diff --git a/src/libraries/System.Numerics.Tensors/src/System/Numerics/Tensors/netcore/TensorPrimitives.PopCount.cs b/src/libraries/System.Numerics.Tensors/src/System/Numerics/Tensors/netcore/TensorPrimitives.PopCount.cs
index da45bd720b129..1ea141baac4d5 100644
--- a/src/libraries/System.Numerics.Tensors/src/System/Numerics/Tensors/netcore/TensorPrimitives.PopCount.cs
+++ b/src/libraries/System.Numerics.Tensors/src/System/Numerics/Tensors/netcore/TensorPrimitives.PopCount.cs
@@ -52,20 +52,40 @@ public static void PopCount<T>(ReadOnlySpan<T> x, Span<T> destination)
                 // This relies on 64-bit shifts for sizeof(T) == 8, and such shifts aren't accelerated on today's hardware.
                 // Alternative approaches, such as doing two 32-bit operations and combining them were observed to not
                 // provide any meaningfuls speedup over scalar. So for now, we don't vectorize when sizeof(T) == 8.
-                sizeof(T) == 1 || sizeof(T) == 2 || sizeof(T) == 4;
+                (sizeof(T) is 1 or 2 or 4) || (AdvSimd.IsSupported && sizeof(T) == 8);
 
             public static T Invoke(T x) => T.PopCount(x);
 
             [MethodImpl(MethodImplOptions.AggressiveInlining)]
             public static Vector128<T> Invoke(Vector128<T> x)
             {
-                if (sizeof(T) == 1)
+                if (AdvSimd.IsSupported)
                 {
-                    if (AdvSimd.IsSupported)
+                    Vector128<byte> cnt = AdvSimd.PopCount(x.AsByte());
+
+                    if (sizeof(T) == 1)
+                    {
+                        return cnt.As<byte, T>();
+                    }
+
+                    if (sizeof(T) == 2)
+                    {
+                        return AdvSimd.AddPairwiseWidening(cnt).As<ushort, T>();
+                    }
+
+                    if (sizeof(T) == 4)
                     {
-                        return AdvSimd.PopCount(x.AsByte()).As<byte, T>();
+                        return AdvSimd.AddPairwiseWidening(AdvSimd.AddPairwiseWidening(cnt)).As<uint, T>();
                     }
 
+                    if (sizeof(T) == 8)
+                    {
+                        return AdvSimd.AddPairwiseWidening(AdvSimd.AddPairwiseWidening(AdvSimd.AddPairwiseWidening(cnt))).As<ulong, T>();
+                    }
+                }
+
+                if (sizeof(T) == 1)
+                {
                     if (PackedSimd.IsSupported)
                     {
                         return PackedSimd.PopCount(x.AsByte()).As<byte, T>();