);
poly->coeffs = (fmpz *) flint_realloc(poly->coeffs, alloc * sizeof(fmpz));
if (alloc > poly->alloc)
mpn_zero((mp_ptr) (poly->coeffs + poly->alloc),
alloc - poly->alloc);
}
else /* Nothing allocated already so do it now */
{
poly->coeffs = (fmpz *) flint_calloc(alloc, sizeof(fmpz));
}
poly->alloc = alloc;
}
void
nmod_poly_cosh_series(nmod_poly_t g, const nmod_poly_t h, long n)
{
mp_ptr g_coeffs, h_coeffs;
nmod_poly_t t1;
long h_len;
h_len = h->length;
if (h_len > 0 && h->coeffs[0] != 0UL)
{
printf("Exception: nmod_poly_cosh_series: constant term != 0\n");
abort();
}
if (h_len == 1 || n < 2)
{
nmod_poly_zero(g);
if (n > 0)
nmod_poly_set_coeff_ui(g, 0, 1UL);
return;
}
if (h_len < n)
{
h_coeffs = _nmod_vec_init(n);
mpn_copyi(h_coeffs, h->coeffs, h_len);
mpn_zero(h_coeffs + h_len, n - h_len);
}
else
h_coeffs = h->coeffs;
if (h == g && h_len >= n)
{
nmod_poly_init2(t1, h->mod.n, n);
g_coeffs = t1->coeffs;
}
else
{
nmod_poly_fit_length(g, n);
g_coeffs = g->coeffs;
}
_nmod_poly_cosh_series(g_coeffs, h_coeffs, n, h->mod);
if (h == g && h_len >= n)
{
nmod_poly_swap(g, t1);
nmod_poly_clear(t1);
}
g->length = n;
if (h_len < n)
_nmod_vec_free(h_coeffs);
_nmod_poly_normalise(g);
}
void
mpz_limbs_copy (mp_limb_t *xp, mpz_srcptr x, mp_size_t n)
{
mp_size_t xn = mpz_size (x);
assert (xn <= n);
mpn_copyi (xp, mpz_limbs_read (x), xn);
if (xn < n)
mpn_zero (xp + xn, n - xn);
}
void
nmod_poly_revert_series_lagrange(nmod_poly_t Qinv,
const nmod_poly_t Q, long n)
{
mp_ptr Qinv_coeffs, Q_coeffs;
nmod_poly_t t1;
long Qlen;
Qlen = Q->length;
if (Qlen < 2 || Q->coeffs[0] != 0 || Q->coeffs[1] == 0)
{
printf("exception: nmod_poly_revert_series_lagrange: input must have "
"zero constant and an invertible coefficient of x^1");
abort();
}
if (Qlen < n)
{
Q_coeffs = _nmod_vec_init(n);
mpn_copyi(Q_coeffs, Q->coeffs, Qlen);
mpn_zero(Q_coeffs + Qlen, n - Qlen);
}
else
Q_coeffs = Q->coeffs;
if (Q == Qinv && Qlen >= n)
{
nmod_poly_init2(t1, Q->mod.n, n);
Qinv_coeffs = t1->coeffs;
}
else
{
nmod_poly_fit_length(Qinv, n);
Qinv_coeffs = Qinv->coeffs;
}
_nmod_poly_revert_series_lagrange(Qinv_coeffs, Q_coeffs, n, Q->mod);
if (Q == Qinv && Qlen >= n)
{
nmod_poly_swap(Qinv, t1);
nmod_poly_clear(t1);
}
Qinv->length = n;
if (Qlen < n)
_nmod_vec_clear(Q_coeffs);
_nmod_poly_normalise(Qinv);
}
void
nmod_poly_inv_series_basecase(nmod_poly_t Qinv,
const nmod_poly_t Q, long n)
{
mp_ptr Qinv_coeffs, Q_coeffs;
nmod_poly_t t1;
long Qlen;
Qlen = Q->length;
if (n == 0 || Q->length == 0 || Q->coeffs[0] == 0)
{
printf("Exception: division by zero in nmod_poly_inv_series_basecase\n");
abort();
}
if (Qlen < n)
{
Q_coeffs = _nmod_vec_init(n);
mpn_copyi(Q_coeffs, Q->coeffs, Qlen);
mpn_zero(Q_coeffs + Qlen, n - Qlen);
}
else
Q_coeffs = Q->coeffs;
if (Q == Qinv && Qlen >= n)
{
nmod_poly_init2(t1, Q->mod.n, n);
Qinv_coeffs = t1->coeffs;
}
else
{
nmod_poly_fit_length(Qinv, n);
Qinv_coeffs = Qinv->coeffs;
}
_nmod_poly_inv_series_basecase(Qinv_coeffs, Q_coeffs, n, Q->mod);
if (Q == Qinv && Qlen >= n)
{
nmod_poly_swap(Qinv, t1);
nmod_poly_clear(t1);
}
Qinv->length = n;
if (Qlen < n)
_nmod_vec_free(Q_coeffs);
_nmod_poly_normalise(Qinv);
}
void
fmpz_poly_set_coeff_si(fmpz_poly_t poly, long n, long x)
{
fmpz_poly_fit_length(poly, n + 1);
if (n + 1 > poly->length) /* insert zeroes between end of poly and new coeff if needed */
{
mpn_zero((mp_ptr) (poly->coeffs + poly->length), n - poly->length);
poly->length = n + 1;
}
fmpz_set_si(poly->coeffs + n, x);
_fmpz_poly_normalise(poly); /* we may have set leading coefficient to zero */
}
int
