mp_limb_t mpn_sumdiff_n(mp_ptr s,mp_ptr d,mp_srcptr x,mp_srcptr y,mp_size_t n)
{mp_limb_t ret;mp_ptr t;
ASSERT(n>0);
ASSERT_MPN(x,n);ASSERT_MPN(y,n);//ASSERT_SPACE(s,n);ASSERT_SPACE(d,n);
ASSERT(MPN_SAME_OR_SEPARATE_P(s,x,n));
ASSERT(MPN_SAME_OR_SEPARATE_P(s,y,n));
ASSERT(MPN_SAME_OR_SEPARATE_P(d,x,n));
ASSERT(MPN_SAME_OR_SEPARATE_P(d,y,n));
ASSERT(!MPN_OVERLAP_P(s,n,d,n));
if( (s==x && d==y)||(s==y && d==x) )
{t=__GMP_ALLOCATE_FUNC_LIMBS(n);
ret=mpn_sub_n(t,x,y,n);
ret+=2*mpn_add_n(s,x,y,n);
MPN_COPY(d,t,n);
__GMP_FREE_FUNC_LIMBS(t,n);
return ret;}
if(s==x || s==y)
{ret=mpn_sub_n(d,x,y,n);
ret+=2*mpn_add_n(s,x,y,n);
return ret;}
ret=2*mpn_add_n(s,x,y,n);
ret+=mpn_sub_n(d,x,y,n);
return ret;}
void
mpn_sqr_n (mp_ptr p, mp_srcptr a, mp_size_t n)
{
ASSERT (n >= 1);
ASSERT (! MPN_OVERLAP_P (p, 2 * n, a, n));
#if 0
/* FIXME: Can this be removed? */
if (n == 0)
return;
#endif
if (BELOW_THRESHOLD (n, SQR_BASECASE_THRESHOLD))
{ /* mul_basecase is faster than sqr_basecase on small sizes sometimes */
mpn_mul_basecase (p, a, n, a, n);
}
else if (BELOW_THRESHOLD (n, SQR_KARATSUBA_THRESHOLD))
{
mpn_sqr_basecase (p, a, n);
}
else if (BELOW_THRESHOLD (n, SQR_TOOM3_THRESHOLD))
{
/* Allocate workspace of fixed size on stack: fast! */
mp_limb_t ws[MPN_KARA_SQR_N_TSIZE (SQR_TOOM3_THRESHOLD_LIMIT-1)];
ASSERT (SQR_TOOM3_THRESHOLD <= SQR_TOOM3_THRESHOLD_LIMIT);
mpn_kara_sqr_n (p, a, n, ws);
}
#if WANT_FFT || TUNE_PROGRAM_BUILD
else if (BELOW_THRESHOLD (n, SQR_FFT_THRESHOLD))
#else
else if (BELOW_THRESHOLD (n, MPN_TOOM3_MAX_N))
#endif
{
mp_ptr ws;
TMP_SDECL;
TMP_SMARK;
ws = TMP_SALLOC_LIMBS (MPN_TOOM3_SQR_N_TSIZE (n));
mpn_toom3_sqr_n (p, a, n, ws);
TMP_SFREE;
}
else
#if WANT_FFT || TUNE_PROGRAM_BUILD
{
/* The current FFT code allocates its own space. That should probably
change. */
mpn_mul_fft_full (p, a, n, a, n);
}
#else
{
/* Toom3 for large operands. Use workspace from the heap, as stack space
may be limited. Since n is at least MUL_TOOM3_THRESHOLD, multiplication
will take much longer than malloc()/free(). */
mp_ptr ws; mp_size_t ws_size;
ws_size = MPN_TOOM3_SQR_N_TSIZE (n);
ws = __GMP_ALLOCATE_FUNC_LIMBS (ws_size);
mpn_toom3_sqr_n (p, a, n, ws);
__GMP_FREE_FUNC_LIMBS (ws, ws_size);
}
#endif
}
void
mpz_rootrem (mpz_ptr root, mpz_ptr rem, mpz_srcptr u, unsigned long int nth)
{
mp_ptr rootp, up, remp;
mp_size_t us, un, rootn, remn;
up = PTR(u);
us = SIZ(u);
/* even roots of negatives provoke an exception */
if (us < 0 && (nth & 1) == 0)
SQRT_OF_NEGATIVE;
/* root extraction interpreted as c^(1/nth) means a zeroth root should
provoke a divide by zero, do this even if c==0 */
