void
free(void *ptr)
{
size_t usable,size,cnt = 0u;
unsigned int i;
char *p;
/* If ptr is NULL, no operation is performed.*/
if (ptr == NULL){
return real_free(ptr);
}
usable = malloc_usable_size(ptr);
/* At the first, check fixed redzone. If overwritten, following size
info is maybe invalid */
for (p = (char *)P_F_RZ(ptr, usable); p < (char *)P_F_RZ(ptr, usable) + SIZEOF_F_RZ; p++){
if(*p != MAGIC_BYTE)
cnt++;
}
if (cnt == SIZEOF_F_RZ){
ofc_count = cnt; /* for testing */
/* Maybe size info was broken */
OFC_DUMP_COUNT_MAYBE(cnt);
ofc_bt();
real_free(ptr);
return;
}
size = *(size_t *)P_SIZE(ptr, usable);
OFC_DUMP(ptr, usable, size);
p = P_RZ(ptr, usable, size);
for (i = 0; i < SIZEOF_RZ(usable ,size) - SIZEOF_F_RZ; i++) {
if(*(p + i) != MAGIC_BYTE)
cnt++;
}
if (cnt){
ofc_count = cnt; /* for testing */
OFC_DUMP_COUNT(cnt);
OFC_DUMP_INFO(ptr,size);
ofc_bt();
}
real_free(ptr);
return;
}
mp_size_t
mpn_gcdext (mp_ptr gp, mp_ptr s0p, mp_size_t *s0size,
mp_ptr ap, mp_size_t an, mp_ptr bp, mp_size_t n)
{
mp_size_t init_scratch, orig_n = n;
mp_size_t scratch, un, u0n, u1n;
mp_limb_t t;
mp_ptr tp, u0, u1;
int swapped = 0;
struct ngcd_matrix M;
mp_size_t p;
mp_size_t nn;
mp_limb_signed_t a;
int c;
TMP_DECL;
ASSERT (an >= n);
if (an == 1)
{
if (!n)
{
/* shouldn't ever occur, but we include for completeness */
gp[0] = ap[0];
s0p[0] = 1;
*s0size = 1;
return 1;
}
gp[0] = mpn_gcdinv_1(&a, ap[0], bp[0]);
if (a < (mp_limb_signed_t) 0)
{
s0p[0] = -a;
(*s0size) = -1;
} else
{
s0p[0] = a;
(*s0size) = 1 - (s0p[0] == 0);
}
return 1;
}
init_scratch = MPN_NGCD_MATRIX_INIT_ITCH (n-P_SIZE(n));
scratch = mpn_nhgcd_itch ((n+1)/2);
/* Space needed for mpn_ngcd_matrix_adjust */
if (scratch < 2*n)
scratch = 2*n;
if (scratch < an - n + 1) /* the first division can sometimes be selfish!! */
scratch = an - n + 1;
/* Space needed for cofactor adjust */
scratch = MAX(scratch, 2*(n+1) + P_SIZE(n) + 1);
TMP_MARK;
if (5*n + 2 + MPN_GCD_LEHMER_N_ITCH(n) > init_scratch + scratch)
tp = TMP_ALLOC_LIMBS (7*n+4+MPN_GCD_LEHMER_N_ITCH(n)); /* 2n+2 for u0, u1, 5*n+2 + MPN_GCD_LEHMER_N_ITCH(n) for Lehmer
and copies of ap and bp and s (and finally 3*n+1 for t and get_t) */
else
tp = TMP_ALLOC_LIMBS (2*(n+1) + init_scratch + scratch);
if (an > n)
{
mp_ptr qp = tp;
mpn_tdiv_qr (qp, ap, 0, ap, an, bp, n);
an = n;
MPN_NORMALIZE (ap, an);
if (an == 0)
{
MPN_COPY (gp, bp, n);
TMP_FREE;
(*s0size) = 0;
return n;
}
}
if (BELOW_THRESHOLD (n, GCDEXT_THRESHOLD))
{
n = mpn_ngcdext_lehmer (gp, s0p, s0size, ap, bp, n, tp);
TMP_FREE;
return n;
}
u0 = tp; /* Cofactor space */
u1 = tp + n + 1;
MPN_ZERO(tp, 2*(n+1));
tp += 2*(n+1);
