diff --git a/src/ImageSharp/Formats/Webp/Lossy/Vp8Encoding.cs b/src/ImageSharp/Formats/Webp/Lossy/Vp8Encoding.cs index d2b9704ab0..9fe526dbf5 100644 --- a/src/ImageSharp/Formats/Webp/Lossy/Vp8Encoding.cs +++ b/src/ImageSharp/Formats/Webp/Lossy/Vp8Encoding.cs @@ -407,59 +407,56 @@ namespace SixLabors.ImageSharp.Formats.Webp.Lossy #if SUPPORTS_RUNTIME_INTRINSICS if (Sse2.IsSupported) { -#pragma warning disable SA1503 // Braces should not be omitted - fixed (byte* srcRef = src) - fixed (byte* referenceRef = reference) - { - // Load src. - Vector128 src0 = Sse2.LoadScalarVector128((ulong*)srcRef); - Vector128 src1 = Sse2.LoadScalarVector128((ulong*)(srcRef + WebpConstants.Bps)); - Vector128 src2 = Sse2.LoadScalarVector128((ulong*)(srcRef + (WebpConstants.Bps * 2))); - Vector128 src3 = Sse2.LoadScalarVector128((ulong*)(srcRef + (WebpConstants.Bps * 3))); - - // Load ref. - Vector128 ref0 = Sse2.LoadScalarVector128((ulong*)referenceRef); - Vector128 ref1 = Sse2.LoadScalarVector128((ulong*)(referenceRef + WebpConstants.Bps)); - Vector128 ref2 = Sse2.LoadScalarVector128((ulong*)(referenceRef + (WebpConstants.Bps * 2))); - Vector128 ref3 = Sse2.LoadScalarVector128((ulong*)(referenceRef + (+WebpConstants.Bps * 3))); - - // Convert both to 16 bit. - Vector128 srcLow0 = Sse2.UnpackLow(src0.AsByte(), Vector128.Zero); - Vector128 srcLow1 = Sse2.UnpackLow(src1.AsByte(), Vector128.Zero); - Vector128 srcLow2 = Sse2.UnpackLow(src2.AsByte(), Vector128.Zero); - Vector128 srcLow3 = Sse2.UnpackLow(src3.AsByte(), Vector128.Zero); - Vector128 refLow0 = Sse2.UnpackLow(ref0.AsByte(), Vector128.Zero); - Vector128 refLow1 = Sse2.UnpackLow(ref1.AsByte(), Vector128.Zero); - Vector128 refLow2 = Sse2.UnpackLow(ref2.AsByte(), Vector128.Zero); - Vector128 refLow3 = Sse2.UnpackLow(ref3.AsByte(), Vector128.Zero); - - // Compute difference. -> 00 01 02 03 00' 01' 02' 03' - Vector128 diff0 = Sse2.Subtract(srcLow0.AsInt16(), refLow0.AsInt16()); - Vector128 diff1 = Sse2.Subtract(srcLow1.AsInt16(), refLow1.AsInt16()); - Vector128 diff2 = Sse2.Subtract(srcLow2.AsInt16(), refLow2.AsInt16()); - Vector128 diff3 = Sse2.Subtract(srcLow3.AsInt16(), refLow3.AsInt16()); - - // Unpack and shuffle. - // 00 01 02 03 0 0 0 0 - // 10 11 12 13 0 0 0 0 - // 20 21 22 23 0 0 0 0 - // 30 31 32 33 0 0 0 0 - Vector128 shuf01l = Sse2.UnpackLow(diff0.AsInt32(), diff1.AsInt32()); - Vector128 shuf23l = Sse2.UnpackLow(diff2.AsInt32(), diff3.AsInt32()); - Vector128 shuf01h = Sse2.UnpackHigh(diff0.AsInt32(), diff1.AsInt32()); - Vector128 shuf23h = Sse2.UnpackHigh(diff2.AsInt32(), diff3.AsInt32()); - - // First pass. - FTransformPass1SSE2(shuf01l.AsInt16(), shuf23l.AsInt16(), out Vector128 v01l, out Vector128 v32l); - FTransformPass1SSE2(shuf01h.AsInt16(), shuf23h.AsInt16(), out Vector128 v01h, out Vector128 v32h); - - // Second pass. - FTransformPass2SSE2(v01l, v32l, output); - FTransformPass2SSE2(v01h, v32h, output2); - } + ref byte srcRef = ref MemoryMarshal.GetReference(src); + ref