void tc4_divexact_by15(mp_ptr rp, mp_size_t * rn, mp_ptr x, mp_size_t xn)
{
if (xn)
{
mp_size_t xu = ABS(xn);
mpn_divexact_byfobm1(rp, x, xu, CNST_LIMB(15), CNST_LIMB((~0)/15)); /* works for 32 and 64 bits */
if (xn > 0)
{
if (rp[xu - 1] == CNST_LIMB(0)) *rn = xn - 1;
else *rn = xn;
} else
{
if (rp[xu - 1] == CNST_LIMB(0)) *rn = xn + 1;
else *rn = xn;
}
} else *rn = 0;
}
/*
Toom 4 interpolation. Interpolates the value at 2^(sn*B) of a
polynomial p(x) with 7 coefficients given the values
p(oo), p(2), p(1), p(-1), 2^6*p(1/2), 2^6*p(-1/2), p(0).
The output is placed in rp and the final number of limbs of the
output is given in rpn.
The 4th and 6th values may be negative, and if so, n4 and n6
should be set to a negative value respectively.
To save space we pass r3, r5, r7 in place in the output rp.
The other r's are stored separately in space tp.
The low limb of r3 is stored in r30, as it will be overwritten
by the high limb of r5.
rp rp1 rp2 rp3 rp4 rp5 rp6 rp7
<----------- r7-----------><------------r5-------------->
<-------------r3------------->
We assume that r1 is stored at tp, r2 at (tp + t4), r4 at (tp + 2*t4)
and r6 (tp + 3*t4). Each of these r's has t4 = s4 + 1 limbs allocated.
*/
void mpn_toom4_interpolate(mp_ptr rp, mp_size_t * rpn, mp_size_t sn,
mp_ptr tp, mp_size_t s4, mp_size_t n4, mp_size_t n6, mp_limb_t r30)
{
mp_size_t n1, n2, n3, n5, n7, t4;
mp_limb_t saved, saved2, cy;
t4 = s4 + 1;
mpn_add_n(r2, r2, r5, s4);
if (n6 < 0)
mpn_add_n(r6, r5, r6, s4);
else
mpn_sub_n(r6, r5, r6, s4);
/* r6 is now in twos complement format */
saved = r3[0];
r3[0] = r30;
if (n4 < 0)
mpn_add_n(r4, r3, r4, s4);
else
mpn_sub_n(r4, r3, r4, s4);
r3[0] = saved;
/* r4 is now in twos complement format */
mpn_sub_n(r5, r5, r1, s4);
#if HAVE_NATIVE_mpn_sublsh_n
r5[s4-1] -= mpn_sublsh_n(r5, r5, r7, s4-1, 6);
#else
r5[s4-1] -= mpn_submul_1(r5, r7, s4-1, 64);
#endif
TC4_RSHIFT1(r4, s4);
saved = r3[0];
r3[0] = r30;
mpn_sub_n(r3, r3, r4, s4);
r30 = r3[0];
r3[0] = saved;
mpn_double(r5, s4);
mpn_sub_n(r5, r5, r6, s4);
saved = r3[0];
r3[0] = r30;
mpn_submul_1(r2, r3, s4, 65);
r3[0] = saved;
saved2 = r7[s4-1];
r7[s4-1] = CNST_LIMB(0); // r7 is always positive so no sign extend needed
saved = r3[0];
r3[0] = r30;
#if HAVE_NATIVE_mpn_subadd_n
mpn_subadd_n(r3, r3, r7, r1, s4);
#else
mpn_sub_n(r3, r3, r7, s4);
mpn_sub_n(r3, r3, r1, s4);
#endif
r7[s4-1] = saved2;
r30 = r3[0];
mpn_addmul_1(r2, r3, s4, 45);
#if HAVE_NATIVE_mpn_sublsh_n
cy = mpn_sublsh_n(r5, r5, r3, s4 - 1, 3);
#else
cy = mpn_submul_1(r5, r3, s4 - 1, 8);
#endif
r3[0] = saved;
r3[0] -= (cy + 8*r3[s4-1]);
mpn_rshift(r5, r5, s4, 3);
