static inline void kernel_log(const A0& a0,
A0& dk,
A0& hfsq,
A0& s,
A0& R,
A0& f)
{
typedef typename meta::as_integer<A0, signed>::type int_type;
typedef typename meta::scalar_of<A0>::type sA0;
A0 x;
int_type k;
boost::fusion::tie(x, k) = fast_frexp(a0);
const int_type x_lt_sqrthf = simd::native_cast<int_type>(gt(Sqrt_2o_2<A0>(), x));
k = k+x_lt_sqrthf;
f = minusone(x+b_and(x, x_lt_sqrthf));
dk = tofloat(k);
s = f/add(Two<A0>(),f);
A0 z = sqr(s);
A0 w = sqr(z);
A0 t1= w*horner<NT2_HORNER_COEFF_T(sA0, 3,
(0x3fc39a09d078c69fll,
0x3fcc71c51d8e78afll,
0x3fd999999997fa04ll)
)> (w);
A0 t2= z*horner<NT2_HORNER_COEFF_T(sA0, 4,
(0x3fc2f112df3e5244ll,
0x3fc7466496cb03dell,
0x3fd2492494229359ll,
0x3fe5555555555593ll)
)> (w);
R = t2+t1;
hfsq = mul(Half<A0>(), sqr(f));
}
inline void kernel_log(const float& a0,
float& fe,
float& x,
float& x2,
float& y)
{
typedef float A0;
typedef meta::as_integer<A0, signed>::type int_type;
int_type e;
boost::fusion::tie(x, e) = fast_frexp(a0);
int_type x_lt_sqrthf = -(Const<float, 0x3f3504f3>() > x);
e += x_lt_sqrthf;
// if (x_lt_sqrthf) x+= x;
// x += Const<float, 0xbf800000>();
x += b_and(x, genmask<float>(x_lt_sqrthf))+Const<float,0xbf800000>();
x2 = sqr(x);
A0 y1 = fma(Const<float, 0x3d9021bb>() ,x2,Const<float, 0x3def251a>() );
A0 y2 = fma(Const<float, 0xbdebd1b8>() ,x2,Const<float, 0xbdfe5d4f>() );
y1 = fma(y1,x2,Const<float, 0x3e11e9bf>() );
y2 = fma(y2,x2,Const<float, 0xbe2aae50>() );
y1 = fma(y1,x2,Const<float, 0x3e4cceac>() );
y2 = fma(y2,x2,Const<float, 0xbe7ffffc>() );
y1 = fma(y1,x2,Const<float, 0x3eaaaaaa>() );
y = fma(x,y2,y1)*x*x2;
fe = tofloat(e);
}
static inline void kernel_log(const A0& a0,
A0& dk,
A0& hfsq,
A0& s,
A0& R,
A0& f)
{
A0 x;
int_type k(fast_frexp(a0, x));
const int_type x_lt_sqrthf = -isgt(Sqrt_2o_2<A0>(), x);
k += x_lt_sqrthf;
f = minusone(x+b_and(x, genmask<A0>(x_lt_sqrthf)));
dk = tofloat(k);
s = f/add(Two<A0>(),f);
A0 z = sqr(s);
A0 w = sqr(z);
A0 t1= w*horner<NT2_HORNER_COEFF_T(A0, 3,
(0x3fc39a09d078c69fll,
0x3fcc71c51d8e78afll,
0x3fd999999997fa04ll)
)> (w);
A0 t2= z*horner<NT2_HORNER_COEFF_T(A0, 4,
(0x3fc2f112df3e5244ll,
0x3fc7466496cb03dell,
0x3fd2492494229359ll,
0x3fe5555555555593ll)
)> (w);
R = t2+t1;
hfsq = mul(Half<A0>(), sqr(f));
}
static inline void kernel_log(const A0& a0,
A0& fe,
A0& x,
A0& x2,
A0& y)
{
typedef typename meta::as_integer<A0, signed>::type int_type;
int_type e;
boost::fusion::tie(x, e) = fast_frexp(a0);
// std::cout << "x " << x << " e " << e << std::endl;
// bf::tie(x, e) = frexp(a0);
// std::cout << "x " << x << " e " << e << std::endl;
int_type x_lt_sqrthf = nt2::simd::native_cast<int_type>(gt(single_constant<A0, 0x3f3504f3>(),x));
e = e+x_lt_sqrthf;
x = x+b_and(x, x_lt_sqrthf)+single_constant<A0, 0xbf800000>();
x2 = sqr(x);
A0 y1 = madd(single_constant<A0, 0x3d9021bb>() ,x2,single_constant<A0, 0x3def251a>() );
