int
mpn_toom_eval_dgr3_pm1 (mp_ptr xp1, mp_ptr xm1,
mp_srcptr xp, mp_size_t n, mp_size_t x3n, mp_ptr tp)
{
int neg;
ASSERT (x3n > 0);
ASSERT (x3n <= n);
xp1[n] = mpn_add_n (xp1, xp, xp + 2*n, n);
tp[n] = mpn_add (tp, xp + n, n, xp + 3*n, x3n);
neg = (mpn_cmp (xp1, tp, n + 1) < 0) ? ~0 : 0;
#if HAVE_NATIVE_mpn_add_n_sub_n
if (neg)
mpn_add_n_sub_n (xp1, xm1, tp, xp1, n + 1);
else
mpn_add_n_sub_n (xp1, xm1, xp1, tp, n + 1);
#else
if (neg)
mpn_sub_n (xm1, tp, xp1, n + 1);
else
mpn_sub_n (xm1, xp1, tp, n + 1);
mpn_add_n (xp1, xp1, tp, n + 1);
#endif
ASSERT (xp1[n] <= 3);
ASSERT (xm1[n] <= 1);
return neg;
}
int
main (int argc, char **argv)
{
mp_ptr r1p, r2p, s1p, s2p;
double t;
mp_size_t n;
n = strtol (argv[1], 0, 0);
r1p = malloc (n * GMP_LIMB_BYTES);
r2p = malloc (n * GMP_LIMB_BYTES);
s1p = malloc (n * GMP_LIMB_BYTES);
s2p = malloc (n * GMP_LIMB_BYTES);
TIME (t,(mpn_add_n(r1p,s1p,s2p,n),mpn_sub_n(r1p,s1p,s2p,n)));
printf (" separate add and sub: %.3f\n", t);
TIME (t,mpn_add_n_sub_n(r1p,r2p,s1p,s2p,n));
printf ("combined addsub separate variables: %.3f\n", t);
TIME (t,mpn_add_n_sub_n(r1p,r2p,r1p,s2p,n));
printf (" combined addsub r1 overlap: %.3f\n", t);
TIME (t,mpn_add_n_sub_n(r1p,r2p,r1p,s2p,n));
printf (" combined addsub r2 overlap: %.3f\n", t);
TIME (t,mpn_add_n_sub_n(r1p,r2p,r1p,r2p,n));
printf (" combined addsub in-place: %.3f\n", t);
return 0;
}
/* Needs n+1 limbs of temporary storage. */
int
mpn_toom_eval_dgr3_pm2 (mp_ptr xp2, mp_ptr xm2,
mp_srcptr xp, mp_size_t n, mp_size_t x3n, mp_ptr tp)
{
mp_limb_t cy;
int neg;
ASSERT (x3n > 0);
ASSERT (x3n <= n);
/* (x0 + 4 * x2) +/- (2 x1 + 8 x_3) */
#if HAVE_NATIVE_mpn_addlsh_n || HAVE_NATIVE_mpn_addlsh2_n
#if HAVE_NATIVE_mpn_addlsh2_n
xp2[n] = mpn_addlsh2_n (xp2, xp, xp + 2*n, n);
cy = mpn_addlsh2_n (tp, xp + n, xp + 3*n, x3n);
#else /* HAVE_NATIVE_mpn_addlsh_n */
xp2[n] = mpn_addlsh_n (xp2, xp, xp + 2*n, n, 2);
cy = mpn_addlsh_n (tp, xp + n, xp + 3*n, x3n, 2);
#endif
if (x3n < n)
cy = mpn_add_1 (tp + x3n, xp + n + x3n, n - x3n, cy);
tp[n] = cy;
#else
cy = mpn_lshift (tp, xp + 2*n, n, 2);
xp2[n] = cy + mpn_add_n (xp2, tp, xp, n);
tp[x3n] = mpn_lshift (tp, xp + 3*n, x3n, 2);
if (x3n < n)
tp[n] = mpn_add (tp, xp + n, n, tp, x3n + 1);
else
tp[n] += mpn_add_n (tp, xp + n, tp, n);
#endif
