Skip to content

Optimize jpeg decoder based on traces. #1303

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
JimBobSquarePants merged 3 commits into from
Aug 4, 2020
Merged

Optimize jpeg decoder based on traces. #1303

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
53d078d
Optimize based on traces.
JimBobSquarePants Aug 3, 2020
c6d4dd9
Update src/ImageSharp/Common/Helpers/SimdUtils.cs
JimBobSquarePants Aug 4, 2020
1eaaf1a
Merge branch 'master' into js/jpeg-decode-perf
JimBobSquarePants Aug 4, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 23 additions & 8 deletions src/ImageSharp/Common/Helpers/SimdUtils.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,10 @@
using System.Numerics;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
#if SUPPORTS_RUNTIME_INTRINSICS
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;
#endif

namespace SixLabors.ImageSharp
{
Expand All @@ -28,7 +32,7 @@ internal static partial class SimdUtils
[MethodImpl(MethodImplOptions.AggressiveInlining)]
internal static Vector4 PseudoRound(this Vector4 v)
{
var sign = Vector4Utilities.FastClamp(v, new Vector4(-1), new Vector4(1));
Vector4 sign = Vector4Utilities.FastClamp(v, new Vector4(-1), new Vector4(1));

return v + (sign * 0.5f);
}
Expand All @@ -44,13 +48,24 @@ internal static Vector4 PseudoRound(this Vector4 v)
[MethodImpl(MethodImplOptions.AggressiveInlining)]
internal static Vector<float> FastRound(this Vector<float> v)
{
var magic0 = new Vector<int>(int.MinValue); // 0x80000000
Vector<float> sgn0 = Vector.AsVectorSingle(magic0);
Vector<float> and0 = Vector.BitwiseAnd(sgn0, v);
Vector<float> or0 = Vector.BitwiseOr(and0, new Vector<float>(8388608.0f));
Vector<float> add0 = Vector.Add(v, or0);
Vector<float> sub0 = Vector.Subtract(add0, or0);
return sub0;
#if SUPPORTS_RUNTIME_INTRINSICS

if (Avx2.IsSupported)
{
ref Vector256<float> v256 = ref Unsafe.As<Vector<float>, Vector256<float>>(ref v);
Vector256<float> vRound = Avx.RoundToNearestInteger(v256);
return Unsafe.As<Vector256<float>, Vector<float>>(ref vRound);
}
else
#endif
{
var magic0 = new Vector<int>(int.MinValue); // 0x80000000
var sgn0 = Vector.AsVectorSingle(magic0);
var and0 = Vector.BitwiseAnd(sgn0, v);
var or0 = Vector.BitwiseOr(and0, new Vector<float>(8388608.0f));
var add0 = Vector.Add(v, or0);
return Vector.Subtract(add0, or0);
}
}

/// <summary>
Expand Down
23 changes: 11 additions & 12 deletions src/ImageSharp/Formats/Jpeg/Components/Decoder/HuffmanScanBuffer.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -93,25 +93,24 @@ public void FillBuffer()
public unsafe int DecodeHuffman(ref HuffmanTable h)
{
this.CheckBits();
int v = this.PeekBits(JpegConstants.Huffman.LookupBits);
int symbol = h.LookaheadValue[v];
int size = h.LookaheadSize[v];
int index = this.PeekBits(JpegConstants.Huffman.LookupBits);
int size = h.LookaheadSize[index];

if (size == JpegConstants.Huffman.SlowBits)
if (size < JpegConstants.Huffman.SlowBits)
{
ulong x = this.data << (JpegConstants.Huffman.RegisterSize - this.remainingBits);
while (x > h.MaxCode[size])
{
size++;
}
this.remainingBits -= size;
return h.LookaheadValue[index];
}

v = (int)(x >> (JpegConstants.Huffman.RegisterSize - size));
symbol = h.Values[(h.ValOffset[size] + v) & 0xFF];
ulong x = this.data << (JpegConstants.Huffman.RegisterSize - this.remainingBits);
while (x > h.MaxCode[size])
{
size++;
}

this.remainingBits -= size;

return symbol;
return h.Values[(h.ValOffset[size] + (int)(x >> (JpegConstants.Huffman.RegisterSize - size))) & 0xFF];
}

