Enable the vectorization of adds and mults of fp16

2026-04-10 11:34:33 +08:00 · 2016-06-07 18:22:18 -07:00
parent d6d39c7ddb
commit 8fd57a97f2
1 changed files with 2 additions and 0 deletions
--- a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
+++ b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
@@ -28,6 +28,8 @@ template<> struct packet_traits<Eigen::half> : default_packet_traits
    AlignedOnScalar = 1,
    size=2,
    HasHalfPacket = 0,
+    HasAdd    = 1,
+    HasMul    = 1,
    HasDiv    = 1,
    HasSqrt   = 1,
    HasRsqrt  = 1,