diff --git a/src/ImageSharp/Common/Helpers/TensorPrimitives_.Add.cs b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Add.cs
new file mode 100644
index 000000000..f2ef214cf
--- /dev/null
+++ b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Add.cs
@@ -0,0 +1,84 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.Intrinsics;
+
+namespace SixLabors.ImageSharp.Common.Helpers;
+
+internal static partial class TensorPrimitives_
+{
+    /// <summary>
+    /// Computes the element-wise sum of the values in <paramref name="x"/> and <paramref name="y"/>.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    /// <param name="x">The first addends.</param>
+    /// <param name="y">The second addends.</param>
+    /// <param name="destination">The destination for the sums.</param>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public static void Add<T>(ReadOnlySpan<T> x, ReadOnlySpan<T> y, Span<T> destination)
+        where T : IAdditionOperators<T, T, T>, IAdditiveIdentity<T, T>
+        => InvokeSpanSpanIntoSpan<T, AddOperator<T>>(x, y, destination);
+
+    /// <summary>
+    /// Computes the element-wise sum of the values in <paramref name="x"/> and the scalar <paramref name="y"/>.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    /// <param name="x">The first addends.</param>
+    /// <param name="y">The scalar second addend.</param>
+    /// <param name="destination">The destination for the sums.</param>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public static void Add<T>(ReadOnlySpan<T> x, T y, Span<T> destination)
+        where T : IAdditionOperators<T, T, T>, IAdditiveIdentity<T, T>
+        => InvokeSpanScalarIntoSpan<T, AddOperator<T>>(x, y, destination);
+
+    /// <summary>
+    /// Adds corresponding values.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    private readonly struct AddOperator<T> : IBinaryOperator<T>
+        where T : IAdditionOperators<T, T, T>, IAdditiveIdentity<T, T>
+    {
+        /// <summary>
+        /// Gets a value indicating whether this operation supports vector execution.
+        /// </summary>
+        public static bool Vectorizable => true;
+
+        /// <summary>
+        /// Adds scalar values.
+        /// </summary>
+        /// <param name="x">The first addend.</param>
+        /// <param name="y">The second addend.</param>
+        /// <returns>The sum.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static T Invoke(T x, T y) => x + y;
+
+        /// <summary>
+        /// Adds 128-bit vectors.
+        /// </summary>
+        /// <param name="x">The first addends.</param>
+        /// <param name="y">The second addends.</param>
+        /// <returns>The sums.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> y) => x + y;
+
+        /// <summary>
+        /// Adds 256-bit vectors.
+        /// </summary>
+        /// <param name="x">The first addends.</param>
+        /// <param name="y">The second addends.</param>
+        /// <returns>The sums.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> y) => x + y;
+
+        /// <summary>
+        /// Adds 512-bit vectors.
+        /// </summary>
+        /// <param name="x">The first addends.</param>
+        /// <param name="y">The second addends.</param>
+        /// <returns>The sums.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> y) => x + y;
+    }
+}
diff --git a/src/ImageSharp/Common/Helpers/TensorPrimitives_.Clamp.cs b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Clamp.cs
new file mode 100644
index 000000000..e2cc9097e
--- /dev/null
+++ b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Clamp.cs
@@ -0,0 +1,318 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.Intrinsics;
+
+namespace SixLabors.ImageSharp.Common.Helpers;
+
+internal static partial class TensorPrimitives_
+{
+    /// <summary>
+    /// Computes the element-wise result of clamping <paramref name="x"/> to the inclusive range specified
+    /// by <paramref name="min"/> and <paramref name="max"/>.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    /// <param name="x">The values to clamp.</param>
+    /// <param name="min">The inclusive lower bound.</param>
+    /// <param name="max">The inclusive upper bound.</param>
+    /// <param name="destination">The destination for the clamped values.</param>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public static void Clamp<T>(ReadOnlySpan<T> x, T min, T max, Span<T> destination)
+        where T : INumber<T>
+        => InvokeSpanScalarScalarIntoSpan<T, ClampOperator<T>>(x, min, max, destination);
+
+    /// <summary>
+    /// Clamps single-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="value">The values to clamp.</param>
+    /// <param name="min">The inclusive lower bounds.</param>
+    /// <param name="max">The inclusive upper bounds.</param>
+    /// <returns>The clamped values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector128<float> ClampSingle(
+        Vector128<float> value,
+        Vector128<float> min,
+        Vector128<float> max)
+    {
+        // Unlike the native x86 min/max instructions, the normalized runtime operations propagate a NaN in the
+        // first operand and select negative zero when equal values have different signs.
+        Vector128<float> maximum = Vector128.ConditionalSelect(
+            Vector128.LessThan(min, value)
+            | ~Vector128.Equals(value, value)
+            | (Vector128.Equals(value, min) & (min.AsInt32() >> 31).AsSingle()),
+            value,
+            min);
+
+        return Vector128.ConditionalSelect(
+            Vector128.LessThan(maximum, max)
+            | ~Vector128.Equals(maximum, maximum)
+            | (Vector128.Equals(maximum, max) & (maximum.AsInt32() >> 31).AsSingle()),
+            maximum,
+            max);
+    }
+
+    /// <summary>
+    /// Clamps single-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="value">The values to clamp.</param>
+    /// <param name="min">The inclusive lower bounds.</param>
+    /// <param name="max">The inclusive upper bounds.</param>
+    /// <returns>The clamped values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector256<float> ClampSingle(
+        Vector256<float> value,
+        Vector256<float> min,
+        Vector256<float> max)
+    {
+        Vector256<float> maximum = Vector256.ConditionalSelect(
+            Vector256.LessThan(min, value)
+            | ~Vector256.Equals(value, value)
+            | (Vector256.Equals(value, min) & (min.AsInt32() >> 31).AsSingle()),
+            value,
+            min);
+
+        return Vector256.ConditionalSelect(
+            Vector256.LessThan(maximum, max)
+            | ~Vector256.Equals(maximum, maximum)
+            | (Vector256.Equals(maximum, max) & (maximum.AsInt32() >> 31).AsSingle()),
+            maximum,
+            max);
+    }
+
+    /// <summary>
+    /// Clamps single-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="value">The values to clamp.</param>
+    /// <param name="min">The inclusive lower bounds.</param>
+    /// <param name="max">The inclusive upper bounds.</param>
+    /// <returns>The clamped values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector512<float> ClampSingle(
+        Vector512<float> value,
+        Vector512<float> min,
+        Vector512<float> max)
+    {
+        Vector512<float> maximum = Vector512.ConditionalSelect(
+            Vector512.LessThan(min, value)
+            | ~Vector512.Equals(value, value)
+            | (Vector512.Equals(value, min) & (min.AsInt32() >> 31).AsSingle()),
+            value,
+            min);
+
+        return Vector512.ConditionalSelect(
+            Vector512.LessThan(maximum, max)
+            | ~Vector512.Equals(maximum, maximum)
+            | (Vector512.Equals(maximum, max) & (maximum.AsInt32() >> 31).AsSingle()),
+            maximum,
+            max);
+    }
+
+    /// <summary>
+    /// Clamps double-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="value">The values to clamp.</param>
+    /// <param name="min">The inclusive lower bounds.</param>
+    /// <param name="max">The inclusive upper bounds.</param>
+    /// <returns>The clamped values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector128<double> ClampDouble(
+        Vector128<double> value,
+        Vector128<double> min,
+        Vector128<double> max)
+    {
+        Vector128<double> maximum = Vector128.ConditionalSelect(
+            Vector128.LessThan(min, value)
+            | ~Vector128.Equals(value, value)
+            | (Vector128.Equals(value, min) & (min.AsInt64() >> 63).AsDouble()),
+            value,
+            min);
+
+        return Vector128.ConditionalSelect(
+            Vector128.LessThan(maximum, max)
+            | ~Vector128.Equals(maximum, maximum)
+            | (Vector128.Equals(maximum, max) & (maximum.AsInt64() >> 63).AsDouble()),
+            maximum,
+            max);
+    }
