int
arf_sub_special(arf_t z, const arf_t x, const arf_t y, slong prec, arf_rnd_t rnd)
{
if (arf_is_zero(x))
{
if (arf_is_zero(y))
{
arf_zero(z);
return 0;
}
else
return arf_neg_round(z, y, prec, rnd);
}
else if (arf_is_zero(y))
{
return arf_set_round(z, x, prec, rnd);
}
else if (arf_is_nan(x) || arf_is_nan(y)
|| (arf_is_pos_inf(x) && arf_is_pos_inf(y))
|| (arf_is_neg_inf(x) && arf_is_neg_inf(y)))
{
arf_nan(z);
return 0;
}
else if (arf_is_special(x))
{
arf_set(z, x);
return 0;
}
else
{
arf_neg(z, y);
return 0;
}
}
void
arf_fprint(FILE * file, const arf_t x)
{
if (arf_is_normal(x))
{
fmpz_t man, exp;
fmpz_init(man);
fmpz_init(exp);
arf_get_fmpz_2exp(man, exp, x);
flint_fprintf(file, "(");
fmpz_fprint(file, man);
flint_fprintf(file, " * 2^");
fmpz_fprint(file, exp);
flint_fprintf(file, ")");
fmpz_clear(man);
fmpz_clear(exp);
}
else
{
if (arf_is_zero(x)) flint_fprintf(file, "(0)");
else if (arf_is_pos_inf(x)) flint_fprintf(file, "(+inf)");
else if (arf_is_neg_inf(x)) flint_fprintf(file, "(-inf)");
else flint_fprintf(file, "(nan)");
}
}
void
arf_get_fmpr(fmpr_t y, const arf_t x)
{
if (arf_is_special(x))
{
if (arf_is_zero(x))
fmpr_zero(y);
else if (arf_is_pos_inf(x))
fmpr_pos_inf(y);
else if (arf_is_neg_inf(x))
fmpr_neg_inf(y);
else
fmpr_nan(y);
}
else
{
arf_get_fmpz_2exp(fmpr_manref(y), fmpr_expref(y), x);
}
}
int
arf_cmp_2exp_si(const arf_t x, long e)
{
if (arf_is_special(x))
{
if (arf_is_zero(x)) return -1;
if (arf_is_pos_inf(x)) return 1;
if (arf_is_neg_inf(x)) return -1;
return 0;
}
if (ARF_SGNBIT(x))
return -1;
/* Fast path. */
if (!COEFF_IS_MPZ(ARF_EXP(x)))
{
if (ARF_IS_POW2(x) && (ARF_EXP(x) - 1 == e))
return 0;
else
return (ARF_EXP(x) <= e) ? -1 : 1;
}
if (ARF_IS_POW2(x))
{
fmpz_t t;
fmpz_init(t);
fmpz_one(t);
fmpz_add_si(t, t, e);
if (fmpz_equal(ARF_EXPREF(x), t))
{
fmpz_clear(t);
return 0;
}
fmpz_clear(t);
}
return (fmpz_cmp_si(ARF_EXPREF(x), e) <= 0) ? -1 : 1;
}
int
arf_root(arf_ptr z, arf_srcptr x, ulong k, slong prec, arf_rnd_t rnd)
{
mp_size_t xn, zn, val;
mp_srcptr xptr;
mp_ptr tmp, zptr;
mpfr_t xf, zf;
fmpz_t q, r;
int inexact;
if (k == 0)
{
arf_nan(z);
return 0;
}
if (k == 1)
return arf_set_round(z, x, prec, rnd);
if (k == 2)
return arf_sqrt(z, x, prec, rnd);
if (arf_is_special(x))
{
if (arf_is_neg_inf(x))
arf_nan(z);
else
arf_set(z, x);
return 0;
}
if (ARF_SGNBIT(x))
{
arf_nan(z);
return 0;
}
fmpz_init(q);
fmpz_init(r);
/* x = m * 2^e where e = qk + r */
/* x^(1/k) = (m * 2^(qk+r))^(1/k) */
/* x^(1/k) = (m * 2^r)^(1/k) * 2^q */