mpn_mul_fft(mp_ptr rp, mp_size_t rn, mp_srcptr ap, mp_size_t an,
mp_srcptr bp, mp_size_t bn, int k)
{
mp_ptr rpp, app, bpp, tpp;
mp_size_t t = rn + 1;
TMP_DECL;
TMP_MARK;
rpp = (mp_ptr)TMP_ALLOC_LIMBS(t);
tpp = (mp_ptr)TMP_ALLOC_LIMBS(t);
app = (mp_ptr)TMP_ALLOC_LIMBS(t);
bpp = (mp_ptr)TMP_ALLOC_LIMBS(t);
mpn_copyi(app, ap, an); mpn_zero(app + an, t - an);
mpn_copyi(bpp, bp, bn); mpn_zero(bpp + bn, t - bn);
mpn_mulmod_Bexpp1(rpp, app, bpp, rn, tpp);
mpn_copyi(rp, rpp, rn);
t = rpp[rn];
TMP_FREE;
return t;
}
/* Get a pointer to an n limb area, for read-only operation. n must be
greater or equal to the current size, and the mpz is zero-padded if
needed. */
const mp_limb_t *
mpz_limbs_read_n (mpz_ptr x, mp_size_t n)
{
mp_size_t xn = mpz_size (x);
mp_ptr xp;
assert (xn <= n);
xp = mpz_limbs_modify (x, n);
if (xn < n)
mpn_zero (xp + xn, n - xn);
return xp;
}
static void
table_init (const struct ecc_curve *ecc,
mp_limb_t *table, unsigned bits,
int initial, const mp_limb_t *p,
mp_limb_t *scratch)
{
unsigned size = 1 << bits;
unsigned j;
mpn_zero (TABLE(0), 3*ecc->size);
ecc_a_to_j (ecc, initial, TABLE(1), p);
for (j = 2; j < size; j += 2)
{
ecc_dup_jj (ecc, TABLE(j), TABLE(j/2), scratch);
ecc_add_jja (ecc, TABLE(j+1), TABLE(j), TABLE(1), scratch);
}
}
void
fmpz_poly_bit_pack(fmpz_t f, const fmpz_poly_t poly,
mp_bitcnt_t bit_size)
{
long len;
__mpz_struct * mpz;
long i, d;
int negate;
len = fmpz_poly_length(poly);
if (len == 0 || bit_size == 0)
{
fmpz_zero(f);
return;
}
mpz = _fmpz_promote(f);
mpz_realloc2(mpz, len * bit_size);
d = mpz->_mp_alloc;
mpn_zero(mpz->_mp_d, d);
if (fmpz_sgn(fmpz_poly_lead(poly)) < 0)
negate = -1;
else
negate = 0;
_fmpz_poly_bit_pack(mpz->_mp_d, poly->coeffs, len, bit_size, negate);
for (i = d - 1; i >= 0; i--)
{
if (mpz->_mp_d[i] != 0)
break;
}
d = i + 1;
mpz->_mp_size = d;
_fmpz_demote_val(f);
if (negate)
fmpz_neg(f, f);
}
static void
table_init (const struct ecc_curve *ecc,
mp_limb_t *table, unsigned bits,
const mp_limb_t *p,
mp_limb_t *scratch)
{
unsigned size = 1 << bits;
unsigned j;
mpn_zero (TABLE(0), 3*ecc->p.size);
TABLE(0)[ecc->p.size] = TABLE(0)[2*ecc->p.size] = 1;
ecc_a_to_j (ecc, TABLE(1), p);
for (j = 2; j < size; j += 2)
{
ecc_dup_eh (ecc, TABLE(j), TABLE(j/2), scratch);
ecc_add_ehh (ecc, TABLE(j+1), TABLE(j), TABLE(1), scratch);
}
}
void
ecc_mul_a (const struct ecc_curve *ecc,
int initial, mp_limb_t *r,
const mp_limb_t *np, const mp_limb_t *p,
mp_limb_t *scratch)
{
#define tp scratch
#define pj (scratch + 3*ecc->size)
#define scratch_out (scratch + 6*ecc->size)
int is_zero;
unsigned i;
ecc_a_to_j (ecc, initial, pj, p);
mpn_zero (r, 3*ecc->size);
for (i = ecc->size, is_zero = 1; i-- > 0; )
{
mp_limb_t w = np[i];
mp_limb_t bit;
for (bit = (mp_limb_t) 1 << (GMP_NUMB_BITS - 1);
bit > 0;
bit >>= 1)
{
int digit;
ecc_dup_jj (ecc, r, r, scratch_out);
ecc_add_jja (ecc, tp, r, pj, scratch_out);
digit = (w & bit) > 0;
/* If is_zero is set, r is the zero point,
and ecc_add_jja produced garbage. */
cnd_copy (is_zero, tp, pj, 3*ecc->size);
is_zero &= ~digit;
/* If we had a one-bit, use the sum. */
cnd_copy (digit, r, tp, 3*ecc->size);
}
}
}
void
_nmod_poly_inv_series_basecase(mp_ptr Qinv,
mp_srcptr Q, long n, nmod_t mod)
{
mp_ptr X2n, Qrev;
X2n = _nmod_vec_init(2*n);
Qrev = X2n + n;
_nmod_poly_reverse(Qrev, Q, n, n);
X2n[n - 1] = 1;
mpn_zero(X2n, n - 1);
X2n -= (n - 1);
_nmod_poly_div_divconquer(Qinv, X2n, 2*n - 1, Qrev, n, mod);
_nmod_poly_reverse(Qinv, Qinv, n, n);
_nmod_vec_free(X2n + n - 1);
}
void
ecc_mul_a_eh (const struct ecc_curve *ecc,
mp_limb_t *r,
