Skip to content

Commit

Permalink
Merge pull request #1857 from SixLabors/bp/quantizeblockavx
Browse files Browse the repository at this point in the history 
Add AVX2 version of QuantizeBlock
  • Loading branch information
@brianpopow
brianpopow authored Dec 2, 2021
2 parents f44f11f + 909525c commit 0517fec
Show file tree
Hide file tree
Showing 2 changed files with 76 additions and 4 deletions.
75 changes: 72 additions & 3 deletions src/ImageSharp/Formats/Webp/Lossy/QuantEnc.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,12 @@ internal static unsafe class QuantEnc
#if SUPPORTS_RUNTIME_INTRINSICS
private static readonly Vector128<short> MaxCoeff2047 = Vector128.Create((short)MaxLevel);

private static readonly Vector256<short> MaxCoeff2047Vec256 = Vector256.Create((short)MaxLevel);

private static readonly Vector256<byte> Cst256 = Vector256.Create(0, 1, 2, 3, 8, 9, 254, 255, 10, 11, 4, 5, 6, 7, 12, 13, 2, 3, 8, 9, 10, 11, 4, 5, 254, 255, 6, 7, 12, 13, 14, 15);

private static readonly Vector256<byte> Cst78 = Vector256.Create(254, 255, 254, 255, 254, 255, 254, 255, 14, 15, 254, 255, 254, 255, 254, 255, 254, 255, 254, 255, 254, 255, 0, 1, 254, 255, 254, 255, 254, 255, 254, 255);

private static readonly Vector128<byte> CstLo = Vector128.Create(0, 1, 2, 3, 8, 9, 254, 255, 10, 11, 4, 5, 6, 7, 12, 13);

private static readonly Vector128<byte> Cst7 = Vector128.Create(254, 255, 254, 255, 254, 255, 254, 255, 14, 15, 254, 255, 254, 255, 254, 255);
Expand Down Expand Up @@ -531,7 +537,70 @@ public static int Quantize2Blocks(Span<short> input, Span<short> output, ref Vp8
public static int QuantizeBlock(Span<short> input, Span<short> output, ref Vp8Matrix mtx)
{
#if SUPPORTS_RUNTIME_INTRINSICS
if (Sse41.IsSupported)
if (Avx2.IsSupported)
{
// Load all inputs.
Vector256<short> input0 = Unsafe.As<short, Vector256<short>>(ref MemoryMarshal.GetReference(input));
Vector256<ushort> iq0 = Unsafe.As<ushort, Vector256<ushort>>(ref mtx.IQ[0]);
Vector256<ushort> q0 = Unsafe.As<ushort, Vector256<ushort>>(ref mtx.Q[0]);

// coeff = abs(in)
Vector256<ushort> coeff0 = Avx2.Abs(input0);

// coeff = abs(in) + sharpen
Vector256<short> sharpen0 = Unsafe.As<short, Vector256<short>>(ref mtx.Sharpen[0]);
Avx2.Add(coeff0.AsInt16(), sharpen0);

// out = (coeff * iQ + B) >> QFIX
// doing calculations with 32b precision (QFIX=17)
// out = (coeff * iQ)
Vector256<ushort> coeffiQ0H = Avx2.MultiplyHigh(coeff0, iq0);
Vector256<ushort> coeffiQ0L = Avx2.MultiplyLow(coeff0, iq0);
Vector256<ushort> out00 = Avx2.UnpackLow(coeffiQ0L, coeffiQ0H);
Vector256<ushort> out08 = Avx2.UnpackHigh(coeffiQ0L, coeffiQ0H);

// out = (coeff * iQ + B)
Vector256<uint> bias00 = Unsafe.As<uint, Vector256<uint>>(ref mtx.Bias[0]);
Vector256<uint> bias08 = Unsafe.As<uint, Vector256<uint>>(ref mtx.Bias[8]);
out00 = Avx2.Add(out00.AsInt32(), bias00.AsInt32()).AsUInt16();
out08 = Avx2.Add(out08.AsInt32(), bias08.AsInt32()).AsUInt16();