fmpz_set_ui(r, k);
fmpz_fdiv_qr(q, r, ARF_EXPREF(x), r);
ARF_GET_MPN_READONLY(xptr, xn, x);
zn = (prec + FLINT_BITS - 1) / FLINT_BITS;
zf->_mpfr_d = tmp = flint_malloc(zn * sizeof(mp_limb_t));
zf->_mpfr_prec = prec;
zf->_mpfr_sign = 1;
zf->_mpfr_exp = 0;
xf->_mpfr_d = (mp_ptr) xptr;
xf->_mpfr_prec = xn * FLINT_BITS;
xf->_mpfr_sign = 1;
xf->_mpfr_exp = fmpz_get_ui(r);
inexact = mpfr_root(zf, xf, k, arf_rnd_to_mpfr(rnd));
inexact = (inexact != 0);
val = 0;
while (tmp[val] == 0)
val++;
ARF_GET_MPN_WRITE(zptr, zn - val, z);
flint_mpn_copyi(zptr, tmp + val, zn - val);
fmpz_add_si(ARF_EXPREF(z), q, zf->_mpfr_exp);
flint_free(tmp);
fmpz_clear(q);
fmpz_clear(r);
return inexact;
}
void
arb_atan_arf(arb_t z, const arf_t x, slong prec)
{
if (arf_is_special(x))
{
if (arf_is_zero(x))
{
arb_zero(z);
}
else if (arf_is_pos_inf(x))
{
arb_const_pi(z, prec);
arb_mul_2exp_si(z, z, -1);
}
else if (arf_is_neg_inf(x))
{
arb_const_pi(z, prec);
arb_mul_2exp_si(z, z, -1);
arb_neg(z, z);
}
else
{
arb_indeterminate(z);
}
}
else if (COEFF_IS_MPZ(*ARF_EXPREF(x)))
{
if (fmpz_sgn(ARF_EXPREF(x)) < 0)
arb_atan_eps(z, x, prec);
else
arb_atan_inf_eps(z, x, prec);
}
else
{
slong exp, wp, wn, N, r;
mp_srcptr xp;
mp_size_t xn, tn;
mp_ptr tmp, w, t, u;
mp_limb_t p1, q1bits, p2, q2bits, error, error2;
int negative, inexact, reciprocal;
TMP_INIT;
exp = ARF_EXP(x);
negative = ARF_SGNBIT(x);
if (exp < -(prec/2) - 2 || exp > prec + 2)
{
if (exp < 0)
arb_atan_eps(z, x, prec);
else
arb_atan_inf_eps(z, x, prec);
return;
}
ARF_GET_MPN_READONLY(xp, xn, x);
/* Special case: +/- 1 (we require |x| != 1 later on) */
if (exp == 1 && xn == 1 && xp[xn-1] == LIMB_TOP)
{
arb_const_pi(z, prec);
arb_mul_2exp_si(z, z, -2);
if (negative)
arb_neg(z, z);
return;
}
/* Absolute working precision (NOT rounded to a limb multiple) */
wp = prec - FLINT_MIN(0, exp) + 4;
/* Too high precision to use table */
if (wp > ARB_ATAN_TAB2_PREC)
{
arb_atan_arf_bb(z, x, prec);
return;
}
/* Working precision in limbs */
wn = (wp + FLINT_BITS - 1) / FLINT_BITS;
TMP_START;
tmp = TMP_ALLOC_LIMBS(4 * wn + 3);
w = tmp; /* requires wn+1 limbs */
t = w + wn + 1; /* requires wn+1 limbs */
u = t + wn + 1; /* requires 2wn+1 limbs */
/* ----------------------------------------------------------------- */
/* Convert x or 1/x to a fixed-point number |w| < 1 */
/* ----------------------------------------------------------------- */
if (exp <= 0) /* |x| < 1 */
{
reciprocal = 0;
/* todo: just zero top */
flint_mpn_zero(w, wn);
/* w = x as a fixed-point number */