mpn_lshift (tp, tp, n+1, 1);
neg = (mpn_cmp (xp2, tp, n + 1) < 0) ? ~0 : 0;
#if HAVE_NATIVE_mpn_add_n_sub_n
if (neg)
mpn_add_n_sub_n (xp2, xm2, tp, xp2, n + 1);
else
mpn_add_n_sub_n (xp2, xm2, xp2, tp, n + 1);
#else
if (neg)
mpn_sub_n (xm2, tp, xp2, n + 1);
else
mpn_sub_n (xm2, xp2, tp, n + 1);
mpn_add_n (xp2, xp2, tp, n + 1);
#endif
ASSERT (xp2[n] < 15);
ASSERT (xm2[n] < 10);
return neg;
}
/* Evaluates a polynomial of degree k > 3, in the points +1 and -1. */
int
mpn_toom_eval_pm1 (mp_ptr xp1, mp_ptr xm1, unsigned k,
mp_srcptr xp, mp_size_t n, mp_size_t hn, mp_ptr tp)
{
unsigned i;
int neg;
ASSERT (k >= 4);
ASSERT (hn > 0);
ASSERT (hn <= n);
/* The degree k is also the number of full-size coefficients, so
* that last coefficient, of size hn, starts at xp + k*n. */
xp1[n] = mpn_add_n (xp1, xp, xp + 2*n, n);
for (i = 4; i < k; i += 2)
ASSERT_NOCARRY (mpn_add (xp1, xp1, n+1, xp+i*n, n));
tp[n] = mpn_add_n (tp, xp + n, xp + 3*n, n);
for (i = 5; i < k; i += 2)
ASSERT_NOCARRY (mpn_add (tp, tp, n+1, xp+i*n, n));
if (k & 1)
ASSERT_NOCARRY (mpn_add (tp, tp, n+1, xp+k*n, hn));
else
ASSERT_NOCARRY (mpn_add (xp1, xp1, n+1, xp+k*n, hn));
neg = (mpn_cmp (xp1, tp, n + 1) < 0) ? ~0 : 0;
#if HAVE_NATIVE_mpn_add_n_sub_n
if (neg)
mpn_add_n_sub_n (xp1, xm1, tp, xp1, n + 1);
else
mpn_add_n_sub_n (xp1, xm1, xp1, tp, n + 1);
#else
if (neg)
mpn_sub_n (xm1, tp, xp1, n + 1);
else
mpn_sub_n (xm1, xp1, tp, n + 1);
mpn_add_n (xp1, xp1, tp, n + 1);
#endif
ASSERT (xp1[n] <= k);
ASSERT (xm1[n] <= k/2 + 1);
return neg;
}
void
mpn_toom3_sqr (mp_ptr pp,
mp_srcptr ap, mp_size_t an,
mp_ptr scratch)
{
mp_size_t n, s;
mp_limb_t cy, vinf0;
mp_ptr gp;
mp_ptr as1, asm1, as2;
#define a0 ap
#define a1 (ap + n)
#define a2 (ap + 2*n)
n = (an + 2) / (size_t) 3;
s = an - 2 * n;
ASSERT (0 < s && s <= n);
as1 = scratch + 4 * n + 4;
asm1 = scratch + 2 * n + 2;
as2 = pp + n + 1;
gp = scratch;
/* Compute as1 and asm1. */
cy = mpn_add (gp, a0, n, a2, s);
#if HAVE_NATIVE_mpn_add_n_sub_n
if (cy == 0 && mpn_cmp (gp, a1, n) < 0)
{
cy = mpn_add_n_sub_n (as1, asm1, a1, gp, n);
as1[n] = cy >> 1;
asm1[n] = 0;
}
void
mpn_toom33_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, vinf0;
mp_ptr gp;
mp_ptr as1, asm1, as2;
mp_ptr bs1, bsm1, bs2;
#define a0 ap
#define a1 (ap + n)
#define a2 (ap + 2*n)
#define b0 bp