[MethodImpl(InliningOptions.ShortMethod)]
Expand Down
35 changes: 15 additions & 20 deletions tests/ImageSharp.Benchmarks/Codecs/Jpeg/DecodeJpegParseStreamOnly.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,6 @@
// Copyright (c) Six Labors.
// Licensed under the Apache License, Version 2.0.

using System.Drawing;
using System.IO;
using BenchmarkDotNet.Attributes;

Expand All @@ -15,7 +14,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
[Config(typeof(Config.ShortClr))]
public class DecodeJpegParseStreamOnly
{
[Params(TestImages.Jpeg.Baseline.Jpeg420Exif)]
[Params(TestImages.Jpeg.BenchmarkSuite.Lake_Small444YCbCr)]
Copy link
Member Author

@JimBobSquarePants JimBobSquarePants Aug 3, 2020

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Updated to match the full decoding benchmark so we can more easily see the difference.

public string TestImage { get; set; }

private string TestImageFullPath => Path.Combine(TestEnvironment.InputImagesDirectoryFullPath, this.TestImage);
Expand All @@ -37,7 +36,7 @@ public SDSize JpegSystemDrawing()
}

[Benchmark(Description = "JpegDecoderCore.ParseStream")]
public void ParseStreamPdfJs()
public void ParseStream()
{
using var memoryStream = new MemoryStream(this.jpegBytes);
using var bufferedStream = new BufferedReadStream(Configuration.Default, memoryStream);
Expand All @@ -46,22 +45,18 @@ public void ParseStreamPdfJs()
decoder.ParseStream(bufferedStream);
decoder.Dispose();
}

// RESULTS (2019 April 23):
//
// BenchmarkDotNet=v0.11.3, OS=Windows 10.0.17763.437 (1809/October2018Update/Redstone5)
// Intel Core i7-6600U CPU 2.60GHz (Skylake), 1 CPU, 4 logical and 2 physical cores
// .NET Core SDK=2.2.202
// [Host] : .NET Core 2.1.9 (CoreCLR 4.6.27414.06, CoreFX 4.6.27415.01), 64bit RyuJIT
// Clr : .NET Framework 4.7.2 (CLR 4.0.30319.42000), 64bit RyuJIT-v4.7.3362.0
// Core : .NET Core 2.1.9 (CoreCLR 4.6.27414.06, CoreFX 4.6.27415.01), 64bit RyuJIT
//
// | Method | Job | Runtime | TestImage | Mean | Error | StdDev | Ratio | RatioSD | Gen 0 | Gen 1 | Gen 2 | Allocated |
// |---------------------------- |----- |-------- |--------------------- |---------:|---------:|----------:|------:|--------:|---------:|------:|------:|----------:|
// | 'System.Drawing FULL' | Clr | Clr | Jpg/b(...)f.jpg [28] | 18.69 ms | 8.273 ms | 0.4535 ms | 1.00 | 0.00 | 343.7500 | - | - | 757.89 KB |
// | JpegDecoderCore.ParseStream | Clr | Clr | Jpg/b(...)f.jpg [28] | 15.76 ms | 4.266 ms | 0.2339 ms | 0.84 | 0.03 | - | - | - | 11.83 KB |
// | | | | | | | | | | | | | |
// | 'System.Drawing FULL' | Core | Core | Jpg/b(...)f.jpg [28] | 17.68 ms | 2.711 ms | 0.1486 ms | 1.00 | 0.00 | 343.7500 | - | - | 757.04 KB |
// | JpegDecoderCore.ParseStream | Core | Core | Jpg/b(...)f.jpg [28] | 14.27 ms | 3.671 ms | 0.2012 ms | 0.81 | 0.00 | - | - | - | 11.76 KB |
}

