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

Avoid pinning, avoid using LoadScalarVector128

pull/1852/head
Brian Popow 5 years ago
parent
cb08407728
commit
0215e99696
1 changed files with 91 additions and 97 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. 188
      src/ImageSharp/Formats/Webp/Lossy/Vp8Encoding.cs

188
src/ImageSharp/Formats/Webp/Lossy/Vp8Encoding.cs
View File

@ -407,59 +407,56 @@ namespace SixLabors.ImageSharp.Formats.Webp.Lossy
#if SUPPORTS_RUNTIME_INTRINSICS #if SUPPORTS_RUNTIME_INTRINSICS
if (Sse2.IsSupported) if (Sse2.IsSupported)
{ {
#pragma warning disable SA1503 // Braces should not be omitted ref byte srcRef = ref MemoryMarshal.GetReference(src);
fixed (byte* srcRef = src) ref byte referenceRef = ref MemoryMarshal.GetReference(reference);
fixed (byte* referenceRef = reference)
{ // Load src.
// Load src. var src0 = Vector128.Create(Unsafe.As<byte, long>(ref srcRef), 0);
Vector128<ulong> src0 = Sse2.LoadScalarVector128((ulong*)srcRef); var src1 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref srcRef, WebpConstants.Bps)), 0);
Vector128<ulong> src1 = Sse2.LoadScalarVector128((ulong*)(srcRef + WebpConstants.Bps)); var src2 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref srcRef, WebpConstants.Bps * 2)), 0);
Vector128<ulong> src2 = Sse2.LoadScalarVector128((ulong*)(srcRef + (WebpConstants.Bps * 2))); var src3 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref srcRef, WebpConstants.Bps * 3)), 0);
Vector128<ulong> src3 = Sse2.LoadScalarVector128((ulong*)(srcRef + (WebpConstants.Bps * 3)));
// Load ref.
// Load ref. var ref0 = Vector128.Create(Unsafe.As<byte, long>(ref referenceRef), 0);
Vector128<ulong> ref0 = Sse2.LoadScalarVector128((ulong*)referenceRef); var ref1 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps)), 0);
Vector128<ulong> ref1 = Sse2.LoadScalarVector128((ulong*)(referenceRef + WebpConstants.Bps)); var ref2 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps * 2)), 0);
Vector128<ulong> ref2 = Sse2.LoadScalarVector128((ulong*)(referenceRef + (WebpConstants.Bps * 2))); var ref3 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps * 3)), 0);
Vector128<ulong> ref3 = Sse2.LoadScalarVector128((ulong*)(referenceRef + (+WebpConstants.Bps * 3)));
// Convert both to 16 bit.
// Convert both to 16 bit. Vector128<byte> srcLow0 = Sse2.UnpackLow(src0.AsByte(), Vector128<byte>.Zero);
Vector128<byte> srcLow0 = Sse2.UnpackLow(src0.AsByte(), Vector128<byte>.Zero); Vector128<byte> srcLow1 = Sse2.UnpackLow(src1.AsByte(), Vector128<byte>.Zero);
Vector128<byte> srcLow1 = Sse2.UnpackLow(src1.AsByte(), Vector128<byte>.Zero); Vector128<byte> srcLow2 = Sse2.UnpackLow(src2.AsByte(), Vector128<byte>.Zero);
Vector128<byte> srcLow2 = Sse2.UnpackLow(src2.AsByte(), Vector128<byte>.Zero); Vector128<byte> srcLow3 = Sse2.UnpackLow(src3.AsByte(), Vector128<byte>.Zero);
Vector128<byte> srcLow3 = Sse2.UnpackLow(src3.AsByte(), Vector128<byte>.Zero); Vector128<byte> refLow0 = Sse2.UnpackLow(ref0.AsByte(), Vector128<byte>.Zero);
Vector128<byte> refLow0 = Sse2.UnpackLow(ref0.AsByte(), Vector128<byte>.Zero); Vector128<byte> refLow1 = Sse2.UnpackLow(ref1.AsByte(), Vector128<byte>.Zero);
Vector128<byte> refLow1 = Sse2.UnpackLow(ref1.AsByte(), Vector128<byte>.Zero); Vector128<byte> refLow2 = Sse2.UnpackLow(ref2.AsByte(), Vector128<byte>.Zero);
Vector128<byte> refLow2 = Sse2.UnpackLow(ref2.AsByte(), Vector128<byte>.Zero); Vector128<byte> refLow3 = Sse2.UnpackLow(ref3.AsByte(), Vector128<byte>.Zero);