const mp_limb_t *np, const mp_limb_t *p,
mp_limb_t *scratch)
{
#define pe scratch
#define tp (scratch + 3*ecc->p.size)
#define scratch_out (scratch + 6*ecc->p.size)
unsigned i;
ecc_a_to_j (ecc, pe, p);
/* x = 0, y = 1, z = 1 */
mpn_zero (r, 3*ecc->p.size);
r[ecc->p.size] = r[2*ecc->p.size] = 1;
for (i = ecc->p.size; i-- > 0; )
{
mp_limb_t w = np[i];
mp_limb_t bit;
for (bit = (mp_limb_t) 1 << (GMP_NUMB_BITS - 1);
bit > 0;
bit >>= 1)
{
int digit;
ecc_dup_eh (ecc, r, r, scratch_out);
ecc_add_ehh (ecc, tp, r, pe, scratch_out);
digit = (w & bit) > 0;
/* If we had a one-bit, use the sum. */
cnd_copy (digit, r, tp, 3*ecc->p.size);
}
}
}
void
nmod_poly_tanh_series(nmod_poly_t g, const nmod_poly_t h, long n)
{
mp_ptr h_coeffs;
long h_len = h->length;
if (h_len > 0 && h->coeffs[0] != 0UL)
{
printf("Exception: nmod_poly_tanh_series: constant term != 0\n");
abort();
}
if (h_len == 1 || n < 2)
{
nmod_poly_zero(g);
return;
}
nmod_poly_fit_length(g, n);
if (h_len < n)
{
h_coeffs = _nmod_vec_init(n);
mpn_copyi(h_coeffs, h->coeffs, h_len);
mpn_zero(h_coeffs + h_len, n - h_len);
}
else
h_coeffs = h->coeffs;
_nmod_poly_tanh_series(g->coeffs, h_coeffs, n, h->mod);
if (h_len < n)
_nmod_vec_free(h_coeffs);
g->length = n;
_nmod_poly_normalise(g);
}
void
test_main (void)
{
unsigned i;
for (i = 0; ecc_curves[i]; i++)
{
const struct ecc_curve *ecc = ecc_curves[i];
mp_limb_t *g = xalloc_limbs (ecc_size_j (ecc));
mp_limb_t *g2 = xalloc_limbs (ecc_size_j (ecc));
mp_limb_t *g3 = xalloc_limbs (ecc_size_j (ecc));
mp_limb_t *p = xalloc_limbs (ecc_size_j (ecc));
mp_limb_t *scratch = xalloc_limbs (ECC_ADD_JJJ_ITCH(ecc->p.size));
if (ecc->p.bit_size == 255)
{
mp_limb_t *z = xalloc_limbs (ecc_size_j (ecc));
/* Zero point has x = 0, y = 1, z = 1 */
mpn_zero (z, 3*ecc->p.size);
z[ecc->p.size] = z[2*ecc->p.size] = 1;
ecc_a_to_j (ecc, g, ecc->g);
ecc_add_ehh (ecc, p, z, z, scratch);
test_ecc_mul_h (i, 0, p);
ecc_add_eh (ecc, p, z, z, scratch);
test_ecc_mul_h (i, 0, p);
ecc_add_ehh (ecc, p, g, p, scratch);
test_ecc_mul_h (i, 1, p);
ecc_add_eh (ecc, p, z, g, scratch);
test_ecc_mul_h (i, 1, p);
ecc_add_ehh (ecc, g2, g, p, scratch);
test_ecc_mul_h (i, 2, g2);
ecc_add_eh (ecc, g2, g, g, scratch);
test_ecc_mul_h (i, 2, g2);
ecc_add_ehh (ecc, g3, g, g2, scratch);
test_ecc_mul_h (i, 3, g3);
ecc_add_eh (ecc, g3, g2, g, scratch);
test_ecc_mul_h (i, 3, g3);
ecc_add_ehh (ecc, p, g, g3, scratch);
test_ecc_mul_h (i, 4, p);
ecc_add_eh (ecc, p, g3, g, scratch);
test_ecc_mul_h (i, 4, p);
ecc_add_ehh (ecc, p, g2, g2, scratch);
test_ecc_mul_h (i, 4, p);
free (z);
}
else
{
ecc_a_to_j (ecc, g, ecc->g);
ecc_dup_jj (ecc, g2, g, scratch);
test_ecc_mul_h (i, 2, g2);
ecc_add_jjj (ecc, g3, g, g2, scratch);
test_ecc_mul_h (i, 3, g3);
ecc_add_jjj (ecc, g3, g2, g, scratch);
test_ecc_mul_h (i, 3, g3);
ecc_add_jjj (ecc, p, g, g3, scratch);
test_ecc_mul_h (i, 4, p);
ecc_add_jjj (ecc, p, g3, g, scratch);
test_ecc_mul_h (i, 4, p);
ecc_dup_jj (ecc, p, g2, scratch);
test_ecc_mul_h (i, 4, p);
}
free (g);
free (g2);
free (g3);
free (p);
free (scratch);
}
}
void
test_main (void)
{
gmp_randstate_t rands;
mpz_t r;
unsigned i;
gmp_randinit_default (rands);
mpz_init (r);
for (i = 0; ecc_curves[i]; i++)
{
const struct ecc_curve *ecc = ecc_curves[i];
mp_size_t size = ecc_size (ecc);
mp_limb_t *p = xalloc_limbs (ecc_size_j (ecc));
mp_limb_t *q = xalloc_limbs (ecc_size_j (ecc));
mp_limb_t *n = xalloc_limbs (size);
mp_limb_t *scratch = xalloc_limbs (ecc->mul_g_itch);
mpn_zero (n, size);
n[0] = 1;
ecc->mul_g (ecc, p, n, scratch);
ecc->h_to_a (ecc, 0, p, p, scratch);