mpn_divexact_by3(r5, r5, s4);
mpn_sub_n(r6, r6, r2, s4);
#if HAVE_NATIVE_mpn_sublsh_n
mpn_sublsh_n(r2, r2, r4, s4, 4);
#else
mpn_submul_1(r2, r4, s4, 16);
#endif
mpn_rshift(r2, r2, s4, 1);
mpn_divexact_by3(r2, r2, s4);
mpn_divexact_by3(r2, r2, s4);
saved = r3[0];
r3[0] = r30;
cy = mpn_sub_n(r3, r3, r5, s4 - 1);
r30 = r3[0];
r3[0] = saved;
r3[s4-1] -= (cy + r5[s4-1]);
mpn_sub_n(r4, r4, r2, s4);
mpn_addmul_1(r6, r2, s4, 30);
mpn_divexact_byfobm1(r6, r6, s4, CNST_LIMB(15), CNST_LIMB(~0/15));
mpn_rshift(r6, r6, s4, 2);
mpn_sub_n(r2, r2, r6, s4);
TC4_NORM(r1, n1, s4);
TC4_NORM(r2, n2, s4);
(*rpn) = 6*sn+1;
cy = mpn_add_1(r3, r3, *rpn - 4*sn, r30); /* don't forget to add r3[0] back in */
if (cy)
{
rp[*rpn] = cy;
(*rpn)++;
}
tc4_copy(rp, rpn, 5*sn, r2, n2);
tc4_copy(rp, rpn, 6*sn, r1, n1);
tc4_copy(rp, rpn, sn, r6, s4);
tc4_copy(rp, rpn, 3*sn, r4, s4);
}
void
check_functions (void)
{
mp_limb_t wp[2], wp2[2], xp[2], yp[2], r;
int i;
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 123;
yp[0] = 456;
mpn_add_n (wp, xp, yp, (mp_size_t) 1);
ASSERT_ALWAYS (wp[0] == 579);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 123;
wp[0] = 456;
r = mpn_addmul_1 (wp, xp, (mp_size_t) 1, CNST_LIMB(2));
ASSERT_ALWAYS (wp[0] == 702);
ASSERT_ALWAYS (r == 0);
}
#if HAVE_NATIVE_mpn_copyd
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 123;
xp[1] = 456;
mpn_copyd (xp+1, xp, (mp_size_t) 1);
ASSERT_ALWAYS (xp[1] == 123);
}
#endif
#if HAVE_NATIVE_mpn_copyi
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 123;
xp[1] = 456;
mpn_copyi (xp, xp+1, (mp_size_t) 1);
ASSERT_ALWAYS (xp[0] == 456);
}
#endif
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 1605;
mpn_divexact_1 (wp, xp, (mp_size_t) 1, CNST_LIMB(5));
ASSERT_ALWAYS (wp[0] == 321);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 1296;
r = mpn_divexact_by3c (wp, xp, (mp_size_t) 1, CNST_LIMB(0));
ASSERT_ALWAYS (wp[0] == 432);
ASSERT_ALWAYS (r == 0);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 578;
r = mpn_divexact_byfobm1 (wp, xp, (mp_size_t) 1, CNST_LIMB(17),CNST_LIMB(-1)/CNST_LIMB(17));
ASSERT_ALWAYS (wp[0] == 34);
ASSERT_ALWAYS (r == 0);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 287;
r = mpn_divrem_1 (wp, (mp_size_t) 1, xp, (mp_size_t) 1, CNST_LIMB(7));
ASSERT_ALWAYS (wp[1] == 41);
ASSERT_ALWAYS (wp[0] == 0);
ASSERT_ALWAYS (r == 0);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 290;
r = mpn_divrem_euclidean_qr_1 (wp, 0, xp, (mp_size_t) 1, CNST_LIMB(7));
ASSERT_ALWAYS (wp[0] == 41);
ASSERT_ALWAYS (r == 3);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 12;
