static inline A0_n asin(const A0_n a0_n)
{
const A0 a0 = { a0_n };
A0 sign, x;
x = nt2::abs(a0);
sign = bitofsign(a0);
const bA0 x_smaller_1e_4 = lt(x, single_constant<A0, 0x38d1b717>()); //1.0e-4f;
const bA0 x_larger_05 = gt(x, Half<A0>());
const bA0 x_else = logical_or(x_smaller_1e_4, x_larger_05);
A0 a = if_else_zero(x_smaller_1e_4, x);
const A0 b = if_else_zero(x_larger_05, Half<A0>()*oneminus(x));
A0 z = b_or(b_or(if_zero_else(x_else, sqr(x)), a), b);
x = if_zero_else(x_else, x);
a = if_else_zero(x_larger_05, sqrt(z));
x = b_or(a, x);
A0 z1 = madd(z, single_constant<A0, 0x3d2cb352>(), single_constant<A0, 0x3cc617e3>());
z1 = madd(z1, z, single_constant<A0, 0x3d3a3ec7>());
z1 = madd(z1, z, single_constant<A0, 0x3d9980f6>());
z1 = madd(z1, z, single_constant<A0, 0x3e2aaae4>());
z1 = madd(z1, z*x, x);
z = select(x_smaller_1e_4, z, z1);
z1 = z+z;
z1 = Pio_2<A0>()-z1;
z = select(x_larger_05, z1, z);
return b_xor(z, sign);
}
static inline A0 sincosa(const A0& a0, A0& c, const regular&)
{
A0 s;
if (nt2::is_invalid(a0)) { c = nt2::Nan<A0>(); return c; }
const A0 x = nt2::abs(a0);
static const sint_type de = static_cast<sint_type>(sizeof(sint_type)*8-1);
A0 xr;
const int_type n = redu_t::reduce(x, xr);
const int_type swap_bit = n&One<int_type>();
const A0 z = nt2::sqr(xr);
const int_type cos_sign_bit = shli(nt2::bitwise_xor(swap_bit, (n&nt2::Two<int_type>())>>1), de);
const A0 sin_sign_bit = nt2::bitwise_xor(bitofsign(a0), nt2::shli(n&Two<int_type>(), de-1));
if (nt2::is_nez(swap_bit))
{
c = eval_t::sin_eval(z, xr);
s = eval_t::cos_eval(z);
}
else
{
c = eval_t::cos_eval(z);
s = eval_t::sin_eval(z, xr);
}
c = nt2::bitwise_xor(c,cos_sign_bit);
return nt2::bitwise_xor(s,sin_sign_bit);
}
static inline A0 asin(const A0& a0)
{
A0 sign, x;
// bf::tie(sign, x) = sign_and_abs(a0);
x = abs(a0);
sign = bitofsign(a0);
A0 x_smaller_1e_4 = lt(x, single_constant<A0, 0x38d1b717>()); //1.0e-4f;
A0 x_larger_05 = gt(x, Half<A0>());
A0 x_else = b_or(x_smaller_1e_4, x_larger_05);
A0 a = b_and(x, x_smaller_1e_4);
A0 b = b_and(Half<A0>()*oneminus(x), x_larger_05);
A0 z = b_or(b_or(b_notand(x_else, sqr(x)), a), b);
x = b_notand(x_else, x);
a = b_and(sqrt(z), x_larger_05);
x = b_or(a, x);
A0 z1 = madd(z, single_constant<A0, 0x3d2cb352>(), single_constant<A0, 0x3cc617e3>());
z1 = madd(z1, z, single_constant<A0, 0x3d3a3ec7>());
z1 = madd(z1, z, single_constant<A0, 0x3d9980f6>());
z1 = madd(z1, z, single_constant<A0, 0x3e2aaae4>());
z1 = madd(z1, z*x, x);
z = select(x_smaller_1e_4, z, z1);
z1 = z+z;
z1 = Pio_2<A0>()-z1;
z = select(x_larger_05, z1, z);
return b_xor(z, sign);
}
static inline A0 sina(const A0& a0)
{
static const sint_type de = sizeof(sint_type)*8-1;
if (is_invalid(a0)) return Nan<A0>();
A0 x = nt2::abs(a0);
if (redu_t::replacement_needed(x))
{
return redu_t::sin_replacement(a0);
}
else
{
A0 xr = Nan<A0>(), xc;
int_type n = redu_t::reduce(x, xr, xc);
int_type swap_bit = n&One<int_type>();
A0 sign_bit = b_xor(bitofsign(a0), shli(n&Two<int_type>(), de-1));
A0 z = sqr(xr);
if (is_eqz(swap_bit))
{
