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

benchmarks & scalar implementation

js/color-alpha-handling
Anton Firszov 5 years ago
parent
dd0447ef4c
commit
e402700a44
5 changed files with 367 additions and 25 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. 3
      .gitattributes
  2. 99
      src/ImageSharp/Common/Helpers/SimdUtils.Pack.cs
  3. 28
      src/ImageSharp/Common/Helpers/SimdUtils.cs
  4. 10
      src/ImageSharp/ImageSharp.csproj
  5. 252
      tests/ImageSharp.Benchmarks/General/PixelConversion/PixelConversion_PackFromRgbPlanes.cs

3
.gitattributes
View File

@ -80,8 +80,11 @@
*.pvr binary *.pvr binary
*.snk binary *.snk binary
*.tga binary *.tga binary
*.tif binary
*.tiff binary
*.ttc binary *.ttc binary
*.ttf binary *.ttf binary
*.wbmp binary
*.webp binary *.webp binary
*.woff binary *.woff binary
*.woff2 binary *.woff2 binary

99
src/ImageSharp/Common/Helpers/SimdUtils.Pack.cs
View File

@ -0,0 +1,99 @@
using System;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using SixLabors.ImageSharp.PixelFormats;
namespace SixLabors.ImageSharp
{
internal static partial class SimdUtils
{
[MethodImpl(InliningOptions.ShortMethod)]
internal static void PackFromRgbPlanes(
Configuration configuration,
ReadOnlySpan<byte> redChannel,
ReadOnlySpan<byte> greenChannel,
ReadOnlySpan<byte> blueChannel,
Span<Rgb24> destination)
{
PackFromRgbPlanesScalarBatchedReduce(ref redChannel, ref greenChannel, ref blueChannel, ref destination);
PackFromRgbPlanesRemainder(redChannel, greenChannel, blueChannel, destination);
}
[MethodImpl(InliningOptions.ShortMethod)]
internal static void PackFromRgbPlanes(
Configuration configuration,
ReadOnlySpan<byte> redChannel,
ReadOnlySpan<byte> greenChannel,
ReadOnlySpan<byte> blueChannel,
Span<Rgba32> destination)
{
}
private static void PackFromRgbPlanesScalarBatchedReduce(
ref ReadOnlySpan<byte> redChannel,
ref ReadOnlySpan<byte> greenChannel,
ref ReadOnlySpan<byte> blueChannel,
ref Span<Rgb24> destination)
{
ref ByteTuple4 r = ref Unsafe.As<byte, ByteTuple4>(ref MemoryMarshal.GetReference(redChannel));
ref ByteTuple4 g = ref Unsafe.As<byte, ByteTuple4>(ref MemoryMarshal.GetReference(greenChannel));
ref ByteTuple4 b = ref Unsafe.As<byte, ByteTuple4>(ref MemoryMarshal.GetReference(blueChannel));
ref Rgb24 rgb = ref MemoryMarshal.GetReference(destination);
int count = destination.Length / 4;
for (int i = 0; i < count; i++)
{
ref Rgb24 d0 = ref Unsafe.Add(ref rgb, i * 4);
ref Rgb24 d1 = ref Unsafe.Add(ref d0, 1);
ref Rgb24 d2 = ref Unsafe.Add(ref d0, 2);
ref Rgb24 d3 = ref Unsafe.Add(ref d0, 3);
ref ByteTuple4 rr = ref Unsafe.Add(ref r, i);
ref ByteTuple4 gg = ref Unsafe.Add(ref g, i);
ref ByteTuple4 bb = ref Unsafe.Add(ref b, i);
d0.R = rr.V0;
d0.G = gg.V0;
d0.B = bb.V0;
d1.R = rr.V1;
d1.G = gg.V1;
d1.B = bb.V1;
d2.R = rr.V2;
d2.G = gg.V2;
d2.B = bb.V2;
d3.R = rr.V3;
d3.G = gg.V3;
d3.B = bb.V3;
}
int finished = count * 4;
redChannel = redChannel.Slice(finished);
greenChannel = greenChannel.Slice(finished);
blueChannel = blueChannel.Slice(finished);
destination = destination.Slice(finished);
}
private static void PackFromRgbPlanesRemainder(
ReadOnlySpan<byte> redChannel,
ReadOnlySpan<byte> greenChannel,
ReadOnlySpan<byte> blueChannel,
Span<Rgb24> destination)
{
ref byte r = ref MemoryMarshal.GetReference(redChannel);
ref byte g = ref MemoryMarshal.GetReference(greenChannel);
ref byte b = ref MemoryMarshal.GetReference(blueChannel);
ref Rgb24 rgb = ref MemoryMarshal.GetReference(destination);
for (int i = 0; i < destination.Length; i++)
{
ref Rgb24 d = ref Unsafe.Add(ref rgb, i);
d.R = Unsafe.Add(ref r, i);
d.G = Unsafe.Add(ref g, i);
d.B = Unsafe.Add(ref b, i);
}
}
}
}