if (mpn_cmp (p, ecc->g, 2*size != 0))
{
fprintf (stderr, "ecc->mul_g with n = 1 failed.\n");
abort ();
}
for (n[0] = 2; n[0] <= 4; n[0]++)
{
ecc->mul_g (ecc, p, n, scratch);
test_ecc_mul_h (i, n[0], p);
}
/* (order - 1) * g = - g */
mpn_sub_1 (n, ecc->q.m, size, 1);
ecc->mul_g (ecc, p, n, scratch);
ecc->h_to_a (ecc, 0, p, p, scratch);
if (ecc->p.bit_size == 255)
/* For edwards curves, - (x,y ) == (-x, y). FIXME: Swap x and
y, to get identical negation? */
mpn_sub_n (p, ecc->p.m, p, size);
else
mpn_sub_n (p + size, ecc->p.m, p + size, size);
if (mpn_cmp (p, ecc->g, 2*size) != 0)
{
fprintf (stderr, "ecc->mul_g with n = order - 1 failed.\n");
abort ();
}
free (n);
free (p);
free (q);
free (scratch);
}
mpz_clear (r);
gmp_randclear (rands);
}
void
nmod_poly_compose_mod_horner(nmod_poly_t res,
const nmod_poly_t poly1, const nmod_poly_t poly2,
const nmod_poly_t poly3)
{
long len1 = poly1->length;
long len2 = poly2->length;
long len3 = poly3->length;
long len = len3 - 1;
mp_ptr ptr2;
if (len3 == 0)
{
printf("exception: division by zero in nmod_poly_compose_mod_horner\n");
abort();
}
if (len1 == 0 || len3 == 1)
{
nmod_poly_zero(res);
return;
}
if (len1 == 1)
{
nmod_poly_set(res, poly1);
return;
}
if (res == poly3 || res == poly1)
{
nmod_poly_t tmp;
nmod_poly_init_preinv(tmp, res->mod.n, res->mod.ninv);
nmod_poly_compose_mod_horner(tmp, poly1, poly2, poly3);
nmod_poly_swap(tmp, res);
nmod_poly_clear(tmp);
return;
}
ptr2 = _nmod_vec_init(len);
if (len2 <= len)
{
mpn_copyi(ptr2, poly2->coeffs, len2);
mpn_zero(ptr2 + len2, len - len2);
}
else
{
_nmod_poly_rem(ptr2, poly2->coeffs, len2,
poly3->coeffs, len3, res->mod);
}
nmod_poly_fit_length(res, len);
_nmod_poly_compose_mod_horner(res->coeffs,
poly1->coeffs, len1, ptr2, poly3->coeffs, len3, res->mod);
res->length = len;
_nmod_poly_normalise(res);
_nmod_vec_clear(ptr2);
}
void
nmod_poly_compose_mod_brent_kung(nmod_poly_t res,
const nmod_poly_t poly1, const nmod_poly_t poly2,
const nmod_poly_t poly3)
{
long len1 = poly1->length;
long len2 = poly2->length;
long len3 = poly3->length;
long len = len3 - 1;
mp_ptr ptr2;
if (len3 == 0)
{
printf("exception: division by zero in "
"nmod_poly_compose_mod_brent_kung\n");
abort();
}
if (len1 >= len3)
{
printf("exception: nmod_poly_compose_brent_kung: the degree of the"
" first polynomial must be smaller than that of the modulus\n");
abort();
}
if (len1 == 0 || len3 == 1)
{
nmod_poly_zero(res);
return;
}
if (len1 == 1)
{
nmod_poly_set(res, poly1);
return;
}
if (res == poly3 || res == poly1)
{
nmod_poly_t tmp;
nmod_poly_init_preinv(tmp, res->mod.n, res->mod.ninv);
nmod_poly_compose_mod_brent_kung(tmp, poly1, poly2, poly3);
nmod_poly_swap(tmp, res);
nmod_poly_clear(tmp);
return;
}
ptr2 = _nmod_vec_init(len);
if (len2 <= len)
{
mpn_copyi(ptr2, poly2->coeffs, len2);
mpn_zero(ptr2 + len2, len - len2);
}
else
{
_nmod_poly_rem(ptr2, poly2->coeffs, len2,
poly3->coeffs, len3, res->mod);
}
nmod_poly_fit_length(res, len);
_nmod_poly_compose_mod_brent_kung(res->coeffs,
poly1->coeffs, len1, ptr2, poly3->coeffs, len3, res->mod);
res->length = len;
_nmod_poly_normalise(res);
_nmod_vec_clear(ptr2);
}
void mpir_fft_mulmod_2expp1(mp_ptr r1, mp_srcptr i1, mp_srcptr i2,
mp_size_t r_limbs, mp_bitcnt_t depth, mp_bitcnt_t w)
{
mp_size_t n = (((mp_size_t)1)<<depth);
mp_bitcnt_t bits1 = (r_limbs*GMP_LIMB_BITS)/(2*n);
mp_size_t limb_add, limbs = (n*w)/GMP_LIMB_BITS;
mp_size_t size = limbs + 1;
mp_size_t i, j, ll;
mp_limb_t * ptr;
mp_limb_t ** ii, ** jj, *tt, *t1, *t2, *s1, *r, *ii0, *jj0;
mp_limb_t c;
TMP_DECL;
TMP_MARK;
