From ced1a45f82684ab18cbe4d830f6832407cabadb3 Mon Sep 17 00:00:00 2001
From: Gael Guennebaud <g.gael@free.fr>
Date: Tue, 20 Jul 2010 14:24:01 +0200
Subject: [PATCH] add NEON ploaddup and pcplxflip functions

---
 Eigen/src/Core/arch/NEON/Complex.h    |  5 +++++
 Eigen/src/Core/arch/NEON/PacketMath.h | 27 +++++++++++++++++++--------
 2 files changed, 24 insertions(+), 8 deletions(-)

diff --git a/Eigen/src/Core/arch/NEON/Complex.h b/Eigen/src/Core/arch/NEON/Complex.h
index 6d9e8da85..9678040e7 100644
--- a/Eigen/src/Core/arch/NEON/Complex.h
+++ b/Eigen/src/Core/arch/NEON/Complex.h
@@ -141,6 +141,11 @@ template<> EIGEN_STRONG_INLINE Packet2cf ei_preverse(const Packet2cf& a)
   return Packet2cf(a_r128);
 }
 
+EIGEN_STRONG_INLINE Packet2cf ei_pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
+{
+  return Packet2cf(vrev64q_f32(a.v));
+}
+
 template<> EIGEN_STRONG_INLINE std::complex<float> ei_predux<Packet2cf>(const Packet2cf& a)
 {
   float32x2_t a1, a2;
diff --git a/Eigen/src/Core/arch/NEON/PacketMath.h b/Eigen/src/Core/arch/NEON/PacketMath.h
index b899fece1..8220ed07c 100644
--- a/Eigen/src/Core/arch/NEON/PacketMath.h
+++ b/Eigen/src/Core/arch/NEON/PacketMath.h
@@ -180,6 +180,21 @@ template<> EIGEN_STRONG_INLINE Packet4i ei_pload<int>(const int*     from) { EIG
 template<> EIGEN_STRONG_INLINE Packet4f ei_ploadu(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f32(from); }
 template<> EIGEN_STRONG_INLINE Packet4i ei_ploadu(const int* from)   { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_s32(from); }
 
+template<> EIGEN_STRONG_INLINE Packet4f ei_ploaddup<Packet4f>(const float* from)
+{
+  float32x2_t lo, ho;
+  lo = vdup_n_f32(*from);
+  hi = vdup_n_f32(*from);
+  return vcombine_f32(lo, hi);
+}
+template<> EIGEN_STRONG_INLINE Packet4i ei_ploaddup<Packet4i>(const float* from)
+{
+  int32x2_t lo, ho;
+  lo = vdup_n_s32(*from);
+  hi = vdup_n_s32(*from);
+  return vcombine_s32(lo, hi);
+}
+
 template<> EIGEN_STRONG_INLINE void ei_pstore<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f32(to, from); }
 template<> EIGEN_STRONG_INLINE void ei_pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_s32(to, from); }
 
@@ -195,25 +210,21 @@ template<> EIGEN_STRONG_INLINE int    ei_pfirst<Packet4i>(const Packet4i& a) { i
 
 template<> EIGEN_STRONG_INLINE Packet4f ei_preverse(const Packet4f& a) {
   float32x2_t a_lo, a_hi;
-  Packet4f a_r64, a_r128;
+  Packet4f a_r64;
 
   a_r64 = vrev64q_f32(a);
   a_lo = vget_low_f32(a_r64);
   a_hi = vget_high_f32(a_r64);
-  a_r128 = vcombine_f32(a_hi, a_lo);
-
-  return a_r128;
+  return vcombine_f32(a_hi, a_lo);
 }
 template<> EIGEN_STRONG_INLINE Packet4i ei_preverse(const Packet4i& a) {
   int32x2_t a_lo, a_hi;
-  Packet4i a_r64, a_r128;
+  Packet4i a_r64;
 
   a_r64 = vrev64q_s32(a);
   a_lo = vget_low_s32(a_r64);
   a_hi = vget_high_s32(a_r64);
-  a_r128 = vcombine_s32(a_hi, a_lo);
-
-  return a_r128;
+  return vcombine_s32(a_hi, a_lo);
 }
 template<> EIGEN_STRONG_INLINE Packet4f ei_pabs(const Packet4f& a) { return vabsq_f32(a); }
 template<> EIGEN_STRONG_INLINE Packet4i ei_pabs(const Packet4i& a) { return vabsq_s32(a); }