if (nth == 0)
DIVIDE_BY_ZERO;
if (us == 0)
{
if (root != NULL)
SIZ(root) = 0;
SIZ(rem) = 0;
return;
}
un = ABS (us);
rootn = (un - 1) / nth + 1;
if (root != NULL)
{
rootp = MPZ_REALLOC (root, rootn);
up = PTR(u);
}
else
{
rootp = __GMP_ALLOCATE_FUNC_LIMBS (rootn);
}
MPZ_REALLOC (rem, un);
remp = PTR(rem);
if (nth == 1)
{
MPN_COPY (rootp, up, un);
remn = 0;
}
else
{
remn = mpn_rootrem (rootp, remp, up, un, nth);
}
if (root != NULL)
SIZ(root) = us >= 0 ? rootn : -rootn;
else
__GMP_FREE_FUNC_LIMBS (rootp, rootn);
SIZ(rem) = remn;
}
static void
ref_mpn_mul (mp_ptr wp, mp_srcptr up, mp_size_t un, mp_srcptr vp, mp_size_t vn)
{
mp_ptr tp;
mp_size_t tn;
mp_limb_t cy;
if (vn < TOOM3_THRESHOLD)
{
/* In the mpn_mul_basecase and mpn_kara_mul_n range, use our own
mul_basecase. */
if (vn != 0)
mul_basecase (wp, up, un, vp, vn);
else
MPN_ZERO (wp, un);
return;
}
if (vn < FFT_THRESHOLD)
{
/* In the mpn_toom3_mul_n and mpn_toom4_mul_n range, use mpn_kara_mul_n. */
tn = 2 * vn + MPN_KARA_MUL_N_TSIZE (vn);
tp = __GMP_ALLOCATE_FUNC_LIMBS (tn);
mpn_kara_mul_n (tp, up, vp, vn, tp + 2 * vn);
}
else
{
/* Finally, for the largest operands, use mpn_toom3_mul_n. */
/* The "- 63 + 255" tweaks the allocation to allow for huge operands.
See the definition of this macro in gmp-impl.h to understand this. */
tn = 2 * vn + MPN_TOOM3_MUL_N_TSIZE (vn) - 63 + 255;
tp = __GMP_ALLOCATE_FUNC_LIMBS (tn);
mpn_toom3_mul_n (tp, up, vp, vn, tp + 2 * vn);
}
if (un != vn)
{
if (un - vn < vn)
ref_mpn_mul (wp + vn, vp, vn, up + vn, un - vn);
else
ref_mpn_mul (wp + vn, up + vn, un - vn, vp, vn);
MPN_COPY (wp, tp, vn);
cy = mpn_add_n (wp + vn, wp + vn, tp + vn, vn);
mpn_incr_u (wp + 2 * vn, cy);
}
else
{
MPN_COPY (wp, tp, 2 * vn);
}
__GMP_FREE_FUNC_LIMBS (tp, tn);
}
void
mpn_sqr_n (mp_ptr prodp,
mp_srcptr up, mp_size_t un)
{
ASSERT (un >= 1);
ASSERT (! MPN_OVERLAP_P (prodp, 2*un, up, un));
/* FIXME: Can this be removed? */
if (un == 0)
return;
if (BELOW_THRESHOLD (un, SQR_BASECASE_THRESHOLD))
{ /* mul_basecase is faster than sqr_basecase on small sizes sometimes */
mpn_mul_basecase (prodp, up, un, up, un);
}
else if (BELOW_THRESHOLD (un, SQR_KARATSUBA_THRESHOLD))
{ /* plain schoolbook multiplication */
mpn_sqr_basecase (prodp, up, un);
}
else if (BELOW_THRESHOLD (un, SQR_TOOM3_THRESHOLD))
{ /* karatsuba multiplication */
mp_ptr tspace;
TMP_DECL (marker);
TMP_MARK (marker);
tspace = TMP_ALLOC_LIMBS (MPN_KARA_SQR_N_TSIZE (un));
mpn_kara_sqr_n (prodp, up, un, tspace);
TMP_FREE (marker);
}
#if WANT_FFT || TUNE_PROGRAM_BUILD
else if (BELOW_THRESHOLD (un, SQR_FFT_THRESHOLD))
#else
else
#endif
{ /* Toom3 multiplication.