#define b1 (bp + n)
#define b2 (bp + 2*n)
n = (an + 2) / (size_t) 3;
s = an - 2 * n;
t = bn - 2 * n;
ASSERT (an >= bn);
ASSERT (0 < s && s <= n);
ASSERT (0 < t && t <= n);
as1 = scratch + 4 * n + 4;
asm1 = scratch + 2 * n + 2;
as2 = pp + n + 1;
bs1 = pp;
bsm1 = scratch + 3 * n + 3; /* we need 4n+4 <= 4n+s+t */
bs2 = pp + 2 * n + 2;
gp = scratch;
vm1_neg = 0;
/* Compute as1 and asm1. */
cy = mpn_add (gp, a0, n, a2, s);
#if HAVE_NATIVE_mpn_add_n_sub_n
if (cy == 0 && mpn_cmp (gp, a1, n) < 0)
{
cy = mpn_add_n_sub_n (as1, asm1, a1, gp, n);
as1[n] = cy >> 1;
asm1[n] = 0;
vm1_neg = 1;
}
/* Evaluates a polynomial of degree k > 2, in the points +2^shift and -2^shift. */
int
mpn_toom_eval_pm2exp (mp_ptr xp2, mp_ptr xm2, unsigned k,
mp_srcptr xp, mp_size_t n, mp_size_t hn, unsigned shift,
mp_ptr tp)
{
unsigned i;
int neg;
#ifdef HAVE_NATIVE_mpn_addlsh_n
mp_limb_t cy;
#endif
ASSERT (k >= 3);
ASSERT (shift*k < GMP_NUMB_BITS);
ASSERT (hn > 0);
ASSERT (hn <= n);
/* The degree k is also the number of full-size coefficients, so
* that last coefficient, of size hn, starts at xp + k*n. */
#ifdef HAVE_NATIVE_mpn_addlsh_n
xp2[n] = mpn_addlsh_n (xp2, xp, xp + 2*n, n, 2*shift);
for (i = 4; i < k; i += 2)
xp2[n] += mpn_addlsh_n (xp2, xp2, xp + i*n, n, i*shift);
tp[n] = mpn_lshift (tp, xp+n, n, shift);
for (i = 3; i < k; i+= 2)
tp[n] += mpn_addlsh_n (tp, tp, xp+i*n, n, i*shift);
if (k & 1)
{
cy = mpn_addlsh_n (tp, tp, xp+k*n, hn, k*shift);
MPN_INCR_U (tp + hn, n+1 - hn, cy);
}
else
{
cy = mpn_addlsh_n (xp2, xp2, xp+k*n, hn, k*shift);
MPN_INCR_U (xp2 + hn, n+1 - hn, cy);
}
#else /* !HAVE_NATIVE_mpn_addlsh_n */
xp2[n] = mpn_lshift (tp, xp+2*n, n, 2*shift);
xp2[n] += mpn_add_n (xp2, xp, tp, n);
for (i = 4; i < k; i += 2)
{
xp2[n] += mpn_lshift (tp, xp + ((mp_size_t) i)*n, n, i*shift);
xp2[n] += mpn_add_n (xp2, xp2, tp, n);
}
tp[n] = mpn_lshift (tp, xp+n, n, shift);
for (i = 3; i < k; i+= 2)
{
tp[n] += mpn_lshift (xm2, xp + ((mp_size_t) i)*n, n, i*shift);
tp[n] += mpn_add_n (tp, tp, xm2, n);
}
xm2[hn] = mpn_lshift (xm2, xp + ((mp_size_t) k)*n, hn, k*shift);
if (k & 1)
mpn_add (tp, tp, n+1, xm2, hn+1);
else
mpn_add (xp2, xp2, n+1, xm2, hn+1);
#endif /* !HAVE_NATIVE_mpn_addlsh_n */
neg = (mpn_cmp (xp2, tp, n + 1) < 0) ? ~0 : 0;
#ifdef HAVE_NATIVE_mpn_add_n_sub_n
if (neg)
