diff --git a/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs b/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
index 8a0b5460c..29b569a80 100644
--- a/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
+++ b/src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
@@ -119,6 +119,39 @@ namespace SixLabors.ImageSharp
}
}
+ ///
+ /// Shuffles then slices 8-bit integers within 128-bit lanes in
+ /// using the control and store the results in .
+ ///
+ /// The source span of bytes.
+ /// The destination span of bytes.
+ /// The byte control.
+ [MethodImpl(InliningOptions.ShortMethod)]
+ public static unsafe void Shuffle4Slice3Reduce(
+ ref ReadOnlySpan source,
+ ref Span dest,
+ byte control)
+ {
+ if (Ssse3.IsSupported)
+ {
+ int remainder = ImageMaths.ModuloP2(dest.Length, Vector128.Count);
+
+ int adjustedCount = dest.Length - remainder;
+ int destSlice = (int)(adjustedCount * (3 / 4F));
+
+ if (adjustedCount > 0)
+ {
+ Shuffle4Slice3(
+ source.Slice(0, adjustedCount),
+ dest.Slice(0, adjustedCount),
+ control);
+
+ source = source.Slice(adjustedCount);
+ dest = dest.Slice(destSlice);
+ }
+ }
+ }
+
[MethodImpl(InliningOptions.ShortMethod)]
private static void Shuffle4(
ReadOnlySpan source,
@@ -316,6 +349,50 @@ namespace SixLabors.ImageSharp
}
}
+ [MethodImpl(InliningOptions.ShortMethod)]
+ private static unsafe void Shuffle4Slice3(
+ ReadOnlySpan source,
+ Span dest,
+ byte control)
+ {
+ if (Ssse3.IsSupported)
+ {
+ Vector128 sliceMask = Vector128.Create(0, 1, 2, 4, 5, 6, 8, 9, 10, 12, 13, 14, -1, -1, -1, -1).AsByte();
+
+ Span bytes = stackalloc byte[Vector128.Count];
+ Shuffle.MmShuffleSpan(ref bytes, control);
+ Vector128 vcm = Unsafe.As>(ref MemoryMarshal.GetReference(bytes));
+
+ // var control = MmShuffle(3, 0, 1, 2);
+ // Span bytes = stackalloc byte[Vector128.Count];
+ // MmShuffleSpan(ref bytes, control);
+ // Vector128 vcm = Unsafe.As>(ref MemoryMarshal.GetReference(bytes));
+ //
+ // Vector128 s0 = Vector128.Create((byte)1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1).Dump("s0");
+ // Vector128 padded = Ssse3.Shuffle(s0, padMask).Dump("padded");
+ //
+ // padded = Sse3.Or(Vector128.Create(0xff000000u).AsByte(), padded).Dump("0r");
+ //
+ // var shuffled = Ssse3.Shuffle(padded, vcm).Dump("shuffled");
+ // var d0 = Ssse3.Shuffle(shuffled, sliceMask).Dump("d0");
+ fixed (byte* sBase = &source.GetPinnableReference())
+ fixed (byte* dBase = &dest.GetPinnableReference())
+ {
+ byte* s = sBase;
+ byte* d = dBase;
+
+ for (int i = 0; i < source.Length; i += 16)
+ {
+ Vector128 vs0 = Ssse3.Shuffle(Sse2.LoadVector128(s), vcm);
+ Sse2.Store(d, Ssse3.Shuffle(vs0, sliceMask));
+
+ s += 16;
+ d += 12;
+ }
+ }
+ }
+ }
+
///
/// Performs a multiplication and an addition of the .