z = eval_t::sin_eval(z, xr, xc);
}
else
{
z = eval_t::cos_eval(z, xr, xc);
}
return b_xor(z,sign_bit);
}
}
static inline A0 asin(const A0& a0)
{
A0 sign, x, z;
// bf::tie(sign, x) = sign_and_abs(a0);
x = nt2::abs(a0);
sign = bitofsign(a0);
if ((x < single_constant<A0,0x38d1b717>())) return a0;
if ((x > One<A0>())) return Nan<A0>();
bool bx_larger_05 = (x > Half<A0>());
if (bx_larger_05)
{
z = Half<A0>()*oneminus(x);
x = sqrt(z);
}
else
{
z = sqr(x);
}
A0 z1 = madd(z, single_constant<A0,0x3d2cb352>(),
single_constant<A0,0x3cc617e3>());
z1 = madd(z1, z, single_constant<A0,0x3d3a3ec7>());
z1 = madd(z1, z, single_constant<A0,0x3d9980f6>());
z1 = madd(z1, z, single_constant<A0,0x3e2aaae4>());
z1 = madd(z1, z*x, x);
if(bx_larger_05)
{
z1 = z1+z1;
z1 = Pio_2<A0>()-z1;
}
return b_xor(z1, sign);
}
static inline void sincosa(const A0& a0, A0& s, A0& c)
{
if (is_invalid(a0)) { s = c = Nan<A0>(); return; }
A0 x = nt2::abs(a0);
if (redu_t::replacement_needed(x))
{
redu_t::sincos_replacement(a0, s, c);
}
else
{
static const sint_type de = sizeof(sint_type)*8-1;
A0 xr, xc;
int_type n = redu_t::reduce(x, xr, xc);
int_type swap_bit = n&One<int_type>();
A0 z = sqr(xr);
int_type cos_sign_bit = shli(b_xor(swap_bit, (n&Two<int_type>())>>1), de);
A0 sin_sign_bit = b_xor(bitofsign(a0), shli(n&Two<int_type>(), de-1));
if (is_nez(swap_bit))
{
c = eval_t::sin_eval(z, xr, xc);
s = eval_t::cos_eval(z, xr, xc);
}
else
{
c = eval_t::cos_eval(z, xr, xc);
s = eval_t::sin_eval(z, xr, xc);
}
c = b_xor(c,cos_sign_bit);
s = b_xor(s,sin_sign_bit);
}
}
static inline A0_n asin(const A0_n a0_n)
{
const A0 a0 = { a0_n };
typedef typename meta::scalar_of<A0>::type sA0;
A0 x = nt2::abs(a0);
const A0 pio4 = Pio_4<A0>();
const bA0 small= lt(x, Sqrteps<A0>());
const A0 morebits = double_constant<A0, 0xbc91a62633145c07ll>();
const A0 ct1 = double_constant<A0, 0x3fe4000000000000ll>();
A0 zz1 = oneminus(x);
const A0 vp = zz1*horner< NT2_HORNER_COEFF_T(sA0, 5,
(0x3f684fc3988e9f08ll,
0xbfe2079259f9290fll,
0x401bdff5baf33e6all,
0xc03991aaac01ab68ll,
0x403c896240f3081dll)
)>(zz1)/
horner< NT2_HORNER_COEFF_T(sA0, 5,
(0x3ff0000000000000ll,
0xc035f2a2b6bf5d8cll,
0x40626219af6a7f42ll,
0xc077fe08959063eell,
0x40756709b0b644bell)
)>(zz1);
zz1 = sqrt(zz1+zz1);
A0 z = pio4-zz1;
zz1 = fma(zz1, vp, morebits);
z = z-zz1;
zz1 = z+pio4;
A0 zz2 = sqr(a0);
z = zz2*horner< NT2_HORNER_COEFF_T(sA0, 6,
(0x3f716b9b0bd48ad3ll,
0xbfe34341333e5c16ll,
0x4015c74b178a2dd9ll,
0xc0304331de27907bll,
0x40339007da779259ll,
0xc020656c06ceafd5ll)
)>(zz2)/
horner< NT2_HORNER_COEFF_T(sA0, 6,
(0x3ff0000000000000ll,
0xc02d7b590b5e0eabll,
0x40519fc025fe9054ll,
0xc06265bb6d3576d7ll,
0x4061705684ffbf9dll,
0xc04898220a3607acll)
)>(zz2);
zz2 = x*z+x;
return if_nan_else( gt(x, One<A0>())
, b_xor ( select( small
, x
, select( gt(x, ct1)
, zz1
, zz2
)
)
, bitofsign(a0)
)
);
}
static inline A0 cota(const A0& a0, const regular&)
{
if (nt2::is_invalid(a0)||redu_t::cot_invalid(a0)) return Nan<A0>();
const A0 x = nt2::abs(a0);
const A0 bos = bitofsign(a0);
if (!a0) return b_or(Inf<A0>(), bos);