error = _arf_get_integer_mpn(w, xp, xn, exp + wn * FLINT_BITS);
}
else /* |x| > 1 */
{
slong one_exp, one_limbs, one_bits;
mp_ptr one;
reciprocal = 1;
one_exp = xn * FLINT_BITS + wn * FLINT_BITS - exp;
flint_mpn_zero(w, wn);
/* 1/x becomes zero */
if (one_exp >= FLINT_BITS - 1)
{
/* w = 1/x */
one_limbs = one_exp / FLINT_BITS;
one_bits = one_exp % FLINT_BITS;
if (one_limbs + 1 >= xn)
{
one = TMP_ALLOC_LIMBS(one_limbs + 1);
flint_mpn_zero(one, one_limbs);
one[one_limbs] = UWORD(1) << one_bits;
/* todo: only zero necessary part */
flint_mpn_zero(w, wn);
mpn_tdiv_q(w, one, one_limbs + 1, xp, xn);
/* Now w must be < 1 since x > 1 and we rounded down; thus
w[wn] must be zero */
}
}
/* todo: moderate powers of two would be exact... */
error = 1;
}
/* ----------------------------------------------------------------- */
/* Table-based argument reduction */
/* ----------------------------------------------------------------- */
/* choose p such that p/q <= x < (p+1)/q */
if (wp <= ARB_ATAN_TAB1_PREC)
q1bits = ARB_ATAN_TAB1_BITS;
else
q1bits = ARB_ATAN_TAB21_BITS;
p1 = w[wn-1] >> (FLINT_BITS - q1bits);
/* atan(w) = atan(p/q) + atan(w2) */
/* where w2 = (q*w-p)/(q+p*w) */
if (p1 != 0)
{
t[wn] = (UWORD(1) << q1bits) + mpn_mul_1(t, w, wn, p1);
flint_mpn_zero(u, wn);
u[2 * wn] = mpn_lshift(u + wn, w, wn, q1bits) - p1;
mpn_tdiv_q(w, u, 2 * wn + 1, t, wn + 1);
error++; /* w2 is computed with 1 ulp error */
}
/* Do a second round of argument reduction */
if (wp <= ARB_ATAN_TAB1_PREC)
{
p2 = 0;
}
else
{
q2bits = ARB_ATAN_TAB21_BITS + ARB_ATAN_TAB22_BITS;
p2 = w[wn-1] >> (FLINT_BITS - q2bits);
if (p2 != 0)
{
t[wn] = (UWORD(1) << q2bits) + mpn_mul_1(t, w, wn, p2);
flint_mpn_zero(u, wn);
u[2 * wn] = mpn_lshift(u + wn, w, wn, q2bits) - p2;
mpn_tdiv_q(w, u, 2 * wn + 1, t, wn + 1);
error++;
}
}
/* |w| <= 2^-r */
r = _arb_mpn_leading_zeros(w, wn);
/* N >= (wp-r)/(2r) */
N = (wp - r + (2*r-1)) / (2*r);
/* Evaluate Taylor series */
_arb_atan_taylor_rs(t, &error2, w, wn, N, 1);
/* Taylor series evaluation error */
error += error2;
/* Size of output number */
tn = wn;
/* First table lookup */
if (p1 != 0)
{
if (wp <= ARB_ATAN_TAB1_PREC)
mpn_add_n(t, t, arb_atan_tab1[p1] + ARB_ATAN_TAB1_LIMBS - tn, tn);
else
mpn_add_n(t, t, arb_atan_tab21[p1] + ARB_ATAN_TAB2_LIMBS - tn, tn);
error++;
}
/* Second table lookup */
if (p2 != 0)
{
mpn_add_n(t, t, arb_atan_tab22[p2] + ARB_ATAN_TAB2_LIMBS - tn, tn);
error++;
}
/* pi/2 - atan(1/x) */
if (reciprocal)
{
t[tn] = LIMB_ONE - mpn_sub_n(t,