///
diff --git a/src/ImageSharp/Common/Helpers/SimdUtils.Shuffle.cs b/src/ImageSharp/Common/Helpers/SimdUtils.Shuffle.cs
index 81d77d655..f3946361b 100644
--- a/src/ImageSharp/Common/Helpers/SimdUtils.Shuffle.cs
+++ b/src/ImageSharp/Common/Helpers/SimdUtils.Shuffle.cs
@@ -69,7 +69,7 @@ namespace SixLabors.ImageSharp
Span dest,
byte control)
{
- VerifyPadShuffleSpanInput(source, dest);
+ VerifyPad3Shuffle4SpanInput(source, dest);
#if SUPPORTS_RUNTIME_INTRINSICS
HwIntrinsics.Pad3Shuffle4Reduce(ref source, ref dest, control);
@@ -82,6 +82,25 @@ namespace SixLabors.ImageSharp
}
}
+ [MethodImpl(InliningOptions.ShortMethod)]
+ public static void Shuffle4Slice3(
+ ReadOnlySpan source,
+ Span dest,
+ byte control)
+ {
+ VerifyShuffle4Slice3SpanInput(source, dest);
+
+#if SUPPORTS_RUNTIME_INTRINSICS
+ HwIntrinsics.Shuffle4Slice3Reduce(ref source, ref dest, control);
+#endif
+
+ // Deal with the remainder:
+ if (source.Length > 0)
+ {
+ Shuffle4Slice3Remainder(source, dest, control);
+ }
+ }
+
public static void Shuffle4Remainder(
ReadOnlySpan source,
Span dest,
@@ -118,6 +137,23 @@ namespace SixLabors.ImageSharp
}
}
+ public static void Shuffle4Slice3Remainder(
+ ReadOnlySpan source,
+ Span dest,
+ byte control)
+ {
+ ref byte sBase = ref MemoryMarshal.GetReference(source);
+ ref byte dBase = ref MemoryMarshal.GetReference(dest);
+ Shuffle.InverseMmShuffle(control, out int _, out int p2, out int p1, out int p0);
+
+ for (int i = 0, j = 0; i < dest.Length; i += 3, j += 4)
+ {
+ Unsafe.Add(ref dBase, i) = Unsafe.Add(ref sBase, p0 + j);
+ Unsafe.Add(ref dBase, i + 1) = Unsafe.Add(ref sBase, p1 + j);
+ Unsafe.Add(ref dBase, i + 2) = Unsafe.Add(ref sBase, p2 + j);
+ }
+ }
+
[Conditional("DEBUG")]
private static void VerifyShuffleSpanInput(ReadOnlySpan source, Span dest)
where T : struct
@@ -130,21 +166,45 @@ namespace SixLabors.ImageSharp
DebugGuard.IsTrue(
source.Length % 4 == 0,
nameof(source),
- "Input spans must be divisiable by 4!");
+ "Input spans must be divisable by 4!");
}
[Conditional("DEBUG")]
- private static void VerifyPadShuffleSpanInput(ReadOnlySpan source, Span dest)
+ private static void VerifyPad3Shuffle4SpanInput(ReadOnlySpan source, Span dest)
{
+ DebugGuard.IsTrue(
+ source.Length % 3 == 0,
+ nameof(source),
+ "Input span must be divisable by 3!");
+
+ DebugGuard.IsTrue(
+ dest.Length % 4 == 0,
+ nameof(dest),
+ "Output span must be divisable by 4!");
+
DebugGuard.IsTrue(
source.Length == (int)(dest.Length * 3 / 4F),
nameof(source),
- "Input spans must be 3/4 the length of the output span!");
+ "Input span must be 3/4 the length of the output span!");
+ }
+ [Conditional("DEBUG")]
+ private static void VerifyShuffle4Slice3SpanInput(ReadOnlySpan source, Span dest)
+ {
DebugGuard.IsTrue(
- source.Length % 3 == 0,
+ source.Length % 4 == 0,
+ nameof(source),
+ "Input span must be divisable by 4!");
+
+ DebugGuard.IsTrue(
+ dest.Length % 3 == 0,
+ nameof(dest),
+ "Output span must be divisable by 3!");
+
+ DebugGuard.IsTrue(
+ source.Length == (int)(dest.Length * 4 / 3F),
nameof(source),
- "Input spans must be divisiable by 3!");
+ "Output span must be 3/4 the length of the input span!");
}
public static class Shuffle