/* First iteration, setup u0 and u1 */
p = P_SIZE(n);
mpn_ngcd_matrix_init (&M, n - p, tp);
ASSERT(tp + init_scratch > M.p[1][1] + M.n);
nn = mpn_nhgcd (ap + p, bp + p, n - p, &M, tp + init_scratch);
if (nn > 0)
{
n = mpn_ngcd_matrix_adjust (&M, p + nn, ap, bp, p, tp + init_scratch);
/*
(ap'', bp'')^T = M^-1(ap', bp')^T
and (ap', bp') = (1*ap + ?*bp, 0*ap + ?*bp)
We let u0 be minus the factor of ap appearing
in the expression for bp'' and u1 be the
factor of ap appearing in the expression for ap''
*/
MPN_COPY(u0, M.p[1][0], M.n);
MPN_COPY(u1, M.p[1][1], M.n);
un = M.n;
while ((u0[un-1] == 0) && (u1[un-1] == 0)) un--; /* normalise u0, u1, both cannot be zero as det = 1*/
}
else
{
mp_size_t gn;
un = 1;
u0[0] = 0; /* bp = 0*ap + ?*bp, thus u0 = -0 */
u1[0] = 1; /* ap = 1*ap + ?*bp, thus u1 = 1 */
n = mpn_ngcdext_subdiv_step (gp, &gn, s0p, u0, u1, &un, ap, bp, n, tp);
if (n == 0)
{
/* never observed to occur */
(*s0size) = un;
ASSERT(s0p[*s0size - 1] != 0);
TMP_FREE;
return gn;
}
}
while (ABOVE_THRESHOLD (n, GCDEXT_THRESHOLD))
{
struct ngcd_matrix M;
mp_size_t p = P_SIZE(n);
mp_size_t nn;
mpn_ngcd_matrix_init (&M, n - p, tp);
nn = mpn_nhgcd (ap + p, bp + p, n - p, &M, tp + init_scratch);
if (nn > 0)
{
n = mpn_ngcd_matrix_adjust (&M, p + nn, ap, bp, p, tp + init_scratch);
ngcdext_cofactor_adjust(u0, u1, &un, &M, tp + init_scratch);
/*
(ap'', bp'')^T = M^-1(ap', bp')^T
and (ap', bp') = (u1*ap + ?*bp, -u0*ap + ?*bp)
So we need u0' = -(-c*u1 + a*-u0) = a*u0 + c*u1
and we need u1' = (d*u1 -b*-u0) = b*u0 + d*u1
*/
ASSERT(un <= orig_n + 1);
} else
{
mp_size_t gn;
n = mpn_ngcdext_subdiv_step (gp, &gn, s0p, u0, u1, &un, ap, bp, n, tp);
ASSERT(un <= orig_n + 1);
if (n == 0)
{
(*s0size) = un;
ASSERT(((*s0size) == 0) || (s0p[ABS(*s0size) - 1] != 0));
TMP_FREE;
return gn;
}
}
}
ASSERT (ap[n-1] > 0 || bp[n-1] > 0);
ASSERT (u0[un-1] > 0 || u1[un-1] > 0);
if (ap[n-1] < bp[n-1])
{
MP_PTR_SWAP (ap, bp);
MP_PTR_SWAP (u0, u1);
swapped = 1;
}
an = n; /* {ap, an} and {bp, bn} are normalised, {ap, an} >= {bp, bn} */
MPN_NORMALIZE (bp, n);
if (n == 0)
{
/* If bp == 0 then gp = ap
with cofactor u1
If we swapped then cofactor is -u1
This case never seems to happen
*/
MPN_COPY (gp, ap, an);
MPN_NORMALIZE(u1, un);
MPN_COPY(s0p, u1, un);
(*s0size) = un;
if (swapped) (*s0size) = -(*s0size);
TMP_FREE;
return an;
}
/*
If at this point we have s*ap' + t*bp' = gp where gp is the gcd
and (ap', bp') = (u1*ap + ?*bp, -u0*ap + ?*bp)
then gp = s*u1*ap - t*u0*ap + ?*bp
and the cofactor we want is (s*u1-t*u0).
First there is the special case u0 = 0, u1 = 1 in which case we do not need
to compute t...