mpn_add_n_sub_n (xp2, xm2, tp, xp2, n + 1);
else
mpn_add_n_sub_n (xp2, xm2, xp2, tp, n + 1);
#else /* !HAVE_NATIVE_mpn_add_n_sub_n */
if (neg)
mpn_sub_n (xm2, tp, xp2, n + 1);
else
mpn_sub_n (xm2, xp2, tp, n + 1);
mpn_add_n (xp2, xp2, tp, n + 1);
#endif /* !HAVE_NATIVE_mpn_add_n_sub_n */
/* FIXME: the following asserts are useless if (k+1)*shift >= GMP_LIMB_BITS */
ASSERT ((k+1)*shift >= GMP_LIMB_BITS ||
xp2[n] < ((CNST_LIMB(1)<<((k+1)*shift))-1)/((CNST_LIMB(1)<<shift)-1));
ASSERT ((k+2)*shift >= GMP_LIMB_BITS ||
xm2[n] < ((CNST_LIMB(1)<<((k+2)*shift))-((k&1)?(CNST_LIMB(1)<<shift):1))/((CNST_LIMB(1)<<(2*shift))-1));
return neg;
}
void
mpn_toom_interpolate_12pts (mp_ptr pp, mp_ptr r1, mp_ptr r3, mp_ptr r5,
mp_size_t n, mp_size_t spt, int half, mp_ptr wsi)
{
mp_limb_t cy;
mp_size_t n3;
mp_size_t n3p1;
n3 = 3 * n;
n3p1 = n3 + 1;
#define r4 (pp + n3) /* 3n+1 */
#define r2 (pp + 7 * n) /* 3n+1 */
#define r0 (pp +11 * n) /* s+t <= 2*n */
/******************************* interpolation *****************************/
if (half != 0) {
cy = mpn_sub_n (r3, r3, r0, spt);
MPN_DECR_U (r3 + spt, n3p1 - spt, cy);
cy = DO_mpn_sublsh_n (r2, r0, spt, 10, wsi);
MPN_DECR_U (r2 + spt, n3p1 - spt, cy);
DO_mpn_subrsh(r5, n3p1, r0, spt, 2, wsi);
cy = DO_mpn_sublsh_n (r1, r0, spt, 20, wsi);
MPN_DECR_U (r1 + spt, n3p1 - spt, cy);
DO_mpn_subrsh(r4, n3p1, r0, spt, 4, wsi);
};
r4[n3] -= DO_mpn_sublsh_n (r4 + n, pp, 2 * n, 20, wsi);
DO_mpn_subrsh(r1 + n, 2 * n + 1, pp, 2 * n, 4, wsi);
#if HAVE_NATIVE_mpn_add_n_sub_n
mpn_add_n_sub_n (r1, r4, r4, r1, n3p1);
#else
ASSERT_NOCARRY(mpn_add_n (wsi, r1, r4, n3p1));
mpn_sub_n (r4, r4, r1, n3p1); /* can be negative */
MP_PTR_SWAP(r1, wsi);
#endif
r5[n3] -= DO_mpn_sublsh_n (r5 + n, pp, 2 * n, 10, wsi);
DO_mpn_subrsh(r2 + n, 2 * n + 1, pp, 2 * n, 2, wsi);
#if HAVE_NATIVE_mpn_add_n_sub_n
mpn_add_n_sub_n (r2, r5, r5, r2, n3p1);
#else
mpn_sub_n (wsi, r5, r2, n3p1); /* can be negative */
ASSERT_NOCARRY(mpn_add_n (r2, r2, r5, n3p1));
MP_PTR_SWAP(r5, wsi);
#endif
r3[n3] -= mpn_sub_n (r3+n, r3+n, pp, 2 * n);
#if AORSMUL_FASTER_AORS_AORSLSH
mpn_submul_1 (r4, r5, n3p1, 257); /* can be negative */
#else
mpn_sub_n (r4, r4, r5, n3p1); /* can be negative */
DO_mpn_sublsh_n (r4, r5, n3p1, 8, wsi); /* can be negative */
#endif