+
+    /// <summary>
+    /// Clamps double-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="value">The values to clamp.</param>
+    /// <param name="min">The inclusive lower bounds.</param>
+    /// <param name="max">The inclusive upper bounds.</param>
+    /// <returns>The clamped values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector256<double> ClampDouble(
+        Vector256<double> value,
+        Vector256<double> min,
+        Vector256<double> max)
+    {
+        Vector256<double> maximum = Vector256.ConditionalSelect(
+            Vector256.LessThan(min, value)
+            | ~Vector256.Equals(value, value)
+            | (Vector256.Equals(value, min) & (min.AsInt64() >> 63).AsDouble()),
+            value,
+            min);
+
+        return Vector256.ConditionalSelect(
+            Vector256.LessThan(maximum, max)
+            | ~Vector256.Equals(maximum, maximum)
+            | (Vector256.Equals(maximum, max) & (maximum.AsInt64() >> 63).AsDouble()),
+            maximum,
+            max);
+    }
+
+    /// <summary>
+    /// Clamps double-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="value">The values to clamp.</param>
+    /// <param name="min">The inclusive lower bounds.</param>
+    /// <param name="max">The inclusive upper bounds.</param>
+    /// <returns>The clamped values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector512<double> ClampDouble(
+        Vector512<double> value,
+        Vector512<double> min,
+        Vector512<double> max)
+    {
+        Vector512<double> maximum = Vector512.ConditionalSelect(
+            Vector512.LessThan(min, value)
+            | ~Vector512.Equals(value, value)
+            | (Vector512.Equals(value, min) & (min.AsInt64() >> 63).AsDouble()),
+            value,
+            min);
+
+        return Vector512.ConditionalSelect(
+            Vector512.LessThan(maximum, max)
+            | ~Vector512.Equals(maximum, maximum)
+            | (Vector512.Equals(maximum, max) & (maximum.AsInt64() >> 63).AsDouble()),
+            maximum,
+            max);
+    }
+
+    /// <summary>
+    /// Clamps values using the complete runtime tensor contract, including signed-zero correction.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    private readonly struct ClampOperator<T> : ITernaryOperator<T>
+        where T : INumber<T>
+    {
+        /// <summary>
+        /// Gets a value indicating whether this operation supports vector execution.
+        /// </summary>
+        public static bool Vectorizable => true;
+
+        /// <summary>
+        /// Clamps a scalar value.
+        /// </summary>
+        /// <param name="x">The value.</param>
+        /// <param name="min">The inclusive lower bound.</param>
+        /// <param name="max">The inclusive upper bound.</param>
+        /// <returns>The clamped value.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static T Invoke(T x, T min, T max)
+            => Vector128<T>.IsSupported ? T.Min(T.Max(x, min), max) : T.Clamp(x, min, max);
+
+        /// <summary>
+        /// Clamps a 128-bit vector.
+        /// </summary>
+        /// <param name="x">The values.</param>
+        /// <param name="min">The inclusive lower bounds.</param>
+        /// <param name="max">The inclusive upper bounds.</param>
+        /// <returns>The clamped values.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> min, Vector128<T> max)
+        {
+            if (typeof(T) == typeof(float))
+            {
+                Vector128<float> result = ClampSingle(
+                    Unsafe.As<Vector128<T>, Vector128<float>>(ref x),
+                    Unsafe.As<Vector128<T>, Vector128<float>>(ref min),
+                    Unsafe.As<Vector128<T>, Vector128<float>>(ref max));
+
+                return Unsafe.As<Vector128<float>, Vector128<T>>(ref result);
+            }
+
+            if (typeof(T) == typeof(double))
+            {
+                Vector128<double> result = ClampDouble(
+                    Unsafe.As<Vector128<T>, Vector128<double>>(ref x),
+                    Unsafe.As<Vector128<T>, Vector128<double>>(ref min),
+                    Unsafe.As<Vector128<T>, Vector128<double>>(ref max));
+
+                return Unsafe.As<Vector128<double>, Vector128<T>>(ref result);
+            }
+
+            return Vector128_.Clamp(x, min, max);
+        }
+
+        /// <summary>
+        /// Clamps a 256-bit vector.
+        /// </summary>
+        /// <param name="x">The values.</param>
+        /// <param name="min">The inclusive lower bounds.</param>
+        /// <param name="max">The inclusive upper bounds.</param>
+        /// <returns>The clamped values.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> min, Vector256<T> max)
+        {
+            if (typeof(T) == typeof(float))
+            {
+                Vector256<float> result = ClampSingle(
+                    Unsafe.As<Vector256<T>, Vector256<float>>(ref x),
+                    Unsafe.As<Vector256<T>, Vector256<float>>(ref min),
+                    Unsafe.As<Vector256<T>, Vector256<float>>(ref max));
+
+                return Unsafe.As<Vector256<float>, Vector256<T>>(ref result);
+            }
+
+            if (typeof(T) == typeof(double))
+            {
+                Vector256<double> result = ClampDouble(
+                    Unsafe.As<Vector256<T>, Vector256<double>>(ref x),
+                    Unsafe.As<Vector256<T>, Vector256<double>>(ref min),
+                    Unsafe.As<Vector256<T>, Vector256<double>>(ref max));
+
+                return Unsafe.As<Vector256<double>, Vector256<T>>(ref result);
+            }
+
+            return Vector256_.Clamp(x, min, max);
+        }
+
+        /// <summary>
+        /// Clamps a 512-bit vector.
+        /// </summary>
+        /// <param name="x">The values.</param>
+        /// <param name="min">The inclusive lower bounds.</param>
+        /// <param name="max">The inclusive upper bounds.</param>
+        /// <returns>The clamped values.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> min, Vector512<T> max)
+        {
+            if (typeof(T) == typeof(float))
+            {
+                Vector512<float> result = ClampSingle(
+                    Unsafe.As<Vector512<T>, Vector512<float>>(ref x),
+                    Unsafe.As<Vector512<T>, Vector512<float>>(ref min),
+                    Unsafe.As<Vector512<T>, Vector512<float>>(ref max));
+
+                return Unsafe.As<Vector512<float>, Vector512<T>>(ref result);
+            }
+
+            if (typeof(T) == typeof(double))
+            {
+                Vector512<double> result = ClampDouble(
+                    Unsafe.As<Vector512<T>, Vector512<double>>(ref x),
+                    Unsafe.As<Vector512<T>, Vector512<double>>(ref min),
+                    Unsafe.As<Vector512<T>, Vector512<double>>(ref max));
+
+                return Unsafe.As<Vector512<double>, Vector512<T>>(ref result);
+            }
+
+            return Vector512_.Clamp(x, min, max);
+        }
+    }
+}
diff --git a/src/ImageSharp/Common/Helpers/TensorPrimitives_.Divide.cs b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Divide.cs
new file mode 100644
index 000000000..e40a26cea
--- /dev/null
+++ b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Divide.cs
@@ -0,0 +1,83 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.Intrinsics;
+
+namespace SixLabors.ImageSharp.Common.Helpers;
+
+internal static partial class TensorPrimitives_
+{
+    /// <summary>
+    /// Computes the element-wise result of dividing the values in <paramref name="x"/> by <paramref name="y"/>.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    /// <param name="x">The dividend values.</param>
+    /// <param name="y">The divisor.</param>
+    /// <param name="destination">The destination for the quotient values.</param>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public static void Divide<T>(ReadOnlySpan<T> x, T y, Span<T> destination)
+        where T : IDivisionOperators<T, T, T>
+        => InvokeSpanScalarIntoSpanForDivision<T, DivideOperator<T>>(x, y, destination);
+
+    /// <summary>
+    /// Determines whether <typeparamref name="T"/> has the same vector division support as <see cref="int"/>.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    /// <returns><see langword="true"/> when <typeparamref name="T"/> is a 32-bit signed native integer type.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static bool IsInt32Like<T>()
+        => typeof(T) == typeof(int) || (IntPtr.Size == 4 && typeof(T) == typeof(nint));
+
+    /// <summary>
+    /// Divides values by a scalar.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    private readonly struct DivideOperator<T> : IBinaryOperator<T>
+        where T : IDivisionOperators<T, T, T>
+    {
+        /// <summary>
+        /// Gets a value indicating whether this operation supports vector execution.
