Skip to content
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

Jump to bottom

SSE2 / AVX2 of webp AlphaDecoding #1930

Merged
merged 10 commits into from
Jan 18, 2022
Merged

SSE2 / AVX2 of webp AlphaDecoding #1930

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
2d9ac58
AVX2 version of VerticalUnfilter
brianpopow Jan 6, 2022
4289b8c
Fix indenting
brianpopow Jan 6, 2022
d7ecfb5
SSE2 version of HorizontalUnfilter
brianpopow Jan 6, 2022
dae6503
Add tests for horizontal and vertical filter without hardware intrinsics
brianpopow Jan 7, 2022
6bd756e
Merge branch 'master' into bp/alphadecodingsse
brianpopow Jan 8, 2022
2e659fa
Merge branch 'master' into bp/alphadecodingsse
JimBobSquarePants Jan 11, 2022
d8db653
Apply suggestions from code review
brianpopow Jan 15, 2022
9d2a462
Merge branch 'master' into bp/alphadecodingsse
brianpopow Jan 15, 2022
6fdf643
Fix build errors
brianpopow Jan 15, 2022
4c4f27d
Merge branch 'master' into bp/alphadecodingsse
JimBobSquarePants Jan 18, 2022
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
82 changes: 73 additions & 9 deletions src/ImageSharp/Formats/Webp/AlphaDecoder.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,10 @@
using SixLabors.ImageSharp.Formats.Webp.BitReader;
using SixLabors.ImageSharp.Formats.Webp.Lossless;
using SixLabors.ImageSharp.Memory;
#if SUPPORTS_RUNTIME_INTRINSICS
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.X86;
#endif

namespace SixLabors.ImageSharp.Formats.Webp
{
Expand Down Expand Up @@ -307,34 +311,94 @@ private static void ColorIndexInverseTransformAlpha(

private static void HorizontalUnfilter(Span<byte> prev, Span<byte> input, Span<byte> dst, int width)
{
byte pred = (byte)(prev == null ? 0 : prev[0]);
#if SUPPORTS_RUNTIME_INTRINSICS
if (Sse2.IsSupported)
{
dst[0] = (byte)(input[0] + (prev.IsEmpty ? 0 : prev[0]));
if (width <= 1)
{
return;
}

nint i;
Vector128<int> last = Vector128<int>.Zero.WithElement(0, dst[0]);
ref byte srcRef = ref MemoryMarshal.GetReference(input);
for (i = 1; i + 8 <= width; i += 8)
{
var a0 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref srcRef, i)), 0);
Vector128<byte> a1 = Sse2.Add(a0.AsByte(), last.AsByte());
Vector128<byte> a2 = Sse2.ShiftLeftLogical128BitLane(a1, 1);
Vector128<byte> a3 = Sse2.Add(a1, a2);
Vector128<byte> a4 = Sse2.ShiftLeftLogical128BitLane(a3, 2);
Vector128<byte> a5 = Sse2.Add(a3, a4);
Vector128<byte> a6 = Sse2.ShiftLeftLogical128BitLane(a5, 4);
Vector128<byte> a7 = Sse2.Add(a5, a6);

ref byte outputRef = ref Unsafe.Add(ref MemoryMarshal.GetReference(dst), i);
Unsafe.As<byte, Vector64<byte>>(ref outputRef) = a7.GetLower();
last = Sse2.ShiftRightLogical(a7.AsInt64(), 56).AsInt32();
}

for (int i = 0; i < width; i++)
for (; i < width; ++i)
{
dst[(int)i] = (byte)(input[(int)i] + dst[(int)i - 1]);
}
}
else
#endif
{
byte val = (byte)(pred + input[i]);
pred = val;
dst[i] = val;
byte pred = (byte)(prev.IsEmpty ? 0 : prev[0]);

for (int i = 0; i < width; i++)
{
byte val = (byte)(pred + input[i]);
pred = val;
dst[i] = val;
}
}
}

private static void VerticalUnfilter(Span<byte> prev, Span<byte> input, Span<byte> dst, int width)
{
if (prev == null)
if (prev.IsEmpty)
{
HorizontalUnfilter(null, input, dst, width);
}
else
{
for (int i = 0; i < width; i++)
#if SUPPORTS_RUNTIME_INTRINSICS
if (Avx2.IsSupported)
{
dst[i] = (byte)(prev[i] + input[i]);
nint i;
int maxPos = width & ~31;
for (i = 0; i < maxPos; i += 32)
{
Vector256<int> a0 = Unsafe.As<byte, Vector256<int>>(ref Unsafe.Add(ref MemoryMarshal.GetReference(input), i));
Vector256<int> b0 = Unsafe.As<byte, Vector256<int>>(ref Unsafe.Add(ref MemoryMarshal.GetReference(prev), i));
Vector256<byte> c0 = Avx2.Add(a0.AsByte(), b0.AsByte());
ref byte outputRef = ref Unsafe.Add(ref MemoryMarshal.GetReference(dst), i);
Unsafe.As<byte, Vector256<byte>>(ref outputRef) = c0;
}

