diff options
| author |  Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-12-21 20:55:07 -0800 |
|---|---|---|
| committer | Benoit Steiner <benoit.steiner.goog@gmail.com> | 2016-12-21 20:55:07 -0800 |
| commit | 354baa0fb1bb6c2d137288a61339e8cf0e702ad7 (patch) | |
| tree | 43ad35f004b018db6a8b55d46083a8093a7503dc /Eigen | |
| parent | d7825b6707367c92f5d2cba742373059779c4e0f (diff) | |
Avoid using horizontal adds since they're not very efficient.
Diffstat (limited to 'Eigen')
| -rw-r--r-- | Eigen/src/Core/arch/AVX512/PacketMath.h | 10 |
1 files changed, 4 insertions, 6 deletions
diff --git a/Eigen/src/Core/arch/AVX512/PacketMath.h b/Eigen/src/Core/arch/AVX512/PacketMath.h index 14ef93c55..e46a60472 100644 --- a/Eigen/src/Core/arch/AVX512/PacketMath.h +++ b/Eigen/src/Core/arch/AVX512/PacketMath.h @@ -628,8 +628,8 @@ EIGEN_STRONG_INLINE Packet8d pabs(const Packet8d& a) { #ifdef EIGEN_VECTORIZE_AVX512DQ // AVX512F does not define _mm512_extractf32x8_ps to extract _m256 from _m512 #define EIGEN_EXTRACT_8f_FROM_16f(INPUT, OUTPUT) \ - __m256 OUTPUT##_0 = _mm512_extractf32x8_ps(INPUT, 0) __m256 OUTPUT##_1 = \ - _mm512_extractf32x8_ps(INPUT, 1) + __m256 OUTPUT##_0 = _mm512_extractf32x8_ps(INPUT, 0); \ + __m256 OUTPUT##_1 = _mm512_extractf32x8_ps(INPUT, 1) #else #define EIGEN_EXTRACT_8f_FROM_16f(INPUT, OUTPUT) \ __m256 OUTPUT##_0 = _mm256_insertf128_ps( \ @@ -847,10 +847,8 @@ EIGEN_STRONG_INLINE float predux<Packet16f>(const Packet16f& a) { #ifdef EIGEN_VECTORIZE_AVX512DQ __m256 lane0 = _mm512_extractf32x8_ps(a, 0); __m256 lane1 = _mm512_extractf32x8_ps(a, 1); - __m256 sum = _mm256_add_ps(lane0, lane1); - __m256 tmp0 = _mm256_hadd_ps(sum, _mm256_permute2f128_ps(a, a, 1)); - tmp0 = _mm256_hadd_ps(tmp0, tmp0); - return _mm_cvtss_f32(_mm256_castps256_ps128(_mm256_hadd_ps(tmp0, tmp0))); + Packet8f x = _mm256_add_ps(lane0, lane1); + return predux<Packet8f>(x); #else __m128 lane0 = _mm512_extractf32x4_ps(a, 0); __m128 lane1 = _mm512_extractf32x4_ps(a, 1); |