+        /// </summary>
+        public static bool Vectorizable => typeof(T) == typeof(float)
+                                        || typeof(T) == typeof(double)
+                                        || (Vector256.IsHardwareAccelerated && IsInt32Like<T>());
+
+        /// <summary>
+        /// Divides scalar values.
+        /// </summary>
+        /// <param name="x">The dividend.</param>
+        /// <param name="y">The divisor.</param>
+        /// <returns>The quotient.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static T Invoke(T x, T y) => x / y;
+
+        /// <summary>
+        /// Divides 128-bit vectors.
+        /// </summary>
+        /// <param name="x">The dividends.</param>
+        /// <param name="y">The divisors.</param>
+        /// <returns>The quotients.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> y) => x / y;
+
+        /// <summary>
+        /// Divides 256-bit vectors.
+        /// </summary>
+        /// <param name="x">The dividends.</param>
+        /// <param name="y">The divisors.</param>
+        /// <returns>The quotients.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> y) => x / y;
+
+        /// <summary>
+        /// Divides 512-bit vectors.
+        /// </summary>
+        /// <param name="x">The dividends.</param>
+        /// <param name="y">The divisors.</param>
+        /// <returns>The quotients.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> y) => x / y;
+    }
+}
diff --git a/src/ImageSharp/Common/Helpers/TensorPrimitives.cs b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Helpers.cs
similarity index 55%
rename from src/ImageSharp/Common/Helpers/TensorPrimitives.cs
rename to src/ImageSharp/Common/Helpers/TensorPrimitives_.Helpers.cs
index 4be4ab3aa..3ba1b61af 100644
--- a/src/ImageSharp/Common/Helpers/TensorPrimitives.cs
+++ b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Helpers.cs
@@ -15,9 +15,7 @@ namespace SixLabors.ImageSharp.Common.Helpers;
 /// The API shape follows <c>System.Numerics.Tensors.TensorPrimitives</c> so call sites can move to the runtime
 /// implementation when ImageSharp no longer supports target frameworks that predate it.
 /// </remarks>
-#pragma warning disable SA1649 // File name should match first type name
 internal static partial class TensorPrimitives_
-#pragma warning restore SA1649 // File name should match first type name
 {
     /// <summary>
     /// Defines an element-wise binary operation.
@@ -111,80 +109,6 @@ internal static partial class TensorPrimitives_
         public static abstract Vector512<T> Invoke(Vector512<T> x, Vector512<T> y, Vector512<T> z);
     }
 
-    /// <summary>
-    /// Computes the element-wise result of clamping <paramref name="x"/> to the inclusive range specified
-    /// by <paramref name="min"/> and <paramref name="max"/>.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    /// <param name="x">The values to clamp.</param>
-    /// <param name="min">The inclusive lower bound.</param>
-    /// <param name="max">The inclusive upper bound.</param>
-    /// <param name="destination">The destination for the clamped values.</param>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static void Clamp<T>(ReadOnlySpan<T> x, T min, T max, Span<T> destination)
-        where T : INumber<T>
-        => InvokeSpanScalarScalarIntoSpan<T, ClampOperator<T>>(x, min, max, destination);
-
-    /// <summary>
-    /// Computes the element-wise sum of the values in <paramref name="x"/> and <paramref name="y"/>.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    /// <param name="x">The first addends.</param>
-    /// <param name="y">The second addends.</param>
-    /// <param name="destination">The destination for the sums.</param>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static void Add<T>(ReadOnlySpan<T> x, ReadOnlySpan<T> y, Span<T> destination)
-        where T : IAdditionOperators<T, T, T>, IAdditiveIdentity<T, T>
-        => InvokeSpanSpanIntoSpan<T, AddOperator<T>>(x, y, destination);
-
-    /// <summary>
-    /// Computes the element-wise sum of the values in <paramref name="x"/> and the scalar <paramref name="y"/>.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    /// <param name="x">The first addends.</param>
-    /// <param name="y">The scalar second addend.</param>
-    /// <param name="destination">The destination for the sums.</param>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static void Add<T>(ReadOnlySpan<T> x, T y, Span<T> destination)
-        where T : IAdditionOperators<T, T, T>, IAdditiveIdentity<T, T>
-        => InvokeSpanScalarIntoSpan<T, AddOperator<T>>(x, y, destination);
-
-    /// <summary>
-    /// Computes the element-wise result of dividing the values in <paramref name="x"/> by <paramref name="y"/>.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    /// <param name="x">The dividend values.</param>
-    /// <param name="y">The divisor.</param>
-    /// <param name="destination">The destination for the quotient values.</param>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static void Divide<T>(ReadOnlySpan<T> x, T y, Span<T> destination)
-        where T : IDivisionOperators<T, T, T>
-        => InvokeSpanScalarIntoSpanForDivision<T, DivideOperator<T>>(x, y, destination);
-
-    /// <summary>
-    /// Computes the element-wise maximum of the values in <paramref name="x"/> and <paramref name="y"/>.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    /// <param name="x">The values to compare.</param>
-    /// <param name="y">The value to compare with each element.</param>
-    /// <param name="destination">The destination for the maximum values.</param>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static void Max<T>(ReadOnlySpan<T> x, T y, Span<T> destination)
-        where T : INumber<T>
-        => InvokeSpanScalarIntoSpan<T, MaxOperator<T>>(x, y, destination);
-
-    /// <summary>
-    /// Computes the element-wise product of the values in <paramref name="x"/> and <paramref name="y"/>.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    /// <param name="x">The multiplicands.</param>
-    /// <param name="y">The multiplier.</param>
-    /// <param name="destination">The destination for the products.</param>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    public static void Multiply<T>(ReadOnlySpan<T> x, T y, Span<T> destination)
-        where T : IMultiplyOperators<T, T, T>, IMultiplicativeIdentity<T, T>
-        => InvokeSpanScalarIntoSpan<T, MultiplyOperator<T>>(x, y, destination);
-
     /// <summary>
     /// Performs an element-wise binary operation between two spans.
     /// </summary>
@@ -204,15 +128,13 @@ internal static partial class TensorPrimitives_
         ref T yRef = ref MemoryMarshal.GetReference(y);
         ref T destinationRef = ref MemoryMarshal.GetReference(destination);
         nuint length = (uint)x.Length;
-        nuint vector512Threshold = Unsafe.SizeOf<T>() == 1 ? (uint)Vector512<T>.Count : 512;
 
-        // Match the runtime's AVX-512 selection for byte-sized elements once one complete vector is available.
