devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Added support for fast reciprocal square root computation.

Browse Source
This commit is contained in:
Benoit Steiner
2015-02-26 09:42:41 -08:00
parent 8e817b65d0
commit f41b1f1666
8 changed files with 72 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
Eigen/src/Core/arch/AVX/MathFunctions.h
View File

@@ -300,16 +300,29 @@ psqrt<Packet8f>(const Packet8f& _x) {
return pmul(_x, x);
}
#else
template <>
EIGEN_STRONG_INLINE Packet8f psqrt<Packet8f>(const Packet8f& x) {
template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet8f psqrt<Packet8f>(const Packet8f& x) {
return _mm256_sqrt_ps(x);
}
#endif
template <>
EIGEN_STRONG_INLINE Packet4d psqrt<Packet4d>(const Packet4d& x) {
template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet4d psqrt<Packet4d>(const Packet4d& x) {
return _mm256_sqrt_pd(x);
}
template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet8f prsqrt<Packet8f>(const Packet8f& x) {
_EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
return _mm256_div_ps(p8f_one, _mm256_sqrt_ps(x));
}
#endif
template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet4d prsqrt<Packet4d>(const Packet4d& x) {
_EIGEN_DECLARE_CONST_Packet4d(one, 1.0);
return _mm256_div_pd(p4d_one, _mm256_sqrt_pd(x));
}
} // end namespace internal
} // end namespace Eigen

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

@@ -65,6 +65,7 @@ template<> struct packet_traits<float> : default_packet_traits
HasLog = 1,
HasExp = 1,
HasSqrt = 1,
HasRsqrt = 1,
HasBlend = 1
};
};
@@ -81,6 +82,7 @@ template<> struct packet_traits<double> : default_packet_traits
HasDiv = 1,
HasExp = 0,
HasSqrt = 1,
HasRsqrt = 1,
HasBlend = 1
};
};

18
Eigen/src/Core/arch/SSE/MathFunctions.h
View File

@@ -462,11 +462,25 @@ Packet4f psqrt<Packet4f>(const Packet4f& _x)
#else
template<> EIGEN_STRONG_INLINE Packet4f psqrt<Packet4f>(const Packet4f& x) { return _mm_sqrt_ps(x); }
template<>EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet4f psqrt<Packet4f>(const Packet4f& x) { return _mm_sqrt_ps(x); }
#endif
template<> EIGEN_STRONG_INLINE Packet2d psqrt<Packet2d>(const Packet2d& x) { return _mm_sqrt_pd(x); }
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet2d psqrt<Packet2d>(const Packet2d& x) { return _mm_sqrt_pd(x); }
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet4f prsqrt<Packet4f>(const Packet4f& x) {
// Unfortunately we can't use the much faster mm_rqsrt_ps since it only provides an approximation.
return _mm_div_ps(pset1<Packet4f>(1.0f), _mm_sqrt_ps(x));
}
template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
Packet2d prsqrt<Packet2d>(const Packet2d& x) {
// Unfortunately we can't use the much faster mm_rqsrt_pd since it only provides an approximation.
return _mm_div_pd(pset1<Packet2d>(1.0), _mm_sqrt_pd(x));
}
} // end namespace internal

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

@@ -108,6 +108,7 @@ template<> struct packet_traits<float> : default_packet_traits
HasLog = 1,
HasExp = 1,
HasSqrt = 1,
HasRsqrt = 1,
HasBlend = 1
};
};
@@ -124,6 +125,7 @@ template<> struct packet_traits<double> : default_packet_traits
HasDiv = 1,
HasExp = 1,
HasSqrt = 1,
HasRsqrt = 1,
HasBlend = 1
};
};