Skip to content

Add AVX2 version of CollectHistogram #1882

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
merged 4 commits into from
Dec 12, 2021
Merged

Add AVX2 version of CollectHistogram #1882

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2a48c94
Add AVX2 version of CollectHistogram
brianpopow Dec 8, 2021
5f2b207
Add collect histogram tests
brianpopow Dec 8, 2021
629d585
Merge branch 'master' into bp/collecthistogramavx
brianpopow Dec 9, 2021
5cc83fb
Merge branch 'master' into bp/collecthistogramavx
brianpopow Dec 12, 2021
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
44 changes: 40 additions & 4 deletions src/ImageSharp/Formats/Webp/Lossy/Vp8Histogram.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,11 @@

using System;
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.Formats.Webp.Lossy
{
Expand All @@ -19,6 +24,10 @@ internal sealed class Vp8Histogram
/// </summary>
private const int MaxCoeffThresh = 31;

#if SUPPORTS_RUNTIME_INTRINSICS
private static readonly Vector256<short> MaxCoeffThreshVec = Vector256.Create((short)MaxCoeffThresh);
#endif

private int maxValue;

private int lastNonZero;
Expand Down Expand Up @@ -52,11 +61,38 @@ public void CollectHistogram(Span<byte> reference, Span<byte> pred, int startBlo
Vp8Encoding.FTransform(reference.Slice(WebpLookupTables.Vp8DspScan[j]), pred.Slice(WebpLookupTables.Vp8DspScan[j]), this.output, this.scratch);

// Convert coefficients to bin.
for (int k = 0; k < 16; ++k)
#if SUPPORTS_RUNTIME_INTRINSICS
if (Avx2.IsSupported)
{
int v = Math.Abs(this.output[k]) >> 3;
int clippedValue = ClipMax(v, MaxCoeffThresh);
++this.distribution[clippedValue];
// Load.
ref short outputRef = ref MemoryMarshal.GetReference<short>(this.output);
Copy link
Contributor

@gfoidl gfoidl Dec 9, 2021

Choose a reason for hiding this comment

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

Suggested change
ref short outputRef = ref MemoryMarshal.GetReference<short>(this.output);
ref short outputRef = ref MemoryMarshal.GetArrayDataReference(this.output);

Avoids the intermediate array -> Span step.

But this API is only available from .NET 5 onwards 😢

So maybe just leave the code as is, and remember this API for other uses 😉

Copy link
Collaborator Author

@brianpopow brianpopow Dec 9, 2021

Choose a reason for hiding this comment

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

We should definitely start using this once we get rid of supporting the older frameworks

Vector256<byte> out0 = Unsafe.As<short, Vector256<byte>>(ref outputRef);

// v = abs(out) >> 3
Vector256<ushort> abs0 = Avx2.Abs(out0.AsInt16());
Vector256<short> v0 = Avx2.ShiftRightArithmetic(abs0.AsInt16(), 3);

// bin = min(v, MAX_COEFF_THRESH)
Vector256<short> min0 = Avx2.Min(v0, MaxCoeffThreshVec);

// Store.
Unsafe.As<short, Vector256<short>>(ref outputRef) = min0;

// Convert coefficients to bin.
for (int k = 0; k < 16; ++k)
{
++this.distribution[this.output[k]];
}
}
else
#endif
{
for (int k = 0; k < 16; ++k)
{
int v = Math.Abs(this.output[k]) >> 3;
int clippedValue = ClipMax(v, MaxCoeffThresh);
++this.distribution[clippedValue];
}
}
}

Expand Down
111 changes: 111 additions & 0 deletions tests/ImageSharp.Tests/Formats/WebP/Vp8HistogramTests.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,7 @@

using System.Collections.Generic;
using SixLabors.ImageSharp.Formats.Webp.Lossy;
using SixLabors.ImageSharp.Tests.TestUtilities;
using Xunit;

namespace SixLabors.ImageSharp.Tests.Formats.Webp
Expand Down Expand Up @@ -67,6 +68,108 @@ public static IEnumerable<object[]> Data
}
}

private static void RunCollectHistogramTest()
{
// arrange
var histogram = new Vp8Histogram();

