diff options
Diffstat (limited to 'Eigen/src/Core/arch/AVX')
-rw-r--r-- | Eigen/src/Core/arch/AVX/MathFunctions.h | 23 |
1 files changed, 13 insertions, 10 deletions
diff --git a/Eigen/src/Core/arch/AVX/MathFunctions.h b/Eigen/src/Core/arch/AVX/MathFunctions.h index 5fe2cff32..c0d362fa7 100644 --- a/Eigen/src/Core/arch/AVX/MathFunctions.h +++ b/Eigen/src/Core/arch/AVX/MathFunctions.h @@ -97,33 +97,36 @@ pexp<Packet4d>(const Packet4d& _x) { // For detail see here: http://www.beyond3d.com/content/articles/8/ #if EIGEN_FAST_MATH template <> -EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f -psqrt<Packet8f>(const Packet8f& _x) { - Packet8f half = pmul(_x, pset1<Packet8f>(.5f)); - Packet8f denormal_mask = _mm256_and_ps( - _mm256_cmp_ps(_x, pset1<Packet8f>((std::numeric_limits<float>::min)()), - _CMP_LT_OQ), - _mm256_cmp_ps(_x, _mm256_setzero_ps(), _CMP_GE_OQ)); +EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED +Packet8f psqrt<Packet8f>(const Packet8f& _x) { + Packet8f minus_half_x = pmul(_x, pset1<Packet8f>(-0.5f)); + Packet8f denormal_mask = pandnot( + pcmp_lt(_x, pset1<Packet8f>((std::numeric_limits<float>::min)())), + pcmp_lt(_x, pzero(_x))); // Compute approximate reciprocal sqrt. Packet8f x = _mm256_rsqrt_ps(_x); // Do a single step of Newton's iteration. - x = pmul(x, psub(pset1<Packet8f>(1.5f), pmul(half, pmul(x,x)))); + x = pmul(x, pmadd(minus_half_x, pmul(x,x), pset1<Packet8f>(1.5f))); // Flush results for denormals to zero. - return _mm256_andnot_ps(denormal_mask, pmul(_x,x)); + return pandnot(pmul(_x,x), denormal_mask); } + #else + template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f psqrt<Packet8f>(const Packet8f& _x) { return _mm256_sqrt_ps(_x); } + #endif + template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d psqrt<Packet4d>(const Packet4d& _x) { return _mm256_sqrt_pd(_x); } -#if EIGEN_FAST_MATH +#if EIGEN_FAST_MATH template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f prsqrt<Packet8f>(const Packet8f& _x) { _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(inf, 0x7f800000); |