// out = QUANTDIV(coeff, iQ, B, QFIX)
out00 = Avx2.ShiftRightArithmetic(out00.AsInt32(), WebpConstants.QFix).AsUInt16();
out08 = Avx2.ShiftRightArithmetic(out08.AsInt32(), WebpConstants.QFix).AsUInt16();

// Pack result as 16b.
Vector256<short> out0 = Avx2.PackSignedSaturate(out00.AsInt32(), out08.AsInt32());

// if (coeff > 2047) coeff = 2047
out0 = Avx2.Min(out0, MaxCoeff2047Vec256);

// Put the sign back.
out0 = Avx2.Sign(out0, input0);

// in = out * Q
input0 = Avx2.MultiplyLow(out0, q0.AsInt16());
ref short inputRef = ref MemoryMarshal.GetReference(input);
Unsafe.As<short, Vector256<short>>(ref inputRef) = input0;

// zigzag the output before storing it.
Vector256<byte> tmp256 = Avx2.Shuffle(out0.AsByte(), Cst256);
Vector256<byte> tmp78 = Avx2.Shuffle(out0.AsByte(), Cst78);

// Reverse the order of the 16-byte lanes.
Vector256<byte> tmp87 = Avx2.Permute2x128(tmp78, tmp78, 1);
Vector256<short> outZ = Avx2.Or(tmp256, tmp87).AsInt16();

ref short outputRef = ref MemoryMarshal.GetReference(output);
Unsafe.As<short, Vector256<short>>(ref outputRef) = outZ;

Vector256<sbyte> packedOutput = Avx2.PackSignedSaturate(outZ, outZ);

// Detect if all 'out' values are zeros or not.
Vector256<sbyte> cmpeq = Avx2.CompareEqual(packedOutput, Vector256<sbyte>.Zero);
return Avx2.MoveMask(cmpeq) != -1 ? 1 : 0;
}
else if (Sse41.IsSupported)
{
// Load all inputs.
Vector128<short> input0 = Unsafe.As<short, Vector128<short>>(ref MemoryMarshal.GetReference(input));
Expand Down Expand Up @@ -579,15 +648,15 @@ public static int QuantizeBlock(Span<short> input, Span<short> output, ref Vp8Ma
out08 = Sse2.ShiftRightArithmetic(out08.AsInt32(), WebpConstants.QFix).AsUInt16();
out12 = Sse2.ShiftRightArithmetic(out12.AsInt32(), WebpConstants.QFix).AsUInt16();

// pack result as 16b
// Pack result as 16b.
Vector128<short> out0 = Sse2.PackSignedSaturate(out00.AsInt32(), out04.AsInt32());
Vector128<short> out8 = Sse2.PackSignedSaturate(out08.AsInt32(), out12.AsInt32());

// if (coeff > 2047) coeff = 2047
out0 = Sse2.Min(out0, MaxCoeff2047);
out8 = Sse2.Min(out8, MaxCoeff2047);

// put sign back
// Put the sign back.
out0 = Ssse3.Sign(out0, input0);
out8 = Ssse3.Sign(out8, input8);

Expand Down
5 changes: 4 additions & 1 deletion tests/ImageSharp.Tests/Formats/WebP/QuantEncTests.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -47,7 +47,10 @@ private static unsafe void RunQuantizeBlockTest()
public void QuantizeBlock_WithHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunQuantizeBlockTest, HwIntrinsics.AllowAll);

[Fact]
public void QuantizeBlock_WithoutHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunQuantizeBlockTest, HwIntrinsics.DisableHWIntrinsic);
public void QuantizeBlock_WithoutSSE2_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunQuantizeBlockTest, HwIntrinsics.DisableSSE2);

[Fact]
public void QuantizeBlock_WithoutAVX2_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunQuantizeBlockTest, HwIntrinsics.DisableAVX2);
#endif
}
}

0 comments on commit 0517fec

Please sign in to comment.