/*
| Method | Job | Runtime | TestImage | Mean | Error | StdDev | Ratio | Gen 0 | Gen 1 | Gen 2 | Allocated |
|---------------------------- |----------- |-------------- |--------------------- |---------:|----------:|----------:|------:|--------:|------:|------:|----------:|
| 'System.Drawing FULL' | Job-HITJFX | .NET 4.7.2 | Jpg/b(...)e.jpg [21] | 5.828 ms | 0.9885 ms | 0.0542 ms | 1.00 | 46.8750 | - | - | 211566 B |
| JpegDecoderCore.ParseStream | Job-HITJFX | .NET 4.7.2 | Jpg/b(...)e.jpg [21] | 5.833 ms | 0.2923 ms | 0.0160 ms | 1.00 | - | - | - | 12416 B |
| | | | | | | | | | | | |
| 'System.Drawing FULL' | Job-WPSKZD | .NET Core 2.1 | Jpg/b(...)e.jpg [21] | 6.018 ms | 2.1374 ms | 0.1172 ms | 1.00 | 46.8750 | - | - | 210768 B |
| JpegDecoderCore.ParseStream | Job-WPSKZD | .NET Core 2.1 | Jpg/b(...)e.jpg [21] | 4.382 ms | 0.9009 ms | 0.0494 ms | 0.73 | - | - | - | 12360 B |
| | | | | | | | | | | | |
| 'System.Drawing FULL' | Job-ZLSNRP | .NET Core 3.1 | Jpg/b(...)e.jpg [21] | 5.714 ms | 0.4078 ms | 0.0224 ms | 1.00 | - | - | - | 176 B |
| JpegDecoderCore.ParseStream | Job-ZLSNRP | .NET Core 3.1 | Jpg/b(...)e.jpg [21] | 4.239 ms | 1.0943 ms | 0.0600 ms | 0.74 | - | - | - | 12406 B |
*/
}
52 changes: 12 additions & 40 deletions tests/ImageSharp.Benchmarks/Codecs/Jpeg/DecodeJpeg_ImageSpecific.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -90,45 +90,17 @@ public Size JpegImageSharp()
}
}

// RESULTS (2018 November 4):
//
// BenchmarkDotNet=v0.10.14, OS=Windows 10.0.17134
// Intel Core i7-7700HQ CPU 2.80GHz (Kaby Lake), 1 CPU, 8 logical and 4 physical cores
// Frequency=2742191 Hz, Resolution=364.6719 ns, Timer=TSC
// .NET Core SDK=2.1.403
// [Host] : .NET Core 2.1.5 (CoreCLR 4.6.26919.02, CoreFX 4.6.26919.02), 64bit RyuJIT
//
// Method | TestImage | Mean | Error | StdDev | Scaled | ScaledSD | Gen 0 | Gen 1 | Gen 2 | Allocated |
// ------------------------------- |-------------------------------------------- |-----------:|-----------:|----------:|-------:|---------:|----------:|---------:|---------:|------------:|
// 'Decode Jpeg - System.Drawing' | Jpg/baseline/Lake.jpg | 6.117 ms | 0.3923 ms | 0.0222 ms | 1.00 | 0.00 | 62.5000 | - | - | 205.83 KB |
// 'Decode Jpeg - ImageSharp' | Jpg/baseline/Lake.jpg | 18.126 ms | 0.6023 ms | 0.0340 ms | 2.96 | 0.01 | - | - | - | 19.97 KB |
// | | | | | | | | | | |
// 'Decode Jpeg - System.Drawing' | Jpg/baseline/jpeg420exif.jpg | 17.063 ms | 2.6096 ms | 0.1474 ms | 1.00 | 0.00 | 218.7500 | - | - | 757.04 KB |
// 'Decode Jpeg - ImageSharp' | Jpg/baseline/jpeg420exif.jpg | 41.366 ms | 1.0115 ms | 0.0572 ms | 2.42 | 0.02 | - | - | - | 21.94 KB |
// | | | | | | | | | | |
// 'Decode Jpeg - System.Drawing' | Jpg/issues/Issue518-Bad-RST-Progressive.jpg | 428.282 ms | 94.9163 ms | 5.3629 ms | 1.00 | 0.00 | 2375.0000 | - | - | 7403.76 KB |
// 'Decode Jpeg - ImageSharp' | Jpg/issues/Issue518-Bad-RST-Progressive.jpg | 386.698 ms | 33.0065 ms | 1.8649 ms | 0.90 | 0.01 | 125.0000 | 125.0000 | 125.0000 | 35186.97 KB |
// | | | | | | | | | | |
// 'Decode Jpeg - System.Drawing' | Jpg/issues/issue750-exif-tranform.jpg | 95.192 ms | 3.1762 ms | 0.1795 ms | 1.00 | 0.00 | 1750.0000 | - | - | 5492.63 KB |
// 'Decode Jpeg - ImageSharp' | Jpg/issues/issue750-exif-tranform.jpg | 230.158 ms | 48.8128 ms | 2.7580 ms | 2.42 | 0.02 | 312.5000 | 312.5000 | 312.5000 | 58834.66 KB |

