Remove restrictions from vector utilities

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

@@ -66,9 +66,9 @@ internal static partial class SimdUtils
             ref Span<float> destination,
             [ConstantExpected] byte control)
         {
-            if ((Vector512.IsHardwareAccelerated && Vector512_.SupportsShuffleNativeFloat) ||
-                (Vector256.IsHardwareAccelerated && Vector256_.SupportsShuffleNativeFloat) ||
-                 Vector128.IsHardwareAccelerated)
+            if (Vector512.IsHardwareAccelerated ||
+                Vector256.IsHardwareAccelerated ||
+                Vector128.IsHardwareAccelerated)
             {
                 int remainder = 0;
                 if (Vector512.IsHardwareAccelerated)
@@ -112,8 +112,8 @@ internal static partial class SimdUtils
             ref Span<byte> destination,
             [ConstantExpected] byte control)
         {
-            if ((Vector512.IsHardwareAccelerated && Vector512_.SupportsShuffleNativeByte) ||
-                (Vector256.IsHardwareAccelerated && Vector256_.SupportsShuffleNativeByte) ||
+            if (Vector512.IsHardwareAccelerated ||
+                Vector256.IsHardwareAccelerated ||
                 Vector128.IsHardwareAccelerated)
             {
                 int remainder = 0;
@@ -249,7 +249,7 @@ internal static partial class SimdUtils
             Span<float> destination,
             [ConstantExpected] byte control)
         {
-            if (Vector512.IsHardwareAccelerated && Vector512_.SupportsShuffleNativeFloat)
+            if (Vector512.IsHardwareAccelerated)
             {
                 ref Vector512<float> sourceBase = ref Unsafe.As<float, Vector512<float>>(ref MemoryMarshal.GetReference(source));
                 ref Vector512<float> destinationBase = ref Unsafe.As<float, Vector512<float>>(ref MemoryMarshal.GetReference(destination));
@@ -277,7 +277,7 @@ internal static partial class SimdUtils
                     }
                 }
             }
-            else if (Vector256.IsHardwareAccelerated && Vector256_.SupportsShuffleNativeFloat)
+            else if (Vector256.IsHardwareAccelerated)
             {
                 ref Vector256<float> sourceBase = ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(source));
                 ref Vector256<float> destinationBase = ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(destination));
@@ -341,7 +341,7 @@ internal static partial class SimdUtils
             Span<byte> destination,
             [ConstantExpected] byte control)
         {
-            if (Vector512.IsHardwareAccelerated && Vector512_.SupportsShuffleNativeByte)
+            if (Vector512.IsHardwareAccelerated)
             {
                 Span<byte> temp = stackalloc byte[Vector512<byte>.Count];
                 Shuffle.MMShuffleSpan(ref temp, control);
@@ -373,7 +373,7 @@ internal static partial class SimdUtils
                     }
                 }
             }
-            else if (Vector256.IsHardwareAccelerated && Vector256_.SupportsShuffleNativeByte)
+            else if (Vector256.IsHardwareAccelerated)
             {
                 Span<byte> temp = stackalloc byte[Vector256<byte>.Count];
                 Shuffle.MMShuffleSpan(ref temp, control);

src/ImageSharp/Common/Helpers/Vector128Utilities.cs

@@ -1323,7 +1323,9 @@ internal static class Vector128_
         {
             // https://stackoverflow.com/questions/11870910/sse-mm-movemask-epi8-equivalent-method-for-arm-neon
             Vector128<byte> powers = Vector128.Create(1, 2, 4, 8, 16, 32, 64, 128, 1, 2, 4, 8, 16, 32, 64, 128);
-            Vector128<byte> masked = value & powers;
+            Vector128<byte> msbMask = Vector128.Create((byte)0x80);
+            Vector128<byte> normalized = AdvSimd.CompareEqual(value & msbMask, msbMask); // 0xFF or 0x00
+            Vector128<byte> masked = normalized & powers;
 
             Vector128<ushort> sum8 = AdvSimd.AddPairwiseWidening(masked);
             Vector128<uint> sum16 = AdvSimd.AddPairwiseWidening(sum8);

src/ImageSharp/Common/Helpers/Vector256Utilities.cs

@@ -1,7 +1,6 @@
 // Copyright (c) Six Labors.
 // Licensed under the Six Labors Split License.
 
