Avoid promotion of Arm __fp16 to float in Neon PacketMath

Using overloaded arithmetic operators for Arm __fp16 always
causes a promotion to float. We replace operator* by vmulh_f16
to avoid this.

1 files changed, 2 insertions, 2 deletions

diff --git a/Eigen/src/Core/arch/NEON/PacketMath.h b/Eigen/src/Core/arch/NEON/PacketMath.h
index a51fc88c6..30edd7097 100644
--- a/Eigen/src/Core/arch/NEON/PacketMath.h
+++ b/Eigen/src/Core/arch/NEON/PacketMath.h
@@ -4355,7 +4355,7 @@ EIGEN_STRONG_INLINE Eigen::half predux_mul<Packet8hf>(const Packet8hf& a) {
   prod = vmul_f16(prod, vrev64_f16(prod));
 
   Eigen::half h;
-  h.x = vget_lane_f16(prod, 0) * vget_lane_f16(prod, 1);
+  h.x = vmulh_f16(vget_lane_f16(prod, 0), vget_lane_f16(prod, 1));
   return h;
 }
 
@@ -4364,7 +4364,7 @@ EIGEN_STRONG_INLINE Eigen::half predux_mul<Packet4hf>(const Packet4hf& a) {
   float16x4_t prod;
   prod = vmul_f16(a, vrev64_f16(a));
   Eigen::half h;
-  h.x = vget_lane_f16(prod, 0) * vget_lane_f16(prod, 1);
+  h.x = vmulh_f16(vget_lane_f16(prod, 0), vget_lane_f16(prod, 1));
   return h;
 }