ii = TMP_BALLOC_MP_PTRS(2*(n + n*size) + 4*n + 5*size);
for (i = 0, ptr = (mp_ptr) ii + 2*n; i < 2*n; i++, ptr += size)
{
ii[i] = ptr;
}
ii0 = ptr;
t1 = ii0 + 2*n;
t2 = t1 + size;
s1 = t2 + size;
r = s1 + size;
tt = r + 2*n;
if (i1 != i2)
{
jj = TMP_BALLOC_MP_PTRS(2*(n + n*size) + 2*n);
for (i = 0, ptr = (mp_ptr) jj + 2*n; i < 2*n; i++, ptr += size)
{
jj[i] = ptr;
}
jj0 = ptr;
} else
{
jj = ii;
jj0 = ii0;
}
j = mpir_fft_split_bits(ii, i1, r_limbs, bits1, limbs);
for ( ; j < 2*n; j++)
mpn_zero(ii[j], limbs + 1);
for (i = 0; i < 2*n; i++)
ii0[i] = ii[i][0];
mpir_fft_negacyclic(ii, n, w, &t1, &t2, &s1);
for (j = 0; j < 2*n; j++)
mpn_normmod_2expp1(ii[j], limbs);
if (i1 != i2)
{
j = mpir_fft_split_bits(jj, i2, r_limbs, bits1, limbs);
for ( ; j < 2*n; j++)
mpn_zero(jj[j], limbs + 1);
for (i = 0; i < 2*n; i++)
jj0[i] = jj[i][0];
mpir_fft_negacyclic(jj, n, w, &t1, &t2, &s1);
}
for (j = 0; j < 2*n; j++)
{
if (i1 != i2) mpn_normmod_2expp1(jj[j], limbs);
c = 2*ii[j][limbs] + jj[j][limbs];
ii[j][limbs] = mpn_mulmod_2expp1_basecase(ii[j], ii[j], jj[j], c, n*w, tt);
}
mpir_ifft_negacyclic(ii, n, w, &t1, &t2, &s1);
mpir_fft_naive_convolution_1(r, ii0, jj0, 2*n);
for (j = 0; j < 2*n; j++)
{
mp_limb_t t, cy2;
mpn_div_2expmod_2expp1(ii[j], ii[j], limbs, depth + 1);
mpn_normmod_2expp1(ii[j], limbs);
t = ii[j][limbs];
ii[j][limbs] = r[j] - ii[j][0];
cy2 = mpn_add_1(ii[j], ii[j], limbs + 1, ii[j][limbs]);
add_ssaaaa(r[j], ii[j][limbs], 0, ii[j][limbs], 0, t);
if (cy2) r[j]++;
}
mpn_zero(r1, r_limbs + 1);
mpir_fft_combine_bits(r1, ii, 2*n - 1, bits1, limbs + 1, r_limbs + 1);
/*
as the negacyclic convolution has effectively done subtractions
some of the coefficients will be negative, so need to subtract p
*/
ll = 0;
limb_add = bits1/GMP_LIMB_BITS;
for (j = 0; j < 2*n - 2; j++)
{
if (r[j])
mpn_sub_1(r1 + ll + 1, r1 + ll + 1, r_limbs - ll, 1);
else if ((mp_limb_signed_t) ii[j][limbs] < 0) /* coefficient was -ve */
{
mpn_sub_1(r1 + ll + 1, r1 + ll + 1, r_limbs - ll, 1);
mpn_sub_1(r1 + ll + limbs + 1, r1 + ll + limbs + 1, r_limbs - limbs - ll, 1);
}
ll += limb_add;
}
/* penultimate coefficient, top bit was already ignored */
if (r[j] || (mp_limb_signed_t) ii[j][limbs] < 0) /* coefficient was -ve */
mpn_sub_1(r1 + ll + 1, r1 + ll + 1, r_limbs - ll, 1);
/* final coefficient wraps around */
if (limb_add)
r1[r_limbs] += mpn_add_n(r1 + r_limbs - limb_add, r1 + r_limbs - limb_add, ii[2*n - 1], limb_add);
c = mpn_sub_n(r1, r1, ii[2*n - 1] + limb_add, limbs + 1 - limb_add);
mpn_addmod_2expp1_1(r1 + limbs + 1 - limb_add, r_limbs - limbs - 1 + limb_add, -c);
mpn_normmod_2expp1(r1, r_limbs);
TMP_FREE;
}
/* Assumes len1 != 0 != len2 */
int
_fmpz_poly_gcd_heuristic(fmpz * res, const fmpz * poly1, long len1,
const fmpz * poly2, long len2)
{
ulong bits1, bits2, max_bits, pack_bits, bound_bits, bits_G, bits_Q;
ulong limbs1, limbs2, limbsg, pack_limbs, qlimbs;
ulong log_glen, log_length;
long sign1, sign2, glen, qlen;
fmpz_t ac, bc, d, gc;
fmpz * A, * B, * G, * Q, * t;
mp_ptr array1, array2, arrayg, q, temp;
int divides;
fmpz_init(ac);
fmpz_init(bc);
fmpz_init(d);
/* compute gcd of content of poly1 and poly2 */
_fmpz_poly_content(ac, poly1, len1);
_fmpz_poly_content(bc, poly2, len2);
fmpz_gcd(d, ac, bc);
/* special case, one of the polys is a constant */
if (len2 == 1) /* if len1 == 1 then so does len2 */
{
fmpz_set(res, d);
fmpz_clear(ac);
fmpz_clear(bc);
fmpz_clear(d);
return 1;
}
/* divide poly1 and poly2 by their content */
A = _fmpz_vec_init(len1);
B = _fmpz_vec_init(len2);
_fmpz_vec_scalar_divexact_fmpz(A, poly1, len1, ac);