Use workspace from the heap, as stack may be limited. Since n is
at least MUL_TOOM3_THRESHOLD, the multiplication will take much
longer than malloc()/free(). */
mp_ptr tspace;
mp_size_t tsize;
tsize = MPN_TOOM3_SQR_N_TSIZE (un);
tspace = __GMP_ALLOCATE_FUNC_LIMBS (tsize);
mpn_toom3_sqr_n (prodp, up, un, tspace);
__GMP_FREE_FUNC_LIMBS (tspace, tsize);
}
#if WANT_FFT || TUNE_PROGRAM_BUILD
else
{
static void
refmpz_mul (mpz_t w, const mpz_t u, const mpz_t v)
{
mp_size_t usize = u->_mp_size;
mp_size_t vsize = v->_mp_size;
mp_size_t wsize;
mp_size_t sign_product;
mp_ptr up, vp;
mp_ptr wp;
mp_size_t talloc;
sign_product = usize ^ vsize;
usize = ABS (usize);
vsize = ABS (vsize);
if (usize == 0 || vsize == 0)
{
SIZ (w) = 0;
return;
}
talloc = usize + vsize;
up = u->_mp_d;
vp = v->_mp_d;
wp = __GMP_ALLOCATE_FUNC_LIMBS (talloc);
if (usize > vsize)
refmpn_mul (wp, up, usize, vp, vsize);
else
refmpn_mul (wp, vp, vsize, up, usize);
wsize = usize + vsize;
wsize -= wp[wsize - 1] == 0;
MPZ_REALLOC (w, wsize);
MPN_COPY (PTR(w), wp, wsize);
SIZ(w) = sign_product < 0 ? -wsize : wsize;
__GMP_FREE_FUNC_LIMBS (wp, talloc);
}
void
mpn_toom4_sqr_n (mp_ptr rp, mp_srcptr up, mp_size_t n)
{
mp_size_t len1, ind;
mp_limb_t cy, r30, r31;
mp_ptr tp;
mp_size_t a0n, a1n, a2n, a3n, sn, n1, n2, n3, n4, n5, n6, n7, n8, n9, rpn, t4;
len1 = n;
ASSERT (n >= 1);
MPN_NORMALIZE(up, len1);
sn = (n - 1) / 4 + 1;
/* a0 - a3 are defined in mpn_toom4_mul_n above */
TC4_NORM(a0, a0n, sn);
TC4_NORM(a1, a1n, sn);
TC4_NORM(a2, a2n, sn);
TC4_NORM(a3, a3n, n - 3*sn);
t4 = 2*sn+2; // allows mult of 2 integers of sn + 1 limbs
tp = __GMP_ALLOCATE_FUNC_LIMBS(4*t4 + 4*(sn + 1));
tc4_add_unsigned(u5, &n5, a3, a3n, a1, a1n);
tc4_add_unsigned(u4, &n4, a2, a2n, a0, a0n);
tc4_add_unsigned(u2, &n2, u4, n4, u5, n5);
tc4_sub(u3, &n3, u4, n4, u5, n5);
SQR_TC4(r4, n4, u3, n3);
SQR_TC4_UNSIGNED(r3, n3, u2, n2);
tc4_lshift(r1, &n1, a0, a0n, 3);
tc4_addlsh1_unsigned(r1, &n1, a2, a2n);
tc4_lshift(r2, &n8, a1, a1n, 2);
tc4_add(r2, &n8, r2, n8, a3, a3n);
tc4_add(u4, &n9, r1, n1, r2, n8);
tc4_sub(u5, &n5, r1, n1, r2, n8);
r30 = r3[0];
if (!n3) r30 = CNST_LIMB(0);
r31 = r3[1];
SQR_TC4(r6, n6, u5, n5);
SQR_TC4_UNSIGNED(r5, n5, u4, n9);
r3[1] = r31;
tc4_lshift(u2, &n8, a3, a3n, 3);
tc4_addmul_1(u2, &n8, a2, a2n, 4);
tc4_addlsh1_unsigned(u2, &n8, a1, a1n);
tc4_add(u2, &n8, u2, n8, a0, a0n);
SQR_TC4_UNSIGNED(r2, n2, u2, n8);
SQR_TC4_UNSIGNED(r1, n1, a3, a3n);
SQR_TC4_UNSIGNED(r7, n7, a0, a0n);
TC4_DENORM(r1, n1, t4 - 1);
TC4_DENORM(r2, n2, t4 - 1);
if (n3)
TC4_DENORM(r3, n3, t4 - 1);
else {
/* MPN_ZERO defeats gcc 4.1.2 here, hence the explicit for loop */
for (ind = 1 ; ind < t4 - 1; ind++)
(r3)[ind] = CNST_LIMB(0);
}
TC4_DENORM(r4, n4, t4 - 1);
TC4_DENORM(r5, n5, t4 - 1);
TC4_DENORM(r6, n6, t4 - 1);
TC4_DENORM(r7, n7, t4 - 2); // we treat r7 differently (it cannot exceed t4-2 in length)
/* rp rp1 rp2 rp3 rp4 rp5 rp6 rp7
<----------- r7-----------><------------r5-------------->
<-------------r3------------->
<-------------r6-------------> < -----------r2------------>{ }
