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

Unroll loops

js/color-alpha-handling
James Jackson-South 6 years ago
parent
84a1d1a28b
commit
956d1a3c77
2 changed files with 88 additions and 19 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. 105
      src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
  2. 2
      src/ImageSharp/Common/Helpers/SimdUtils.Shuffle.cs

105
src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs
View File

@ -106,34 +106,72 @@ namespace SixLabors.ImageSharp
{ {
if (Avx.IsSupported) if (Avx.IsSupported)
{ {
int n = dest.Length / Vector256<float>.Count;
ref Vector256<float> sourceBase = ref Vector256<float> sourceBase =
ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(source)); ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(source));
ref Vector256<float> destBase = ref Vector256<float> destBase =
ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(dest)); ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(dest));
for (int i = 0; i < n; i++) int n = dest.Length / Vector256<float>.Count;
int m = ImageMaths.Modulo4(n);
int u = n - m;
for (int i = 0; i < u; i += 4)
{
ref Vector256<float> vd0 = ref Unsafe.Add(ref destBase, i);
ref Vector256<float> vs0 = ref Unsafe.Add(ref sourceBase, i);
vd0 = Avx.Permute(vs0, control);
Unsafe.Add(ref vd0, 1) = Avx.Permute(Unsafe.Add(ref vs0, 1), control);
Unsafe.Add(ref vd0, 2) = Avx.Permute(Unsafe.Add(ref vs0, 2), control);
Unsafe.Add(ref vd0, 3) = Avx.Permute(Unsafe.Add(ref vs0, 3), control);
}
if (m > 0)
{ {
Unsafe.Add(ref destBase, i) = Avx.Permute(Unsafe.Add(ref sourceBase, i), control); for (int i = u; i < n; i++)
{
Unsafe.Add(ref destBase, i) = Avx.Permute(Unsafe.Add(ref sourceBase, i), control);
}
} }
} }
else else
{ {
// Sse // Sse
int n = dest.Length / Vector128<float>.Count;
ref Vector128<float> sourceBase = ref Vector128<float> sourceBase =
ref Unsafe.As<float, Vector128<float>>(ref MemoryMarshal.GetReference(source)); ref Unsafe.As<float, Vector128<float>>(ref MemoryMarshal.GetReference(source));
ref Vector128<float> destBase = ref Vector128<float> destBase =
ref Unsafe.As<float, Vector128<float>>(ref MemoryMarshal.GetReference(dest)); ref Unsafe.As<float, Vector128<float>>(ref MemoryMarshal.GetReference(dest));
for (int i = 0; i < n; i++) int n = dest.Length / Vector128<float>.Count;
int m = ImageMaths.Modulo4(n);
int u = n - m;
for (int i = 0; i < u; i += 4)
{ {
Vector128<float> vs = Unsafe.Add(ref sourceBase, i); ref Vector128<float> vd0 = ref Unsafe.Add(ref destBase, i);
Unsafe.Add(ref destBase, i) = Sse.Shuffle(vs, vs, control); ref Vector128<float> vs0 = ref Unsafe.Add(ref sourceBase, i);
vd0 = Sse.Shuffle(vs0, vs0, control);
Vector128<float> vs1 = Unsafe.Add(ref vs0, 1);
Unsafe.Add(ref vd0, 1) = Sse.Shuffle(vs1, vs1, control);
Vector128<float> vs2 = Unsafe.Add(ref vs0, 2);
Unsafe.Add(ref vd0, 2) = Sse.Shuffle(vs2, vs2, control);
Vector128<float> vs3 = Unsafe.Add(ref vs0, 3);
Unsafe.Add(ref vd0, 3) = Sse.Shuffle(vs3, vs3, control);
}
if (m > 0)
{
for (int i = u; i < n; i++)
{
Vector128<float> vs = Unsafe.Add(ref sourceBase, i);
Unsafe.Add(ref destBase, i) = Sse.Shuffle(vs, vs, control);
}
} }
} }
} }
@ -146,8 +184,6 @@ namespace SixLabors.ImageSharp
{ {
if (Avx2.IsSupported) if (Avx2.IsSupported)
{ {
int n = dest.Length / Vector256<byte>.Count;
// I've chosen to do this for convenience while we determine what // I've chosen to do this for convenience while we determine what
