Move eigen_packet_wrapper to GenericPacketMath.h and use it for SSE/AVX/AVX512 as it is already used for NEON.

This will allow us to define multiple packet types backed by the same vector type, e.g., __m128i.
Use this machanism to define packets for half and clean up the packet op implementations.

1 files changed, 2 insertions, 18 deletions

diff --git a/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h
index d04013e7e..6a4dd5253 100755
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -36,29 +36,15 @@ namespace internal {
 // One solution is to increase ABI version using -fabi-version=4 (or greater).
 // Otherwise, we workaround this inconvenience by wrapping 128bit types into the following helper
 // structure:
-template<typename T>
-struct eigen_packet_wrapper
-{
-  EIGEN_ALWAYS_INLINE operator T&() { return m_val; }
-  EIGEN_ALWAYS_INLINE operator const T&() const { return m_val; }
-  EIGEN_ALWAYS_INLINE eigen_packet_wrapper() {}
-  EIGEN_ALWAYS_INLINE eigen_packet_wrapper(const T &v) : m_val(v) {}
-  EIGEN_ALWAYS_INLINE eigen_packet_wrapper& operator=(const T &v) {
-    m_val = v;
-    return *this;
-  }
-
-  T m_val;
-};
 typedef eigen_packet_wrapper<__m128>  Packet4f;
-typedef eigen_packet_wrapper<__m128i> Packet4i;
 typedef eigen_packet_wrapper<__m128d> Packet2d;
 #else
 typedef __m128  Packet4f;
-typedef __m128i Packet4i;
 typedef __m128d Packet2d;
 #endif
 
+typedef eigen_packet_wrapper<__m128i, 0> Packet4i;
+
 template<> struct is_arithmetic<__m128>  { enum { value = true }; };
 template<> struct is_arithmetic<__m128i> { enum { value = true }; };
 template<> struct is_arithmetic<__m128d> { enum { value = true }; };
@@ -400,7 +386,6 @@ template<> EIGEN_STRONG_INLINE Packet2d pcmp_eq(const Packet2d& a, const Packet2
 template<> EIGEN_STRONG_INLINE Packet4i pcmp_lt(const Packet4i& a, const Packet4i& b) { return _mm_cmplt_epi32(a,b); }
 template<> EIGEN_STRONG_INLINE Packet4i pcmp_eq(const Packet4i& a, const Packet4i& b) { return _mm_cmpeq_epi32(a,b); }
 
-
 template<> EIGEN_STRONG_INLINE Packet4i ptrue<Packet4i>(const Packet4i& a) { return _mm_cmpeq_epi32(a, a); }
 template<> EIGEN_STRONG_INLINE Packet4f
 ptrue<Packet4f>(const Packet4f& a) {
@@ -521,7 +506,6 @@ template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
   return _mm_loadu_si128(reinterpret_cast<const __m128i*>(from));
 }
 
-
 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
 {
   return vec4f_swizzle1(_mm_castpd_ps(_mm_load_sd(reinterpret_cast<const double*>(from))), 0, 0, 1, 1);