r = mpn_gcd_1 (xp, (mp_size_t) 1, CNST_LIMB(9));
ASSERT_ALWAYS (r == 3);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 0x1001;
mpn_lshift (wp, xp, (mp_size_t) 1, 1);
ASSERT_ALWAYS (wp[0] == 0x2002);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 14;
r = mpn_mod_1 (xp, (mp_size_t) 1, CNST_LIMB(4));
ASSERT_ALWAYS (r == 2);
}
#if (GMP_NUMB_BITS % 4) == 0
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
int bits = (GMP_NUMB_BITS / 4) * 3;
mp_limb_t mod = (CNST_LIMB(1) << bits) - 1;
mp_limb_t want = GMP_NUMB_MAX % mod;
xp[0] = GMP_NUMB_MAX;
r = mpn_mod_34lsub1 (xp, (mp_size_t) 1);
ASSERT_ALWAYS (r % mod == want);
}
#endif
// DECL_modexact_1c_odd ((*modexact_1c_odd));
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 14;
r = mpn_mul_1 (wp, xp, (mp_size_t) 1, CNST_LIMB(4));
ASSERT_ALWAYS (wp[0] == 56);
ASSERT_ALWAYS (r == 0);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 5;
yp[0] = 7;
mpn_mul_basecase (wp, xp, (mp_size_t) 1, yp, (mp_size_t) 1);
ASSERT_ALWAYS (wp[0] == 35);
ASSERT_ALWAYS (wp[1] == 0);
}
#if HAVE_NATIVE_mpn_preinv_divrem_1 && GMP_NAIL_BITS == 0
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 0x101;
r = mpn_preinv_divrem_1 (wp, (mp_size_t) 1, xp, (mp_size_t) 1,
GMP_LIMB_HIGHBIT,
refmpn_invert_limb (GMP_LIMB_HIGHBIT), 0);
ASSERT_ALWAYS (wp[0] == 0x202);
ASSERT_ALWAYS (wp[1] == 0);
ASSERT_ALWAYS (r == 0);
}
#endif
#if GMP_NAIL_BITS == 0
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = GMP_LIMB_HIGHBIT+123;
r = mpn_preinv_mod_1 (xp, (mp_size_t) 1, GMP_LIMB_HIGHBIT,
refmpn_invert_limb (GMP_LIMB_HIGHBIT));
ASSERT_ALWAYS (r == 123);
}
#endif
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 5;
modlimb_invert(r,xp[0]);
r=-r;
yp[0]=43;
yp[1]=75;
mpn_redc_1 (wp, yp, xp, (mp_size_t) 1,r);
ASSERT_ALWAYS (wp[0] == 78);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0]=5;
yp[0]=3;
mpn_sumdiff_n (wp, wp2,xp, yp,1);
ASSERT_ALWAYS (wp[0] == 8);
ASSERT_ALWAYS (wp2[0] == 2);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 0x8008;
mpn_rshift (wp, xp, (mp_size_t) 1, 1);
ASSERT_ALWAYS (wp[0] == 0x4004);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 5;
mpn_sqr_basecase (wp, xp, (mp_size_t) 1);
ASSERT_ALWAYS (wp[0] == 25);
ASSERT_ALWAYS (wp[1] == 0);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 999;
yp[0] = 666;
mpn_sub_n (wp, xp, yp, (mp_size_t) 1);
ASSERT_ALWAYS (wp[0] == 333);
}
memcpy (&__gmpn_cpuvec, &initial_cpuvec, sizeof (__gmpn_cpuvec));
for (i = 0; i < 2; i++)
{
xp[0] = 123;
wp[0] = 456;
r = mpn_submul_1 (wp, xp, (mp_size_t) 1, CNST_LIMB(2));
ASSERT_ALWAYS (wp[0] == 210);
ASSERT_ALWAYS (r == 0);
}
}