diff --git a/tests/ImageSharp.Benchmarks/Color/Bulk/Pad3Shuffle4Channel.cs b/tests/ImageSharp.Benchmarks/Color/Bulk/Pad3Shuffle4Channel.cs
index c529b2af1..8286fea0e 100644
--- a/tests/ImageSharp.Benchmarks/Color/Bulk/Pad3Shuffle4Channel.cs
+++ b/tests/ImageSharp.Benchmarks/Color/Bulk/Pad3Shuffle4Channel.cs
@@ -9,6 +9,7 @@ namespace SixLabors.ImageSharp.Benchmarks.ColorSpaces.Bulk
[Config(typeof(Config.HwIntrinsics_SSE_AVX))]
public class Pad3Shuffle4Channel
{
+ private static readonly byte Control = default(WXYZShuffle4).Control;
private byte[] source;
private byte[] destination;
@@ -17,20 +18,20 @@ namespace SixLabors.ImageSharp.Benchmarks.ColorSpaces.Bulk
{
this.source = new byte[this.Count];
new Random(this.Count).NextBytes(this.source);
- this.destination = new byte[this.Count];
+ this.destination = new byte[(int)(this.Count * (4 / 3F))];
}
[Params(96, 384, 768, 1536)]
public int Count { get; set; }
[Benchmark]
- public void Shuffle4Channel()
+ public void Pad3Shuffle4()
{
- SimdUtils.Shuffle4(this.source, this.destination, default);
+ SimdUtils.Pad3Shuffle4(this.source, this.destination, Control);
}
}
- // 2020-10-29
+ // 2020-10-30
// ##########
//
// BenchmarkDotNet=v0.12.1, OS=Windows 10.0.19041.572 (2004/?/20H1)
@@ -43,25 +44,21 @@ namespace SixLabors.ImageSharp.Benchmarks.ColorSpaces.Bulk
//
// Runtime=.NET Core 3.1
//
- // | Method | Job | EnvironmentVariables | Count | Mean | Error | StdDev | Ratio | RatioSD | Gen 0 | Gen 1 | Gen 2 | Allocated |
- // |---------------- |------------------- |-------------------------------------------------- |------ |----------:|---------:|---------:|------:|--------:|------:|------:|------:|----------:|
- // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 128 | 17.39 ns | 0.187 ns | 0.175 ns | 1.00 | 0.00 | - | - | - | - |
- // | Shuffle4Channel | 2. AVX | Empty | 128 | 21.72 ns | 0.299 ns | 0.279 ns | 1.25 | 0.02 | - | - | - | - |
- // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 128 | 18.10 ns | 0.346 ns | 0.289 ns | 1.04 | 0.02 | - | - | - | - |
- // | | | | | | | | | | | | | |
- // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 256 | 35.51 ns | 0.711 ns | 0.790 ns | 1.00 | 0.00 | - | - | - | - |
- // | Shuffle4Channel | 2. AVX | Empty | 256 | 23.90 ns | 0.508 ns | 0.820 ns | 0.69 | 0.02 | - | - | - | - |
- // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 256 | 20.40 ns | 0.133 ns | 0.111 ns | 0.57 | 0.01 | - | - | - | - |
- // | | | | | | | | | | | | | |
- // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 512 | 73.39 ns | 0.310 ns | 0.259 ns | 1.00 | 0.00 | - | - | - | - |
- // | Shuffle4Channel | 2. AVX | Empty | 512 | 26.10 ns | 0.418 ns | 0.391 ns | 0.36 | 0.01 | - | - | - | - |
- // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 512 | 27.59 ns | 0.556 ns | 0.571 ns | 0.38 | 0.01 | - | - | - | - |
- // | | | | | | | | | | | | | |
- // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 1024 | 150.64 ns | 2.903 ns | 2.716 ns | 1.00 | 0.00 | - | - | - | - |
- // | Shuffle4Channel | 2. AVX | Empty | 1024 | 38.67 ns | 0.801 ns | 1.889 ns | 0.24 | 0.02 | - | - | - | - |
- // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 1024 | 47.13 ns | 0.948 ns | 1.054 ns | 0.31 | 0.01 | - | - | - | - |
- // | | | | | | | | | | | | | |
- // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 2048 | 315.29 ns | 5.206 ns | 6.583 ns | 1.00 | 0.00 | - | - | - | - |
- // | Shuffle4Channel | 2. AVX | Empty | 2048 | 57.37 ns | 1.152 ns | 1.078 ns | 0.18 | 0.01 | - | - | - | - |
- // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 2048 | 65.75 ns | 1.198 ns | 1.600 ns | 0.21 | 0.01 | - | - | - | - |
+ // | Method | Job | EnvironmentVariables | Count | Mean | Error | StdDev | Ratio | RatioSD | Gen 0 | Gen 1 | Gen 2 | Allocated |
+ // |------------- |------------------- |-------------------------------------------------- |------ |----------:|---------:|---------:|------:|--------:|------:|------:|------:|----------:|
+ // | Pad3Shuffle4 | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 96 | 62.91 ns | 1.240 ns | 1.569 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Pad3Shuffle4 | 2. AVX | Empty | 96 | 44.34 ns | 0.371 ns | 0.329 ns | 0.70 | 0.02 | - | - | - | - |
+ // | Pad3Shuffle4 | 3. SSE | COMPlus_EnableAVX=0 | 96 | 44.46 ns | 0.617 ns | 0.515 ns | 0.70 | 0.02 | - | - | - | - |
+ // | | | | | | | | | | | | | |
+ // | Pad3Shuffle4 | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 384 | 247.93 ns | 2.640 ns | 2.470 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Pad3Shuffle4 | 2. AVX | Empty | 384 | 92.91 ns | 1.204 ns | 1.127 ns | 0.37 | 0.01 | - | - | - | - |
+ // | Pad3Shuffle4 | 3. SSE | COMPlus_EnableAVX=0 | 384 | 91.42 ns | 1.234 ns | 1.094 ns | 0.37 | 0.01 | - | - | - | - |
+ // | | | | | | | | | | | | | |
+ // | Pad3Shuffle4 | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 768 | 444.79 ns | 5.094 ns | 4.254 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Pad3Shuffle4 | 2. AVX | Empty | 768 | 162.92 ns | 1.046 ns | 0.873 ns | 0.37 | 0.00 | - | - | - | - |
+ // | Pad3Shuffle4 | 3. SSE | COMPlus_EnableAVX=0 | 768 | 166.22 ns | 1.728 ns | 1.443 ns | 0.37 | 0.00 | - | - | - | - |
+ // | | | | | | | | | | | | | |
+ // | Pad3Shuffle4 | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 1536 | 882.51 ns | 6.936 ns | 5.792 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Pad3Shuffle4 | 2. AVX | Empty | 1536 | 309.72 ns | 3.777 ns | 3.533 ns | 0.35 | 0.01 | - | - | - | - |
+ // | Pad3Shuffle4 | 3. SSE | COMPlus_EnableAVX=0 | 1536 | 323.18 ns | 4.079 ns | 3.816 ns | 0.37 | 0.00 | - | - | - | - |
}
diff --git a/tests/ImageSharp.Benchmarks/Color/Bulk/Shuffle4Slice3Channel.cs b/tests/ImageSharp.Benchmarks/Color/Bulk/Shuffle4Slice3Channel.cs
new file mode 100644
index 000000000..b64379959
--- /dev/null
+++ b/tests/ImageSharp.Benchmarks/Color/Bulk/Shuffle4Slice3Channel.cs
@@ -0,0 +1,68 @@
+// Copyright (c) Six Labors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
+using BenchmarkDotNet.Attributes;
+
+namespace SixLabors.ImageSharp.Benchmarks.ColorSpaces.Bulk
+{
+ [Config(typeof(Config.HwIntrinsics_SSE_AVX))]
+ public class Shuffle4Slice3Channel
+ {
+ private static readonly byte Control = default(WXYZShuffle4).Control;
+ private byte[] source;
+ private byte[] destination;
+
+ [GlobalSetup]
+ public void Setup()
+ {
+ this.source = new byte[this.Count];
+ new Random(this.Count).NextBytes(this.source);
+ this.destination = new byte[(int)(this.Count * (3 / 4F))];
