tsai
/
ImageSharp
mirror of https://github.com/SixLabors/ImageSharp
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
Browse Source

Initial 3padshuffle4

js/color-alpha-handling
James Jackson-South 6 years ago
parent
65b5a47c1a
commit
c4f7f67fd5
8 changed files with 274 additions and 29 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 84
      src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
  2. 63
      src/ImageSharp/Common/Helpers/SimdUtils.Shuffle.cs
  3. 12
      src/ImageSharp/PixelFormats/Utils/PixelConverter.cs
  4. 67
      tests/ImageSharp.Benchmarks/Color/Bulk/Pad3Shuffle4Channel.cs
  5. 2
      tests/ImageSharp.Benchmarks/Color/Bulk/ShuffleByte4Channel.cs
  6. 2
      tests/ImageSharp.Benchmarks/Color/Bulk/ShuffleFloat4Channel.cs
  7. 71
      tests/ImageSharp.Tests/Common/SimdUtilsTests.Shuffle.cs
  8. 2
      tests/ImageSharp.Tests/Common/SimdUtilsTests.cs

84
src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
View File

@ -26,7 +26,7 @@ namespace SixLabors.ImageSharp
/// <param name="dest">The destination span of floats.</param> /// <param name="dest">The destination span of floats.</param>
/// <param name="control">The byte control.</param> /// <param name="control">The byte control.</param>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void Shuffle4ChannelReduce( public static void Shuffle4Reduce(
ref ReadOnlySpan<float> source, ref ReadOnlySpan<float> source,
ref Span<float> dest, ref Span<float> dest,
byte control) byte control)
@ -41,7 +41,7 @@ namespace SixLabors.ImageSharp
if (adjustedCount > 0) if (adjustedCount > 0)
{ {
Shuffle4Channel( Shuffle4(
source.Slice(0, adjustedCount), source.Slice(0, adjustedCount),
dest.Slice(0, adjustedCount), dest.Slice(0, adjustedCount),
control); control);
@ -53,14 +53,14 @@ namespace SixLabors.ImageSharp
} }
/// <summary> /// <summary>
/// Shuffle 8-bit integers in a within 128-bit lanes in <paramref name="source"/> /// Shuffle 8-bit integers within 128-bit lanes in <paramref name="source"/>
/// using the control and store the results in <paramref name="dest"/>. /// using the control and store the results in <paramref name="dest"/>.
/// </summary> /// </summary>
/// <param name="source">The source span of bytes.</param> /// <param name="source">The source span of bytes.</param>
/// <param name="dest">The destination span of bytes.</param> /// <param name="dest">The destination span of bytes.</param>
/// <param name="control">The byte control.</param> /// <param name="control">The byte control.</param>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void Shuffle4ChannelReduce( public static void Shuffle4Reduce(
ref ReadOnlySpan<byte> source, ref ReadOnlySpan<byte> source,
ref Span<byte> dest, ref Span<byte> dest,
byte control) byte control)
@ -75,7 +75,7 @@ namespace SixLabors.ImageSharp
if (adjustedCount > 0) if (adjustedCount > 0)
{ {
Shuffle4Channel( Shuffle4(
source.Slice(0, adjustedCount), source.Slice(0, adjustedCount),
