aboutsummaryrefslogtreecommitdiffhomepage
path: root/Eigen/src/Core/arch/AVX512/Complex.h
diff options
context:
space:
mode:
authorGravatar Rasmus Munk Larsen <rmlarsen@google.com>2019-01-11 10:28:52 -0800
committerGravatar Rasmus Munk Larsen <rmlarsen@google.com>2019-01-11 10:28:52 -0800
commitfff5a5b5791d6ad0adeeff7049227680b858f884 (patch)
treeb24863eda14d50dc115a82c060e1814ff9a4fc78 /Eigen/src/Core/arch/AVX512/Complex.h
parente9936cf2b9665fab9f9aeb87a8fd54474d7ce32a (diff)
parent74882471d0012596e78bffa7ad4e74d940be5059 (diff)
Resolve.
Diffstat (limited to 'Eigen/src/Core/arch/AVX512/Complex.h')
-rw-r--r--Eigen/src/Core/arch/AVX512/Complex.h32
1 files changed, 30 insertions, 2 deletions
diff --git a/Eigen/src/Core/arch/AVX512/Complex.h b/Eigen/src/Core/arch/AVX512/Complex.h
index b7e68d2ab..f2034a713 100644
--- a/Eigen/src/Core/arch/AVX512/Complex.h
+++ b/Eigen/src/Core/arch/AVX512/Complex.h
@@ -406,12 +406,40 @@ template<> EIGEN_STRONG_INLINE Packet4cd pcplxflip<Packet4cd>(const Packet4cd& x
EIGEN_DEVICE_FUNC inline void
ptranspose(PacketBlock<Packet8cf,4>& kernel) {
- ptranspose(reinterpret_cast<PacketBlock<Packet8d,4>&>(kernel));
+ PacketBlock<Packet8d,4> pb;
+
+ pb.packet[0] = _mm512_castps_pd(kernel.packet[0].v);
+ pb.packet[1] = _mm512_castps_pd(kernel.packet[1].v);
+ pb.packet[2] = _mm512_castps_pd(kernel.packet[2].v);
+ pb.packet[3] = _mm512_castps_pd(kernel.packet[3].v);
+ ptranspose(pb);
+ kernel.packet[0].v = _mm512_castpd_ps(pb.packet[0]);
+ kernel.packet[1].v = _mm512_castpd_ps(pb.packet[1]);
+ kernel.packet[2].v = _mm512_castpd_ps(pb.packet[2]);
+ kernel.packet[3].v = _mm512_castpd_ps(pb.packet[3]);
}
EIGEN_DEVICE_FUNC inline void
ptranspose(PacketBlock<Packet8cf,8>& kernel) {
- ptranspose(reinterpret_cast<PacketBlock<Packet8d,8>&>(kernel));
+ PacketBlock<Packet8d,8> pb;
+
+ pb.packet[0] = _mm512_castps_pd(kernel.packet[0].v);
+ pb.packet[1] = _mm512_castps_pd(kernel.packet[1].v);
+ pb.packet[2] = _mm512_castps_pd(kernel.packet[2].v);
+ pb.packet[3] = _mm512_castps_pd(kernel.packet[3].v);
+ pb.packet[4] = _mm512_castps_pd(kernel.packet[4].v);
+ pb.packet[5] = _mm512_castps_pd(kernel.packet[5].v);
+ pb.packet[6] = _mm512_castps_pd(kernel.packet[6].v);
+ pb.packet[7] = _mm512_castps_pd(kernel.packet[7].v);
+ ptranspose(pb);
+ kernel.packet[0].v = _mm512_castpd_ps(pb.packet[0]);
+ kernel.packet[1].v = _mm512_castpd_ps(pb.packet[1]);
+ kernel.packet[2].v = _mm512_castpd_ps(pb.packet[2]);
+ kernel.packet[3].v = _mm512_castpd_ps(pb.packet[3]);
+ kernel.packet[4].v = _mm512_castpd_ps(pb.packet[4]);
+ kernel.packet[5].v = _mm512_castpd_ps(pb.packet[5]);
+ kernel.packet[6].v = _mm512_castpd_ps(pb.packet[6]);
+ kernel.packet[7].v = _mm512_castpd_ps(pb.packet[7]);
}
EIGEN_DEVICE_FUNC inline void