A0 xr = Nan<A0>(), xc;
const int_type n = redu_t::reduce(x, xr, xc);
const A0 y = eval_t::cot_eval(xr, xc, 1-((n&1)<<1));
return b_xor(y, bos);
}
static inline A0 tana(const A0& a0, const regular&)
{
if (is_invalid(a0)||redu_t::tan_invalid(a0)) return Nan<A0>();
if (is_eqz(a0)) return a0;
const A0 x = nt2::abs(a0);
A0 xr = Nan<A0>(), xc, y;
const int_type n = redu_t::reduce(x, xr, xc);
y = eval_t::tan_eval(xr, xc, 1-((n&1)<<1));
// 1 -- n even
//-1 -- n odd
return b_xor(y, bitofsign(a0));
}
static inline A0 asin(const A0& a0)
{
A0 x = nt2::abs(a0);
if ((x > One<A0>())) return Nan<A0>();
if ((x < Sqrteps<A0>())) return a0;
A0 zz;
if((x > double_constant<double,0x3fe4000000000000ll> ())) //0.625;
{
zz = oneminus(x);
const A0 vp = zz*horner< NT2_HORNER_COEFF_T(stype, 5,
(0x3f684fc3988e9f08ll,
0xbfe2079259f9290fll,
0x401bdff5baf33e6all,
0xc03991aaac01ab68ll,
0x403c896240f3081dll)
)>(zz)/
horner< NT2_HORNER_COEFF_T(stype, 5,
(0x3ff0000000000000ll,
0xc035f2a2b6bf5d8cll,
0x40626219af6a7f42ll,
0xc077fe08959063eell,
0x40756709b0b644bell)
)>(zz);
zz = sqrt(zz+zz);
A0 z = Pio_4<A0>()-zz;
zz = madd(zz, vp, double_constant<double,0xbc91a62633145c07ll>());
z = z-zz;
zz = z+Pio_4<A0>();
}
else
{
zz = sqr(x);
A0 z = zz*horner< NT2_HORNER_COEFF_T(stype, 6,
(0x3f716b9b0bd48ad3ll,
0xbfe34341333e5c16ll,
0x4015c74b178a2dd9ll,
0xc0304331de27907bll,
0x40339007da779259ll,
0xc020656c06ceafd5ll)
)>(zz)/
horner< NT2_HORNER_COEFF_T(stype, 6,
(0x3ff0000000000000ll,
0xc02d7b590b5e0eabll,
0x40519fc025fe9054ll,
0xc06265bb6d3576d7ll,
0x4061705684ffbf9dll,
0xc04898220a3607acll)
)>(zz);
zz = x*z+x;
}
return b_xor(bitofsign(a0), zz);
}
static inline A0 cota(const A0& a0)
{
if (nt2::is_invalid(a0)||redu_t::cot_invalid(a0)) return Nan<A0>();
A0 x = nt2::abs(a0);
if (redu_t::replacement_needed(x))
{
return redu_t::cot_replacement(a0);
}
else
{
const A0 bos = bitofsign(a0);
if (!a0) return b_or(Inf<A0>(), bos);
A0 xr = Nan<A0>(), xc, y;
int_type n = redu_t::reduce(x, xr, xc);
y = eval_t::cot_eval(xr, xc, 1-((n&1)<<1));
return b_xor(y, bos);
}
}
static inline A0 atan(const A0& a0)
{
A0 x, sign;
x = nt2::abs(a0);
sign = bitofsign(a0);
// bf::tie(sign, x) = sign_and_abs(a0);
const A0 flag1 = lt(x, single_constant<A0, 0x401a827a>()); //tan3pio8);
const A0 flag2 = b_and(ge(x, single_constant<A0, 0x3ed413cd>()), flag1);
A0 yy = b_notand(flag1, Pio_2<A0>());
yy = select(flag2, Pio_4<A0>(), yy);
A0 xx = select(flag1, x, -rec(x));
xx = select(flag2, (minusone(x)/oneplus(x)),xx);
const A0 z = sqr(xx);
A0 z1 = madd(z, single_constant<A0, 0x3da4f0d1>(),single_constant<A0, 0xbe0e1b85>());
A0 z2 = madd(z, single_constant<A0, 0x3e4c925f>(),single_constant<A0, 0xbeaaaa2a>());
z1 = madd(z1, sqr(z), z2);
yy = add(yy, madd(xx, mul( z1, z), xx));
return b_xor(yy, sign);
}
static inline A0 tana(const A0& a0)
{
if (is_invalid(a0)||redu_t::tan_invalid(a0)) return Nan<A0>();
if (is_eqz(a0)) return a0;
A0 x = nt2::abs(a0);
if (redu_t::replacement_needed(x))
{
return redu_t::tan_replacement(a0);
}
else
{
A0 xr = Nan<A0>(), xc, y;