dest.Slice(0, adjustedCount), dest.Slice(0, adjustedCount),
control); control);
@ -86,8 +86,41 @@ namespace SixLabors.ImageSharp
} }
} }
/// <summary>
/// Pads then shuffles 8-bit integers within 128-bit lanes in <paramref name="source"/>
/// using the control and store the results in <paramref name="dest"/>.
/// </summary>
/// <param name="source">The source span of bytes.</param>
/// <param name="dest">The destination span of bytes.</param>
/// <param name="control">The byte control.</param>
[MethodImpl(InliningOptions.ShortMethod)]
public static unsafe void Pad3Shuffle4Reduce(
ref ReadOnlySpan<byte> source,
ref Span<byte> dest,
byte control)
{
if (Ssse3.IsSupported)
{
int remainder = ImageMaths.ModuloP2(source.Length, Vector128<byte>.Count);
int adjustedCount = source.Length - remainder;
int sourceSlice = (int)(adjustedCount * (3 / 4F));
if (adjustedCount > 0)
{
Pad3Shuffle4(
source.Slice(0, adjustedCount),
dest.Slice(0, adjustedCount),
control);
source = source.Slice(sourceSlice);
dest = dest.Slice(adjustedCount);
}
}
}
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
private static void Shuffle4Channel( private static void Shuffle4(
ReadOnlySpan<float> source, ReadOnlySpan<float> source,
Span<float> dest, Span<float> dest,
byte control) byte control)
@ -165,7 +198,7 @@ namespace SixLabors.ImageSharp
} }
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
private static void Shuffle4Channel( private static void Shuffle4(
ReadOnlySpan<byte> source, ReadOnlySpan<byte> source,
Span<byte> dest, Span<byte> dest,
byte control) byte control)
@ -246,6 +279,43 @@ namespace SixLabors.ImageSharp
} }
} }
[MethodImpl(InliningOptions.ShortMethod)]
private static unsafe void Pad3Shuffle4(
ReadOnlySpan<byte> source,
Span<byte> dest,
byte control)
{
if (Ssse3.IsSupported)
{
Vector128<byte> wMask = Vector128.Create(0xff000000u).AsByte();
Vector128<byte> padMask = Vector128.Create(0, 1, 2, -1, 3, 4, 5, -1, 6, 7, 8, -1, 9, 10, 11, -1).AsByte();
Span<byte> bytes = stackalloc byte[Vector128<byte>.Count];
Shuffle.MmShuffleSpan(ref bytes, control);
Vector128<byte> vcm = Unsafe.As<byte, Vector128<byte>>(ref MemoryMarshal.GetReference(bytes));
fixed (byte* sBase = &source.GetPinnableReference())
fixed (byte* dBase = &dest.GetPinnableReference())
{
byte* s = sBase;
byte* d = dBase;
// TODO: Consider unrolling and shuffling 4 at a time using Ssse3.AlignRight
// See https://stackoverflow.com/questions/2973708/fast-24-bit-array-32-bit-array-conversion
for (int i = 0; i < source.Length; i += 16)
{
Vector128<byte> vs0 = Sse2.LoadVector128(s);
Vector128<byte> val = Sse2.Or(wMask, Ssse3.Shuffle(vs0, padMask));
val = Ssse3.Shuffle(val, vcm);
Sse2.Store(d, val);
s += 12;
d += 16;
}
}
}
}
/// <summary> /// <summary>
/// Performs a multiplication and an addition of the <see cref="Vector256{T}"/>. /// Performs a multiplication and an addition of the <see cref="Vector256{T}"/>.
/// </summary> /// </summary>