// shuffle controls to add to the library. // shuffle controls to add to the library.
// We can add static ROS instances if need be in the future. // We can add static ROS instances if need be in the future.
@ -161,16 +197,32 @@ namespace SixLabors.ImageSharp
ref Vector256<byte> destBase = ref Vector256<byte> destBase =
ref Unsafe.As<byte, Vector256<byte>>(ref MemoryMarshal.GetReference(dest)); ref Unsafe.As<byte, Vector256<byte>>(ref MemoryMarshal.GetReference(dest));
for (int i = 0; i < n; i++) int n = dest.Length / Vector256<byte>.Count;
int m = ImageMaths.Modulo4(n);
int u = n - m;
for (int i = 0; i < u; i += 4)
{ {
Unsafe.Add(ref destBase, i) = Avx2.Shuffle(Unsafe.Add(ref sourceBase, i), vcm); ref Vector256<byte> vs0 = ref Unsafe.Add(ref sourceBase, i);
ref Vector256<byte> vd0 = ref Unsafe.Add(ref destBase, i);
vd0 = Avx2.Shuffle(vs0, vcm);
Unsafe.Add(ref vd0, 1) = Avx2.Shuffle(Unsafe.Add(ref vs0, 1), vcm);
Unsafe.Add(ref vd0, 2) = Avx2.Shuffle(Unsafe.Add(ref vs0, 2), vcm);
Unsafe.Add(ref vd0, 3) = Avx2.Shuffle(Unsafe.Add(ref vs0, 3), vcm);
}
if (m > 0)
{
for (int i = u; i < n; i++)
{
Unsafe.Add(ref destBase, i) = Avx2.Shuffle(Unsafe.Add(ref sourceBase, i), vcm);
}
} }
} }
else else
{ {
// Ssse3 // Ssse3
int n = dest.Length / Vector128<byte>.Count;
Span<byte> bytes = stackalloc byte[Vector128<byte>.Count]; Span<byte> bytes = stackalloc byte[Vector128<byte>.Count];
Shuffle.MmShuffleSpan(ref bytes, control); Shuffle.MmShuffleSpan(ref bytes, control);
Vector128<byte> vcm = Unsafe.As<byte, Vector128<byte>>(ref MemoryMarshal.GetReference(bytes)); Vector128<byte> vcm = Unsafe.As<byte, Vector128<byte>>(ref MemoryMarshal.GetReference(bytes));
@ -181,10 +233,27 @@ namespace SixLabors.ImageSharp
ref Vector128<byte> destBase = ref Vector128<byte> destBase =
ref Unsafe.As<byte, Vector128<byte>>(ref MemoryMarshal.GetReference(dest)); ref Unsafe.As<byte, Vector128<byte>>(ref MemoryMarshal.GetReference(dest));
for (int i = 0; i < n; i++) int n = dest.Length / Vector128<byte>.Count;
int m = ImageMaths.Modulo4(n);
int u = n - m;
for (int i = 0; i < u; i += 4)
{ {
Vector128<byte> vs = Unsafe.Add(ref sourceBase, i); ref Vector128<byte> vs0 = ref Unsafe.Add(ref sourceBase, i);
Unsafe.Add(ref destBase, i) = Ssse3.Shuffle(vs, vcm); ref Vector128<byte> vd0 = ref Unsafe.Add(ref destBase, i);
vd0 = Ssse3.Shuffle(vs0, vcm);
Unsafe.Add(ref vd0, 1) = Ssse3.Shuffle(Unsafe.Add(ref vs0, 1), vcm);
Unsafe.Add(ref vd0, 2) = Ssse3.Shuffle(Unsafe.Add(ref vs0, 2), vcm);
Unsafe.Add(ref vd0, 3) = Ssse3.Shuffle(Unsafe.Add(ref vs0, 3), vcm);
}
if (m > 0)
{
for (int i = u; i < n; i++)
{
Unsafe.Add(ref destBase, i) = Ssse3.Shuffle(Unsafe.Add(ref sourceBase, i), vcm);
}
} }
} }
} }

2
src/ImageSharp/Common/Helpers/SimdUtils.Shuffle.cs
View File

@ -229,7 +229,7 @@ namespace SixLabors.ImageSharp
public const byte ZYXW = (3 << 6) | (0 << 4) | (1 << 2) | 2; public const byte ZYXW = (3 << 6) | (0 << 4) | (1 << 2) | 2;
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]
public static byte MmShuffle(int p3, int p2, int p1, int p0) public static byte MmShuffle(byte p3, byte p2, byte p1, byte p0)
=> (byte)((p3 << 6) | (p2 << 4) | (p1 << 2) | p0); => (byte)((p3 << 6) | (p2 << 4) | (p1 << 2) | p0);
[MethodImpl(InliningOptions.ShortMethod)] [MethodImpl(InliningOptions.ShortMethod)]

Write Preview
Loading…
Cancel
Save