_fmpz_vec_scalar_divexact_fmpz(B, poly2, len2, bc);
fmpz_clear(ac);
fmpz_clear(bc);
/* special case, one of the polys is length 2 */
if (len2 == 2) /* if len1 == 2 then so does len2 */
{
Q = _fmpz_vec_init(len1 - len2 + 1);
if (_fmpz_poly_divides(Q, A, len1, B, 2))
{
_fmpz_vec_scalar_mul_fmpz(res, B, 2, d);
if (fmpz_sgn(res + 1) < 0)
_fmpz_vec_neg(res, res, 2);
}
else
{
fmpz_set(res, d);
fmpz_zero(res + 1);
}
fmpz_clear(d);
_fmpz_vec_clear(A, len1);
_fmpz_vec_clear(B, len2);
_fmpz_vec_clear(Q, len1 - len2 + 1);
return 1;
}
/*
Determine how many bits (pack_bits) to pack into. The bound
bound_bits ensures that if G | A and G | B with G primitive
then G is the gcd of A and B. The bound is taken from
http://arxiv.org/abs/cs/0206032v1
*/
bits1 = FLINT_ABS(_fmpz_vec_max_bits(A, len1));
bits2 = FLINT_ABS(_fmpz_vec_max_bits(B, len2));
max_bits = FLINT_MAX(bits1, bits2);
bound_bits = FLINT_MIN(bits1, bits2) + 6;
pack_bits = FLINT_MAX(bound_bits, max_bits); /* need to pack original polys */
pack_limbs = (pack_bits - 1)/FLINT_BITS + 1;
if (pack_bits >= 32) /* pack into multiples of limbs if >= 32 bits */
pack_bits = FLINT_BITS*pack_limbs;
/* allocate space to pack into */
limbs1 = (pack_bits*len1 - 1)/FLINT_BITS + 1;
limbs2 = (pack_bits*len2 - 1)/FLINT_BITS + 1;
array1 = flint_calloc(limbs1, sizeof(mp_limb_t));
array2 = flint_calloc(limbs2, sizeof(mp_limb_t));
arrayg = flint_calloc(limbs2, sizeof(mp_limb_t));
/* pack first poly and normalise */
sign1 = (long) fmpz_sgn(A + len1 - 1);
_fmpz_poly_bit_pack(array1, A, len1, pack_bits, sign1);
while (array1[limbs1 - 1] == 0) limbs1--;
/* pack second poly and normalise */
sign2 = (long) fmpz_sgn(B + len2 - 1);
_fmpz_poly_bit_pack(array2, B, len2, pack_bits, sign2);
while (array2[limbs2 - 1] == 0) limbs2--;
/* compute integer GCD */
limbsg = mpn_gcd_full(arrayg, array1, limbs1, array2, limbs2);
/*
Make space for unpacked gcd. May have one extra coeff due to
1 0 -x being packed as 0 -1 -x.
*/
glen = FLINT_MIN((limbsg*FLINT_BITS)/pack_bits + 1, len2);
G = _fmpz_vec_init(glen);
/* unpack gcd */
_fmpz_poly_bit_unpack(G, glen, arrayg, pack_bits, 0);
while (G[glen - 1] == 0) glen--;
/* divide by any content */
fmpz_init(gc);
_fmpz_poly_content(gc, G, glen);
if (!fmpz_is_one(gc))
limbsg = mpn_tdiv_q_fmpz_inplace(arrayg, limbsg, gc);
/* make space for quotient and remainder of first poly by gcd */
qlimbs = limbs1 - limbsg + 1;
qlen = FLINT_MIN(len1, (qlimbs*FLINT_BITS)/pack_bits + 1);
qlimbs = (qlen*pack_bits - 1)/FLINT_BITS + 1;
q = flint_calloc(qlimbs, sizeof(mp_limb_t));
temp = flint_malloc(limbsg*sizeof(mp_limb_t));
divides = 0;
if (mpn_divides(q, array1, limbs1, arrayg, limbsg, temp))
{
/* unpack quotient of first poly by gcd */
Q = _fmpz_vec_init(len1);
t = _fmpz_vec_init(len1 + glen);
_fmpz_poly_bit_unpack(Q, qlen, q, pack_bits, 0);
while (Q[qlen - 1] == 0) qlen--;
/* divide by content */
_fmpz_vec_scalar_divexact_fmpz(G, G, glen, gc);
/* check if we really need to multiply out to check for exact quotient */
bits_G = FLINT_ABS(_fmpz_vec_max_bits(G, glen));
bits_Q = FLINT_ABS(_fmpz_vec_max_bits(Q, qlen));
log_glen = FLINT_BIT_COUNT(glen);
log_length = FLINT_MIN(log_glen, FLINT_BIT_COUNT(qlen));
divides = (bits_G + bits_Q + log_length < pack_bits);
if (!divides) /* need to multiply out to check exact quotient */
divides = multiplies_out(A, len1, Q, qlen, G, glen, sign1, t);
if (divides) /* quotient really was exact */
{
mpn_zero(q, qlimbs);
if (mpn_divides(q, array2, limbs2, arrayg, limbsg, temp))
{
/* unpack quotient of second poly by gcd */