63
src/ImageSharp/Common/Helpers/SimdUtils.Shuffle.cs
View File

@ -18,7 +18,7 @@ namespace SixLabors.ImageSharp
/// <param name="dest">The destination span of floats.</param> /// <param name="dest">The destination span of floats.</param>
/// <param name="control">The byte control.</param> /// <param name="control">The byte control.</param>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void Shuffle4Channel( public static void Shuffle4(
ReadOnlySpan<float> source, ReadOnlySpan<float> source,
Span<float> dest, Span<float> dest,
byte control) byte control)
@ -26,13 +26,13 @@ namespace SixLabors.ImageSharp
VerifyShuffleSpanInput(source, dest); VerifyShuffleSpanInput(source, dest);
#if SUPPORTS_RUNTIME_INTRINSICS #if SUPPORTS_RUNTIME_INTRINSICS
HwIntrinsics.Shuffle4ChannelReduce(ref source, ref dest, control); HwIntrinsics.Shuffle4Reduce(ref source, ref dest, control);
#endif #endif
// Deal with the remainder: // Deal with the remainder:
if (source.Length > 0) if (source.Length > 0)
{ {
ShuffleRemainder4Channel(source, dest, control); Shuffle4Remainder(source, dest, control);
} }
} }
@ -44,7 +44,7 @@ namespace SixLabors.ImageSharp
/// <param name="dest">The destination span of bytes.</param> /// <param name="dest">The destination span of bytes.</param>
/// <param name="shuffle">The type of shuffle to perform.</param> /// <param name="shuffle">The type of shuffle to perform.</param>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void Shuffle4Channel<TShuffle>( public static void Shuffle4<TShuffle>(
ReadOnlySpan<byte> source, ReadOnlySpan<byte> source,
Span<byte> dest, Span<byte> dest,
TShuffle shuffle) TShuffle shuffle)
@ -53,7 +53,7 @@ namespace SixLabors.ImageSharp
VerifyShuffleSpanInput(source, dest); VerifyShuffleSpanInput(source, dest);
#if SUPPORTS_RUNTIME_INTRINSICS #if SUPPORTS_RUNTIME_INTRINSICS
HwIntrinsics.Shuffle4ChannelReduce(ref source, ref dest, shuffle.Control); HwIntrinsics.Shuffle4Reduce(ref source, ref dest, shuffle.Control);
#endif #endif
// Deal with the remainder: // Deal with the remainder:
@ -63,7 +63,26 @@ namespace SixLabors.ImageSharp
} }
} }
public static void ShuffleRemainder4Channel( [MethodImpl(InliningOptions.ShortMethod)]
public static void Pad3Shuffle4(
ReadOnlySpan<byte> source,
Span<byte> dest,
byte control)
{
VerifyPadShuffleSpanInput(source, dest);
#if SUPPORTS_RUNTIME_INTRINSICS
HwIntrinsics.Pad3Shuffle4Reduce(ref source, ref dest, control);
#endif
// Deal with the remainder:
if (source.Length > 0)
{
Pad3Shuffle4Remainder(source, dest, control);
}
}
public static void Shuffle4Remainder(
ReadOnlySpan<float> source, ReadOnlySpan<float> source,
Span<float> dest, Span<float> dest,
byte control) byte control)
@ -81,6 +100,24 @@ namespace SixLabors.ImageSharp
} }
} }
public static void Pad3Shuffle4Remainder(
ReadOnlySpan<byte> source,
Span<byte> dest,
byte control)
{
ref byte sBase = ref MemoryMarshal.GetReference(source);
ref byte dBase = ref MemoryMarshal.GetReference(dest);
Shuffle.InverseMmShuffle(control, out int p3, out int p2, out int p1, out int p0);
for (int i = 0, j = 0; i < dest.Length; i += 4, j += 3)
{
Unsafe.Add(ref dBase, p0 + i) = Unsafe.Add(ref sBase, j);
Unsafe.Add(ref dBase, p1 + i) = Unsafe.Add(ref sBase, j + 1);
Unsafe.Add(ref dBase, p2 + i) = Unsafe.Add(ref sBase, j + 2);
Unsafe.Add(ref dBase, p3 + i) = byte.MaxValue;
}
}
[Conditional("DEBUG")] [Conditional("DEBUG")]
private static void VerifyShuffleSpanInput<T>(ReadOnlySpan<T> source, Span<T> dest) private static void VerifyShuffleSpanInput<T>(ReadOnlySpan<T> source, Span<T> dest)
where T : struct where T : struct
@ -96,6 +133,20 @@ namespace SixLabors.ImageSharp
"Input spans must be divisiable by 4!"); "Input spans must be divisiable by 4!");
} }
[Conditional("DEBUG")]
private static void VerifyPadShuffleSpanInput(ReadOnlySpan<byte> source, Span<byte> dest)
{
DebugGuard.IsTrue(
source.Length == (int)(dest.Length * 3 / 4F),
nameof(source),
"Input spans must be 3/4 the length of the output span!");
DebugGuard.IsTrue(
source.Length % 3 == 0,
nameof(source),
"Input spans must be divisiable by 3!");
}
public static class Shuffle public static class Shuffle
{ {
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]