arb_atan_pi2_minus_one + ARB_ATAN_TAB2_LIMBS - tn, t, tn);
/* result can be >= 1 */
tn += (t[tn] != 0);
/* error of pi/2 */
error++;
}
/* The accumulated arithmetic error */
mag_set_ui_2exp_si(arb_radref(z), error, -wn * FLINT_BITS);
/* Truncation error from the Taylor series */
mag_add_ui_2exp_si(arb_radref(z), arb_radref(z), 1, -r*(2*N+1));
/* Set the midpoint */
inexact = _arf_set_mpn_fixed(arb_midref(z), t, tn, wn, negative, prec, ARB_RND);
if (inexact)
arf_mag_add_ulp(arb_radref(z), arb_radref(z), arb_midref(z), prec);
TMP_END;
}
}
int
arf_sum(arf_t s, arf_srcptr terms, long len, long prec, arf_rnd_t rnd)
{
arf_ptr blocks;
long i, j, used;
int have_merged, res;
/* first check if the result is inf or nan */
{
int have_pos_inf = 0;
int have_neg_inf = 0;
for (i = 0; i < len; i++)
{
if (arf_is_pos_inf(terms + i))
{
if (have_neg_inf)
{
arf_nan(s);
return 0;
}
have_pos_inf = 1;
}
else if (arf_is_neg_inf(terms + i))
{
if (have_pos_inf)
{
arf_nan(s);
return 0;
}
have_neg_inf = 1;
}
else if (arf_is_nan(terms + i))
{
arf_nan(s);
return 0;
}
}
if (have_pos_inf)
{
arf_pos_inf(s);
return 0;
}
if (have_neg_inf)
{
arf_neg_inf(s);
return 0;
}
}
blocks = flint_malloc(sizeof(arf_struct) * len);
for (i = 0; i < len; i++)
arf_init(blocks + i);
/* put all terms into blocks */
used = 0;
for (i = 0; i < len; i++)
{
if (!arf_is_zero(terms + i))
{
arf_set(blocks + used, terms + i);
used++;
}
}
/* merge blocks until all are well separated */
have_merged = 1;
while (used >= 2 && have_merged)
{
have_merged = 0;
for (i = 0; i < used && !have_merged; i++)
{
for (j = i + 1; j < used && !have_merged; j++)
{
if (_arf_are_close(blocks + i, blocks + j, prec))
{
arf_add(blocks + i, blocks + i, blocks + j,
ARF_PREC_EXACT, ARF_RND_DOWN);
/* remove the merged block */
arf_swap(blocks + j, blocks + used - 1);
used--;
/* remove the updated block if the sum is zero */
if (arf_is_zero(blocks + i))
{
arf_swap(blocks + i, blocks + used - 1);
used--;
}
have_merged = 1;
}
}
}
}
if (used == 0)
{
arf_zero(s);
res = 0;
}
else if (used == 1)
{
res = arf_set_round(s, blocks + 0, prec, rnd);
}
else
{
/* find the two largest blocks */
for (i = 1; i < used; i++)
if (arf_cmpabs(blocks + 0, blocks + i) < 0)
arf_swap(blocks + 0, blocks + i);
for (i = 2; i < used; i++)
if (arf_cmpabs(blocks + 1, blocks + i) < 0)
arf_swap(blocks + 1, blocks + i);
res = _arf_add_eps(s, blocks + 0, arf_sgn(blocks + 1), prec, rnd);
}
for (i = 0; i < len; i++)
arf_clear(blocks + i);
flint_free(blocks);
return res;
}