<-------------r4--------------> <--------------r1---->
*/
mpn_toom4_interpolate(rp, &rpn, sn, tp, t4 - 1, n4, n6, r30);
if (rpn != 2*n)
{
MPN_ZERO((rp + rpn), 2*n - rpn);
}
__GMP_FREE_FUNC_LIMBS (tp, 4*t4 + 4*(sn+1));
}
void
mpz_clear (mpz_ptr x)
{
if (ALLOC (x))
__GMP_FREE_FUNC_LIMBS (PTR (x), ALLOC(x));
}
int
main (int argc, char **argv)
{
gmp_randstate_ptr rands;
unsigned long maxnbits, maxdbits, nbits, dbits;
mpz_t n, d, tz;
mp_size_t maxnn, maxdn, nn, dn, clearn, i;
mp_ptr np, dp, qp, rp;
mp_limb_t rh;
mp_limb_t t;
mp_limb_t dinv;
int count = COUNT;
mp_ptr scratch;
mp_limb_t ran;
mp_size_t alloc, itch;
mp_limb_t rran0, rran1, qran0, qran1;
TMP_DECL;
if (argc > 1)
{
char *end;
count = strtol (argv[1], &end, 0);
if (*end || count <= 0)
{
fprintf (stderr, "Invalid test count: %s.\n", argv[1]);
return 1;
}
}
maxdbits = MAX_DN;
maxnbits = MAX_NN;
tests_start ();
rands = RANDS;
mpz_init (n);
mpz_init (d);
mpz_init (tz);
maxnn = maxnbits / GMP_NUMB_BITS + 1;
maxdn = maxdbits / GMP_NUMB_BITS + 1;
TMP_MARK;
qp = TMP_ALLOC_LIMBS (maxnn + 2) + 1;
rp = TMP_ALLOC_LIMBS (maxnn + 2) + 1;
alloc = 1;
scratch = __GMP_ALLOCATE_FUNC_LIMBS (alloc);
for (test = 0; test < count;)
{
nbits = random_word (rands) % (maxnbits - GMP_NUMB_BITS) + 2 * GMP_NUMB_BITS;
if (maxdbits > nbits)
dbits = random_word (rands) % nbits + 1;
else
dbits = random_word (rands) % maxdbits + 1;
#if RAND_UNIFORM
#define RANDFUNC mpz_urandomb
#else
#define RANDFUNC mpz_rrandomb
#endif
do
{
RANDFUNC (n, rands, nbits);
do
{
RANDFUNC (d, rands, dbits);
}
while (mpz_sgn (d) == 0);
np = PTR (n);
dp = PTR (d);
nn = SIZ (n);
dn = SIZ (d);
}
while (nn < dn);
dp[0] |= 1;
mpz_urandomb (tz, rands, 32);
t = mpz_get_ui (tz);
if (t % 17 == 0)
dp[0] = GMP_NUMB_MAX;
switch ((int) t % 16)
{
case 0:
clearn = random_word (rands) % nn;
for (i = 0; i <= clearn; i++)
np[i] = 0;
break;
case 1:
mpn_sub_1 (np + nn - dn, dp, dn, random_word (rands));
break;
case 2:
mpn_add_1 (np + nn - dn, dp, dn, random_word (rands));
break;
}
test++;
binvert_limb (dinv, dp[0]);
rran0 = random_word (rands);
rran1 = random_word (rands);
qran0 = random_word (rands);
qran1 = random_word (rands);
qp[-1] = qran0;
qp[nn - dn + 1] = qran1;
rp[-1] = rran0;
ran = random_word (rands);
if ((double) (nn - dn) * dn < 1e5)
{
if (nn > dn)
{
/* Test mpn_sbpi1_bdiv_qr */
MPN_ZERO (qp, nn - dn);
MPN_ZERO (rp, dn);
MPN_COPY (rp, np, nn);
rh = mpn_sbpi1_bdiv_qr (qp, rp, nn, dp, dn, -dinv);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0);
check_one (qp, rp + nn - dn, rh, np, nn, dp, dn, "mpn_sbpi1_bdiv_qr");
}
if (nn > dn)
{
/* Test mpn_sbpi1_bdiv_q */
MPN_COPY (rp, np, nn);
MPN_ZERO (qp, nn - dn);
mpn_sbpi1_bdiv_q (qp, rp, nn - dn, dp, MIN(dn,nn-dn), -dinv);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0);
check_one (qp, NULL, 0, np, nn, dp, dn, "mpn_sbpi1_bdiv_q");
}
}
if (dn >= 4 && nn - dn >= 2)
{
/* Test mpn_dcpi1_bdiv_qr */
MPN_COPY (rp, np, nn);
MPN_ZERO (qp, nn - dn);
rh = mpn_dcpi1_bdiv_qr (qp, rp, nn, dp, dn, -dinv);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0);