Vector128<byte> refLow3 = Sse2.UnpackLow(ref3.AsByte(), Vector128<byte>.Zero);
// Compute difference. -> 00 01 02 03 00' 01' 02' 03'
// Compute difference. -> 00 01 02 03 00' 01' 02' 03' Vector128<short> diff0 = Sse2.Subtract(srcLow0.AsInt16(), refLow0.AsInt16());
Vector128<short> diff0 = Sse2.Subtract(srcLow0.AsInt16(), refLow0.AsInt16()); Vector128<short> diff1 = Sse2.Subtract(srcLow1.AsInt16(), refLow1.AsInt16());
Vector128<short> diff1 = Sse2.Subtract(srcLow1.AsInt16(), refLow1.AsInt16()); Vector128<short> diff2 = Sse2.Subtract(srcLow2.AsInt16(), refLow2.AsInt16());
Vector128<short> diff2 = Sse2.Subtract(srcLow2.AsInt16(), refLow2.AsInt16()); Vector128<short> diff3 = Sse2.Subtract(srcLow3.AsInt16(), refLow3.AsInt16());
Vector128<short> diff3 = Sse2.Subtract(srcLow3.AsInt16(), refLow3.AsInt16());
// Unpack and shuffle.
// Unpack and shuffle. // 00 01 02 03 0 0 0 0
// 00 01 02 03 0 0 0 0 // 10 11 12 13 0 0 0 0
// 10 11 12 13 0 0 0 0 // 20 21 22 23 0 0 0 0
// 20 21 22 23 0 0 0 0 // 30 31 32 33 0 0 0 0
// 30 31 32 33 0 0 0 0 Vector128<int> shuf01l = Sse2.UnpackLow(diff0.AsInt32(), diff1.AsInt32());
Vector128<int> shuf01l = Sse2.UnpackLow(diff0.AsInt32(), diff1.AsInt32()); Vector128<int> shuf23l = Sse2.UnpackLow(diff2.AsInt32(), diff3.AsInt32());
Vector128<int> shuf23l = Sse2.UnpackLow(diff2.AsInt32(), diff3.AsInt32()); Vector128<int> shuf01h = Sse2.UnpackHigh(diff0.AsInt32(), diff1.AsInt32());
Vector128<int> shuf01h = Sse2.UnpackHigh(diff0.AsInt32(), diff1.AsInt32()); Vector128<int> shuf23h = Sse2.UnpackHigh(diff2.AsInt32(), diff3.AsInt32());
Vector128<int> shuf23h = Sse2.UnpackHigh(diff2.AsInt32(), diff3.AsInt32());
// First pass.
// First pass. FTransformPass1SSE2(shuf01l.AsInt16(), shuf23l.AsInt16(), out Vector128<int> v01l, out Vector128<int> v32l);
FTransformPass1SSE2(shuf01l.AsInt16(), shuf23l.AsInt16(), out Vector128<int> v01l, out Vector128<int> v32l); FTransformPass1SSE2(shuf01h.AsInt16(), shuf23h.AsInt16(), out Vector128<int> v01h, out Vector128<int> v32h);
FTransformPass1SSE2(shuf01h.AsInt16(), shuf23h.AsInt16(), out Vector128<int> v01h, out Vector128<int> v32h);
// Second pass.
// Second pass. FTransformPass2SSE2(v01l, v32l, output);
FTransformPass2SSE2(v01l, v32l, output); FTransformPass2SSE2(v01h, v32h, output2);
FTransformPass2SSE2(v01h, v32h, output2);
}
} }
else else
#pragma warning restore SA1503 // Braces should not be omitted
#endif #endif
{ {
FTransform(src, reference, output, scratch); FTransform(src, reference, output, scratch);
@ -472,52 +469,49 @@ namespace SixLabors.ImageSharp.Formats.Webp.Lossy
#if SUPPORTS_RUNTIME_INTRINSICS #if SUPPORTS_RUNTIME_INTRINSICS
if (Sse2.IsSupported) if (Sse2.IsSupported)
{ {
#pragma warning disable SA1503 // Braces should not be omitted ref byte srcRef = ref MemoryMarshal.GetReference(src);
fixed (byte* srcRef = src) ref byte referenceRef = ref MemoryMarshal.GetReference(reference);
fixed (byte* referenceRef = reference)
{ // Load src.
// Load src. var src0 = Vector128.Create(Unsafe.As<byte, long>(ref srcRef), 0);
Vector128<ulong> src0 = Sse2.LoadScalarVector128((ulong*)srcRef); var src1 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref srcRef, WebpConstants.Bps)), 0);
Vector128<ulong> src1 = Sse2.LoadScalarVector128((ulong*)(srcRef + WebpConstants.Bps)); var src2 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref srcRef, WebpConstants.Bps * 2)), 0);