/* A division by 2835x4 follows. Warning: the operand can be negative! */
mpn_divexact_by2835x4(r4, r4, n3p1);
if ((r4[n3] & (GMP_NUMB_MAX << (GMP_NUMB_BITS-3))) != 0)
r4[n3] |= (GMP_NUMB_MAX << (GMP_NUMB_BITS-2));
#if AORSMUL_FASTER_2AORSLSH
mpn_addmul_1 (r5, r4, n3p1, 60); /* can be negative */
#else
DO_mpn_sublsh_n (r5, r4, n3p1, 2, wsi); /* can be negative */
DO_mpn_addlsh_n (r5, r4, n3p1, 6, wsi); /* can give a carry */
#endif
mpn_divexact_by255(r5, r5, n3p1);
ASSERT_NOCARRY(DO_mpn_sublsh_n (r2, r3, n3p1, 5, wsi));
#if AORSMUL_FASTER_3AORSLSH
ASSERT_NOCARRY(mpn_submul_1 (r1, r2, n3p1, 100));
#else
ASSERT_NOCARRY(DO_mpn_sublsh_n (r1, r2, n3p1, 6, wsi));
ASSERT_NOCARRY(DO_mpn_sublsh_n (r1, r2, n3p1, 5, wsi));
ASSERT_NOCARRY(DO_mpn_sublsh_n (r1, r2, n3p1, 2, wsi));
#endif
ASSERT_NOCARRY(DO_mpn_sublsh_n (r1, r3, n3p1, 9, wsi));
mpn_divexact_by42525(r1, r1, n3p1);
#if AORSMUL_FASTER_AORS_2AORSLSH
ASSERT_NOCARRY(mpn_submul_1 (r2, r1, n3p1, 225));
#else
ASSERT_NOCARRY(mpn_sub_n (r2, r2, r1, n3p1));
ASSERT_NOCARRY(DO_mpn_addlsh_n (r2, r1, n3p1, 5, wsi));
ASSERT_NOCARRY(DO_mpn_sublsh_n (r2, r1, n3p1, 8, wsi));
#endif
mpn_divexact_by9x4(r2, r2, n3p1);
ASSERT_NOCARRY(mpn_sub_n (r3, r3, r2, n3p1));
mpn_sub_n (r4, r2, r4, n3p1);
ASSERT_NOCARRY(mpn_rshift(r4, r4, n3p1, 1));
ASSERT_NOCARRY(mpn_sub_n (r2, r2, r4, n3p1));
mpn_add_n (r5, r5, r1, n3p1);
ASSERT_NOCARRY(mpn_rshift(r5, r5, n3p1, 1));
/* last interpolation steps... */
ASSERT_NOCARRY(mpn_sub_n (r3, r3, r1, n3p1));
ASSERT_NOCARRY(mpn_sub_n (r1, r1, r5, n3p1));
/* ... could be mixed with recomposition
||H-r5|M-r5|L-r5| ||H-r1|M-r1|L-r1|
*/
/***************************** recomposition *******************************/
/*
pp[] prior to operations:
|M r0|L r0|___||H r2|M r2|L r2|___||H r4|M r4|L r4|____|H_r6|L r6|pp
summation scheme for remaining operations:
|__12|n_11|n_10|n__9|n__8|n__7|n__6|n__5|n__4|n__3|n__2|n___|n___|pp
|M r0|L r0|___||H r2|M r2|L r2|___||H r4|M r4|L r4|____|H_r6|L r6|pp
||H r1|M r1|L r1| ||H r3|M r3|L r3| ||H_r5|M_r5|L_r5|
*/
cy = mpn_add_n (pp + n, pp + n, r5, n);
cy = mpn_add_1 (pp + 2 * n, r5 + n, n, cy);
#if HAVE_NATIVE_mpn_add_nc
cy = r5[n3] + mpn_add_nc(pp + n3, pp + n3, r5 + 2 * n, n, cy);
#else
MPN_INCR_U (r5 + 2 * n, n + 1, cy);
cy = r5[n3] + mpn_add_n (pp + n3, pp + n3, r5 + 2 * n, n);
#endif