-        // Wider elements retain the measured crossover point where their 512-bit setup cost becomes worthwhile.
+        // Runtime main selects the widest supported pipeline once one complete vector is available.
         // Each pipeline preloads its final inputs when a tail overlaps so same-start in-place operation remains correct.
         if (TOperator.Vectorizable
             && Vector512.IsHardwareAccelerated
             && Vector512<T>.IsSupported
-            && length >= vector512Threshold)
+            && length >= (uint)Vector512<T>.Count)
         {
             InvokeVectorized512<T, TOperator>(ref xRef, ref yRef, ref destinationRef, length);
             return;
@@ -255,12 +177,11 @@ internal static partial class TensorPrimitives_
         ref T destinationRef = ref MemoryMarshal.GetReference(destination);
         nuint length = (uint)x.Length;
 
-        // The runtime-style unrolled 512-bit pipeline wins on large inputs, but its setup cost regresses the
-        // shorter JPEG and ICC buffers. Measurements put the crossover safely below 512 elements.
+        // Runtime main selects the widest supported pipeline once one complete vector is available.
         if (TOperator.Vectorizable
             && Vector512.IsHardwareAccelerated
             && Vector512<T>.IsSupported
-            && length >= 512)
+            && length >= (uint)Vector512<T>.Count)
         {
             InvokeVectorized512<T, TOperator>(ref xRef, y, ref destinationRef, length);
             return;
@@ -285,7 +206,7 @@ internal static partial class TensorPrimitives_
     }
 
     /// <summary>
-    /// Performs element-wise division using thresholds measured for ImageSharp normalization workloads.
+    /// Performs element-wise division using the runtime tensor width-selection order.
     /// </summary>
     /// <typeparam name="T">The element type.</typeparam>
     /// <typeparam name="TOperator">The division operation to apply.</typeparam>
@@ -303,13 +224,11 @@ internal static partial class TensorPrimitives_
         ref T destinationRef = ref MemoryMarshal.GetReference(destination);
         nuint length = (uint)x.Length;
 
-        // AVX-512 only wins once there is enough work to amortize its wider dispatch and division latency.
-        // Eight vectors is also the runtime pipeline's unrolled-loop boundary, while shorter inputs retain
-        // the lower setup cost of 256-bit vectors.
+        // Runtime main selects the widest supported pipeline once one complete vector is available.
         if (TOperator.Vectorizable
             && Vector512.IsHardwareAccelerated
             && Vector512<T>.IsSupported
-            && length >= (uint)(Vector512<T>.Count * 8))
+            && length >= (uint)Vector512<T>.Count)
         {
             InvokeVectorized512<T, TOperator>(ref xRef, y, ref destinationRef, length);
             return;
@@ -726,116 +645,6 @@ internal static partial class TensorPrimitives_
         }
     }
 
-    /// <summary>
-    /// Selects maximum single-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="x">The first values.</param>
-    /// <param name="y">The second values.</param>
-    /// <returns>The maximum values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector128<float> MaxSingle(Vector128<float> x, Vector128<float> y)
-    {
-        // The .NET 8 operation already handles ordered unequal values. Correct its second-operand result for a
-        // first-operand NaN, then use bitwise AND for equal values so positive zero wins regardless of operand order.
-        Vector128<float> result = Vector128.Max(x, y);
-        result = Vector128.ConditionalSelect(~Vector128.Equals(x, x), x, result);
-
-        return Vector128.ConditionalSelect(
-            Vector128.Equals(x, y),
-            x & y,
-            result);
-    }
-
-    /// <summary>
-    /// Selects maximum single-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="x">The first values.</param>
-    /// <param name="y">The second values.</param>
-    /// <returns>The maximum values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector256<float> MaxSingle(Vector256<float> x, Vector256<float> y)
-    {
-        Vector256<float> result = Vector256.Max(x, y);
-        result = Vector256.ConditionalSelect(~Vector256.Equals(x, x), x, result);
-
-        return Vector256.ConditionalSelect(
-            Vector256.Equals(x, y),
-            x & y,
-            result);
-    }
-
-    /// <summary>
-    /// Selects maximum single-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="x">The first values.</param>
-    /// <param name="y">The second values.</param>
-    /// <returns>The maximum values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector512<float> MaxSingle(Vector512<float> x, Vector512<float> y)
-    {
-        Vector512<float> result = Vector512.Max(x, y);
-        result = Vector512.ConditionalSelect(~Vector512.Equals(x, x), x, result);
-
-        return Vector512.ConditionalSelect(
-            Vector512.Equals(x, y),
-            x & y,
-            result);
-    }
-
-    /// <summary>
-    /// Selects maximum double-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="x">The first values.</param>
-    /// <param name="y">The second values.</param>
-    /// <returns>The maximum values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector128<double> MaxDouble(Vector128<double> x, Vector128<double> y)
-    {
-        Vector128<double> result = Vector128.Max(x, y);
-        result = Vector128.ConditionalSelect(~Vector128.Equals(x, x), x, result);
-
-        return Vector128.ConditionalSelect(
-            Vector128.Equals(x, y),
-            x & y,
-            result);
-    }
-
-    /// <summary>
-    /// Selects maximum double-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="x">The first values.</param>
-    /// <param name="y">The second values.</param>
-    /// <returns>The maximum values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector256<double> MaxDouble(Vector256<double> x, Vector256<double> y)
-    {
-        Vector256<double> result = Vector256.Max(x, y);
-        result = Vector256.ConditionalSelect(~Vector256.Equals(x, x), x, result);
-
-        return Vector256.ConditionalSelect(
-            Vector256.Equals(x, y),
-            x & y,
-            result);
-    }
-
-    /// <summary>
-    /// Selects maximum double-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="x">The first values.</param>
-    /// <param name="y">The second values.</param>
-    /// <returns>The maximum values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector512<double> MaxDouble(Vector512<double> x, Vector512<double> y)
-    {
-        Vector512<double> result = Vector512.Max(x, y);
-        result = Vector512.ConditionalSelect(~Vector512.Equals(x, x), x, result);
-
-        return Vector512.ConditionalSelect(
-            Vector512.Equals(x, y),
-            x & y,
-            result);
-    }
-
     /// <summary>
     /// Applies a ternary operation with 128-bit vectors.
     /// </summary>
@@ -1012,567 +821,4 @@ internal static partial class TensorPrimitives_
             end.StoreUnsafe(ref destinationRef, length - vectorCount);
         }
     }
-
-    /// <summary>
-    /// Clamps single-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="value">The values to clamp.</param>
-    /// <param name="min">The inclusive lower bounds.</param>
-    /// <param name="max">The inclusive upper bounds.</param>
-    /// <returns>The clamped values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector128<float> ClampSingle(
-        Vector128<float> value,
-        Vector128<float> min,
-        Vector128<float> max)
-    {
-        // Unlike the native x86 min/max instructions, the normalized runtime operations propagate a NaN in the
-        // first operand and select negative zero when equal values have different signs.