28
src/ImageSharp/Common/Helpers/SimdUtils.cs
View File

@ -148,26 +148,6 @@ namespace SixLabors.ImageSharp
} }
} }
[MethodImpl(InliningOptions.ShortMethod)]
internal static void PackFromRgbPlanes(
Configuration configuration,
ReadOnlySpan<byte> redChannel,
ReadOnlySpan<byte> greenChannel,
ReadOnlySpan<byte> blueChannel,
Span<Rgb24> destination)
{
}
[MethodImpl(InliningOptions.ShortMethod)]
internal static void PackFromRgbPlanes(
Configuration configuration,
ReadOnlySpan<byte> redChannel,
ReadOnlySpan<byte> greenChannel,
ReadOnlySpan<byte> blueChannel,
Span<Rgba32> destination)
{
}
[MethodImpl(InliningOptions.ColdPath)] [MethodImpl(InliningOptions.ColdPath)]
private static void ConvertByteToNormalizedFloatRemainder(ReadOnlySpan<byte> source, Span<float> dest) private static void ConvertByteToNormalizedFloatRemainder(ReadOnlySpan<byte> source, Span<float> dest)
{ {
@ -241,5 +221,13 @@ namespace SixLabors.ImageSharp
nameof(source), nameof(source),
$"length should be divisible by {shouldBeDivisibleBy}!"); $"length should be divisible by {shouldBeDivisibleBy}!");
} }
private struct ByteTuple4
{
public byte V0;
public byte V1;
public byte V2;
public byte V3;
}
} }
} }

10
src/ImageSharp/ImageSharp.csproj
View File

@ -25,16 +25,16 @@
</ItemGroup> </ItemGroup>
<ItemGroup Condition=" $(TargetFramework.StartsWith('netstandard')) OR '$(TargetFramework)' == 'net472'"> <ItemGroup Condition=" $(TargetFramework.StartsWith('netstandard')) OR '$(TargetFramework)' == 'net472'">
<PackageReference Include="System.Numerics.Vectors" Version="4.5.0"/> <PackageReference Include="System.Numerics.Vectors" Version="4.5.0" />
<PackageReference Include="System.Buffers" Version="4.5.1" /> <PackageReference Include="System.Buffers" Version="4.5.1" />
<PackageReference Include="System.Memory" Version="4.5.4"/> <PackageReference Include="System.Memory" Version="4.5.4" />
</ItemGroup> </ItemGroup>
<ItemGroup Condition=" '$(TargetFramework)' == 'netstandard1.3'"> <ItemGroup Condition=" '$(TargetFramework)' == 'netstandard1.3'">
<PackageReference Include="System.IO.Compression" Version="4.3.0"/> <PackageReference Include="System.IO.Compression" Version="4.3.0" />
<PackageReference Include="System.IO.UnmanagedMemoryStream" Version="4.3.0" /> <PackageReference Include="System.IO.UnmanagedMemoryStream" Version="4.3.0" />
<PackageReference Include="System.Threading.Tasks.Parallel" Version="4.3.0"/> <PackageReference Include="System.Threading.Tasks.Parallel" Version="4.3.0" />
<PackageReference Include="System.ValueTuple" Version="4.5.0"/> <PackageReference Include="System.ValueTuple" Version="4.5.0" />
</ItemGroup> </ItemGroup>
<ItemGroup> <ItemGroup>

252
tests/ImageSharp.Benchmarks/General/PixelConversion/PixelConversion_PackFromRgbPlanes.cs
View File

