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
mpn_sqr (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);
}
else if (BELOW_THRESHOLD (n, SQR_TOOM4_THRESHOLD))
{
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 (BELOW_THRESHOLD (n, SQR_TOOM8_THRESHOLD))
{
mpn_toom4_sqr_n (p, a, n);
}
#if WANT_FFT || TUNE_PROGRAM_BUILD
else if (BELOW_THRESHOLD (n, SQR_FFT_FULL_THRESHOLD))
#else
else
#endif
{
mpn_toom8_sqr_n (p, a, n);
}
#if WANT_FFT || TUNE_PROGRAM_BUILD
else
{
// (rp,2n)=(xp,n)^2 with temp space (tp,2*n+C)
void mpn_kara_sqr_n(mp_ptr rp,mp_srcptr xp,mp_size_t n,mp_ptr tp)
{mp_size_t n2,n3;mp_srcptr xl,xh;mp_ptr dx;mp_limb_t c;
n2=n>>1;
xl=xp;xh=xp+n2;n3=n-n2;
dx=rp+2*n2;
if((n&1)==0)
{if(mpn_cmp(xh,xl,n2)>=0){mpn_sub_n(dx,xh,xl,n2);}else{mpn_sub_n(dx,xl,xh,n2);}}
else
{if(xh[n2]!=0 || mpn_cmp(xh,xl,n2)>=0){c=mpn_sub_n(dx,xh,xl,n2);dx[n2]=xh[n2]-c;}else{mpn_sub_n(dx,xl,xh,n2);dx[n2]=0;}}
if(BELOW_THRESHOLD(n3,MUL_KARATSUBA_THRESHOLD))
{mpn_mul_basecase(rp,xl,n2,xl,n2);
mpn_mul_basecase(tp,dx,n3,dx,n3);
mpn_mul_basecase(rp+2*n2,xh,n3,xh,n3);}
else if(BELOW_THRESHOLD(n3,SQR_KARATSUBA_THRESHOLD))
{mpn_sqr_basecase(rp,xl,n2);
mpn_sqr_basecase(tp,dx,n3);
mpn_sqr_basecase(rp+2*n2,xh,n3);}
else
{mpn_kara_sqr_n(rp,xl,n2,tp+2*n3);
mpn_kara_sqr_n(tp,dx,n3,tp+2*n3);
mpn_kara_sqr_n(rp+2*n2,xh,n3,tp+2*n3);}
mpn_karasub(rp,tp,n);
return;}
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
{
void
mpn_kara_sqr_n (mp_ptr p, mp_srcptr a, mp_size_t n, mp_ptr ws)
{
mp_limb_t w, w0, w1;
mp_size_t n2;
mp_srcptr x, y;
mp_size_t i;
n2 = n >> 1;
ASSERT (n2 > 0);
if ((n & 1) != 0)
{
/* Odd length. */
mp_size_t n1, n3, nm1;
n3 = n - n2;
w = a[n2];
if (w != 0)
w -= mpn_sub_n (p, a, a + n3, n2);
else
{
i = n2;
do
{
--i;
w0 = a[i];
w1 = a[n3 + i];
}
while (w0 == w1 && i != 0);
if (w0 < w1)
{
x = a + n3;
y = a;
}
else
{
x = a;
y = a + n3;
}
mpn_sub_n (p, x, y, n2);
}
p[n2] = w;
n1 = n + 1;
/* n2 is always either n3 or n3-1 so maybe the two sets of tests here
could be combined. But that's not important, since the tests will
take a miniscule amount of time compared to the function calls. */
if (BELOW_THRESHOLD (n3, SQR_BASECASE_THRESHOLD))
{
mpn_mul_basecase (ws, p, n3, p, n3);
mpn_mul_basecase (p, a, n3, a, n3);
}
else if (BELOW_THRESHOLD (n3, SQR_KARATSUBA_THRESHOLD))
{
mpn_sqr_basecase (ws, p, n3);
mpn_sqr_basecase (p, a, n3);
}
else
{
mpn_kara_sqr_n (ws, p, n3, ws + n1); /* (x-y)^2 */
mpn_kara_sqr_n (p, a, n3, ws + n1); /* x^2 */
}
if (BELOW_THRESHOLD (n2, SQR_BASECASE_THRESHOLD))
mpn_mul_basecase (p + n1, a + n3, n2, a + n3, n2);
else if (BELOW_THRESHOLD (n2, SQR_KARATSUBA_THRESHOLD))
mpn_sqr_basecase (p + n1, a + n3, n2);
else
mpn_kara_sqr_n (p + n1, a + n3, n2, ws + n1); /* y^2 */
/* Since x^2+y^2-(x-y)^2 = 2xy >= 0 there's no need to track the
borrow from mpn_sub_n. If it occurs then it'll be cancelled by a
carry from ws[n]. Further, since 2xy fits in n1 limbs there won't
be any carry out of ws[n] other than cancelling that borrow. */
mpn_sub_n (ws, p, ws, n1); /* x^2-(x-y)^2 */
nm1 = n - 1;
if (mpn_add_n (ws, p + n1, ws, nm1)) /* x^2+y^2-(x-y)^2 = 2xy */
{
mp_limb_t x = (ws[nm1] + 1) & GMP_NUMB_MASK;
ws[nm1] = x;
if (x == 0)
ws[n] = (ws[n] + 1) & GMP_NUMB_MASK;
}
if (mpn_add_n (p + n3, p + n3, ws, n1))
{
mpn_incr_u (p + n1 + n3, 1);
}
}
else
{
/* Even length. */
i = n2;
do
{
--i;
w0 = a[i];
w1 = a[n2 + i];
}
while (w0 == w1 && i != 0);
if (w0 < w1)
{
x = a + n2;
y = a;
}
else
{
x = a;
y = a + n2;
}
mpn_sub_n (p, x, y, n2);
/* Pointwise products. */
if (BELOW_THRESHOLD (n2, SQR_BASECASE_THRESHOLD))
{
mpn_mul_basecase (ws, p, n2, p, n2);
mpn_mul_basecase (p, a, n2, a, n2);
mpn_mul_basecase (p + n, a + n2, n2, a + n2, n2);
}
else if (BELOW_THRESHOLD (n2, SQR_KARATSUBA_THRESHOLD))
{
mpn_sqr_basecase (ws, p, n2);
mpn_sqr_basecase (p, a, n2);
mpn_sqr_basecase (p + n, a + n2, n2);
}
else
{
mpn_kara_sqr_n (ws, p, n2, ws + n);
mpn_kara_sqr_n (p, a, n2, ws + n);
mpn_kara_sqr_n (p + n, a + n2, n2, ws + n);
}
/* Interpolate. */
w = -mpn_sub_n (ws, p, ws, n);
w += mpn_add_n (ws, p + n, ws, n);
w += mpn_add_n (p + n2, p + n2, ws, n);
MPN_INCR_U (p + n2 + n, 2 * n - (n2 + n), w);
}
}
