void
mpn_toom22_mul (mp_ptr pp,
mp_srcptr ap, mp_size_t an,
mp_srcptr bp, mp_size_t bn,
mp_ptr scratch)
{
const int __gmpn_cpuvec_initialized = 1;
mp_size_t n, s, t;
int vm1_neg;
mp_limb_t cy, cy2;
mp_ptr asm1;
mp_ptr bsm1;
#define a0 ap
#define a1 (ap + n)
#define b0 bp
#define b1 (bp + n)
s = an >> 1;
n = an - s;
t = bn - n;
ASSERT (an >= bn);
ASSERT (0 < s && s <= n && s >= n - 1);
ASSERT (0 < t && t <= s);
asm1 = pp;
bsm1 = pp + n;
vm1_neg = 0;
/* Compute asm1. */
if (s == n)
{
if (mpn_cmp (a0, a1, n) < 0)
{
mpn_sub_n (asm1, a1, a0, n);
vm1_neg = 1;
}
else
{
mpn_sub_n (asm1, a0, a1, n);
}
}
else /* n - s == 1 */
{
if (a0[s] == 0 && mpn_cmp (a0, a1, s) < 0)
{
mpn_sub_n (asm1, a1, a0, s);
asm1[s] = 0;
vm1_neg = 1;
}
else
{
asm1[s] = a0[s] - mpn_sub_n (asm1, a0, a1, s);
}
}
/* Compute bsm1. */
if (t == n)
{
if (mpn_cmp (b0, b1, n) < 0)
{
mpn_sub_n (bsm1, b1, b0, n);
vm1_neg ^= 1;
}
else
{
mpn_sub_n (bsm1, b0, b1, n);
}
}
else
{
if (mpn_zero_p (b0 + t, n - t) && mpn_cmp (b0, b1, t) < 0)
{
mpn_sub_n (bsm1, b1, b0, t);
MPN_ZERO (bsm1 + t, n - t);
vm1_neg ^= 1;
}
else
{
mpn_sub (bsm1, b0, n, b1, t);
}
}
#define v0 pp /* 2n */
#define vinf (pp + 2 * n) /* s+t */
#define vm1 scratch /* 2n */
#define scratch_out scratch + 2 * n
/* vm1, 2n limbs */
TOOM22_MUL_N_REC (vm1, asm1, bsm1, n, scratch_out);
if (s > t) TOOM22_MUL_REC (vinf, a1, s, b1, t, scratch_out);
else TOOM22_MUL_N_REC (vinf, a1, b1, s, scratch_out);
/* v0, 2n limbs */
TOOM22_MUL_N_REC (v0, ap, bp, n, scratch_out);
/* H(v0) + L(vinf) */
cy = mpn_add_n (pp + 2 * n, v0 + n, vinf, n);
/* L(v0) + H(v0) */
cy2 = cy + mpn_add_n (pp + n, pp + 2 * n, v0, n);
/* L(vinf) + H(vinf) */
cy += mpn_add (pp + 2 * n, pp + 2 * n, n, vinf + n, s + t - n);
if (vm1_neg)
cy += mpn_add_n (pp + n, pp + n, vm1, 2 * n);
else
cy -= mpn_sub_n (pp + n, pp + n, vm1, 2 * n);
ASSERT (cy + 1 <= 3);
ASSERT (cy2 <= 2);
MPN_INCR_U (pp + 2 * n, s + t, cy2);
if (LIKELY (cy <= 2))
/* if s+t==n, cy is zero, but we should not acces pp[3*n] at all. */
MPN_INCR_U (pp + 3 * n, s + t - n, cy);
else
MPN_DECR_U (pp + 3 * n, s + t - n, 1);
}
void
mpn_toom22_mul (mp_ptr pp,
mp_srcptr ap, mp_size_t an,
mp_srcptr bp, mp_size_t bn,
mp_ptr scratch)
{
mp_size_t n, s, t;
int vm1_neg;
mp_limb_t cy, cy2;
mp_ptr asm1;
mp_ptr bsm1;
#define a0 ap
#define a1 (ap + n)
#define b0 bp
#define b1 (bp + n)
s = an >> 1;
n = an - s;
t = bn - n;
ASSERT (an >= bn);
ASSERT (0 < s && s <= n);
ASSERT (0 < t && t <= s);
asm1 = pp;
bsm1 = pp + n;
vm1_neg = 0;
/* Compute asm1. */
if (s == n)
{
if (mpn_cmp (a0, a1, n) < 0)
{
mpn_sub_n (asm1, a1, a0, n);
vm1_neg = 1;
}
else
{
mpn_sub_n (asm1, a0, a1, n);
}
}
else
{
if (mpn_zero_p (a0 + s, n - s) && mpn_cmp (a0, a1, s) < 0)
{
mpn_sub_n (asm1, a1, a0, s);
MPN_ZERO (asm1 + s, n - s);
vm1_neg = 1;
}
else
{
mpn_sub (asm1, a0, n, a1, s);
}
}
/* Compute bsm1. */
if (t == n)
{
if (mpn_cmp (b0, b1, n) < 0)
{
mpn_sub_n (bsm1, b1, b0, n);
vm1_neg ^= 1;
}
else
{
mpn_sub_n (bsm1, b0, b1, n);
}
}
else
{
if (mpn_zero_p (b0 + t, n - t) && mpn_cmp (b0, b1, t) < 0)
{
mpn_sub_n (bsm1, b1, b0, t);
MPN_ZERO (bsm1 + t, n - t);
vm1_neg ^= 1;
}
else
{
mpn_sub (bsm1, b0, n, b1, t);
}
}
#define v0 pp /* 2n */
#define vinf (pp + 2 * n) /* s+t */
#define vm1 scratch /* 2n */
#define scratch_out scratch + 2 * n
/* vm1, 2n limbs */
TOOM22_MUL_N_REC (vm1, asm1, bsm1, n, scratch_out);
if (s > t) TOOM22_MUL_REC (vinf, a1, s, b1, t, scratch_out);
else TOOM22_MUL_N_REC (vinf, a1, b1, s, scratch_out);
/* v0, 2n limbs */
TOOM22_MUL_N_REC (v0, ap, bp, n, scratch_out);
/* H(v0) + L(vinf) */
cy = mpn_add_n (pp + 2 * n, v0 + n, vinf, n);
/* L(v0) + H(v0) */
cy2 = cy + mpn_add_n (pp + n, pp + 2 * n, v0, n);
/* L(vinf) + H(vinf) */
cy += mpn_add (pp + 2 * n, pp + 2 * n, n, vinf + n, s + t - n);
if (vm1_neg)
cy += mpn_add_n (pp + n, pp + n, vm1, 2 * n);
else
cy -= mpn_sub_n (pp + n, pp + n, vm1, 2 * n);
ASSERT (cy + 1 <= 3);
ASSERT (cy2 <= 2);
mpn_incr_u (pp + 2 * n, cy2);
if (LIKELY (cy <= 2))
mpn_incr_u (pp + 3 * n, cy);
else
mpn_decr_u (pp + 3 * n, 1);
}