*/
ASSERT(u1 + un <= tp);
u0n = un;
MPN_NORMALIZE(u0, u0n); /* {u0, u0n} is now normalised */
if (u0n == 0) /* u1 = 1 case is rare*/
{
mp_size_t gn;
gn = mpn_ngcdext_lehmer (gp, s0p, s0size, ap, bp, n, tp);
if (swapped) (*s0size) = -(*s0size);
TMP_FREE;
return gn;
}
else
{
/* Compute final gcd. */
mp_size_t gn, sn, tn;
mp_ptr s, t;
mp_limb_t cy;
int negate = 0;
/* Save an, bn first as gcdext destroys inputs */
s = tp;
tp += an;
MPN_COPY(tp, ap, an);
MPN_COPY(tp + an, bp, an);
if (mpn_cmp(tp, tp + an, an) == 0)
{
/* gcd is tp or tp + an
return smallest cofactor, either -u0 or u1
*/
gn = an;
MPN_NORMALIZE(tp, gn);
MPN_COPY(gp, tp, gn);
MPN_CMP(c, u0, u1, un);
if (c < (mp_limb_signed_t) 0)
{
MPN_COPY(s0p, u0, u0n);
(*s0size) = -u0n;
} else
{
MPN_NORMALIZE(u1, un);
MPN_COPY(s0p, u1, un);
(*s0size) = un;
}
TMP_FREE;
return gn;
}
gn = mpn_ngcdext_lehmer (gp, s, &sn, tp, tp + an, an, tp + 2*an);
/* Special case, s == 0, t == 1, cofactor = -u0 case is rare*/
if (sn == 0)
{
MPN_COPY(s0p, u0, u0n);
(*s0size) = -u0n;
if (swapped) (*s0size) = -(*s0size);
TMP_FREE;
return gn;
}
/* We'll need the other cofactor t = (gp - s*ap)/bp
*/
t = tp;
tp += (an + 1);
gcdext_get_t(t, &tn, gp, gn, ap, an, bp, n, s, sn, tp);
ASSERT((tn == 0) || (t[tn - 1] > 0)); /* {t, tn} is normalised */
ASSERT(tn <= an + 1);
/* We want to compute s*u1 - t*u0, so if s is negative
t will be positive, so we'd be dealing with negative
numbers. We fix that here.
*/
if (sn < 0)
{
sn = -sn;
negate = 1;
}
/* Now we can deal with the special case u1 = 0 */
u1n = un;
MPN_NORMALIZE(u1, u1n); /* {u1, u1n} is now normalised */
if (u1n == 0) /* case is rare */
{
MPN_COPY(s0p, t, tn);
(*s0size) = -tn;
if (swapped ^ negate) (*s0size) = -(*s0size);
TMP_FREE;
return gn;
}
/* t may be zero, but we need to compute s*u1 anyway */
if (sn >= u1n)
mpn_mul(s0p, s, sn, u1, u1n);
else
mpn_mul(s0p, u1, u1n, s, sn);
(*s0size) = sn + u1n;
(*s0size) -= (s0p[sn + u1n - 1] == 0);
ASSERT(s0p[*s0size - 1] > 0); /* {s0p, *s0size} is normalised now */
if (tn == 0) /* case is rare */
{
if (swapped ^ negate) (*s0size) = -(*s0size);
TMP_FREE;
return gn;
}
/* Now compute the rest of the cofactor, t*u0
and subtract it
We're done with u1 and s which happen to be
consecutive, so use that space
*/
ASSERT(u1 + tn + u0n <= t);
if (tn > u0n)
mpn_mul(u1, t, tn, u0, u0n);
else
mpn_mul(u1, u0, u0n, t, tn);
u1n = tn + u0n;
u1n -= (u1[tn + u0n - 1] == 0);
ASSERT(u1[u1n - 1] > 0);
/* Recall t is now negated so s*u1 - t*u0
involves an *addition*
*/
if ((*s0size) >= u1n)
{
cy = mpn_add(s0p, s0p, *s0size, u1, u1n);
if (cy) s0p[(*s0size)++] = cy;
}
else
{
cy = mpn_add(s0p, u1, u1n, s0p, *s0size);
(*s0size) = u1n;
if (cy) s0p[(*s0size)++] = cy;
}
if (swapped ^ negate) (*s0size) = -(*s0size);
TMP_FREE;
return gn;
}
}