int_type n = redu_t::reduce(x, xr, xc);
y = eval_t::tan_eval(xr, xc, 1-((n&1)<<1));
// 1 -- n even
//-1 -- n odd
return b_xor(y, bitofsign(a0));
}
}
static inline A0 atan(const A0& a0)
{
if (is_eqz(a0)) return a0;
if (is_inf(a0)) return Pio_2<A0>()*sign(a0);
A0 y;
A0 x = nt2::abs(a0);
A0 sgn = bitofsign(a0);
if( x >single_constant<float,0x401a827a>())//2.414213562373095 ) /* tan 3pi/8 */
{
y = Pio_2<A0>();
x = -rec(x);
}
else if( x > single_constant<float,0x3ed413cd>()) //0.4142135623730950f ) /* tan pi/8 */
{
y = Pio_4<A0>();
x = minusone(x)/oneplus(x);
}
else
y = 0.0;
A0 z = sqr(x);
A0 z1 = madd(z, single_constant<A0,0x3da4f0d1>(),single_constant<A0,0xbe0e1b85>());
A0 z2 = madd(z, single_constant<A0,0x3e4c925f>(),single_constant<A0,0xbeaaaa2a>());
z1 = madd(z1, sqr(z), z2);
y = add(y, madd(x, mul( z1, z), x));
// y +=
// ((( 8.05374449538e-2 * z
// - 1.38776856032E-1) * z
// + 1.99777106478E-1) * z
// - 3.33329491539E-1) * z * x
// + x;
return b_xor(sgn, y );
}
static inline A0 atan(const A0& a0)
{
typedef typename meta::scalar_of<A0>::type sA0;
static const A0 tan3pio8 = double_constant<A0, 0x4003504f333f9de6ll>();
static const A0 Twothird = double_constant<A0, 0x3fe51eb851eb851fll>();
static const A0 tanpio8 = double_constant<A0, 0x3fda827999fcef31ll>();
A0 x = abs(a0);
const A0 flag1 = lt(x, double_constant<A0, 0x4003504f333f9de6ll>()); //tan3pio8
const A0 flag2 = b_and(ge(x, double_constant<A0, 0x3fda827999fcef31ll>()), flag1); //tanpio8
A0 yy = b_notand(flag1, Pio_2<A0>());
yy = select(flag2, Pio_4<A0>(), yy);
A0 xx = select(flag1, x, -rec(x));
xx = select(flag2, minusone(x)/oneplus(x),xx);
A0 z = sqr(xx);
z = z*horner< NT2_HORNER_COEFF_T(sA0, 5,
(0xbfec007fa1f72594ll,
0xc03028545b6b807all,
0xc052c08c36880273ll,
0xc05eb8bf2d05ba25ll,
0xc0503669fd28ec8ell)
)>(z)/
horner< NT2_HORNER_COEFF_T(sA0, 6,
(0x3ff0000000000000ll,
0x4038dbc45b14603cll,
0x4064a0dd43b8fa25ll,
0x407b0e18d2e2be3bll,
0x407e563f13b049eall,
0x4068519efbbd62ecll)
)>(z);
z = fma(xx, z, xx);
// static const A0 morebits = double_constant<A0, 0x3c91a62633145c07ll>();
z = seladd(flag2, z, mul(Half<A0>(), double_constant<A0, 0x3c91a62633145c07ll>()));
z = z+b_notand(flag1, double_constant<A0, 0x3c91a62633145c07ll>());
yy = yy + z;
return b_xor(yy, bitofsign(a0));
}
static inline A0_n atan(const A0_n a0_n)
{
const A0 a0 = {a0_n};
const A0 x = {kernel_atan(a0)};
return b_xor(x, bitofsign(a0));
}
(real_<A0>)
)
namespace nt2 { namespace ext
{
template<class Dummy>
struct call<tag::acotd_(tag::real_),
tag::cpu_, Dummy> : callable
{
template<class Sig> struct result;
template<class This,class A0>
struct result<This(A0)> : meta::strip<A0>{};
NT2_FUNCTOR_CALL(1)
{
A0 s = bitofsign(a0);
if(!a0) return b_or(Ninety<A0>(), s);
if(is_inf(a0)) return b_or(Zero<A0>(), s);
return b_or(Ninety<A0>()-atand(abs(a0)), s);
}
};
} }
NT2_REGISTER_DISPATCH(tag::acotd_, tag::cpu_,
(A0),
(arithmetic_<A0>)
)
namespace nt2 { namespace ext
{
static inline A0 atan(const A0& a0)
{
A0 x = kernel_atan(a0);
return b_xor(x, bitofsign(a0));
}