A0 y2 = madd(single_constant<A0, 0xbdebd1b8>() ,x2,single_constant<A0, 0xbdfe5d4f>() );
y1 = madd(y1,x2,single_constant<A0, 0x3e11e9bf>() );
y2 = madd(y2,x2,single_constant<A0, 0xbe2aae50>() );
y1 = madd(y1,x2,single_constant<A0, 0x3e4cceac>() );
y2 = madd(y2,x2,single_constant<A0, 0xbe7ffffc>() );
y1 = madd(y1,x2,single_constant<A0, 0x3eaaaaaa>() );
y = madd(x,y2,y1)*x*x2;
fe = tofloat(e);
}
static inline void kernel_log(const A0& a0,
A0& fe,
A0& x,
A0& x2,
A0& y)
{
typedef typename meta::as_integer<A0, signed>::type int_type;
typedef typename meta::strip<A0>::type stA0;
int_type e;
boost::fusion::vector_tie(x, e) = fast_frexp(a0);
int_type x_lt_sqrthf = -(single_constant<stA0, 0x3f3504f3>() > x);
e += x_lt_sqrthf;
// if (x_lt_sqrthf) x+= x;
// x += single_constant<A0, 0xbf800000>();
x += b_and(x, x_lt_sqrthf)+single_constant<stA0,0xbf800000>();
x2 = sqr(x);
A0 y1 = madd(single_constant<stA0, 0x3d9021bb>() ,x2,single_constant<stA0, 0x3def251a>() );
A0 y2 = madd(single_constant<stA0, 0xbdebd1b8>() ,x2,single_constant<stA0, 0xbdfe5d4f>() );
y1 = madd(y1,x2,single_constant<stA0, 0x3e11e9bf>() );
y2 = madd(y2,x2,single_constant<stA0, 0xbe2aae50>() );
y1 = madd(y1,x2,single_constant<stA0, 0x3e4cceac>() );
y2 = madd(y2,x2,single_constant<stA0, 0xbe7ffffc>() );
y1 = madd(y1,x2,single_constant<stA0, 0x3eaaaaaa>() );
y = madd(x,y2,y1)*x*x2;
fe = tofloat(e);
}
static inline void
kernel_log(const A0& a0,A0& fe, A0& x,A0& x2, A0& y, const A0&)
{
int_type e;
bf::tie(x, e) = fast_frexp(a0);
int_type x_lt_sqrthf = -(Const<float,0x3f3504f3>() > x);
e += x_lt_sqrthf;
x += b_and(x, genmask<float>(x_lt_sqrthf))+Const<float,0xbf800000>();
x2 = sqr(x);
A0 y1 = fma(Const<float, 0x3d9021bb>() ,x2,Const<float, 0x3def251a>() );
A0 y2 = fma(Const<float, 0xbdebd1b8>() ,x2,Const<float, 0xbdfe5d4f>() );
y1 = fma(y1,x2,Const<float, 0x3e11e9bf>() );
y2 = fma(y2,x2,Const<float, 0xbe2aae50>() );
y1 = fma(y1,x2,Const<float, 0x3e4cceac>() );
y2 = fma(y2,x2,Const<float, 0xbe7ffffc>() );
y1 = fma(y1,x2,Const<float, 0x3eaaaaaa>() );
y = fma(x,y2,y1)*x*x2;
fe = tofloat(e);
}
static inline void kernel_log(const A0& a0,
A0& fe,
A0& x,
A0& x2,
A0& y)
{
typedef typename meta::as_integer<A0, signed>::type int_type;
int_type e;
x = fast_frexp(a0, e);
int_type x_lt_sqrthf = if_else_zero(gt(single_constant<A0, 0x3f3504f3>(),x), Mone<int_type>());
e = e+x_lt_sqrthf;
x = x+b_and(x, x_lt_sqrthf)+single_constant<A0, 0xbf800000>();
x2 = sqr(x);
A0 y1 = madd(single_constant<A0, 0x3d9021bb>() ,x2,single_constant<A0, 0x3def251a>() );
A0 y2 = madd(single_constant<A0, 0xbdebd1b8>() ,x2,single_constant<A0, 0xbdfe5d4f>() );
y1 = madd(y1,x2,single_constant<A0, 0x3e11e9bf>() );
y2 = madd(y2,x2,single_constant<A0, 0xbe2aae50>() );
y1 = madd(y1,x2,single_constant<A0, 0x3e4cceac>() );
y2 = madd(y2,x2,single_constant<A0, 0xbe7ffffc>() );
y1 = madd(y1,x2,single_constant<A0, 0x3eaaaaaa>() );
y = madd(x,y2,y1)*x*x2;
fe = tofloat(e);
}