12
src/ImageSharp/PixelFormats/Utils/PixelConverter.cs
View File

@ -27,7 +27,7 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
/// </summary> /// </summary>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void ToArgb32(ReadOnlySpan<byte> source, Span<byte> dest) public static void ToArgb32(ReadOnlySpan<byte> source, Span<byte> dest)
=> SimdUtils.Shuffle4Channel<WXYZShuffle4>(source, dest, default); => SimdUtils.Shuffle4<WXYZShuffle4>(source, dest, default);
/// <summary> /// <summary>
/// Converts a <see cref="ReadOnlySpan{Byte}"/> representing a collection of /// Converts a <see cref="ReadOnlySpan{Byte}"/> representing a collection of
@ -36,7 +36,7 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
/// </summary> /// </summary>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void ToBgra32(ReadOnlySpan<byte> source, Span<byte> dest) public static void ToBgra32(ReadOnlySpan<byte> source, Span<byte> dest)
=> SimdUtils.Shuffle4Channel<ZYXWShuffle4>(source, dest, default); => SimdUtils.Shuffle4<ZYXWShuffle4>(source, dest, default);
} }
public static class FromArgb32 public static class FromArgb32
@ -48,7 +48,7 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
/// </summary> /// </summary>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void ToRgba32(ReadOnlySpan<byte> source, Span<byte> dest) public static void ToRgba32(ReadOnlySpan<byte> source, Span<byte> dest)
=> SimdUtils.Shuffle4Channel<YZWXShuffle4>(source, dest, default); => SimdUtils.Shuffle4<YZWXShuffle4>(source, dest, default);
/// <summary> /// <summary>
/// Converts a <see cref="ReadOnlySpan{Byte}"/> representing a collection of /// Converts a <see cref="ReadOnlySpan{Byte}"/> representing a collection of
@ -57,7 +57,7 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
/// </summary> /// </summary>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void ToBgra32(ReadOnlySpan<byte> source, Span<byte> dest) public static void ToBgra32(ReadOnlySpan<byte> source, Span<byte> dest)
=> SimdUtils.Shuffle4Channel<WZYXShuffle4>(source, dest, default); => SimdUtils.Shuffle4<WZYXShuffle4>(source, dest, default);
} }
public static class FromBgra32 public static class FromBgra32
@ -69,7 +69,7 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
/// </summary> /// </summary>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void ToArgb32(ReadOnlySpan<byte> source, Span<byte> dest) public static void ToArgb32(ReadOnlySpan<byte> source, Span<byte> dest)
=> SimdUtils.Shuffle4Channel<WZYXShuffle4>(source, dest, default); => SimdUtils.Shuffle4<WZYXShuffle4>(source, dest, default);
/// <summary> /// <summary>
/// Converts a <see cref="ReadOnlySpan{Byte}"/> representing a collection of /// Converts a <see cref="ReadOnlySpan{Byte}"/> representing a collection of
@ -78,7 +78,7 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
/// </summary> /// </summary>
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static void ToRgba32(ReadOnlySpan<byte> source, Span<byte> dest) public static void ToRgba32(ReadOnlySpan<byte> source, Span<byte> dest)
=> SimdUtils.Shuffle4Channel<ZYXWShuffle4>(source, dest, default); => SimdUtils.Shuffle4<ZYXWShuffle4>(source, dest, default);
} }
} }
} }

67
tests/ImageSharp.Benchmarks/Color/Bulk/Pad3Shuffle4Channel.cs
View File

@ -0,0 +1,67 @@
// 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 Pad3Shuffle4Channel
{
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[this.Count];
}
[Params(96, 384, 768, 1536)]
public int Count { get; set; }
[Benchmark]
public void Shuffle4Channel()
{
SimdUtils.Shuffle4<WXYZShuffle4>(this.source, this.destination, default);
}
}
// 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 | 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 | - | - | - | - |
}