+ }
+
+ [Params(128, 256, 512, 1024, 2048)]
+ public int Count { get; set; }
+
+ [Benchmark]
+ public void Shuffle4Slice3()
+ {
+ SimdUtils.Shuffle4Slice3(this.source, this.destination, Control);
+ }
+ }
+
+ // 2020-10-29
+ // ##########
+ //
+ // BenchmarkDotNet=v0.12.1, OS=Windows 10.0.19041.572 (2004/?/20H1)
+ // Intel Core i7-8650U CPU 1.90GHz (Kaby Lake R), 1 CPU, 8 logical and 4 physical cores
+ // .NET Core SDK=3.1.403
+ // [Host] : .NET Core 3.1.9 (CoreCLR 4.700.20.47201, CoreFX 4.700.20.47203), X64 RyuJIT
+ // 1. No HwIntrinsics : .NET Core 3.1.9 (CoreCLR 4.700.20.47201, CoreFX 4.700.20.47203), X64 RyuJIT
+ // 2. AVX : .NET Core 3.1.9 (CoreCLR 4.700.20.47201, CoreFX 4.700.20.47203), X64 RyuJIT
+ // 3. SSE : .NET Core 3.1.9 (CoreCLR 4.700.20.47201, CoreFX 4.700.20.47203), X64 RyuJIT
+ //
+ // Runtime=.NET Core 3.1
+ //
+ // | Method | Job | EnvironmentVariables | Count | Mean | Error | StdDev | Median | Ratio | RatioSD | Gen 0 | Gen 1 | Gen 2 | Allocated |
+ // |---------------- |------------------- |-------------------------------------------------- |------ |----------:|---------:|---------:|----------:|------:|--------:|------:|------:|------:|----------:|
+ // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 128 | 50.09 ns | 1.018 ns | 1.460 ns | 49.16 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Shuffle4Channel | 2. AVX | Empty | 128 | 35.28 ns | 0.106 ns | 0.089 ns | 35.30 ns | 0.69 | 0.02 | - | - | - | - |
+ // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 128 | 35.13 ns | 0.247 ns | 0.231 ns | 35.22 ns | 0.69 | 0.02 | - | - | - | - |
+ // | | | | | | | | | | | | | | |
+ // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 256 | 101.48 ns | 0.875 ns | 0.819 ns | 101.60 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Shuffle4Channel | 2. AVX | Empty | 256 | 53.25 ns | 0.518 ns | 0.433 ns | 53.21 ns | 0.52 | 0.00 | - | - | - | - |
+ // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 256 | 57.21 ns | 0.508 ns | 0.451 ns | 57.38 ns | 0.56 | 0.01 | - | - | - | - |
+ // | | | | | | | | | | | | | | |
+ // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 512 | 202.53 ns | 0.884 ns | 0.827 ns | 202.40 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Shuffle4Channel | 2. AVX | Empty | 512 | 82.55 ns | 0.418 ns | 0.391 ns | 82.59 ns | 0.41 | 0.00 | - | - | - | - |
+ // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 512 | 82.89 ns | 1.057 ns | 0.989 ns | 82.48 ns | 0.41 | 0.00 | - | - | - | - |
+ // | | | | | | | | | | | | | | |
+ // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 1024 | 398.79 ns | 7.807 ns | 6.921 ns | 395.67 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Shuffle4Channel | 2. AVX | Empty | 1024 | 144.51 ns | 1.033 ns | 0.966 ns | 144.42 ns | 0.36 | 0.01 | - | - | - | - |
+ // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 1024 | 143.77 ns | 0.820 ns | 0.684 ns | 143.62 ns | 0.36 | 0.01 | - | - | - | - |
+ // | | | | | | | | | | | | | | |
+ // | Shuffle4Channel | 1. No HwIntrinsics | COMPlus_EnableHWIntrinsic=0,COMPlus_FeatureSIMD=0 | 2048 | 798.44 ns | 4.447 ns | 3.472 ns | 799.39 ns | 1.00 | 0.00 | - | - | - | - |
