Add explicit AdvSimd to MultiplyAddAdjacent

src/ImageSharp/Common/Helpers/Vector128Utilities.cs

@@ -436,6 +436,20 @@ internal static class Vector128_
         return Vector128.Narrow(prodLo, prodHi);
     }
 
+    /// <summary>
+    /// Multiply packed signed 16-bit integers in <paramref name="left"/> and <paramref name="right"/>, producing
+    /// intermediate signed 32-bit integers. Horizontally add adjacent pairs of intermediate 32-bit integers, and
+    /// pack the results.
+    /// </summary>
+    /// <param name="left">
+    /// The first vector containing packed signed 16-bit integers to multiply and add.
+    /// </param>
+    /// <param name="right">
+    /// The second vector containing packed signed 16-bit integers to multiply and add.
+    /// </param>
+    /// <returns>
+    /// A vector containing the results of multiplying and adding adjacent pairs of packed signed 16-bit integers
+    /// </returns>
     public static Vector128<int> MultiplyAddAdjacent(Vector128<short> left, Vector128<short> right)
     {
         if (Sse2.IsSupported)
@@ -443,6 +457,18 @@ internal static class Vector128_
             return Sse2.MultiplyAddAdjacent(left, right);
         }
 
+        if (AdvSimd.IsSupported)
+        {
+            Vector128<int> prodLo = AdvSimd.MultiplyWideningLower(left.GetLower(), right.GetLower());
+            Vector128<int> prodHi = AdvSimd.MultiplyWideningLower(left.GetUpper(), right.GetUpper());
+
+            Vector128<long> v0 = AdvSimd.AddPairwiseWidening(prodLo);
+            Vector128<long> v1 = AdvSimd.AddPairwiseWidening(prodHi);
+
+            return Vector128.Narrow(v0, v1);
+        }
+
+        {
             // Widen each half of the short vectors into two int vectors
             (Vector128<int> leftLower, Vector128<int> leftUpper) = Vector128.Widen(left);
             (Vector128<int> rightLower, Vector128<int> rightUpper) = Vector128.Widen(right);
@@ -460,6 +486,7 @@ internal static class Vector128_
 
             return v0 + v1 + v2 + v3;
         }
+    }
 
     /// <summary>
     /// Multiply the packed 16-bit integers in <paramref name="left"/> and <paramref name="right"/>, producing