2
tests/ImageSharp.Benchmarks/Color/Bulk/ShuffleByte4Channel.cs
View File

@ -26,7 +26,7 @@ namespace SixLabors.ImageSharp.Benchmarks.ColorSpaces.Bulk
[Benchmark] [Benchmark]
public void Shuffle4Channel() public void Shuffle4Channel()
{ {
SimdUtils.Shuffle4Channel<WXYZShuffle4>(this.source, this.destination, default); SimdUtils.Shuffle4<WXYZShuffle4>(this.source, this.destination, default);
} }
} }

2
tests/ImageSharp.Benchmarks/Color/Bulk/ShuffleFloat4Channel.cs
View File

@ -27,7 +27,7 @@ namespace SixLabors.ImageSharp.Benchmarks.ColorSpaces.Bulk
[Benchmark] [Benchmark]
public void Shuffle4Channel() public void Shuffle4Channel()
{ {
SimdUtils.Shuffle4Channel(this.source, this.destination, control); SimdUtils.Shuffle4(this.source, this.destination, control);
} }
} }

71
tests/ImageSharp.Tests/Common/SimdUtilsTests.Shuffle.cs
View File

@ -22,7 +22,7 @@ namespace SixLabors.ImageSharp.Tests.Common
TestShuffleFloat4Channel( TestShuffleFloat4Channel(
size, size,
(s, d) => SimdUtils.Shuffle4Channel(s.Span, d.Span, control), (s, d) => SimdUtils.Shuffle4(s.Span, d.Span, control),
control); control);
} }
@ -49,43 +49,43 @@ namespace SixLabors.ImageSharp.Tests.Common
WXYZShuffle4 wxyz = default; WXYZShuffle4 wxyz = default;
TestShuffleByte4Channel( TestShuffleByte4Channel(
size, size,
(s, d) => SimdUtils.Shuffle4Channel(s.Span, d.Span, wxyz), (s, d) => SimdUtils.Shuffle4(s.Span, d.Span, wxyz),
wxyz.Control); wxyz.Control);
WZYXShuffle4 wzyx = default; WZYXShuffle4 wzyx = default;
TestShuffleByte4Channel( TestShuffleByte4Channel(
size, size,
(s, d) => SimdUtils.Shuffle4Channel(s.Span, d.Span, wzyx), (s, d) => SimdUtils.Shuffle4(s.Span, d.Span, wzyx),
wzyx.Control); wzyx.Control);
YZWXShuffle4 yzwx = default; YZWXShuffle4 yzwx = default;
TestShuffleByte4Channel( TestShuffleByte4Channel(
size, size,
(s, d) => SimdUtils.Shuffle4Channel(s.Span, d.Span, yzwx), (s, d) => SimdUtils.Shuffle4(s.Span, d.Span, yzwx),
yzwx.Control); yzwx.Control);
ZYXWShuffle4 zyxw = default; ZYXWShuffle4 zyxw = default;
TestShuffleByte4Channel( TestShuffleByte4Channel(
size, size,
(s, d) => SimdUtils.Shuffle4Channel(s.Span, d.Span, zyxw), (s, d) => SimdUtils.Shuffle4(s.Span, d.Span, zyxw),
zyxw.Control); zyxw.Control);
var xwyz = new DefaultShuffle4(2, 1, 3, 0); var xwyz = new DefaultShuffle4(2, 1, 3, 0);
TestShuffleByte4Channel( TestShuffleByte4Channel(
size, size,
(s, d) => SimdUtils.Shuffle4Channel(s.Span, d.Span, xwyz), (s, d) => SimdUtils.Shuffle4(s.Span, d.Span, xwyz),
xwyz.Control); xwyz.Control);
var yyyy = new DefaultShuffle4(1, 1, 1, 1); var yyyy = new DefaultShuffle4(1, 1, 1, 1);
TestShuffleByte4Channel( TestShuffleByte4Channel(
size, size,
(s, d) => SimdUtils.Shuffle4Channel(s.Span, d.Span, yyyy), (s, d) => SimdUtils.Shuffle4(s.Span, d.Span, yyyy),
yyyy.Control); yyyy.Control);