// RESULTS (2019 April 23):
//
// BenchmarkDotNet=v0.11.5, OS=Windows 10.0.17763.437 (1809/October2018Update/Redstone5)
// Intel Core i7-6600U CPU 2.60GHz (Skylake), 1 CPU, 4 logical and 2 physical cores
// .NET Core SDK=2.2.202
// [Host] : .NET Core 2.1.9 (CoreCLR 4.6.27414.06, CoreFX 4.6.27415.01), 64bit RyuJIT
// Core : .NET Core 2.1.9 (CoreCLR 4.6.27414.06, CoreFX 4.6.27415.01), 64bit RyuJIT
//
// | Method | TestImage | Mean | Error | StdDev | Ratio | RatioSD | Gen 0 | Gen 1 | Gen 2 | Allocated |
// |------------------------------- |--------------------- |-----------:|-----------:|-----------:|------:|--------:|----------:|------:|------:|------------:|
// | 'Decode Jpeg - System.Drawing' | Jpg/b(...)e.jpg [21] | 6.957 ms | 9.618 ms | 0.5272 ms | 1.00 | 0.00 | 93.7500 | - | - | 205.83 KB |
// | 'Decode Jpeg - ImageSharp' | Jpg/b(...)e.jpg [21] | 18.348 ms | 8.876 ms | 0.4865 ms | 2.65 | 0.23 | - | - | - | 14.49 KB |
// | | | | | | | | | | | |
// | 'Decode Jpeg - System.Drawing' | Jpg/b(...)f.jpg [28] | 18.687 ms | 11.632 ms | 0.6376 ms | 1.00 | 0.00 | 343.7500 | - | - | 757.04 KB |
// | 'Decode Jpeg - ImageSharp' | Jpg/b(...)f.jpg [28] | 41.990 ms | 25.514 ms | 1.3985 ms | 2.25 | 0.10 | - | - | - | 15.48 KB |
// | | | | | | | | | | | |
// | 'Decode Jpeg - System.Drawing' | Jpg/i(...)e.jpg [43] | 477.265 ms | 732.126 ms | 40.1303 ms | 1.00 | 0.00 | 3000.0000 | - | - | 7403.76 KB |
// | 'Decode Jpeg - ImageSharp' | Jpg/i(...)e.jpg [43] | 348.545 ms | 91.480 ms | 5.0143 ms | 0.73 | 0.06 | - | - | - | 35177.21 KB |
/*
| Method | TestImage | Mean | Error | StdDev | Ratio | RatioSD | Gen 0 | Gen 1 | Gen 2 | Allocated |
|------------------------------- |--------------------- |-----------:|------------:|-----------:|------:|--------:|------:|------:|------:|-----------:|
| 'Decode Jpeg - System.Drawing' | Jpg/b(...)e.jpg [21] | 5.122 ms | 1.3978 ms | 0.0766 ms | 1.00 | 0.00 | - | - | - | 176 B |
| 'Decode Jpeg - ImageSharp' | Jpg/b(...)e.jpg [21] | 11.991 ms | 0.2514 ms | 0.0138 ms | 2.34 | 0.03 | - | - | - | 15816 B |
| | | | | | | | | | | |
| 'Decode Jpeg - System.Drawing' | Jpg/b(...)f.jpg [28] | 14.943 ms | 1.8410 ms | 0.1009 ms | 1.00 | 0.00 | - | - | - | 176 B |
| 'Decode Jpeg - ImageSharp' | Jpg/b(...)f.jpg [28] | 29.759 ms | 1.5452 ms | 0.0847 ms | 1.99 | 0.01 | - | - | - | 16824 B |
| | | | | | | | | | | |
| 'Decode Jpeg - System.Drawing' | Jpg/i(...)e.jpg [43] | 388.229 ms | 382.8946 ms | 20.9877 ms | 1.00 | 0.00 | - | - | - | 216 B |
| 'Decode Jpeg - ImageSharp' | Jpg/i(...)e.jpg [43] | 276.490 ms | 195.5104 ms | 10.7166 ms | 0.71 | 0.01 | - | - | - | 36022368 B |
*/
}
}