@ -0,0 +1,252 @@
// Copyright (c) Six Labors.
// Licensed under the Apache License, Version 2.0.
using System;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
#if SUPPORTS_RUNTIME_INTRINSICS
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;
#endif
using BenchmarkDotNet.Attributes;
using SixLabors.ImageSharp.PixelFormats;
namespace SixLabors.ImageSharp.Benchmarks.General.PixelConversion
{
public unsafe class PixelConversion_PackFromRgbPlanes
{
private byte[] rBuf;
private byte[] gBuf;
private byte[] bBuf;
private Rgb24[] rgbBuf;
private Rgba32[] rgbaBuf;
private float[] rFloat;
private float[] gFloat;
private float[] bFloat;
private float[] rgbaFloat;
[Params(512)]
public int Count { get; set; }
[GlobalSetup]
public void Setup()
{
this.rBuf = new byte[this.Count];
this.gBuf = new byte[this.Count];
this.bBuf = new byte[this.Count];
this.rgbBuf = new Rgb24[this.Count];
this.rgbaBuf = new Rgba32[this.Count];
this.rFloat = new float[this.Count];
this.gFloat = new float[this.Count];
this.bFloat = new float[this.Count];
this.rgbaFloat = new float[this.Count * 4];
}
// [Benchmark(Baseline = true)]
public void Rgb24_Scalar_PerElement_Pinned()
{
fixed (byte* r = &this.rBuf[0])
fixed (byte* g = &this.gBuf[0])
fixed (byte* b = &this.bBuf[0])
fixed (Rgb24* rgb = &this.rgbBuf[0])
{
for (int i = 0; i < this.Count; i++)
{
Rgb24* d = rgb + i;
d->R = r[i];
d->G = g[i];
d->B = b[i];
}
}
}
[Benchmark]
public void Rgb24_Scalar_PerElement_Span()
{
Span<byte> r = this.rBuf;
Span<byte> g = this.rBuf;
Span<byte> b = this.rBuf;
Span<Rgb24> rgb = this.rgbBuf;
for (int i = 0; i < rgb.Length; i++)
{
ref Rgb24 d = ref rgb[i];
d.R = r[i];
d.G = g[i];
d.B = b[i];
}
}
[Benchmark(Baseline = true)]
public void Rgb24_Scalar_PerElement_Unsafe()
{
ref byte r = ref this.rBuf[0];
ref byte g = ref this.rBuf[0];
ref byte b = ref this.rBuf[0];
ref Rgb24 rgb = ref this.rgbBuf[0];
for (int i = 0; i < this.Count; i++)
{
ref Rgb24 d = ref Unsafe.Add(ref rgb, i);
d.R = Unsafe.Add(ref r, i);
d.G = Unsafe.Add(ref g, i);
d.B = Unsafe.Add(ref b, i);
}
}
[Benchmark]
public void Rgb24_Scalar_PerElement_Batched8()
{
ref Byte8 r = ref Unsafe.As<byte, Byte8>(ref this.rBuf[0]);
ref Byte8 g = ref Unsafe.As<byte, Byte8>(ref this.rBuf[0]);
ref Byte8 b = ref Unsafe.As<byte, Byte8>(ref this.rBuf[0]);
ref Rgb24 rgb = ref this.rgbBuf[0];
int count = this.Count / 8;
for (int i = 0; i < count; i++)
{
ref Rgb24 d0 = ref Unsafe.Add(ref rgb, i * 8);
ref Rgb24 d1 = ref Unsafe.Add(ref d0, 1);
ref Rgb24 d2 = ref Unsafe.Add(ref d0, 2);
ref Rgb24 d3 = ref Unsafe.Add(ref d0, 3);
ref Rgb24 d4 = ref Unsafe.Add(ref d0, 4);
ref Rgb24 d5 = ref Unsafe.Add(ref d0, 5);
ref Rgb24 d6 = ref Unsafe.Add(ref d0, 6);
ref Rgb24 d7 = ref Unsafe.Add(ref d0, 7);
ref Byte8 rr = ref Unsafe.Add(ref r, i);
ref Byte8 gg = ref Unsafe.Add(ref g, i);