byte[] reference =
{
154, 154, 151, 151, 149, 148, 151, 157, 163, 163, 154, 132, 102, 98, 104, 108, 107, 104, 104, 103,
101, 106, 123, 119, 170, 171, 172, 171, 168, 175, 171, 173, 151, 151, 149, 150, 147, 147, 146, 159,
164, 165, 154, 129, 92, 90, 101, 105, 104, 103, 104, 101, 100, 105, 123, 117, 172, 172, 172, 168,
170, 177, 170, 175, 151, 149, 150, 150, 147, 147, 156, 161, 161, 161, 151, 126, 93, 90, 102, 107,
104, 103, 104, 101, 104, 104, 122, 117, 172, 172, 170, 168, 170, 177, 172, 175, 150, 149, 152, 151,
148, 151, 160, 159, 157, 157, 148, 133, 96, 90, 103, 107, 104, 104, 101, 100, 102, 102, 121, 117,
170, 170, 169, 171, 171, 179, 173, 175, 149, 151, 152, 151, 148, 154, 162, 157, 154, 154, 151, 132,
92, 89, 101, 108, 104, 102, 101, 101, 103, 103, 123, 118, 171, 168, 177, 173, 171, 178, 172, 176,
152, 152, 152, 151, 154, 162, 161, 155, 149, 157, 156, 129, 92, 87, 101, 107, 102, 100, 107, 100,
101, 102, 123, 118, 170, 175, 182, 172, 171, 179, 173, 175, 152, 151, 154, 155, 160, 162, 161, 153,
150, 156, 153, 129, 92, 91, 102, 106, 100, 109, 115, 99, 101, 102, 124, 120, 171, 179, 178, 172,
171, 181, 171, 173, 154, 154, 154, 162, 160, 158, 156, 152, 153, 157, 151, 128, 86, 86, 102, 105,
102, 122, 114, 99, 101, 102, 125, 120, 178, 173, 177, 172, 171, 180, 172, 173, 154, 152, 158, 163,
150, 148, 148, 156, 151, 158, 152, 129, 87, 87, 101, 105, 204, 204, 204, 204, 204, 204, 204, 204,
204, 204, 204, 204, 204, 204, 204, 204, 154, 151, 165, 156, 141, 137, 146, 158, 152, 159, 152, 133,
90, 88, 99, 106, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204,
154, 160, 164, 150, 126, 127, 149, 159, 155, 161, 153, 131, 84, 86, 97, 103, 204, 204, 204, 204,
204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 157, 167, 157, 137, 102, 128, 155, 161,
157, 159, 154, 134, 84, 82, 97, 102, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204,
204, 204, 204, 204, 163, 163, 150, 113, 78, 132, 156, 162, 159, 160, 154, 132, 83, 78, 91, 97, 204,
204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 163, 157, 137, 80, 78,
131, 154, 163, 157, 159, 149, 131, 82, 77, 94, 100, 204, 204, 204, 204, 204, 204, 204, 204, 204,
204, 204, 204, 204, 204, 204, 204, 159, 151, 108, 72, 88, 132, 156, 162, 159, 157, 151, 130, 79, 78,
95, 102, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 204, 151, 130,
82, 82, 89, 134, 154, 161, 161, 157, 152, 129, 81, 77, 95, 102, 204, 204, 204, 204, 204, 204, 204,
204, 204, 204, 204, 204, 204, 204, 204, 204
};
byte[] pred =
{
128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128,
128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128,
128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128,
128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128,
128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128,
128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128,
128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128,
128, 128, 128, 128, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127,
127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127,
127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127,
127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127,
127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127,
127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127,
127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127,
127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127,
127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127,
127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129,
129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 127, 127, 127, 127, 127, 127, 127, 127,
127, 127, 127, 127, 127, 127, 127, 127, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129, 129,
129, 129, 129, 129
};
int expectedAlpha = 146;

// act
histogram.CollectHistogram(reference, pred, 0, 10);
int actualAlpha = histogram.GetAlpha();

// assert
Assert.Equal(expectedAlpha, actualAlpha);
}

[Fact]
public void RunCollectHistogramTest_Works() => RunCollectHistogramTest();

[Fact]
public void GetAlpha_WithEmptyHistogram_Works()
{
Expand Down Expand Up @@ -111,5 +214,13 @@ public void Merge_Works(byte[] reference, byte[] pred)
// assert
Assert.Equal(1054, alpha);
}

#if SUPPORTS_RUNTIME_INTRINSICS
[Fact]
public void CollectHistogramTest_WithHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunCollectHistogramTest, HwIntrinsics.AllowAll);

[Fact]
public void CollectHistogramTest_WithoutHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunCollectHistogramTest, HwIntrinsics.DisableHWIntrinsic);
#endif
}
}