-        Vector128<float> maximum = Vector128.ConditionalSelect(
-            Vector128.LessThan(min, value)
-            | ~Vector128.Equals(value, value)
-            | (Vector128.Equals(value, min) & (min.AsInt32() >> 31).AsSingle()),
-            value,
-            min);
-
-        return Vector128.ConditionalSelect(
-            Vector128.LessThan(maximum, max)
-            | ~Vector128.Equals(maximum, maximum)
-            | (Vector128.Equals(maximum, max) & (maximum.AsInt32() >> 31).AsSingle()),
-            maximum,
-            max);
-    }
-
-    /// <summary>
-    /// Clamps single-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="value">The values to clamp.</param>
-    /// <param name="min">The inclusive lower bounds.</param>
-    /// <param name="max">The inclusive upper bounds.</param>
-    /// <returns>The clamped values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector256<float> ClampSingle(
-        Vector256<float> value,
-        Vector256<float> min,
-        Vector256<float> max)
-    {
-        Vector256<float> maximum = Vector256.ConditionalSelect(
-            Vector256.LessThan(min, value)
-            | ~Vector256.Equals(value, value)
-            | (Vector256.Equals(value, min) & (min.AsInt32() >> 31).AsSingle()),
-            value,
-            min);
-
-        return Vector256.ConditionalSelect(
-            Vector256.LessThan(maximum, max)
-            | ~Vector256.Equals(maximum, maximum)
-            | (Vector256.Equals(maximum, max) & (maximum.AsInt32() >> 31).AsSingle()),
-            maximum,
-            max);
-    }
-
-    /// <summary>
-    /// Clamps single-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="value">The values to clamp.</param>
-    /// <param name="min">The inclusive lower bounds.</param>
-    /// <param name="max">The inclusive upper bounds.</param>
-    /// <returns>The clamped values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector512<float> ClampSingle(
-        Vector512<float> value,
-        Vector512<float> min,
-        Vector512<float> max)
-    {
-        Vector512<float> maximum = Vector512.ConditionalSelect(
-            Vector512.LessThan(min, value)
-            | ~Vector512.Equals(value, value)
-            | (Vector512.Equals(value, min) & (min.AsInt32() >> 31).AsSingle()),
-            value,
-            min);
-
-        return Vector512.ConditionalSelect(
-            Vector512.LessThan(maximum, max)
-            | ~Vector512.Equals(maximum, maximum)
-            | (Vector512.Equals(maximum, max) & (maximum.AsInt32() >> 31).AsSingle()),
-            maximum,
-            max);
-    }
-
-    /// <summary>
-    /// Clamps double-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="value">The values to clamp.</param>
-    /// <param name="min">The inclusive lower bounds.</param>
-    /// <param name="max">The inclusive upper bounds.</param>
-    /// <returns>The clamped values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector128<double> ClampDouble(
-        Vector128<double> value,
-        Vector128<double> min,
-        Vector128<double> max)
-    {
-        Vector128<double> maximum = Vector128.ConditionalSelect(
-            Vector128.LessThan(min, value)
-            | ~Vector128.Equals(value, value)
-            | (Vector128.Equals(value, min) & (min.AsInt64() >> 63).AsDouble()),
-            value,
-            min);
-
-        return Vector128.ConditionalSelect(
-            Vector128.LessThan(maximum, max)
-            | ~Vector128.Equals(maximum, maximum)
-            | (Vector128.Equals(maximum, max) & (maximum.AsInt64() >> 63).AsDouble()),
-            maximum,
-            max);
-    }
-
-    /// <summary>
-    /// Clamps double-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="value">The values to clamp.</param>
-    /// <param name="min">The inclusive lower bounds.</param>
-    /// <param name="max">The inclusive upper bounds.</param>
-    /// <returns>The clamped values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector256<double> ClampDouble(
-        Vector256<double> value,
-        Vector256<double> min,
-        Vector256<double> max)
-    {
-        Vector256<double> maximum = Vector256.ConditionalSelect(
-            Vector256.LessThan(min, value)
-            | ~Vector256.Equals(value, value)
-            | (Vector256.Equals(value, min) & (min.AsInt64() >> 63).AsDouble()),
-            value,
-            min);
-
-        return Vector256.ConditionalSelect(
-            Vector256.LessThan(maximum, max)
-            | ~Vector256.Equals(maximum, maximum)
-            | (Vector256.Equals(maximum, max) & (maximum.AsInt64() >> 63).AsDouble()),
-            maximum,
-            max);
-    }
-
-    /// <summary>
-    /// Clamps double-precision values with the normalized runtime semantics.
-    /// </summary>
-    /// <param name="value">The values to clamp.</param>
-    /// <param name="min">The inclusive lower bounds.</param>
-    /// <param name="max">The inclusive upper bounds.</param>
-    /// <returns>The clamped values.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static Vector512<double> ClampDouble(
-        Vector512<double> value,
-        Vector512<double> min,
-        Vector512<double> max)
-    {
-        Vector512<double> maximum = Vector512.ConditionalSelect(
-            Vector512.LessThan(min, value)
-            | ~Vector512.Equals(value, value)
-            | (Vector512.Equals(value, min) & (min.AsInt64() >> 63).AsDouble()),
-            value,
-            min);
-
-        return Vector512.ConditionalSelect(
-            Vector512.LessThan(maximum, max)
-            | ~Vector512.Equals(maximum, maximum)
-            | (Vector512.Equals(maximum, max) & (maximum.AsInt64() >> 63).AsDouble()),
-            maximum,
-            max);
-    }
-
-    /// <summary>
-    /// Determines whether <typeparamref name="T"/> has the same vector division support as <see cref="int"/>.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    /// <returns><see langword="true"/> when <typeparamref name="T"/> is a 32-bit signed native integer type.</returns>
-    [MethodImpl(MethodImplOptions.AggressiveInlining)]
-    private static bool IsInt32Like<T>()
-        => typeof(T) == typeof(int) || (IntPtr.Size == 4 && typeof(T) == typeof(nint));
-
-    /// <summary>
-    /// Adds corresponding values.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    private readonly struct AddOperator<T> : IBinaryOperator<T>
-        where T : IAdditionOperators<T, T, T>, IAdditiveIdentity<T, T>
-    {
-        /// <summary>
-        /// Gets a value indicating whether this operation supports vector execution.
-        /// </summary>
-        public static bool Vectorizable => true;
-
-        /// <summary>
-        /// Adds scalar values.
-        /// </summary>
-        /// <param name="x">The first addend.</param>
-        /// <param name="y">The second addend.</param>
-        /// <returns>The sum.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static T Invoke(T x, T y) => x + y;
-
-        /// <summary>
-        /// Adds 128-bit vectors.
-        /// </summary>
-        /// <param name="x">The first addends.</param>
-        /// <param name="y">The second addends.</param>
-        /// <returns>The sums.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> y) => x + y;
-
-        /// <summary>
-        /// Adds 256-bit vectors.
-        /// </summary>
-        /// <param name="x">The first addends.</param>
-        /// <param name="y">The second addends.</param>
-        /// <returns>The sums.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> y) => x + y;
-
-        /// <summary>
-        /// Adds 512-bit vectors.
-        /// </summary>
-        /// <param name="x">The first addends.</param>
-        /// <param name="y">The second addends.</param>
-        /// <returns>The sums.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> y) => x + y;
-    }
-
-    /// <summary>
-    /// Clamps values using the complete runtime tensor contract, including signed-zero correction.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    private readonly struct ClampOperator<T> : ITernaryOperator<T>
-        where T : INumber<T>
-    {
-        /// <summary>
-        /// Gets a value indicating whether this operation supports vector execution.
-        /// </summary>
-        public static bool Vectorizable => true;
-
-        /// <summary>
-        /// Clamps a scalar value.
-        /// </summary>
-        /// <param name="x">The value.</param>
-        /// <param name="min">The inclusive lower bound.</param>
-        /// <param name="max">The inclusive upper bound.</param>
-        /// <returns>The clamped value.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static T Invoke(T x, T min, T max)
-            => Vector128<T>.IsSupported ? T.Min(T.Max(x, min), max) : T.Clamp(x, min, max);
-
-        /// <summary>
-        /// Clamps a 128-bit vector.
