From 9312a5bf5cd72f45558f402077b0c95683ee0fea Mon Sep 17 00:00:00 2001 From: Rasmus Munk Larsen Date: Wed, 30 Jun 2021 15:53:06 -0700 Subject: Implement a generic vectorized version of Smith's algorithms for complex division. --- Eigen/src/Core/arch/AVX512/Complex.h | 44 ++---------------------------------- 1 file changed, 2 insertions(+), 42 deletions(-) (limited to 'Eigen/src/Core/arch/AVX512') diff --git a/Eigen/src/Core/arch/AVX512/Complex.h b/Eigen/src/Core/arch/AVX512/Complex.h index 49c72b3f1..c11b8d2f8 100644 --- a/Eigen/src/Core/arch/AVX512/Complex.h +++ b/Eigen/src/Core/arch/AVX512/Complex.h @@ -157,11 +157,7 @@ EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet8cf,Packet16f) template<> EIGEN_STRONG_INLINE Packet8cf pdiv(const Packet8cf& a, const Packet8cf& b) { - Packet8cf num = pmul(a, pconj(b)); - __m512 tmp = _mm512_mul_ps(b.v, b.v); - __m512 tmp2 = _mm512_shuffle_ps(tmp,tmp,0xB1); - __m512 denom = _mm512_add_ps(tmp, tmp2); - return Packet8cf(_mm512_div_ps(num.v, denom)); + return pdiv_complex(a, b); } template<> EIGEN_STRONG_INLINE Packet8cf pcplxflip(const Packet8cf& x) @@ -309,47 +305,11 @@ template<> EIGEN_STRONG_INLINE std::complex predux_mul(const Packet2cd(_mm512_extractf64x4_pd(a.v,1)))); } -template<> struct conj_helper -{ - EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const - { return padd(pmul(x,y),c); } - - EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const - { - return internal::pmul(a, pconj(b)); - } -}; - -template<> struct conj_helper -{ - EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const - { return padd(pmul(x,y),c); } - - EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const - { - return internal::pmul(pconj(a), b); - } -}; - -template<> struct conj_helper -{ - EIGEN_STRONG_INLINE Packet4cd pmadd(const Packet4cd& x, const Packet4cd& y, const Packet4cd& c) const - { return padd(pmul(x,y),c); } - - EIGEN_STRONG_INLINE Packet4cd pmul(const Packet4cd& a, const Packet4cd& b) const - { - return pconj(internal::pmul(a, b)); - } -}; - EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cd,Packet8d) template<> EIGEN_STRONG_INLINE Packet4cd pdiv(const Packet4cd& a, const Packet4cd& b) { - Packet4cd num = pmul(a, pconj(b)); - __m512d tmp = _mm512_mul_pd(b.v, b.v); - __m512d denom = padd(_mm512_permute_pd(tmp,0x55), tmp); - return Packet4cd(_mm512_div_pd(num.v, denom)); + return pdiv_complex(a, b); } template<> EIGEN_STRONG_INLINE Packet4cd pcplxflip(const Packet4cd& x) -- cgit v1.2.3