qlimbs = limbs2 - limbsg + 1;
qlen = FLINT_MIN(len2, (qlimbs*FLINT_BITS - 1)/pack_bits + 1);
_fmpz_poly_bit_unpack(Q, qlen, q, pack_bits, 0);
while (Q[qlen - 1] == 0) qlen--;
/* check if we really need to multiply out to check for exact quotient */
bits_Q = FLINT_ABS(_fmpz_vec_max_bits(Q, qlen));
log_length = FLINT_MIN(log_glen, FLINT_BIT_COUNT(qlen));
divides = (bits_G + bits_Q + log_length < pack_bits);
if (!divides) /* we need to multiply out */
divides = multiplies_out(B, len2, Q, qlen, G, glen, sign1, t);
}
}
_fmpz_vec_clear(t, len1 + glen);
_fmpz_vec_clear(Q, len1);
}
flint_free(q);
flint_free(temp);
flint_free(arrayg);
flint_free(array1);
flint_free(array2);
fmpz_clear(gc);
_fmpz_vec_clear(A, len1);
_fmpz_vec_clear(B, len2);
/* we found the gcd, so multiply by content */
if (divides)
{
_fmpz_vec_zero(res + glen, len2 - glen);
_fmpz_vec_scalar_mul_fmpz(res, G, glen, d);
}
fmpz_clear(d);
_fmpz_vec_clear(G, glen);
return divides;
}
int
main(void)
{
int i, result;
flint_rand_t state;
flint_randinit(state);
printf("rem....");
fflush(stdout);
/* Check result of rem */
for (i = 0; i < 1000; i++)
{
nmod_poly_t a, b, q, r, prod;
mp_limb_t n;
do n = n_randtest_not_zero(state);
while (!n_is_probabprime(n));
nmod_poly_init(a, n);
nmod_poly_init(b, n);
nmod_poly_init(q, n);
nmod_poly_init(r, n);
nmod_poly_init(prod, n);
nmod_poly_randtest(a, state, n_randint(state, 2000));
do nmod_poly_randtest(b, state, n_randint(state, 2000));
while (b->length == 0);
nmod_poly_div(q, a, b);
nmod_poly_rem(r, a, b);
nmod_poly_mul(prod, q, b);
nmod_poly_add(prod, prod, r);
result = (nmod_poly_equal(a, prod));
if (!result)
{
printf("FAIL:\n");
nmod_poly_print(a), printf("\n\n");
nmod_poly_print(prod), printf("\n\n");
nmod_poly_print(q), printf("\n\n");
nmod_poly_print(r), printf("\n\n");
printf("n = %ld\n", n);
abort();
}
nmod_poly_clear(a);
nmod_poly_clear(b);
nmod_poly_clear(q);
nmod_poly_clear(r);
nmod_poly_clear(prod);
}
/* Check aliasing of a and r */
for (i = 0; i < 1000; i++)
{
nmod_poly_t a, b, r;
mp_limb_t n;
do n = n_randtest(state);
while (!n_is_probabprime(n));
nmod_poly_init(a, n);
nmod_poly_init(b, n);
nmod_poly_init(r, n);
nmod_poly_randtest(a, state, n_randint(state, 2000));
do nmod_poly_randtest(b, state, n_randint(state, 2000));
while (b->length == 0);
nmod_poly_rem(r, a, b);
nmod_poly_rem(a, a, b);
result = (nmod_poly_equal(a, r));
if (!result)
{
printf("FAIL:\n");
nmod_poly_print(a), printf("\n\n");
nmod_poly_print(b), printf("\n\n");
nmod_poly_print(r), printf("\n\n");
printf("n = %ld\n", n);
abort();
}
nmod_poly_clear(a);
nmod_poly_clear(b);
nmod_poly_clear(r);
}
/* Check aliasing of b and r */
for (i = 0; i < 1000; i++)
{
nmod_poly_t a, b, r;
mp_limb_t n;
do n = n_randtest(state);
while (!n_is_probabprime(n));
nmod_poly_init(a, n);
nmod_poly_init(b, n);
nmod_poly_init(r, n);
nmod_poly_randtest(a, state, n_randint(state, 2000));
do nmod_poly_randtest(b, state, n_randint(state, 2000));
while (b->length == 0);
nmod_poly_rem(r, a, b);
nmod_poly_rem(b, a, b);
result = (nmod_poly_equal(b, r));
if (!result)
{
printf("FAIL:\n");
nmod_poly_print(a), printf("\n\n");
nmod_poly_print(b), printf("\n\n");
nmod_poly_print(r), printf("\n\n");
printf("n = %ld\n", n);
abort();
}
nmod_poly_clear(a);
nmod_poly_clear(b);
nmod_poly_clear(r);
}
/* Check result of rem_q1 */
for (i = 0; i < 5000; i++)
{
nmod_poly_t a, b, q0, r0, r;
mp_limb_t n = n_randprime(state, n_randint(state,FLINT_BITS-1)+2, 0);
nmod_poly_init(a, n);
nmod_poly_init(b, n);
nmod_poly_init(q0, n);
nmod_poly_init(r0, n);
nmod_poly_init(r, n);