-        /// </summary>
-        /// <param name="x">The values.</param>
-        /// <param name="min">The inclusive lower bounds.</param>
-        /// <param name="max">The inclusive upper bounds.</param>
-        /// <returns>The clamped values.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> min, Vector128<T> max)
-        {
-            if (typeof(T) == typeof(float))
-            {
-                Vector128<float> result = ClampSingle(
-                    Unsafe.As<Vector128<T>, Vector128<float>>(ref x),
-                    Unsafe.As<Vector128<T>, Vector128<float>>(ref min),
-                    Unsafe.As<Vector128<T>, Vector128<float>>(ref max));
-
-                return Unsafe.As<Vector128<float>, Vector128<T>>(ref result);
-            }
-
-            if (typeof(T) == typeof(double))
-            {
-                Vector128<double> result = ClampDouble(
-                    Unsafe.As<Vector128<T>, Vector128<double>>(ref x),
-                    Unsafe.As<Vector128<T>, Vector128<double>>(ref min),
-                    Unsafe.As<Vector128<T>, Vector128<double>>(ref max));
-
-                return Unsafe.As<Vector128<double>, Vector128<T>>(ref result);
-            }
-
-            return Vector128_.Clamp(x, min, max);
-        }
-
-        /// <summary>
-        /// Clamps a 256-bit vector.
-        /// </summary>
-        /// <param name="x">The values.</param>
-        /// <param name="min">The inclusive lower bounds.</param>
-        /// <param name="max">The inclusive upper bounds.</param>
-        /// <returns>The clamped values.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> min, Vector256<T> max)
-        {
-            if (typeof(T) == typeof(float))
-            {
-                Vector256<float> result = ClampSingle(
-                    Unsafe.As<Vector256<T>, Vector256<float>>(ref x),
-                    Unsafe.As<Vector256<T>, Vector256<float>>(ref min),
-                    Unsafe.As<Vector256<T>, Vector256<float>>(ref max));
-
-                return Unsafe.As<Vector256<float>, Vector256<T>>(ref result);
-            }
-
-            if (typeof(T) == typeof(double))
-            {
-                Vector256<double> result = ClampDouble(
-                    Unsafe.As<Vector256<T>, Vector256<double>>(ref x),
-                    Unsafe.As<Vector256<T>, Vector256<double>>(ref min),
-                    Unsafe.As<Vector256<T>, Vector256<double>>(ref max));
-
-                return Unsafe.As<Vector256<double>, Vector256<T>>(ref result);
-            }
-
-            return Vector256_.Clamp(x, min, max);
-        }
-
-        /// <summary>
-        /// Clamps a 512-bit vector.
-        /// </summary>
-        /// <param name="x">The values.</param>
-        /// <param name="min">The inclusive lower bounds.</param>
-        /// <param name="max">The inclusive upper bounds.</param>
-        /// <returns>The clamped values.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> min, Vector512<T> max)
-        {
-            if (typeof(T) == typeof(float))
-            {
-                Vector512<float> result = ClampSingle(
-                    Unsafe.As<Vector512<T>, Vector512<float>>(ref x),
-                    Unsafe.As<Vector512<T>, Vector512<float>>(ref min),
-                    Unsafe.As<Vector512<T>, Vector512<float>>(ref max));
-
-                return Unsafe.As<Vector512<float>, Vector512<T>>(ref result);
-            }
-
-            if (typeof(T) == typeof(double))
-            {
-                Vector512<double> result = ClampDouble(
-                    Unsafe.As<Vector512<T>, Vector512<double>>(ref x),
-                    Unsafe.As<Vector512<T>, Vector512<double>>(ref min),
-                    Unsafe.As<Vector512<T>, Vector512<double>>(ref max));
-
-                return Unsafe.As<Vector512<double>, Vector512<T>>(ref result);
-            }
-
-            return Vector512_.Clamp(x, min, max);
-        }
-    }
-
-    /// <summary>
-    /// Selects the maximum corresponding values.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    private readonly struct MaxOperator<T> : IBinaryOperator<T>
-        where T : INumber<T>
-    {
-        /// <summary>
-        /// Gets a value indicating whether this operation supports vector execution.
-        /// </summary>
-        public static bool Vectorizable => true;
-
-        /// <summary>
-        /// Selects the maximum scalar value.
-        /// </summary>
-        /// <param name="x">The first value.</param>
-        /// <param name="y">The second value.</param>
-        /// <returns>The maximum value.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static T Invoke(T x, T y) => T.Max(x, y);
-
-        /// <summary>
-        /// Selects the maximum values from 128-bit vectors.
-        /// </summary>
-        /// <param name="x">The first values.</param>
-        /// <param name="y">The second values.</param>
-        /// <returns>The maximum values.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> y)
-        {
-            if (typeof(T) == typeof(float))
-            {
-                Vector128<float> result = MaxSingle(
-                    Unsafe.As<Vector128<T>, Vector128<float>>(ref x),
-                    Unsafe.As<Vector128<T>, Vector128<float>>(ref y));
-
-                return Unsafe.As<Vector128<float>, Vector128<T>>(ref result);
-            }
-
-            if (typeof(T) == typeof(double))
-            {
-                Vector128<double> result = MaxDouble(
-                    Unsafe.As<Vector128<T>, Vector128<double>>(ref x),
-                    Unsafe.As<Vector128<T>, Vector128<double>>(ref y));
-
-                return Unsafe.As<Vector128<double>, Vector128<T>>(ref result);
-            }
-
-            return Vector128.Max(x, y);
-        }
-
-        /// <summary>
-        /// Selects the maximum values from 256-bit vectors.
-        /// </summary>
-        /// <param name="x">The first values.</param>
-        /// <param name="y">The second values.</param>
-        /// <returns>The maximum values.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> y)
-        {
-            if (typeof(T) == typeof(float))
-            {
-                Vector256<float> result = MaxSingle(
-                    Unsafe.As<Vector256<T>, Vector256<float>>(ref x),
-                    Unsafe.As<Vector256<T>, Vector256<float>>(ref y));
-
-                return Unsafe.As<Vector256<float>, Vector256<T>>(ref result);
-            }
-
-            if (typeof(T) == typeof(double))
-            {
-                Vector256<double> result = MaxDouble(
-                    Unsafe.As<Vector256<T>, Vector256<double>>(ref x),
-                    Unsafe.As<Vector256<T>, Vector256<double>>(ref y));
-
-                return Unsafe.As<Vector256<double>, Vector256<T>>(ref result);
-            }
-
-            return Vector256.Max(x, y);
-        }
-
-        /// <summary>
-        /// Selects the maximum values from 512-bit vectors.
-        /// </summary>
-        /// <param name="x">The first values.</param>
-        /// <param name="y">The second values.</param>
-        /// <returns>The maximum values.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> y)
-        {
-            if (typeof(T) == typeof(float))
-            {
-                Vector512<float> result = MaxSingle(
-                    Unsafe.As<Vector512<T>, Vector512<float>>(ref x),
-                    Unsafe.As<Vector512<T>, Vector512<float>>(ref y));
-
-                return Unsafe.As<Vector512<float>, Vector512<T>>(ref result);
-            }
-
-            if (typeof(T) == typeof(double))
-            {
-                Vector512<double> result = MaxDouble(
-                    Unsafe.As<Vector512<T>, Vector512<double>>(ref x),
-                    Unsafe.As<Vector512<T>, Vector512<double>>(ref y));
-
-                return Unsafe.As<Vector512<double>, Vector512<T>>(ref result);
-            }
-
-            return Vector512.Max(x, y);
-        }
-    }
-
-    /// <summary>
-    /// Multiplies corresponding values.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    private readonly struct MultiplyOperator<T> : IBinaryOperator<T>
-        where T : IMultiplyOperators<T, T, T>, IMultiplicativeIdentity<T, T>
-    {
-        /// <summary>
-        /// Gets a value indicating whether this operation supports vector execution.