ref Byte8 bb = ref Unsafe.Add(ref b, i);
d0.R = rr.V0;
d0.G = gg.V0;
d0.B = bb.V0;
d1.R = rr.V1;
d1.G = gg.V1;
d1.B = bb.V1;
d2.R = rr.V2;
d2.G = gg.V2;
d2.B = bb.V2;
d3.R = rr.V3;
d3.G = gg.V3;
d3.B = bb.V3;
d4.R = rr.V4;
d4.G = gg.V4;
d4.B = bb.V4;
d5.R = rr.V5;
d5.G = gg.V5;
d5.B = bb.V5;
d6.R = rr.V6;
d6.G = gg.V6;
d6.B = bb.V6;
d7.R = rr.V7;
d7.G = gg.V7;
d7.B = bb.V7;
}
}
[Benchmark]
public void Rgb24_Scalar_PerElement_Batched4()
{
ref Byte4 r = ref Unsafe.As<byte, Byte4>(ref this.rBuf[0]);
ref Byte4 g = ref Unsafe.As<byte, Byte4>(ref this.rBuf[0]);
ref Byte4 b = ref Unsafe.As<byte, Byte4>(ref this.rBuf[0]);
ref Rgb24 rgb = ref this.rgbBuf[0];
int count = this.Count / 4;
for (int i = 0; i < count; i++)
{
ref Rgb24 d0 = ref Unsafe.Add(ref rgb, i * 4);
ref Rgb24 d1 = ref Unsafe.Add(ref d0, 1);
ref Rgb24 d2 = ref Unsafe.Add(ref d0, 2);
ref Rgb24 d3 = ref Unsafe.Add(ref d0, 3);
ref Byte4 rr = ref Unsafe.Add(ref r, i);
ref Byte4 gg = ref Unsafe.Add(ref g, i);
ref Byte4 bb = ref Unsafe.Add(ref b, i);
d0.R = rr.V0;
d0.G = gg.V0;
d0.B = bb.V0;
d1.R = rr.V1;
d1.G = gg.V1;
d1.B = bb.V1;
d2.R = rr.V2;
d2.G = gg.V2;
d2.B = bb.V2;
d3.R = rr.V3;
d3.G = gg.V3;
d3.B = bb.V3;
}
}
#if SUPPORTS_RUNTIME_INTRINSICS
[Benchmark]
public void Rgba32_Vector_Float()
{
ref Vector256<float> rBase = ref Unsafe.As<float, Vector256<float>>(ref this.rFloat[0]);
ref Vector256<float> gBase = ref Unsafe.As<float, Vector256<float>>(ref this.gFloat[0]);
ref Vector256<float> bBase = ref Unsafe.As<float, Vector256<float>>(ref this.bFloat[0]);
ref Vector256<float> resultBase = ref Unsafe.As<float, Vector256<float>>(ref this.rgbaFloat[0]);
int count = this.Count / Vector256<float>.Count;
ref byte control = ref MemoryMarshal.GetReference(SimdUtils.HwIntrinsics.PermuteMaskEvenOdd8x32);
Vector256<int> vcontrol = Unsafe.As<byte, Vector256<int>>(ref control);
var va = Vector256.Create(1F);
for (int i = 0; i < count; i++)
{
Vector256<float> r = Unsafe.Add(ref rBase, i);
Vector256<float> g = Unsafe.Add(ref gBase, i);
Vector256<float> b = Unsafe.Add(ref bBase, i);
r = Avx2.PermuteVar8x32(r, vcontrol);
g = Avx2.PermuteVar8x32(g, vcontrol);
b = Avx2.PermuteVar8x32(b, vcontrol);
Vector256<float> vte = Avx.UnpackLow(r, b);
Vector256<float> vto = Avx.UnpackLow(g, va);
ref Vector256<float> destination = ref Unsafe.Add(ref resultBase, i * 4);
destination = Avx.UnpackLow(vte, vto);
Unsafe.Add(ref destination, 1) = Avx.UnpackHigh(vte, vto);
vte = Avx.UnpackHigh(r, b);
vto = Avx.UnpackHigh(g, va);
Unsafe.Add(ref destination, 2) = Avx.UnpackLow(vte, vto);
Unsafe.Add(ref destination, 3) = Avx.UnpackHigh(vte, vto);
}
}
#endif
#pragma warning disable SA1132
private struct Byte8
{
public byte V0, V1, V2, V3, V4, V5, V6, V7;
}
private struct Byte4
{
public byte V0, V1, V2, V3;
}
#pragma warning restore
}
}
Write Preview
Loading…
Cancel
Save