do nmod_poly_randtest(a, state, n_randint(state, 1000));
while (a->length < 2);
nmod_poly_fit_length(b, a->length - 1);
mpn_zero(b->coeffs, a->length - 1);
nmod_poly_randtest_not_zero(b, state, n_randint(state, 1000) + 1);
do b->coeffs[a->length - 2] = n_randint(state, n);
while (b->coeffs[a->length - 2] == 0);
b->length = a->length - 1;
nmod_poly_divrem(q0, r0, a, b);
nmod_poly_rem(r, a, b);
result = (nmod_poly_equal(r0, r));
if (!result)
{
printf("FAIL:\n");
nmod_poly_print(a), printf("\n\n");
nmod_poly_print(b), printf("\n\n");
nmod_poly_print(q0), printf("\n\n");
nmod_poly_print(r0), printf("\n\n");
nmod_poly_print(r), printf("\n\n");
printf("n = %ld\n", n);
abort();
}
nmod_poly_clear(a);
nmod_poly_clear(b);
nmod_poly_clear(q0);
nmod_poly_clear(r0);
nmod_poly_clear(r);
}
flint_randclear(state);
printf("PASS\n");
return 0;
}
void
nmod_poly_exp_series(nmod_poly_t f, const nmod_poly_t h, long n)
{
mp_ptr f_coeffs, h_coeffs;
nmod_poly_t t1;
long hlen, k;
nmod_poly_fit_length(f, n);
hlen = h->length;
if (hlen > 0 && h->coeffs[0] != 0UL)
{
printf("Exception: nmod_poly_exp_series: constant term != 0\n");
abort();
}
if (n <= 1 || hlen == 0)
{
if (n == 0)
{
nmod_poly_zero(f);
}
else
{
f->coeffs[0] = 1UL;
f->length = 1;
}
return;
}
/* Handle monomials */
for (k = 0; h->coeffs[k] == 0UL && k < n - 1; k++);
if (k == hlen - 1 || k == n - 1)
{
hlen = FLINT_MIN(hlen, n);
_nmod_poly_exp_series_monomial_ui(f->coeffs,
h->coeffs[hlen-1], hlen - 1, n, f->mod);
f->length = n;
_nmod_poly_normalise(f);
return;
}
if (n < NMOD_NEWTON_EXP_CUTOFF2)
{
_nmod_poly_exp_series_basecase(f->coeffs, h->coeffs, hlen, n, f->mod);
f->length = n;
_nmod_poly_normalise(f);
return;
}
if (hlen < n)
{
h_coeffs = _nmod_vec_init(n);
mpn_copyi(h_coeffs, h->coeffs, hlen);
mpn_zero(h_coeffs + hlen, n - hlen);
}
else
h_coeffs = h->coeffs;
if (h == f && hlen >= n)
{
nmod_poly_init2(t1, h->mod.n, n);
f_coeffs = t1->coeffs;
}
else
{
nmod_poly_fit_length(f, n);
f_coeffs = f->coeffs;
}
_nmod_poly_exp_series(f_coeffs, h_coeffs, n, f->mod);
if (h == f && hlen >= n)
{
nmod_poly_swap(f, t1);
nmod_poly_clear(t1);
}
f->length = n;
if (hlen < n)
_nmod_vec_free(h_coeffs);
_nmod_poly_normalise(f);
}
void fmpz_mod_poly_radix(fmpz_mod_poly_struct **B,
const fmpz_mod_poly_t F,
const fmpz_mod_poly_radix_t D)
{
const long lenF = F->length;
const long degF = F->length - 1;
const long degR = D->degR;
const long N = degF / degR;
if (N == 0)
{
fmpz_mod_poly_set(B[0], F);
}
else
{
const long k = FLINT_BIT_COUNT(N); /* k := ceil{log{N+1}} */
const long lenG = (1L << k) * degR; /* Padded size */
const long t = (lenG - 1) / degR - N; /* Extra {degR}-blocks */
fmpz *G; /* Padded copy of F */
fmpz *T; /* Additional B[i] */
fmpz **C; /* Enlarged version of B */
fmpz *W; /* Temporary space */
long i;
if (lenF < lenG)
{
G = flint_malloc(lenG * sizeof(fmpz));
for (i = 0; i < lenF; i++)
G[i] = F->coeffs[i];
mpn_zero((mp_ptr) G + lenF, lenG - lenF);
T = t ? _fmpz_vec_init(t * degR) : NULL;
}
else /* lenF == lenG */
{
G = F->coeffs;
T = NULL;
}
C = flint_malloc((N + 1 + t) * sizeof(fmpz *));
for (i = 0; i <= N; i++)
{
fmpz_mod_poly_fit_length(B[i], degR);
C[i] = B[i]->coeffs;
}
for (i = 0; i < t; i++)
{
C[N + 1 + i] = T + i * degR;
}
W = _fmpz_vec_init(lenG);
_fmpz_mod_poly_radix(C, G, D->Rpow, D->Rinv, degR, 0, k-1, W, &(F->p));
_fmpz_vec_clear(W, lenG);
for (i = 0; i <= N; i++)
{
_fmpz_mod_poly_set_length(B[i], degR);
_fmpz_mod_poly_normalise(B[i]);
}
flint_free(C);
if (lenF < lenG)
{
flint_free(G);
}
if (t)
{
_fmpz_vec_clear(T, t * degR);
}
}
}