check_one (qp, rp + nn - dn, rh, np, nn, dp, dn, "mpn_dcpi1_bdiv_qr");
}
if (dn >= 4 && nn - dn >= 2)
{
/* Test mpn_dcpi1_bdiv_q */
MPN_COPY (rp, np, nn);
MPN_ZERO (qp, nn - dn);
mpn_dcpi1_bdiv_q (qp, rp, nn - dn, dp, MIN(dn,nn-dn), -dinv);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0);
check_one (qp, NULL, 0, np, nn, dp, dn, "mpn_dcpi1_bdiv_q");
}
if (nn > dn)
{
/* Test mpn_bdiv_qr */
itch = mpn_bdiv_qr_itch (nn, dn);
if (itch + 1 > alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
MPN_ZERO (qp, nn - dn);
MPN_ZERO (rp, dn);
rp[dn] = rran1;
rh = mpn_bdiv_qr (qp, rp, np, nn, dp, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0); ASSERT_ALWAYS (rp[dn] == rran1);
check_one (qp, rp, rh, np, nn, dp, dn, "mpn_bdiv_qr");
}
if (nn - dn < 2 || dn < 2)
continue;
/* Test mpn_mu_bdiv_qr */
itch = mpn_mu_bdiv_qr_itch (nn, dn);
if (itch + 1 > alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
MPN_ZERO (qp, nn - dn);
MPN_ZERO (rp, dn);
rp[dn] = rran1;
rh = mpn_mu_bdiv_qr (qp, rp, np, nn, dp, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0); ASSERT_ALWAYS (rp[dn] == rran1);
check_one (qp, rp, rh, np, nn, dp, dn, "mpn_mu_bdiv_qr");
/* Test mpn_mu_bdiv_q */
itch = mpn_mu_bdiv_q_itch (nn, dn);
if (itch + 1 > alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
MPN_ZERO (qp, nn - dn + 1);
mpn_mu_bdiv_q (qp, np, nn - dn, dp, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
check_one (qp, NULL, 0, np, nn, dp, dn, "mpn_mu_bdiv_q");
}
__GMP_FREE_FUNC_LIMBS (scratch, alloc);
TMP_FREE;
mpz_clear (n);
mpz_clear (d);
mpz_clear (tz);
tests_end ();
return 0;
}
int
main (int argc, char **argv)
{
gmp_randstate_ptr rands;
unsigned long maxnbits, maxdbits, nbits, dbits;
mpz_t n, d, q, r, tz, junk;
mp_size_t maxnn, maxdn, nn, dn, clearn, i;
mp_ptr np, dup, dnp, qp, rp, junkp;
mp_limb_t t;
gmp_pi1_t dinv;
long count = COUNT;
mp_ptr scratch;
mp_limb_t ran;
mp_size_t alloc, itch;
mp_limb_t rran0, rran1, qran0, qran1;
TMP_DECL;
if (argc > 1)
{
char *end;
count = strtol (argv[1], &end, 0);
if (*end || count <= 0)
{
fprintf (stderr, "Invalid test count: %s.\n", argv[1]);
return 1;
}
}
maxdbits = MAX_DN;
maxnbits = MAX_NN;
tests_start ();
rands = RANDS;
mpz_init (n);
mpz_init (d);
mpz_init (q);
mpz_init (r);
mpz_init (tz);
mpz_init (junk);
maxnn = maxnbits / GMP_NUMB_BITS + 1;
maxdn = maxdbits / GMP_NUMB_BITS + 1;
TMP_MARK;
qp = TMP_ALLOC_LIMBS (maxnn + 2) + 1;
rp = TMP_ALLOC_LIMBS (maxnn + 2) + 1;
dnp = TMP_ALLOC_LIMBS (maxdn);
alloc = 1;
scratch = __GMP_ALLOCATE_FUNC_LIMBS (alloc);
for (test = -300; test < count; test++)
{
nbits = random_word (rands) % (maxnbits - GMP_NUMB_BITS) + 2 * GMP_NUMB_BITS;
if (test < 0)
dbits = (test + 300) % (nbits - 1) + 1;
else
dbits = random_word (rands) % (nbits - 1) % maxdbits + 1;
#if RAND_UNIFORM
#define RANDFUNC mpz_urandomb
#else
#define RANDFUNC mpz_rrandomb
#endif
do
RANDFUNC (d, rands, dbits);
while (mpz_sgn (d) == 0);
dn = SIZ (d);
dup = PTR (d);
MPN_COPY (dnp, dup, dn);
dnp[dn - 1] |= GMP_NUMB_HIGHBIT;
if (test % 2 == 0)
{
RANDFUNC (n, rands, nbits);