-using System.Diagnostics;
 using System.Diagnostics.CodeAnalysis;
 using System.Runtime.CompilerServices;
 using System.Runtime.Intrinsics;
@@ -21,24 +20,6 @@ namespace SixLabors.ImageSharp.Common.Helpers;
 internal static class Vector256_
 #pragma warning restore SA1649 // File name should match first type name
 {
-    /// <summary>
-    /// Gets a value indicating whether shuffle byte operations are supported.
-    /// </summary>
-    public static bool SupportsShuffleNativeFloat
-    {
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        get => Avx.IsSupported;
-    }
-
-    /// <summary>
-    /// Gets a value indicating whether shuffle byte operations are supported.
-    /// </summary>
-    public static bool SupportsShuffleNativeByte
-    {
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        get => Avx2.IsSupported;
-    }
-
     /// <summary>
     /// Creates a new vector by selecting values from an input vector using a set of indices.
     /// </summary>
@@ -47,15 +28,7 @@ internal static class Vector256_
     /// <returns>The <see cref="Vector256{Single}"/>.</returns>
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     public static Vector256<float> ShuffleNative(Vector256<float> vector, [ConstantExpected] byte control)
-    {
-        if (Avx.IsSupported)
-        {
-            return Avx.Shuffle(vector, vector, control);
-        }
-
-        ThrowUnreachableException();
-        return default;
-    }
+        => Avx.Shuffle(vector, vector, control);
 
     /// <summary>
     /// Creates a new vector by selecting values from an input vector using a set of indices.</summary>
@@ -73,8 +46,9 @@ internal static class Vector256_
             return Avx2.Shuffle(vector, indices);
         }
 
-        ThrowUnreachableException();
-        return default;
+        return Vector256.Create(
+            Vector128_.ShuffleNative(vector.GetLower(), indices.GetLower()),
+            Vector128_.ShuffleNative(vector.GetUpper(), indices.GetUpper()));
     }
 
     /// <summary>
@@ -508,7 +482,4 @@ internal static class Vector256_
         int hiMask = Vector128_.MoveMask(value.GetUpper());
         return loMask | (hiMask << 16);
     }
-
-    [DoesNotReturn]
-    private static void ThrowUnreachableException() => throw new UnreachableException();
 }

src/ImageSharp/Common/Helpers/Vector512Utilities.cs

@@ -1,7 +1,6 @@
 // Copyright (c) Six Labors.
 // Licensed under the Six Labors Split License.
 
-using System.Diagnostics;
 using System.Diagnostics.CodeAnalysis;
 using System.Runtime.CompilerServices;
 using System.Runtime.Intrinsics;
@@ -21,24 +20,6 @@ namespace SixLabors.ImageSharp.Common.Helpers;
 internal static class Vector512_
 #pragma warning restore SA1649 // File name should match first type name
 {
-    /// <summary>
-    /// Gets a value indicating whether shuffle float operations are supported.
-    /// </summary>
-    public static bool SupportsShuffleNativeFloat
-    {
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        get => Avx512F.IsSupported;
-    }
-
-    /// <summary>
-    /// Gets a value indicating whether shuffle byte operations are supported.
-    /// </summary>
-    public static bool SupportsShuffleNativeByte
-    {
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        get => Avx512BW.IsSupported;
-    }
-
     /// <summary>
     /// Creates a new vector by selecting values from an input vector using the control.
     /// </summary>
@@ -47,15 +28,7 @@ internal static class Vector512_
     /// <returns>The <see cref="Vector512{Single}"/>.</returns>
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     public static Vector512<float> ShuffleNative(Vector512<float> vector, [ConstantExpected] byte control)
-    {
-        if (Avx512F.IsSupported)
-        {
-            return Avx512F.Shuffle(vector, vector, control);
-        }
-
-        ThrowUnreachableException();
-        return default;
-    }
+        => Avx512F.Shuffle(vector, vector, control);
 
     /// <summary>
     /// Creates a new vector by selecting values from an input vector using a set of indices.
@@ -73,8 +46,9 @@ internal static class Vector512_
             return Avx512BW.Shuffle(vector, indices);
         }
 
-        ThrowUnreachableException();
-        return default;
+        return Vector512.Create(
+            Vector256_.ShuffleNative(vector.GetLower(), indices.GetLower()),
+            Vector256_.ShuffleNative(vector.GetUpper(), indices.GetUpper()));
     }
 
     /// <summary>
@@ -175,7 +149,4 @@ internal static class Vector512_
     [MethodImpl(MethodImplOptions.AggressiveInlining)]
     public static Vector512<T> Clamp<T>(Vector512<T> value, Vector512<T> min, Vector512<T> max)
         => Vector512.Min(Vector512.Max(value, min), max);
-
-    [DoesNotReturn]
-    private static void ThrowUnreachableException() => throw new UnreachableException();
 }

src/ImageSharp/Formats/Webp/Lossless/ColorSpaceTransformUtils.cs

@@ -12,7 +12,7 @@ internal static class ColorSpaceTransformUtils
 {
     public static void CollectColorBlueTransforms(Span<uint> bgra, int stride, int tileWidth, int tileHeight, int greenToBlue, int redToBlue, Span<int> histo)
     {
-        if (Vector256_.SupportsShuffleNativeByte && tileWidth >= 16)
+        if (Vector256.IsHardwareAccelerated && tileWidth >= 16)
         {
             const int span = 16;
             Span<ushort> values = stackalloc ushort[span];