for (; i < width; i++)
{
dst[(int)i] = (byte)(prev[(int)i] + input[(int)i]);
}
}
else
#endif
{
for (int i = 0; i < width; i++)
{
dst[i] = (byte)(prev[i] + input[i]);
}
}
}
}

private static void GradientUnfilter(Span<byte> prev, Span<byte> input, Span<byte> dst, int width)
{
if (prev == null)
if (prev.IsEmpty)
{
HorizontalUnfilter(null, input, dst, width);
}
Expand Down
1 change: 0 additions & 1 deletion src/ImageSharp/Formats/Webp/Lossless/LosslessUtils.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,6 @@
// Licensed under the Apache License, Version 2.0.

using System;
using System.Numerics;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using SixLabors.ImageSharp.Memory;
Expand Down
10 changes: 5 additions & 5 deletions tests/ImageSharp.Tests/Formats/GeneralFormatTests.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -85,11 +85,11 @@ public void DecodeThenEncodeImageFromStreamShouldSucceed()
public static readonly TheoryData<string> QuantizerNames =
new()
{
nameof(KnownQuantizers.Octree),
nameof(KnownQuantizers.WebSafe),
nameof(KnownQuantizers.Werner),
nameof(KnownQuantizers.Wu)
};
nameof(KnownQuantizers.Octree),
nameof(KnownQuantizers.WebSafe),
nameof(KnownQuantizers.Werner),
nameof(KnownQuantizers.Wu)
};

[Theory]
[WithFile(TestImages.Png.CalliphoraPartial, nameof(QuantizerNames), PixelTypes.Rgba32)]
Expand Down
30 changes: 30 additions & 0 deletions tests/ImageSharp.Tests/Formats/WebP/WebpDecoderTests.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,10 @@ public class WebpDecoderTests

private static MagickReferenceDecoder ReferenceDecoder => new();

private static string TestImageLossyHorizontalFilterPath => Path.Combine(TestEnvironment.InputImagesDirectoryFullPath, Lossy.AlphaCompressedHorizontalFilter);

private static string TestImageLossyVerticalFilterPath => Path.Combine(TestEnvironment.InputImagesDirectoryFullPath, Lossy.AlphaCompressedVerticalFilter);

private static string TestImageLossySimpleFilterPath => Path.Combine(TestEnvironment.InputImagesDirectoryFullPath, Lossy.SimpleFilter02);

private static string TestImageLossyComplexFilterPath => Path.Combine(TestEnvironment.InputImagesDirectoryFullPath, Lossy.BikeComplexFilter);
Expand Down Expand Up @@ -365,6 +369,26 @@ public void WebpDecoder_ThrowImageFormatException_OnInvalidImages<TPixel>(TestIm
});

#if SUPPORTS_RUNTIME_INTRINSICS
private static void RunDecodeLossyWithHorizontalFilter()
{
var provider = TestImageProvider<Rgba32>.File(TestImageLossyHorizontalFilterPath);
using (Image<Rgba32> image = provider.GetImage(WebpDecoder))
{
image.DebugSave(provider);
image.CompareToOriginal(provider, ReferenceDecoder);
}
}

private static void RunDecodeLossyWithVerticalFilter()
{
var provider = TestImageProvider<Rgba32>.File(TestImageLossyVerticalFilterPath);
using (Image<Rgba32> image = provider.GetImage(WebpDecoder))
{
image.DebugSave(provider);
image.CompareToOriginal(provider, ReferenceDecoder);
}
}

private static void RunDecodeLossyWithSimpleFilterTest()
{
var provider = TestImageProvider<Rgba32>.File(TestImageLossySimpleFilterPath);
Expand All @@ -385,6 +409,12 @@ private static void RunDecodeLossyWithComplexFilterTest()
}
}

[Fact]
public void DecodeLossyWithHorizontalFilter_WithoutHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunDecodeLossyWithHorizontalFilter, HwIntrinsics.DisableHWIntrinsic);

[Fact]
public void DecodeLossyWithVerticalFilter_WithoutHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunDecodeLossyWithVerticalFilter, HwIntrinsics.DisableHWIntrinsic);

[Fact]
public void DecodeLossyWithSimpleFilterTest_WithoutHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunDecodeLossyWithSimpleFilterTest, HwIntrinsics.DisableHWIntrinsic);

Expand Down