nn = SIZ (n);
ASSERT_ALWAYS (nn >= dn);
}
else
{
do
{
RANDFUNC (q, rands, random_word (rands) % (nbits - dbits + 1));
RANDFUNC (r, rands, random_word (rands) % mpz_sizeinbase (d, 2));
mpz_mul (n, q, d);
mpz_add (n, n, r);
nn = SIZ (n);
}
while (nn > maxnn || nn < dn);
}
ASSERT_ALWAYS (nn <= maxnn);
ASSERT_ALWAYS (dn <= maxdn);
mpz_urandomb (junk, rands, nbits);
junkp = PTR (junk);
np = PTR (n);
mpz_urandomb (tz, rands, 32);
t = mpz_get_ui (tz);
if (t % 17 == 0)
{
dnp[dn - 1] = GMP_NUMB_MAX;
dup[dn - 1] = GMP_NUMB_MAX;
}
switch ((int) t % 16)
{
case 0:
clearn = random_word (rands) % nn;
for (i = clearn; i < nn; i++)
np[i] = 0;
break;
case 1:
mpn_sub_1 (np + nn - dn, dnp, dn, random_word (rands));
break;
case 2:
mpn_add_1 (np + nn - dn, dnp, dn, random_word (rands));
break;
}
if (dn >= 2)
invert_pi1 (dinv, dnp[dn - 1], dnp[dn - 2]);
rran0 = random_word (rands);
rran1 = random_word (rands);
qran0 = random_word (rands);
qran1 = random_word (rands);
qp[-1] = qran0;
qp[nn - dn + 1] = qran1;
rp[-1] = rran0;
ran = random_word (rands);
if ((double) (nn - dn) * dn < 1e5)
{
/* Test mpn_sbpi1_div_qr */
if (dn > 2)
{
MPN_COPY (rp, np, nn);
if (nn > dn)
MPN_COPY (qp, junkp, nn - dn);
qp[nn - dn] = mpn_sbpi1_div_qr (qp, rp, nn, dnp, dn, dinv.inv32);
check_one (qp, rp, np, nn, dnp, dn, "mpn_sbpi1_div_qr", 0);
}
/* Test mpn_sbpi1_divappr_q */
if (dn > 2)
{
MPN_COPY (rp, np, nn);
if (nn > dn)
MPN_COPY (qp, junkp, nn - dn);
qp[nn - dn] = mpn_sbpi1_divappr_q (qp, rp, nn, dnp, dn, dinv.inv32);
check_one (qp, NULL, np, nn, dnp, dn, "mpn_sbpi1_divappr_q", 1);
}
/* Test mpn_sbpi1_div_q */
if (dn > 2)
{
MPN_COPY (rp, np, nn);
if (nn > dn)
MPN_COPY (qp, junkp, nn - dn);
qp[nn - dn] = mpn_sbpi1_div_q (qp, rp, nn, dnp, dn, dinv.inv32);
check_one (qp, NULL, np, nn, dnp, dn, "mpn_sbpi1_div_q", 0);
}
/* Test mpn_sb_div_qr_sec */
itch = 3 * nn + 4;
if (itch + 1 > alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
MPN_COPY (rp, np, nn);
if (nn >= dn)
MPN_COPY (qp, junkp, nn - dn + 1);
mpn_sb_div_qr_sec (qp, rp, nn, dup, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
check_one (qp, rp, np, nn, dup, dn, "mpn_sb_div_qr_sec", 0);
/* Test mpn_sb_div_r_sec */
itch = nn + 2 * dn + 2;
if (itch + 1 > alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
MPN_COPY (rp, np, nn);
mpn_sb_div_r_sec (rp, nn, dup, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
/* Note: Since check_one cannot cope with random-only functions, we
pass qp[] from the previous function, mpn_sb_div_qr_sec. */
check_one (qp, rp, np, nn, dup, dn, "mpn_sb_div_r_sec", 0);
}
/* Test mpn_dcpi1_div_qr */
if (dn >= 6 && nn - dn >= 3)
{
MPN_COPY (rp, np, nn);
if (nn > dn)
MPN_COPY (qp, junkp, nn - dn);
qp[nn - dn] = mpn_dcpi1_div_qr (qp, rp, nn, dnp, dn, &dinv);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0);
check_one (qp, rp, np, nn, dnp, dn, "mpn_dcpi1_div_qr", 0);
}
/* Test mpn_dcpi1_divappr_q */
if (dn >= 6 && nn - dn >= 3)
{
MPN_COPY (rp, np, nn);
if (nn > dn)
MPN_COPY (qp, junkp, nn - dn);