-        /// </summary>
-        public static bool Vectorizable => true;
-
-        /// <summary>
-        /// Multiplies scalar values.
-        /// </summary>
-        /// <param name="x">The multiplicand.</param>
-        /// <param name="y">The multiplier.</param>
-        /// <returns>The product.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static T Invoke(T x, T y) => x * y;
-
-        /// <summary>
-        /// Multiplies 128-bit vectors.
-        /// </summary>
-        /// <param name="x">The multiplicands.</param>
-        /// <param name="y">The multipliers.</param>
-        /// <returns>The products.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> y) => x * y;
-
-        /// <summary>
-        /// Multiplies 256-bit vectors.
-        /// </summary>
-        /// <param name="x">The multiplicands.</param>
-        /// <param name="y">The multipliers.</param>
-        /// <returns>The products.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> y) => x * y;
-
-        /// <summary>
-        /// Multiplies 512-bit vectors.
-        /// </summary>
-        /// <param name="x">The multiplicands.</param>
-        /// <param name="y">The multipliers.</param>
-        /// <returns>The products.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> y) => x * y;
-    }
-
-    /// <summary>
-    /// Divides values by a scalar.
-    /// </summary>
-    /// <typeparam name="T">The element type.</typeparam>
-    private readonly struct DivideOperator<T> : IBinaryOperator<T>
-        where T : IDivisionOperators<T, T, T>
-    {
-        /// <summary>
-        /// Gets a value indicating whether this operation supports vector execution.
-        /// </summary>
-        public static bool Vectorizable => typeof(T) == typeof(float)
-                                        || typeof(T) == typeof(double)
-                                        || (Vector256.IsHardwareAccelerated && IsInt32Like<T>());
-
-        /// <summary>
-        /// Divides scalar values.
-        /// </summary>
-        /// <param name="x">The dividend.</param>
-        /// <param name="y">The divisor.</param>
-        /// <returns>The quotient.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static T Invoke(T x, T y) => x / y;
-
-        /// <summary>
-        /// Divides 128-bit vectors.
-        /// </summary>
-        /// <param name="x">The dividends.</param>
-        /// <param name="y">The divisors.</param>
-        /// <returns>The quotients.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> y) => x / y;
-
-        /// <summary>
-        /// Divides 256-bit vectors.
-        /// </summary>
-        /// <param name="x">The dividends.</param>
-        /// <param name="y">The divisors.</param>
-        /// <returns>The quotients.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> y) => x / y;
-
-        /// <summary>
-        /// Divides 512-bit vectors.
-        /// </summary>
-        /// <param name="x">The dividends.</param>
-        /// <param name="y">The divisors.</param>
-        /// <returns>The quotients.</returns>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> y) => x / y;
-    }
 }
diff --git a/src/ImageSharp/Common/Helpers/TensorPrimitives_.Max.cs b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Max.cs
new file mode 100644
index 000000000..31312dddf
--- /dev/null
+++ b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Max.cs
@@ -0,0 +1,245 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.Intrinsics;
+
+namespace SixLabors.ImageSharp.Common.Helpers;
+
+internal static partial class TensorPrimitives_
+{
+    /// <summary>
+    /// Computes the element-wise maximum of the values in <paramref name="x"/> and <paramref name="y"/>.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    /// <param name="x">The values to compare.</param>
+    /// <param name="y">The value to compare with each element.</param>
+    /// <param name="destination">The destination for the maximum values.</param>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public static void Max<T>(ReadOnlySpan<T> x, T y, Span<T> destination)
+        where T : INumber<T>
+        => InvokeSpanScalarIntoSpan<T, MaxOperator<T>>(x, y, destination);
+
+    /// <summary>
+    /// Selects maximum single-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="x">The first values.</param>
+    /// <param name="y">The second values.</param>
+    /// <returns>The maximum values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector128<float> MaxSingle(Vector128<float> x, Vector128<float> y)
+    {
+        // The .NET 8 operation already handles ordered unequal values. Correct its second-operand result for a
+        // first-operand NaN, then use bitwise AND for equal values so positive zero wins regardless of operand order.
+        Vector128<float> result = Vector128.Max(x, y);
+        result = Vector128.ConditionalSelect(~Vector128.Equals(x, x), x, result);
+
+        return Vector128.ConditionalSelect(
+            Vector128.Equals(x, y),
+            x & y,
+            result);
+    }
+
+    /// <summary>
+    /// Selects maximum single-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="x">The first values.</param>
+    /// <param name="y">The second values.</param>
+    /// <returns>The maximum values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector256<float> MaxSingle(Vector256<float> x, Vector256<float> y)
+    {
+        Vector256<float> result = Vector256.Max(x, y);
+        result = Vector256.ConditionalSelect(~Vector256.Equals(x, x), x, result);
+
+        return Vector256.ConditionalSelect(
+            Vector256.Equals(x, y),
+            x & y,
+            result);
+    }
+
+    /// <summary>
+    /// Selects maximum single-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="x">The first values.</param>
+    /// <param name="y">The second values.</param>
+    /// <returns>The maximum values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector512<float> MaxSingle(Vector512<float> x, Vector512<float> y)
+    {
+        Vector512<float> result = Vector512.Max(x, y);
+        result = Vector512.ConditionalSelect(~Vector512.Equals(x, x), x, result);
+
+        return Vector512.ConditionalSelect(
+            Vector512.Equals(x, y),
+            x & y,
+            result);
+    }
+
+    /// <summary>
+    /// Selects maximum double-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="x">The first values.</param>
+    /// <param name="y">The second values.</param>
+    /// <returns>The maximum values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector128<double> MaxDouble(Vector128<double> x, Vector128<double> y)
+    {
+        Vector128<double> result = Vector128.Max(x, y);
+        result = Vector128.ConditionalSelect(~Vector128.Equals(x, x), x, result);
+
+        return Vector128.ConditionalSelect(
+            Vector128.Equals(x, y),
+            x & y,
+            result);
+    }
+
+    /// <summary>
+    /// Selects maximum double-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="x">The first values.</param>
+    /// <param name="y">The second values.</param>
+    /// <returns>The maximum values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector256<double> MaxDouble(Vector256<double> x, Vector256<double> y)
+    {
+        Vector256<double> result = Vector256.Max(x, y);
+        result = Vector256.ConditionalSelect(~Vector256.Equals(x, x), x, result);
+
+        return Vector256.ConditionalSelect(
+            Vector256.Equals(x, y),
+            x & y,
+            result);
+    }
+
+    /// <summary>
+    /// Selects maximum double-precision values with the normalized runtime semantics.
+    /// </summary>
+    /// <param name="x">The first values.</param>
+    /// <param name="y">The second values.</param>
+    /// <returns>The maximum values.</returns>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static Vector512<double> MaxDouble(Vector512<double> x, Vector512<double> y)
+    {
+        Vector512<double> result = Vector512.Max(x, y);
+        result = Vector512.ConditionalSelect(~Vector512.Equals(x, x), x, result);
+
+        return Vector512.ConditionalSelect(
+            Vector512.Equals(x, y),
+            x & y,
+            result);
+    }
+
+    /// <summary>
+    /// Selects the maximum corresponding values.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    private readonly struct MaxOperator<T> : IBinaryOperator<T>
+        where T : INumber<T>
+    {
+        /// <summary>
+        /// Gets a value indicating whether this operation supports vector execution.
