From f046e326d9e30772725d8fb26dc33328e418d9d3 Mon Sep 17 00:00:00 2001 From: Antonio Sanchez Date: Fri, 17 Sep 2021 12:49:01 -0700 Subject: [PATCH] Fix strict aliasing bug causing product_small failure. Packet loading is skipped due to aliasing violation, leading to nullopt matrix multiplication. Fixes #2327. (cherry picked from commit 3c724c44cff3f9e2e9e35351abff0b5c022b320d) --- Eigen/src/Core/arch/AVX/Complex.h | 4 +++- Eigen/src/Core/arch/AVX512/Complex.h | 4 +++- Eigen/src/Core/arch/SSE/Complex.h | 11 +++-------- 3 files changed, 9 insertions(+), 10 deletions(-) diff --git a/Eigen/src/Core/arch/AVX/Complex.h b/Eigen/src/Core/arch/AVX/Complex.h index ab7bd6c65..e9096c0a1 100644 --- a/Eigen/src/Core/arch/AVX/Complex.h +++ b/Eigen/src/Core/arch/AVX/Complex.h @@ -99,7 +99,9 @@ template<> EIGEN_STRONG_INLINE Packet4cf ploadu(const std::complex EIGEN_STRONG_INLINE Packet4cf pset1(const std::complex& from) { - return Packet4cf(_mm256_castpd_ps(_mm256_broadcast_sd((const double*)(const void*)&from))); + const float re = std::real(from); + const float im = std::imag(from); + return Packet4cf(_mm256_set_ps(im, re, im, re, im, re, im, re)); } template<> EIGEN_STRONG_INLINE Packet4cf ploaddup(const std::complex* from) diff --git a/Eigen/src/Core/arch/AVX512/Complex.h b/Eigen/src/Core/arch/AVX512/Complex.h index 49c72b3f1..074253859 100644 --- a/Eigen/src/Core/arch/AVX512/Complex.h +++ b/Eigen/src/Core/arch/AVX512/Complex.h @@ -97,7 +97,9 @@ template<> EIGEN_STRONG_INLINE Packet8cf ploadu(const std::complex EIGEN_STRONG_INLINE Packet8cf pset1(const std::complex& from) { - return Packet8cf(_mm512_castpd_ps(pload1((const double*)(const void*)&from))); + const float re = std::real(from); + const float im = std::imag(from); + return Packet8cf(_mm512_set_ps(im, re, im, re, im, re, im, re, im, re, im, re, im, re, im, re)); } template<> EIGEN_STRONG_INLINE Packet8cf ploaddup(const std::complex* from) diff --git a/Eigen/src/Core/arch/SSE/Complex.h b/Eigen/src/Core/arch/SSE/Complex.h index 8fe22da46..215bfd7bb 100644 --- a/Eigen/src/Core/arch/SSE/Complex.h +++ b/Eigen/src/Core/arch/SSE/Complex.h @@ -106,14 +106,9 @@ template<> EIGEN_STRONG_INLINE Packet2cf ploadu(const std::complex EIGEN_STRONG_INLINE Packet2cf pset1(const std::complex& from) { - Packet2cf res; -#ifdef EIGEN_VECTORIZE_SSE3 - res.v = _mm_castpd_ps(_mm_loaddup_pd(reinterpret_cast(&from))); -#else - res.v = _mm_castpd_ps(_mm_load_sd(reinterpret_cast(&from))); - res.v = _mm_movelh_ps(res.v, res.v); -#endif - return res; + const float re = std::real(from); + const float im = std::imag(from); + return Packet2cf(_mm_set_ps(im, re, im, re)); } template<> EIGEN_STRONG_INLINE Packet2cf ploaddup(const std::complex* from) { return pset1(*from); } -- 2.37.0 53af63e'>shepherd: Add #:supplementary-groups....Leo Nikkilä 2022-04-11shepherd: 'fork+exec-command/container' always returns a PID....Ludovic Courtès 2022-04-11shepherd: 'exec-command*' has a valid default #:directory....Ludovic Courtès 2022-04-07shepherd: Adjust 'fork+exec-command/container' for the Shepherd 0.9....Ludovic Courtès 2021-08-02build: shepherd: Use autoload to lazily bind Shepherd modules....Maxim Cournoyer