qp[nn - dn] = mpn_dcpi1_divappr_q (qp, rp, nn, dnp, dn, &dinv);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0);
check_one (qp, NULL, np, nn, dnp, dn, "mpn_dcpi1_divappr_q", 1);
}
/* Test mpn_dcpi1_div_q */
if (dn >= 6 && nn - dn >= 3)
{
MPN_COPY (rp, np, nn);
if (nn > dn)
MPN_COPY (qp, junkp, nn - dn);
qp[nn - dn] = mpn_dcpi1_div_q (qp, rp, nn, dnp, dn, &dinv);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0);
check_one (qp, NULL, np, nn, dnp, dn, "mpn_dcpi1_div_q", 0);
}
/* Test mpn_mu_div_qr */
if (nn - dn > 2 && dn >= 2)
{
itch = mpn_mu_div_qr_itch (nn, dn, 0);
if (itch + 1 > alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
MPN_COPY (qp, junkp, nn - dn);
MPN_ZERO (rp, dn);
rp[dn] = rran1;
qp[nn - dn] = mpn_mu_div_qr (qp, rp, np, nn, dnp, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
ASSERT_ALWAYS (rp[-1] == rran0); ASSERT_ALWAYS (rp[dn] == rran1);
check_one (qp, rp, np, nn, dnp, dn, "mpn_mu_div_qr", 0);
}
/* Test mpn_mu_divappr_q */
if (nn - dn > 2 && dn >= 2)
{
itch = mpn_mu_divappr_q_itch (nn, dn, 0);
if (itch + 1 > alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
MPN_COPY (qp, junkp, nn - dn);
qp[nn - dn] = mpn_mu_divappr_q (qp, np, nn, dnp, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
check_one (qp, NULL, np, nn, dnp, dn, "mpn_mu_divappr_q", 4);
}
/* Test mpn_mu_div_q */
if (nn - dn > 2 && dn >= 2)
{
itch = mpn_mu_div_q_itch (nn, dn, 0);
if (itch + 1> alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
MPN_COPY (qp, junkp, nn - dn);
qp[nn - dn] = mpn_mu_div_q (qp, np, nn, dnp, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
check_one (qp, NULL, np, nn, dnp, dn, "mpn_mu_div_q", 0);
}
if (1)
{
itch = nn + 1;
if (itch + 1> alloc)
{
scratch = __GMP_REALLOCATE_FUNC_LIMBS (scratch, alloc, itch + 1);
alloc = itch + 1;
}
scratch[itch] = ran;
mpn_div_q (qp, np, nn, dup, dn, scratch);
ASSERT_ALWAYS (ran == scratch[itch]);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - dn + 1] == qran1);
check_one (qp, NULL, np, nn, dup, dn, "mpn_div_q", 0);
}
if (dn >= 2 && nn >= 2)
{
mp_limb_t qh;
/* mpn_divrem_2 */
MPN_COPY (rp, np, nn);
qp[nn - 2] = qp[nn-1] = qran1;
qh = mpn_divrem_2 (qp, 0, rp, nn, dnp + dn - 2);
ASSERT_ALWAYS (qp[nn - 2] == qran1);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - 1] == qran1);
qp[nn - 2] = qh;
check_one (qp, rp, np, nn, dnp + dn - 2, 2, "mpn_divrem_2", 0);
/* Missing: divrem_2 with fraction limbs. */
/* mpn_div_qr_2 */
qp[nn - 2] = qran1;
qh = mpn_div_qr_2 (qp, rp, np, nn, dup + dn - 2);
ASSERT_ALWAYS (qp[nn - 2] == qran1);
ASSERT_ALWAYS (qp[-1] == qran0); ASSERT_ALWAYS (qp[nn - 1] == qran1);
qp[nn - 2] = qh;
check_one (qp, rp, np, nn, dup + dn - 2, 2, "mpn_div_qr_2", 0);
}
}
__GMP_FREE_FUNC_LIMBS (scratch, alloc);
TMP_FREE;
mpz_clear (n);
mpz_clear (d);
mpz_clear (q);
mpz_clear (r);
mpz_clear (tz);
mpz_clear (junk);
tests_end ();
return 0;
}
/* Multiply {up, un} by {vp, vn} and write the result to
{prodp, un + vn} assuming vn > 2*ceil(un/5).
Note that prodp gets un + vn limbs stored, even if the actual
result only needs un + vn - 1.