MPN_INCR_U (pp + n3 + n, 2 * n + 1, cy);
pp[2 * n3]+= mpn_add_n (pp + 5 * n, pp + 5 * n, r3, n);
cy = mpn_add_1 (pp + 2 * n3, r3 + n, n, pp[2 * n3]);
#if HAVE_NATIVE_mpn_add_nc
cy = r3[n3] + mpn_add_nc(pp + 7 * n, pp + 7 * n, r3 + 2 * n, n, cy);
#else
MPN_INCR_U (r3 + 2 * n, n + 1, cy);
cy = r3[n3] + mpn_add_n (pp + 7 * n, pp + 7 * n, r3 + 2 * n, n);
#endif
MPN_INCR_U (pp + 8 * n, 2 * n + 1, cy);
pp[10*n]+=mpn_add_n (pp + 9 * n, pp + 9 * n, r1, n);
if (half) {
cy = mpn_add_1 (pp + 10 * n, r1 + n, n, pp[10 * n]);
#if HAVE_NATIVE_mpn_add_nc
if (LIKELY (spt > n)) {
cy = r1[n3] + mpn_add_nc(pp + 11 * n, pp + 11 * n, r1 + 2 * n, n, cy);
MPN_INCR_U (pp + 4 * n3, spt - n, cy);
} else {
ASSERT_NOCARRY(mpn_add_nc(pp + 11 * n, pp + 11 * n, r1 + 2 * n, spt, cy));
}
#else
MPN_INCR_U (r1 + 2 * n, n + 1, cy);
if (LIKELY (spt > n)) {
cy = r1[n3] + mpn_add_n (pp + 11 * n, pp + 11 * n, r1 + 2 * n, n);
MPN_INCR_U (pp + 4 * n3, spt - n, cy);
} else {
ASSERT_NOCARRY(mpn_add_n (pp + 11 * n, pp + 11 * n, r1 + 2 * n, spt));
}
#endif
} else {
ASSERT_NOCARRY(mpn_add_1 (pp + 10 * n, r1 + n, spt, pp[10 * n]));
}
#undef r0
#undef r2
#undef r4
}
void
mpn_toom63_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;
mp_limb_t cy;
int sign;
/***************************** decomposition *******************************/
#define a5 (ap + 5 * n)
#define b0 (bp + 0 * n)
#define b1 (bp + 1 * n)
#define b2 (bp + 2 * n)
ASSERT (an >= bn);
n = 1 + (an >= 2 * bn ? (an - 1) / (size_t) 6 : (bn - 1) / (size_t) 3);
s = an - 5 * n;
t = bn - 2 * n;
ASSERT (0 < s && s <= n);
ASSERT (0 < t && t <= n);
/* WARNING! it assumes s+t>=n */
ASSERT ( s + t >= n );
ASSERT ( s + t > 4);
/* WARNING! it assumes n>1 */
ASSERT ( n > 2);
#define r8 pp /* 2n */
#define r7 scratch /* 3n+1 */
#define r5 (pp + 3*n) /* 3n+1 */
#define v0 (pp + 3*n) /* n+1 */
#define v1 (pp + 4*n+1) /* n+1 */
#define v2 (pp + 5*n+2) /* n+1 */
#define v3 (pp + 6*n+3) /* n+1 */
#define r3 (scratch + 3 * n + 1) /* 3n+1 */
#define r1 (pp + 7*n) /* s+t <= 2*n */
#define ws (scratch + 6 * n + 2) /* ??? */
/* Alloc also 3n+1 limbs for ws... mpn_toom_interpolate_8pts may
need all of them, when DO_mpn_sublsh_n usea a scratch */
/* if (scratch == NULL) scratch = TMP_SALLOC_LIMBS (9 * n + 3); */
/********************** evaluation and recursive calls *********************/