var wwww = new DefaultShuffle4(3, 3, 3, 3); var wwww = new DefaultShuffle4(3, 3, 3, 3);
TestShuffleByte4Channel( TestShuffleByte4Channel(
size, size,
(s, d) => SimdUtils.Shuffle4Channel(s.Span, d.Span, wwww), (s, d) => SimdUtils.Shuffle4(s.Span, d.Span, wwww),
wwww.Control); wwww.Control);
} }
} }
@ -97,6 +97,29 @@ namespace SixLabors.ImageSharp.Tests.Common
HwIntrinsics.AllowAll | HwIntrinsics.DisableAVX2 | HwIntrinsics.DisableSSE); HwIntrinsics.AllowAll | HwIntrinsics.DisableAVX2 | HwIntrinsics.DisableSSE);
} }
[Theory]
[MemberData(nameof(ArraySizesDivisibleBy3))]
public void BulkPad3Shuffle4Channel(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<int>(serialized);
byte control = default(WZYXShuffle4).Control;
TestPad3Shuffle4Channel(
size,
(s, d) => SimdUtils.Pad3Shuffle4(s.Span, d.Span, control),
control);
}
FeatureTestRunner.RunWithHwIntrinsicsFeature(
RunTest,
count,
HwIntrinsics.AllowAll | HwIntrinsics.DisableAVX | HwIntrinsics.DisableSSE);
}
private static void TestShuffleFloat4Channel( private static void TestShuffleFloat4Channel(
int count, int count,
Action<Memory<float>, Memory<float>> convert, Action<Memory<float>, Memory<float>> convert,
@ -157,5 +180,37 @@ namespace SixLabors.ImageSharp.Tests.Common
Assert.Equal(expected, result); Assert.Equal(expected, result);
} }
private static void TestPad3Shuffle4Channel(
int count,
Action<Memory<byte>, Memory<byte>> convert,
byte control)
{
byte[] source = new byte[count];
new Random(count).NextBytes(source);
var result = new byte[(int)(count * (4 / 3F))];
byte[] expected = new byte[result.Length];
SimdUtils.Shuffle.InverseMmShuffle(
control,
out int p3,
out int p2,
out int p1,
out int p0);
for (int i = 0, j = 0; i < expected.Length; i += 4, j += 3)
{
expected[p0 + i] = source[j];
expected[p1 + i] = source[j + 1];
expected[p2 + i] = source[j + 2];
expected[p3 + i] = byte.MaxValue;
}
convert(source, result);
Assert.Equal(expected, result);
}
} }
} }

2
tests/ImageSharp.Tests/Common/SimdUtilsTests.cs
View File

@ -163,6 +163,8 @@ namespace SixLabors.ImageSharp.Tests.Common
public static readonly TheoryData<int> ArraySizesDivisibleBy8 = new TheoryData<int> { 0, 8, 16, 1024 }; public static readonly TheoryData<int> ArraySizesDivisibleBy8 = new TheoryData<int> { 0, 8, 16, 1024 };
public static readonly TheoryData<int> ArraySizesDivisibleBy4 = new TheoryData<int> { 0, 4, 8, 28, 1020 }; public static readonly TheoryData<int> ArraySizesDivisibleBy4 = new TheoryData<int> { 0, 4, 8, 28, 1020 };
public static readonly TheoryData<int> ArraySizesDivisibleBy3 = new TheoryData<int> { 0, 3, 9, 36, 957 };
public static readonly TheoryData<int> ArraySizesDivisibleBy32 = new TheoryData<int> { 0, 32, 512 }; public static readonly TheoryData<int> ArraySizesDivisibleBy32 = new TheoryData<int> { 0, 32, 512 };

Write Preview
Loading…
Cancel
Save