1 files changed, 21 insertions, 6 deletions
diff --git a/Eigen/src/Core/arch/AVX/PacketMath.h b/Eigen/src/Core/arch/AVX/PacketMath.h
index e2376bd1f..be66a502a 100644
--- a/Eigen/src/Core/arch/AVX/PacketMath.h
+++ b/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -22,9 +22,9 @@ namespace internal {
 #define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS (2*sizeof(void*))
 #endif
 
-#ifdef EIGEN_VECTORIZE_FMA
-#ifndef EIGEN_HAS_FUSED_MADD
-#define EIGEN_HAS_FUSED_MADD 1
+#ifdef __FMA__
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
 #endif
 #endif
 
@@ -58,7 +58,8 @@ template<> struct packet_traits<float>  : default_packet_traits
     HasCos  = 0,
     HasLog  = 0,
     HasExp  = 0,
-    HasSqrt = 0
+    HasSqrt = 0,
+    HasBlend = 1
   };
  };
 template<> struct packet_traits<double> : default_packet_traits
@@ -72,7 +73,8 @@ template<> struct packet_traits<double> : default_packet_traits
     HasHalfPacket = 1,
 
     HasDiv  = 1,
-    HasExp  = 0
+    HasExp  = 0,
+    HasBlend = 1
   };
 };
 
@@ -133,7 +135,7 @@ template<> EIGEN_STRONG_INLINE Packet8i pdiv<Packet8i>(const Packet8i& /*a*/, co
   return pset1<Packet8i>(0);
 }
 
-#ifdef EIGEN_VECTORIZE_FMA
+#ifdef __FMA__
 template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f& b, const Packet8f& c) {
 #if EIGEN_COMP_GNUC || EIGEN_COMP_CLANG
   // clang stupidly generates a vfmadd213ps instruction plus some vmovaps on registers,
@@ -557,6 +559,19 @@ ptranspose(PacketBlock<Packet4d,4>& kernel) {
   kernel.packet[2] = _mm256_permute2f128_pd(T1, T3, 49);
 }
 
+template<> EIGEN_STRONG_INLINE Packet8f pblend(const Selector<8>& ifPacket, const Packet8f& thenPacket, const Packet8f& elsePacket) {
+  const __m256 zero = _mm256_setzero_ps();
+  const __m256 select = _mm256_set_ps(ifPacket.select[7], ifPacket.select[6], ifPacket.select[5], ifPacket.select[4], ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m256 false_mask = _mm256_cmp_ps(select, zero, _CMP_EQ_UQ);
+  return _mm256_blendv_ps(thenPacket, elsePacket, false_mask);
+}
+template<> EIGEN_STRONG_INLINE Packet4d pblend(const Selector<4>& ifPacket, const Packet4d& thenPacket, const Packet4d& elsePacket) {
+  const __m256d zero = _mm256_setzero_pd();
+  const __m256d select = _mm256_set_pd(ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m256d false_mask = _mm256_cmp_pd(select, zero, _CMP_EQ_UQ);
+  return _mm256_blendv_pd(thenPacket, elsePacket, false_mask);
+}
+
 } // end namespace internal
 
 } // end namespace Eigen