Update based on Tanner's investigationl.

src/ImageSharp/Common/Helpers/SimdUtils.FallbackIntrinsics128.cs

@@ -125,8 +125,7 @@ namespace SixLabors.ImageSharp
                     Vector4 s = Unsafe.Add(ref sBase, i);
                     s *= maxBytes;
                     s += half;
-                    s = Vector4.Max(Vector4.Zero, s);
-                    s = Vector4.Min(maxBytes, s);
+                    s = Vector4Utilities.FastClamp(s, Vector4.Zero, maxBytes);
 
                     ref ByteVector4 d = ref Unsafe.Add(ref dBase, i);
                     d.X = (byte)s.X;

src/ImageSharp/Common/Helpers/Vector4Utilities.cs

@@ -23,7 +23,7 @@ namespace SixLabors.ImageSharp
         /// <returns>The <see cref="Vector4"/>.</returns>
         [MethodImpl(InliningOptions.ShortMethod)]
         public static Vector4 FastClamp(Vector4 x, Vector4 min, Vector4 max)
-            => Vector4.Min(Vector4.Max(min, x), max);
+            => Vector4.Min(Vector4.Max(x, min), max);
 
         /// <summary>
         /// Pre-multiplies the "x", "y", "z" components of a vector by its "w" component leaving the "w" component intact.