Vector128<ulong> src2 = Sse2.LoadScalarVector128((ulong*)(srcRef + (WebpConstants.Bps * 2))); var src3 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref srcRef, WebpConstants.Bps * 3)), 0);
Vector128<ulong> src3 = Sse2.LoadScalarVector128((ulong*)(srcRef + (WebpConstants.Bps * 3)));
// Load ref.
// Load ref. var ref0 = Vector128.Create(Unsafe.As<byte, long>(ref referenceRef), 0);
Vector128<ulong> ref0 = Sse2.LoadScalarVector128((ulong*)referenceRef); var ref1 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps)), 0);
Vector128<ulong> ref1 = Sse2.LoadScalarVector128((ulong*)(referenceRef + WebpConstants.Bps)); var ref2 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps * 2)), 0);
Vector128<ulong> ref2 = Sse2.LoadScalarVector128((ulong*)(referenceRef + (WebpConstants.Bps * 2))); var ref3 = Vector128.Create(Unsafe.As<byte, long>(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps * 3)), 0);
Vector128<ulong> ref3 = Sse2.LoadScalarVector128((ulong*)(referenceRef + (+WebpConstants.Bps * 3)));
// 00 01 02 03 *
// 00 01 02 03 * // 10 11 12 13 *
// 10 11 12 13 * // 20 21 22 23 *
// 20 21 22 23 * // 30 31 32 33 *
// 30 31 32 33 * // Shuffle.
// Shuffle. Vector128<short> srcLow0 = Sse2.UnpackLow(src0.AsInt16(), src1.AsInt16());
Vector128<short> srcLow0 = Sse2.UnpackLow(src0.AsInt16(), src1.AsInt16()); Vector128<short> srcLow1 = Sse2.UnpackLow(src2.AsInt16(), src3.AsInt16());
Vector128<short> srcLow1 = Sse2.UnpackLow(src2.AsInt16(), src3.AsInt16()); Vector128<short> refLow0 = Sse2.UnpackLow(ref0.AsInt16(), ref1.AsInt16());
Vector128<short> refLow0 = Sse2.UnpackLow(ref0.AsInt16(), ref1.AsInt16()); Vector128<short> refLow1 = Sse2.UnpackLow(ref2.AsInt16(), ref3.AsInt16());
Vector128<short> refLow1 = Sse2.UnpackLow(ref2.AsInt16(), ref3.AsInt16());
// 00 01 10 11 02 03 12 13 * * ...
// 00 01 10 11 02 03 12 13 * * ... // 20 21 30 31 22 22 32 33 * * ...
// 20 21 30 31 22 22 32 33 * * ...
// Convert both to 16 bit.
// Convert both to 16 bit. Vector128<byte> src0_16b = Sse2.UnpackLow(srcLow0.AsByte(), Vector128<byte>.Zero);
Vector128<byte> src0_16b = Sse2.UnpackLow(srcLow0.AsByte(), Vector128<byte>.Zero); Vector128<byte> src1_16b = Sse2.UnpackLow(srcLow1.AsByte(), Vector128<byte>.Zero);
Vector128<byte> src1_16b = Sse2.UnpackLow(srcLow1.AsByte(), Vector128<byte>.Zero); Vector128<byte> ref0_16b = Sse2.UnpackLow(refLow0.AsByte(), Vector128<byte>.Zero);
Vector128<byte> ref0_16b = Sse2.UnpackLow(refLow0.AsByte(), Vector128<byte>.Zero); Vector128<byte> ref1_16b = Sse2.UnpackLow(refLow1.AsByte(), Vector128<byte>.Zero);
Vector128<byte> ref1_16b = Sse2.UnpackLow(refLow1.AsByte(), Vector128<byte>.Zero);
// Compute the difference.
// Compute the difference. Vector128<short> row01 = Sse2.Subtract(src0_16b.AsInt16(), ref0_16b.AsInt16());
Vector128<short> row01 = Sse2.Subtract(src0_16b.AsInt16(), ref0_16b.AsInt16()); Vector128<short> row23 = Sse2.Subtract(src1_16b.AsInt16(), ref1_16b.AsInt16());
Vector128<short> row23 = Sse2.Subtract(src1_16b.AsInt16(), ref1_16b.AsInt16());
// First pass.
// First pass FTransformPass1SSE2(row01, row23, out Vector128<int> v01, out Vector128<int> v32);
FTransformPass1SSE2(row01, row23, out Vector128<int> v01, out Vector128<int> v32);
// Second pass.
// Second pass FTransformPass2SSE2(v01, v32, output);
FTransformPass2SSE2(v01, v32, output);
}
#pragma warning restore SA1503 // Braces should not be omitted
} }
else else
#endif #endif

Write Preview
Loading…
Cancel
Save