void
arb_poly_mullow_classical(arb_poly_t res, const arb_poly_t poly1,
const arb_poly_t poly2,
long n, long prec)
{
long len_out;
if (poly1->length == 0 || poly2->length == 0 || n == 0)
{
arb_poly_zero(res);
return;
}
len_out = poly1->length + poly2->length - 1;
if (n > len_out)
n = len_out;
if (res == poly1 || res == poly2)
{
arb_poly_t t;
arb_poly_init2(t, n);
_arb_poly_mullow_classical(t->coeffs, poly1->coeffs, poly1->length,
poly2->coeffs, poly2->length, n, prec);
arb_poly_swap(res, t);
arb_poly_clear(t);
}
else
{
arb_poly_fit_length(res, n);
_arb_poly_mullow_classical(res->coeffs, poly1->coeffs, poly1->length,
poly2->coeffs, poly2->length, n, prec);
}
_arb_poly_set_length(res, n);
_arb_poly_normalise(res);
}
void
_arb_poly_mullow(arb_ptr res,
arb_srcptr poly1, long len1,
arb_srcptr poly2, long len2, long n, long prec)
{
if (n == 1)
{
arb_mul(res, poly1, poly2, prec);
}
else
{
if (n < BLOCK_CUTOFF || len1 < BLOCK_CUTOFF || len2 < BLOCK_CUTOFF)
_arb_poly_mullow_classical(res, poly1, len1, poly2, len2, n, prec);
else
_arb_poly_mullow_block(res, poly1, len1, poly2, len2, n, prec);
}
}
void
_arb_poly_mullow_block(arb_ptr z, arb_srcptr x, slong xlen,
arb_srcptr y, slong ylen, slong n, slong prec)
{
slong xmlen, xrlen, ymlen, yrlen, i;
fmpz *xz, *yz, *zz;
fmpz *xe, *ye;
slong *xblocks, *yblocks;
int squaring;
fmpz_t scale, t;
xlen = FLINT_MIN(xlen, n);
ylen = FLINT_MIN(ylen, n);
squaring = (x == y) && (xlen == ylen);
/* Strip trailing zeros */
xmlen = xrlen = xlen;
while (xmlen > 0 && arf_is_zero(arb_midref(x + xmlen - 1))) xmlen--;
while (xrlen > 0 && mag_is_zero(arb_radref(x + xrlen - 1))) xrlen--;
if (squaring)
{
ymlen = xmlen;
yrlen = xrlen;
}
else
{
ymlen = yrlen = ylen;
while (ymlen > 0 && arf_is_zero(arb_midref(y + ymlen - 1))) ymlen--;
while (yrlen > 0 && mag_is_zero(arb_radref(y + yrlen - 1))) yrlen--;
}
/* We don't know how to deal with infinities or NaNs */
if (!_arb_vec_is_finite(x, xlen) ||
(!squaring && !_arb_vec_is_finite(y, ylen)))
{
_arb_poly_mullow_classical(z, x, xlen, y, ylen, n, prec);
return;
}
xlen = FLINT_MAX(xmlen, xrlen);
ylen = FLINT_MAX(ymlen, yrlen);
/* Start with the zero polynomial */
_arb_vec_zero(z, n);
/* Nothing to do */
if (xlen == 0 || ylen == 0)
return;
n = FLINT_MIN(n, xlen + ylen - 1);
fmpz_init(scale);
fmpz_init(t);
xz = _fmpz_vec_init(xlen);
yz = _fmpz_vec_init(ylen);
zz = _fmpz_vec_init(n);
xe = _fmpz_vec_init(xlen);
ye = _fmpz_vec_init(ylen);
xblocks = flint_malloc(sizeof(slong) * (xlen + 1));
yblocks = flint_malloc(sizeof(slong) * (ylen + 1));
_arb_poly_get_scale(scale, x, xlen, y, ylen);
/* Error propagation */
/* (xm + xr)*(ym + yr) = (xm*ym) + (xr*ym + xm*yr + xr*yr)