/* $\pm4$ */
sign = mpn_toom_eval_pm2exp (v2, v0, 5, ap, n, s, 2, pp);
pp[n] = mpn_lshift (pp, b1, n, 2); /* 4b1 */
/* FIXME: use addlsh */
v3[t] = mpn_lshift (v3, b2, t, 4);/* 16b2 */
if ( n == t )
v3[n]+= mpn_add_n (v3, v3, b0, n); /* 16b2+b0 */
else
v3[n] = mpn_add (v3, b0, n, v3, t+1); /* 16b2+b0 */
sign ^= abs_sub_add_n (v1, v3, pp, n + 1);
TOOM_63_MUL_N_REC(pp, v0, v1, n + 1, ws); /* A(-4)*B(-4) */
TOOM_63_MUL_N_REC(r3, v2, v3, n + 1, ws); /* A(+4)*B(+4) */
mpn_toom_couple_handling (r3, 2*n+1, pp, sign, n, 2, 4);
/* $\pm1$ */
sign = mpn_toom_eval_pm1 (v2, v0, 5, ap, n, s, pp);
/* Compute bs1 and bsm1. Code taken from toom33 */
cy = mpn_add (ws, b0, n, b2, t);
#if HAVE_NATIVE_mpn_add_n_sub_n
if (cy == 0 && mpn_cmp (ws, b1, n) < 0)
{
cy = mpn_add_n_sub_n (v3, v1, b1, ws, n);
v3[n] = cy >> 1;
v1[n] = 0;
sign = ~sign;
}
/* Evaluates a polynomial of degree 2 < k < GMP_NUMB_BITS, in the
points +2 and -2. */
int
mpn_toom_eval_pm2 (mp_ptr xp2, mp_ptr xm2, unsigned k,
mp_srcptr xp, mp_size_t n, mp_size_t hn, mp_ptr tp)
{
int i;
int neg;
mp_limb_t cy;
ASSERT (k >= 3);
ASSERT (k < GMP_NUMB_BITS);
ASSERT (hn > 0);
ASSERT (hn <= n);
/* The degree k is also the number of full-size coefficients, so
* that last coefficient, of size hn, starts at xp + k*n. */
cy = 0;
DO_addlsh2 (xp2, xp + (k-2) * n, xp + k * n, hn, cy);
if (hn != n)
cy = mpn_add_1 (xp2 + hn, xp + (k-2) * n + hn, n - hn, cy);
for (i = k - 4; i >= 0; i -= 2)
DO_addlsh2 (xp2, xp + i * n, xp2, n, cy);
xp2[n] = cy;
k--;
cy = 0;
DO_addlsh2 (tp, xp + (k-2) * n, xp + k * n, n, cy);
for (i = k - 4; i >= 0; i -= 2)
DO_addlsh2 (tp, xp + i * n, tp, n, cy);
tp[n] = cy;
if (k & 1)
ASSERT_NOCARRY(mpn_lshift (tp , tp , n + 1, 1));
else
ASSERT_NOCARRY(mpn_lshift (xp2, xp2, n + 1, 1));
neg = (mpn_cmp (xp2, tp, n + 1) < 0) ? ~0 : 0;
#if HAVE_NATIVE_mpn_add_n_sub_n
if (neg)
mpn_add_n_sub_n (xp2, xm2, tp, xp2, n + 1);
else
mpn_add_n_sub_n (xp2, xm2, xp2, tp, n + 1);
#else /* !HAVE_NATIVE_mpn_add_n_sub_n */
if (neg)
mpn_sub_n (xm2, tp, xp2, n + 1);
else
mpn_sub_n (xm2, xp2, tp, n + 1);
mpn_add_n (xp2, xp2, tp, n + 1);
#endif /* !HAVE_NATIVE_mpn_add_n_sub_n */
ASSERT (xp2[n] < (1<<(k+2))-1);
ASSERT (xm2[n] < ((1<<(k+3))-1 - (1^k&1))/3);
neg ^= ((k & 1) - 1);
return neg;
}