+ // | Shuffle4Channel | 2. AVX | Empty | 2048 | 277.12 ns | 1.723 ns | 1.612 ns | 276.93 ns | 0.35 | 0.00 | - | - | - | - |
+ // | Shuffle4Channel | 3. SSE | COMPlus_EnableAVX=0 | 2048 | 275.70 ns | 1.796 ns | 1.500 ns | 275.51 ns | 0.35 | 0.00 | - | - | - | - ||
+}
diff --git a/tests/ImageSharp.Benchmarks/Color/Bulk/ShuffleFloat4Channel.cs b/tests/ImageSharp.Benchmarks/Color/Bulk/ShuffleFloat4Channel.cs
index 4a2512fea..86b1f766e 100644
--- a/tests/ImageSharp.Benchmarks/Color/Bulk/ShuffleFloat4Channel.cs
+++ b/tests/ImageSharp.Benchmarks/Color/Bulk/ShuffleFloat4Channel.cs
@@ -10,7 +10,7 @@ namespace SixLabors.ImageSharp.Benchmarks.ColorSpaces.Bulk
[Config(typeof(Config.HwIntrinsics_SSE_AVX))]
public class ShuffleFloat4Channel
{
- private static readonly byte control = default(WXYZShuffle4).Control;
+ private static readonly byte Control = default(WXYZShuffle4).Control;
private float[] source;
private float[] destination;
@@ -27,7 +27,7 @@ namespace SixLabors.ImageSharp.Benchmarks.ColorSpaces.Bulk
[Benchmark]
public void Shuffle4Channel()
{
- SimdUtils.Shuffle4(this.source, this.destination, control);
+ SimdUtils.Shuffle4(this.source, this.destination, Control);
}
}
diff --git a/tests/ImageSharp.Tests/Common/SimdUtilsTests.Shuffle.cs b/tests/ImageSharp.Tests/Common/SimdUtilsTests.Shuffle.cs
index 1c456e5a2..f801cd28b 100644
--- a/tests/ImageSharp.Tests/Common/SimdUtilsTests.Shuffle.cs
+++ b/tests/ImageSharp.Tests/Common/SimdUtilsTests.Shuffle.cs
@@ -120,6 +120,29 @@ namespace SixLabors.ImageSharp.Tests.Common
HwIntrinsics.AllowAll | HwIntrinsics.DisableAVX | HwIntrinsics.DisableSSE);
}
+ [Theory]
+ [MemberData(nameof(ArraySizesDivisibleBy4))]
+ public void BulkShuffle4Slice3Channel(int count)
+ {
+ static void RunTest(string serialized)
+ {
+ // No need to test multiple shuffle controls as the
+ // pipeline is always the same.
+ int size = FeatureTestRunner.Deserialize(serialized);
+ byte control = default(WZYXShuffle4).Control;
+
+ TestShuffle4Slice3Channel(
+ size,
+ (s, d) => SimdUtils.Shuffle4Slice3(s.Span, d.Span, control),
+ control);
+ }
+
+ FeatureTestRunner.RunWithHwIntrinsicsFeature(
+ RunTest,
+ count,
+ HwIntrinsics.AllowAll | HwIntrinsics.DisableAVX | HwIntrinsics.DisableSSE);
+ }
+
private static void TestShuffleFloat4Channel(
int count,
Action, Memory> convert,
@@ -210,6 +233,47 @@ namespace SixLabors.ImageSharp.Tests.Common
convert(source, result);
+ for (int i = 0; i < expected.Length; i++)
+ {
+ Assert.Equal(expected[i], result[i]);
+ }
+
+ Assert.Equal(expected, result);
+ }
+
+ private static void TestShuffle4Slice3Channel(
+ int count,
+ Action, Memory> convert,
+ byte control)
+ {
+ byte[] source = new byte[count];
+ new Random(count).NextBytes(source);
+
+ var result = new byte[(int)(count * (3 / 4F))];
+
+ byte[] expected = new byte[result.Length];
+
+ SimdUtils.Shuffle.InverseMmShuffle(
+ control,
+ out int _,
+ out int p2,
+ out int p1,
+ out int p0);
+
+ for (int i = 0, j = 0; i < expected.Length; i += 3, j += 4)
+ {
+ expected[i] = source[p0 + j];
+ expected[i + 1] = source[p1 + j];
+ expected[i + 2] = source[p2 + j];
+ }
+
+ convert(source, result);
+
+ for (int i = 0; i < expected.Length; i++)
+ {
+ Assert.Equal(expected[i], result[i]);
+ }
+
Assert.Equal(expected, result);
}
}