mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-01-24 14:45:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merged in guillaume_michel/eigen (pull request PR-334)

Browse Source 
- Add support for NEON plog PacketMath function
This commit is contained in:
Christoph Hertzberg 2017-10-23 13:22:22 +00:00
parent a6d875bac8
commit 11ddac57e5
2 changed files with 93 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

92
Eigen/src/Core/arch/NEON/MathFunctions.h
View File

@ -84,6 +84,98 @@ Packet4f pexp<Packet4f>(const Packet4f& _x)
return y;
}
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet4f plog<Packet4f>(const Packet4f& _x)
{
Packet4f x = _x;
_EIGEN_DECLARE_CONST_Packet4f(1 , 1.0f);
_EIGEN_DECLARE_CONST_Packet4f(half, 0.5f);
_EIGEN_DECLARE_CONST_Packet4i(0x7f, 0x7f);
_EIGEN_DECLARE_CONST_Packet4i(inv_mant_mask, ~0x7f800000);
/* natural logarithm computed for 4 simultaneous float
return NaN for x <= 0
*/
_EIGEN_DECLARE_CONST_Packet4f(cephes_SQRTHF, 0.707106781186547524f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p0, 7.0376836292E-2f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p1, - 1.1514610310E-1f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p2, 1.1676998740E-1f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p3, - 1.2420140846E-1f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p4, + 1.4249322787E-1f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p5, - 1.6668057665E-1f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p6, + 2.0000714765E-1f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p7, - 2.4999993993E-1f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_p8, + 3.3333331174E-1f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_q1, -2.12194440e-4f);
_EIGEN_DECLARE_CONST_Packet4f(cephes_log_q2, 0.693359375f);
x = vmaxq_f32(x, vdupq_n_f32(0)); /* force flush to zero on denormal values */
Packet4ui invalid_mask = vcleq_f32(x, vdupq_n_f32(0));
Packet4i ux = vreinterpretq_s32_f32(x);
Packet4i emm0 = vshrq_n_s32(ux, 23);
/* keep only the fractional part */
ux = vandq_s32(ux, p4i_inv_mant_mask);
ux = vorrq_s32(ux, vreinterpretq_s32_f32(p4f_half));
x = vreinterpretq_f32_s32(ux);
emm0 = vsubq_s32(emm0, p4i_0x7f);
Packet4f e = vcvtq_f32_s32(emm0);
e = vaddq_f32(e, p4f_1);
/* part2:
if( x < SQRTHF ) {
e -= 1;
x = x + x - 1.0;
} else { x = x - 1.0; }
*/
Packet4ui mask = vcltq_f32(x, p4f_cephes_SQRTHF);
Packet4f tmp = vreinterpretq_f32_u32(vandq_u32(vreinterpretq_u32_f32(x), mask));
x = vsubq_f32(x, p4f_1);
e = vsubq_f32(e, vreinterpretq_f32_u32(vandq_u32(vreinterpretq_u32_f32(p4f_1), mask)));
x = vaddq_f32(x, tmp);
Packet4f z = vmulq_f32(x,x);
Packet4f y = p4f_cephes_log_p0;
y = vmulq_f32(y, x);
y = vaddq_f32(y, p4f_cephes_log_p1);
y = vmulq_f32(y, x);
y = vaddq_f32(y, p4f_cephes_log_p2);
y = vmulq_f32(y, x);
y = vaddq_f32(y, p4f_cephes_log_p3);
y = vmulq_f32(y, x);
y = vaddq_f32(y, p4f_cephes_log_p4);
y = vmulq_f32(y, x);
y = vaddq_f32(y, p4f_cephes_log_p5);
y = vmulq_f32(y, x);
y = vaddq_f32(y, p4f_cephes_log_p6);
y = vmulq_f32(y, x);
y = vaddq_f32(y, p4f_cephes_log_p7);
y = vmulq_f32(y, x);
y = vaddq_f32(y, p4f_cephes_log_p8);
y = vmulq_f32(y, x);
y = vmulq_f32(y, z);
tmp = vmulq_f32(e, p4f_cephes_log_q1);
y = vaddq_f32(y, tmp);
tmp = vmulq_f32(z, p4f_half);
y = vsubq_f32(y, tmp);
tmp = vmulq_f32(e, p4f_cephes_log_q2);
x = vaddq_f32(x, y);
x = vaddq_f32(x, tmp);
x = vreinterpretq_f32_u32(vorrq_u32(vreinterpretq_u32_f32(x), invalid_mask)); // negative arg will be NAN
return x;
}
} // end namespace internal
} // end namespace Eigen

2
Eigen/src/Core/arch/NEON/PacketMath.h
View File

@ -81,7 +81,7 @@ template<> struct packet_traits<float> : default_packet_traits
// FIXME check the Has*
HasSin = 0,
HasCos = 0,
HasLog = 0,
HasLog = 1,
HasExp = 1,
HasSqrt = 0
};