+        /// </summary>
+        public static bool Vectorizable => true;
+
+        /// <summary>
+        /// Selects the maximum scalar value.
+        /// </summary>
+        /// <param name="x">The first value.</param>
+        /// <param name="y">The second value.</param>
+        /// <returns>The maximum value.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static T Invoke(T x, T y) => T.Max(x, y);
+
+        /// <summary>
+        /// Selects the maximum values from 128-bit vectors.
+        /// </summary>
+        /// <param name="x">The first values.</param>
+        /// <param name="y">The second values.</param>
+        /// <returns>The maximum values.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> y)
+        {
+            if (typeof(T) == typeof(float))
+            {
+                Vector128<float> result = MaxSingle(
+                    Unsafe.As<Vector128<T>, Vector128<float>>(ref x),
+                    Unsafe.As<Vector128<T>, Vector128<float>>(ref y));
+
+                return Unsafe.As<Vector128<float>, Vector128<T>>(ref result);
+            }
+
+            if (typeof(T) == typeof(double))
+            {
+                Vector128<double> result = MaxDouble(
+                    Unsafe.As<Vector128<T>, Vector128<double>>(ref x),
+                    Unsafe.As<Vector128<T>, Vector128<double>>(ref y));
+
+                return Unsafe.As<Vector128<double>, Vector128<T>>(ref result);
+            }
+
+            return Vector128.Max(x, y);
+        }
+
+        /// <summary>
+        /// Selects the maximum values from 256-bit vectors.
+        /// </summary>
+        /// <param name="x">The first values.</param>
+        /// <param name="y">The second values.</param>
+        /// <returns>The maximum values.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> y)
+        {
+            if (typeof(T) == typeof(float))
+            {
+                Vector256<float> result = MaxSingle(
+                    Unsafe.As<Vector256<T>, Vector256<float>>(ref x),
+                    Unsafe.As<Vector256<T>, Vector256<float>>(ref y));
+
+                return Unsafe.As<Vector256<float>, Vector256<T>>(ref result);
+            }
+
+            if (typeof(T) == typeof(double))
+            {
+                Vector256<double> result = MaxDouble(
+                    Unsafe.As<Vector256<T>, Vector256<double>>(ref x),
+                    Unsafe.As<Vector256<T>, Vector256<double>>(ref y));
+
+                return Unsafe.As<Vector256<double>, Vector256<T>>(ref result);
+            }
+
+            return Vector256.Max(x, y);
+        }
+
+        /// <summary>
+        /// Selects the maximum values from 512-bit vectors.
+        /// </summary>
+        /// <param name="x">The first values.</param>
+        /// <param name="y">The second values.</param>
+        /// <returns>The maximum values.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> y)
+        {
+            if (typeof(T) == typeof(float))
+            {
+                Vector512<float> result = MaxSingle(
+                    Unsafe.As<Vector512<T>, Vector512<float>>(ref x),
+                    Unsafe.As<Vector512<T>, Vector512<float>>(ref y));
+
+                return Unsafe.As<Vector512<float>, Vector512<T>>(ref result);
+            }
+
+            if (typeof(T) == typeof(double))
+            {
+                Vector512<double> result = MaxDouble(
+                    Unsafe.As<Vector512<T>, Vector512<double>>(ref x),
+                    Unsafe.As<Vector512<T>, Vector512<double>>(ref y));
+
+                return Unsafe.As<Vector512<double>, Vector512<T>>(ref result);
+            }
+
+            return Vector512.Max(x, y);
+        }
+    }
+}
diff --git a/src/ImageSharp/Common/Helpers/TensorPrimitives_.Multiply.cs b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Multiply.cs
new file mode 100644
index 000000000..deb1e922e
--- /dev/null
+++ b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Multiply.cs
@@ -0,0 +1,72 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.Intrinsics;
+
+namespace SixLabors.ImageSharp.Common.Helpers;
+
+internal static partial class TensorPrimitives_
+{
+    /// <summary>
+    /// Computes the element-wise product of the values in <paramref name="x"/> and <paramref name="y"/>.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    /// <param name="x">The multiplicands.</param>
+    /// <param name="y">The multiplier.</param>
+    /// <param name="destination">The destination for the products.</param>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    public static void Multiply<T>(ReadOnlySpan<T> x, T y, Span<T> destination)
+        where T : IMultiplyOperators<T, T, T>, IMultiplicativeIdentity<T, T>
+        => InvokeSpanScalarIntoSpan<T, MultiplyOperator<T>>(x, y, destination);
+
+    /// <summary>
+    /// Multiplies corresponding values.
+    /// </summary>
+    /// <typeparam name="T">The element type.</typeparam>
+    private readonly struct MultiplyOperator<T> : IBinaryOperator<T>
+        where T : IMultiplyOperators<T, T, T>, IMultiplicativeIdentity<T, T>
+    {
+        /// <summary>
+        /// Gets a value indicating whether this operation supports vector execution.
+        /// </summary>
+        public static bool Vectorizable => true;
+
+        /// <summary>
+        /// Multiplies scalar values.
+        /// </summary>
+        /// <param name="x">The multiplicand.</param>
+        /// <param name="y">The multiplier.</param>
+        /// <returns>The product.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static T Invoke(T x, T y) => x * y;
+
+        /// <summary>
+        /// Multiplies 128-bit vectors.
+        /// </summary>
+        /// <param name="x">The multiplicands.</param>
+        /// <param name="y">The multipliers.</param>
+        /// <returns>The products.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector128<T> Invoke(Vector128<T> x, Vector128<T> y) => x * y;
+
+        /// <summary>
+        /// Multiplies 256-bit vectors.
+        /// </summary>
+        /// <param name="x">The multiplicands.</param>
+        /// <param name="y">The multipliers.</param>
+        /// <returns>The products.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector256<T> Invoke(Vector256<T> x, Vector256<T> y) => x * y;
+
+        /// <summary>
+        /// Multiplies 512-bit vectors.
+        /// </summary>
+        /// <param name="x">The multiplicands.</param>
+        /// <param name="y">The multipliers.</param>
+        /// <returns>The products.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector512<T> Invoke(Vector512<T> x, Vector512<T> y) => x * y;
+    }
+}
diff --git a/src/ImageSharp/Common/Helpers/TensorPrimitives_.Negate.cs b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Negate.cs
index f61e4877c..eb90a3903 100644
--- a/src/ImageSharp/Common/Helpers/TensorPrimitives_.Negate.cs
+++ b/src/ImageSharp/Common/Helpers/TensorPrimitives_.Negate.cs
@@ -76,12 +76,11 @@ internal static partial class TensorPrimitives_
         ref T destinationRef = ref MemoryMarshal.GetReference(destination);
         nuint length = (uint)x.Length;
 
-        // The dispatch matches the other compatibility pipelines: AVX-512 is reserved for large spans because
-        // its setup cost is not recovered by the short image-processing buffers that dominate ImageSharp.
+        // Runtime main selects the widest supported pipeline once one complete vector is available.
         if (TOperator.Vectorizable
             && Vector512.IsHardwareAccelerated
             && Vector512<T>.IsSupported
-            && length >= 512)
+            && length >= (uint)Vector512<T>.Count)
         {
             InvokeUnaryVectorized512<T, TOperator>(ref xRef, ref destinationRef, length);
             return;