*/
void
mpn_toom53_mul (mp_ptr rp, mp_srcptr up, mp_size_t un,
mp_srcptr vp, mp_size_t vn)
{
mp_size_t ind;
mp_limb_t cy, r30, r31;
mp_ptr tp;
mp_size_t a0n, a1n, a2n, a3n, a4n, b0n, b1n, b2n, sn, n1, n2, n3, n4, n5, n6, n7, n8, n9, n10, rpn, t4;
sn = (un + 4) / 5;
ASSERT (vn > 2*sn);
#define a0 (up)
#define a1 (up + sn)
#define a2 (up + 2*sn)
#define a3 (up + 3*sn)
#define a4 (up + 4*sn)
#define b0 (vp)
#define b1 (vp + sn)
#define b2 (vp + 2*sn)
TC4_NORM(a0, a0n, sn);
TC4_NORM(a1, a1n, sn);
TC4_NORM(a2, a2n, sn);
TC4_NORM(a3, a3n, sn);
TC4_NORM(a4, a4n, un - 4*sn);
TC4_NORM(b0, b0n, sn);
TC4_NORM(b1, b1n, sn);
TC4_NORM(b2, b2n, vn - 2*sn);
t4 = 2*sn+2; // allows mult of 2 integers of sn + 1 limbs
tp = __GMP_ALLOCATE_FUNC_LIMBS(4*t4 + 4*(sn + 1));
#define u2 (tp + 4*t4)
#define u3 (tp + 4*t4 + (sn+1))
#define u4 (tp + 4*t4 + 2*(sn+1))
#define u5 (tp + 4*t4 + 3*(sn+1))
tc4_add_unsigned(u2, &n2, a3, a3n, a1, a1n);
tc4_add_unsigned(u5, &n5, a2, a2n, a0, a0n);
tc4_add_unsigned(u5, &n5, u5, n5, a4, a4n);
tc4_add_unsigned(u3, &n3, u5, n5, u2, n2);
tc4_sub(u4, &n4, u5, n5, u2, n2);
tc4_add_unsigned(u5, &n5, b2, b2n, b0, b0n);
tc4_add_unsigned(r2, &n8, u5, n5, b1, b1n);
tc4_sub(u5, &n5, u5, n5, b1, b1n);
MUL_TC4_UNSIGNED(r3, n3, u3, n3, r2, n8); /* 1 */
MUL_TC4(r4, n4, u4, n4, u5, n5); /* -1 */
tc4_lshift(r1, &n1, a0, a0n, 4);
tc4_lshift(u3, &n9, a2, a2n, 2);
tc4_add_unsigned(r1, &n1, r1, n1, u3, n9);
tc4_add_unsigned(r1, &n1, r1, n1, a4, a4n);
tc4_lshift(r2, &n8, a1, a1n, 3);
tc4_addlsh1_unsigned(r2, &n8, a3, a3n);
tc4_add_unsigned(u5, &n5, r1, n1, r2, n8);
tc4_sub(u3, &n9, r1, n1, r2, n8);
tc4_lshift(r1, &n1, b0, b0n, 2);
tc4_add_unsigned(r1, &n1, r1, n1, b2, b2n);
tc4_lshift(u4, &n10, b1, b1n, 1);
tc4_add_unsigned(u2, &n2, r1, n1, u4, n10);
tc4_sub(r2, &n8, r1, n1, u4, n10);
r30 = r3[0];
if (!n3) r30 = CNST_LIMB(0);
r31 = r3[1];
MUL_TC4_UNSIGNED(r5, n5, u5, n5, u2, n2); /* 1/2 */
MUL_TC4(r6, n6, u3, n9, r2, n8); /* -1/2 */
r3[1] = r31;
tc4_lshift(u2, &n2, a4, a4n, 4);
tc4_addmul_1(u2, &n2, a3, a3n, 8);
tc4_addmul_1(u2, &n2, a2, a2n, 4);
tc4_addlsh1_unsigned(u2, &n2, a1, a1n);
tc4_add(u2, &n2, u2, n2, a0, a0n);
tc4_lshift(r1, &n1, b2, b2n, 2);
tc4_addlsh1_unsigned(r1, &n1, b1, b1n);
tc4_add(r1, &n1, r1, n1, b0, b0n);
MUL_TC4_UNSIGNED(r2, n2, u2, n2, r1, n1); /* 2 */
MUL_TC4_UNSIGNED(r1, n1, a4, a4n, b2, b2n); /* oo */
MUL_TC4_UNSIGNED(r7, n7, a0, a0n, b0, b0n); /* 0 */
TC4_DENORM(r1, n1, t4 - 1);
TC4_DENORM(r2, n2, t4 - 1);
if (n3)
TC4_DENORM(r3, n3, t4 - 1);
else {
/* MPN_ZERO defeats gcc 4.1.2 here, hence the explicit for loop */
for (ind = 1 ; ind < t4 - 1; ind++)
(r3)[ind] = CNST_LIMB(0);
}
TC4_DENORM(r4, n4, t4 - 1);
TC4_DENORM(r5, n5, t4 - 1);
TC4_DENORM(r6, n6, t4 - 1);
TC4_DENORM(r7, n7, t4 - 2); // we treat r7 differently (it cannot exceed t4-2 in length)
/* rp rp1 rp2 rp3 rp4 rp5 rp6 rp7
<----------- r7-----------><------------r5-------------->
<-------------r3------------->
<-------------r6-------------> < -----------r2------------>{ }
<-------------r4--------------> <--------------r1---->
*/
mpn_toom4_interpolate(rp, &rpn, sn, tp, t4 - 1, n4, n6, r30);
if (rpn != un + vn)
{
MPN_ZERO((rp + rpn), un + vn - rpn);
}
__GMP_FREE_FUNC_LIMBS (tp, 4*t4 + 4*(sn+1));
}