= (xm*ym) + (xm*yr + xr*(ym + yr)) */
if (xrlen != 0 || yrlen != 0)
{
mag_ptr tmp;
double *xdbl, *ydbl;
tmp = _mag_vec_init(FLINT_MAX(xlen, ylen));
xdbl = flint_malloc(sizeof(double) * xlen);
ydbl = flint_malloc(sizeof(double) * ylen);
/* (xm + xr)^2 = (xm*ym) + (xr^2 + 2 xm xr)
= (xm*ym) + xr*(2 xm + xr) */
if (squaring)
{
_mag_vec_get_fmpz_2exp_blocks(xz, xdbl, xe, xblocks, scale, x, NULL, xrlen);
for (i = 0; i < xlen; i++)
{
arf_get_mag(tmp + i, arb_midref(x + i));
mag_mul_2exp_si(tmp + i, tmp + i, 1);
mag_add(tmp + i, tmp + i, arb_radref(x + i));
}
_mag_vec_get_fmpz_2exp_blocks(yz, ydbl, ye, yblocks, scale, NULL, tmp, xlen);
_arb_poly_addmullow_rad(z, zz, xz, xdbl, xe, xblocks, xrlen, yz, ydbl, ye, yblocks, xlen, n);
}
else if (yrlen == 0)
{
/* xr * |ym| */
_mag_vec_get_fmpz_2exp_blocks(xz, xdbl, xe, xblocks, scale, x, NULL, xrlen);
for (i = 0; i < ymlen; i++)
arf_get_mag(tmp + i, arb_midref(y + i));
_mag_vec_get_fmpz_2exp_blocks(yz, ydbl, ye, yblocks, scale, NULL, tmp, ymlen);
_arb_poly_addmullow_rad(z, zz, xz, xdbl, xe, xblocks, xrlen, yz, ydbl, ye, yblocks, ymlen, n);
}
else
{
/* |xm| * yr */
for (i = 0; i < xmlen; i++)
arf_get_mag(tmp + i, arb_midref(x + i));
_mag_vec_get_fmpz_2exp_blocks(xz, xdbl, xe, xblocks, scale, NULL, tmp, xmlen);
_mag_vec_get_fmpz_2exp_blocks(yz, ydbl, ye, yblocks, scale, y, NULL, yrlen);
_arb_poly_addmullow_rad(z, zz, xz, xdbl, xe, xblocks, xmlen, yz, ydbl, ye, yblocks, yrlen, n);
/* xr*(|ym| + yr) */
if (xrlen != 0)
{
_mag_vec_get_fmpz_2exp_blocks(xz, xdbl, xe, xblocks, scale, x, NULL, xrlen);
for (i = 0; i < ylen; i++)
arb_get_mag(tmp + i, y + i);
_mag_vec_get_fmpz_2exp_blocks(yz, ydbl, ye, yblocks, scale, NULL, tmp, ylen);
_arb_poly_addmullow_rad(z, zz, xz, xdbl, xe, xblocks, xrlen, yz, ydbl, ye, yblocks, ylen, n);
}
}
_mag_vec_clear(tmp, FLINT_MAX(xlen, ylen));
flint_free(xdbl);
flint_free(ydbl);
}
/* multiply midpoints */
if (xmlen != 0 && ymlen != 0)
{
_arb_vec_get_fmpz_2exp_blocks(xz, xe, xblocks, scale, x, xmlen, prec);
if (squaring)
{
_arb_poly_addmullow_block(z, zz, xz, xe, xblocks, xmlen, xz, xe, xblocks, xmlen, n, prec, 1);
}
else
{
_arb_vec_get_fmpz_2exp_blocks(yz, ye, yblocks, scale, y, ymlen, prec);
_arb_poly_addmullow_block(z, zz, xz, xe, xblocks, xmlen, yz, ye, yblocks, ymlen, n, prec, 0);
}
}
/* Unscale. */
if (!fmpz_is_zero(scale))
{
fmpz_zero(t);
for (i = 0; i < n; i++)
{
arb_mul_2exp_fmpz(z + i, z + i, t);
fmpz_add(t, t, scale);
}
}
_fmpz_vec_clear(xz, xlen);
_fmpz_vec_clear(yz, ylen);
_fmpz_vec_clear(zz, n);
_fmpz_vec_clear(xe, xlen);
_fmpz_vec_clear(ye, ylen);
flint_free(xblocks);
flint_free(yblocks);
fmpz_clear(scale);
fmpz_clear(t);
}