byte referenceRef = ref MemoryMarshal.GetReference(reference); + + // Load src. + var src0 = Vector128.Create(Unsafe.As(ref srcRef), 0); + var src1 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref srcRef, WebpConstants.Bps)), 0); + var src2 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref srcRef, WebpConstants.Bps * 2)), 0); + var src3 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref srcRef, WebpConstants.Bps * 3)), 0); + + // Load ref. + var ref0 = Vector128.Create(Unsafe.As(ref referenceRef), 0); + var ref1 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps)), 0); + var ref2 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps * 2)), 0); + var ref3 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps * 3)), 0); + + // Convert both to 16 bit. + Vector128 srcLow0 = Sse2.UnpackLow(src0.AsByte(), Vector128.Zero); + Vector128 srcLow1 = Sse2.UnpackLow(src1.AsByte(), Vector128.Zero); + Vector128 srcLow2 = Sse2.UnpackLow(src2.AsByte(), Vector128.Zero); + Vector128 srcLow3 = Sse2.UnpackLow(src3.AsByte(), Vector128.Zero); + Vector128 refLow0 = Sse2.UnpackLow(ref0.AsByte(), Vector128.Zero); + Vector128 refLow1 = Sse2.UnpackLow(ref1.AsByte(), Vector128.Zero); + Vector128 refLow2 = Sse2.UnpackLow(ref2.AsByte(), Vector128.Zero); + Vector128 refLow3 = Sse2.UnpackLow(ref3.AsByte(), Vector128.Zero); + + // Compute difference. -> 00 01 02 03 00' 01' 02' 03' + Vector128 diff0 = Sse2.Subtract(srcLow0.AsInt16(), refLow0.AsInt16()); + Vector128 diff1 = Sse2.Subtract(srcLow1.AsInt16(), refLow1.AsInt16()); + Vector128 diff2 = Sse2.Subtract(srcLow2.AsInt16(), refLow2.AsInt16()); + Vector128 diff3 = Sse2.Subtract(srcLow3.AsInt16(), refLow3.AsInt16()); + + // Unpack and shuffle. + // 00 01 02 03 0 0 0 0 + // 10 11 12 13 0 0 0 0 + // 20 21 22 23 0 0 0 0 + // 30 31 32 33 0 0 0 0 + Vector128 shuf01l = Sse2.UnpackLow(diff0.AsInt32(), diff1.AsInt32()); + Vector128 shuf23l = Sse2.UnpackLow(diff2.AsInt32(), diff3.AsInt32()); + Vector128 shuf01h = Sse2.UnpackHigh(diff0.AsInt32(), diff1.AsInt32()); + Vector128 shuf23h = Sse2.UnpackHigh(diff2.AsInt32(), diff3.AsInt32()); + + // First pass. + FTransformPass1SSE2(shuf01l.AsInt16(), shuf23l.AsInt16(), out Vector128 v01l, out Vector128 v32l); + FTransformPass1SSE2(shuf01h.AsInt16(), shuf23h.AsInt16(), out Vector128 v01h, out Vector128 v32h); + + // Second pass. + FTransformPass2SSE2(v01l, v32l, output); + FTransformPass2SSE2(v01h, v32h, output2); } else -#pragma warning restore SA1503 // Braces should not be omitted #endif { FTransform(src, reference, output, scratch); @@ -472,52 +469,49 @@ namespace SixLabors.ImageSharp.Formats.Webp.Lossy #if SUPPORTS_RUNTIME_INTRINSICS if (Sse2.IsSupported) { -#pragma warning disable SA1503 // Braces should not be omitted - fixed (byte* srcRef = src) - fixed (byte* referenceRef = reference) - { - // Load src. - Vector128 src0 = Sse2.LoadScalarVector128((ulong*)srcRef); - Vector128 src1 = Sse2.LoadScalarVector128((ulong*)(srcRef + WebpConstants.Bps)); - Vector128 src2 = Sse2.LoadScalarVector128((ulong*)(srcRef + (WebpConstants.Bps * 2))); - Vector128 src3 = Sse2.LoadScalarVector128((ulong*)(srcRef + (WebpConstants.Bps * 3))); - - // Load ref. - Vector128 ref0 = Sse2.LoadScalarVector128((ulong*)referenceRef); - Vector128 ref1 = Sse2.LoadScalarVector128((ulong*)(referenceRef + WebpConstants.Bps)); - Vector128 ref2 = Sse2.LoadScalarVector128((ulong*)(referenceRef + (WebpConstants.Bps * 2))); - Vector128 ref3 = Sse2.LoadScalarVector128((ulong*)(referenceRef + (+WebpConstants.Bps * 3))); - - // 00 01 02 03 * - // 10 11 12 13 * - // 20 21 22 23 * - // 30 31 32 33 * - // Shuffle. - Vector128 srcLow0 = Sse2.UnpackLow(src0.AsInt16(), src1.AsInt16()); - Vector128 srcLow1 = Sse2.UnpackLow(src2.AsInt16(), src3.AsInt16()); - Vector128 refLow0 = Sse2.UnpackLow(ref0.AsInt16(), ref1.AsInt16()); - Vector128 refLow1 = Sse2.UnpackLow(ref2.AsInt16(), ref3.AsInt16()); - - // 00 01 10 11 02 03 12 13 * * ... - // 20 21 30 31 22 22 32 33 * * ... - - // Convert both to 16 bit. - Vector128 src0_16b = Sse2.UnpackLow(srcLow0.AsByte(), Vector128.Zero); - Vector128 src1_16b = Sse2.UnpackLow(srcLow1.AsByte(), Vector128.Zero); - Vector128 ref0_16b = Sse2.UnpackLow(refLow0.AsByte(), Vector128.Zero); - Vector128 ref1_16b = Sse2.UnpackLow(refLow1.AsByte(), Vector128.Zero); - - // Compute the difference. - Vector128 row01 = Sse2.Subtract(src0_16b.AsInt16(), ref0_16b.AsInt16()); - Vector128 row23 = Sse2.Subtract(src1_16b.AsInt16(), ref1_16b.AsInt16()); - - // First pass - FTransformPass1SSE2(row01, row23, out Vector128 v01, out Vector128 v32); - - // Second pass - FTransformPass2SSE2(v01, v32, output); - } -#pragma warning restore SA1503 // Braces should not be omitted + ref byte srcRef = ref MemoryMarshal.GetReference(src); + ref byte referenceRef = ref MemoryMarshal.GetReference(reference); + + // Load src. + var src0 = Vector128.Create(Unsafe.As(ref srcRef), 0); + var src1 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref srcRef, WebpConstants.Bps)), 0); + var src2 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref srcRef, WebpConstants.Bps * 2)), 0); + var src3 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref srcRef, WebpConstants.Bps * 3)), 0); + + // Load ref. + var ref0 = Vector128.Create(Unsafe.As(ref referenceRef), 0); + var ref1 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps)), 0); + var ref2 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps * 2)), 0); + var ref3 = Vector128.Create(Unsafe.As(ref Unsafe.Add(ref referenceRef, WebpConstants.Bps * 3)), 0); + + // 00 01 02 03 * + // 10 11 12 13 * + // 20 21 22 23 * + // 30 31 32 33 * + // Shuffle. + Vector128 srcLow0 = Sse2.UnpackLow(src0.AsInt16(), src1.AsInt16()); + Vector128 srcLow1 = Sse2.UnpackLow(src2.AsInt16(), src3.AsInt16()); + Vector128 refLow0 = Sse2.UnpackLow(ref0.AsInt16(), ref1.AsInt16()); + Vector128 refLow1 = Sse2.UnpackLow(ref2.AsInt16(), ref3.AsInt16()); + + // 00 01 10 11 02 03 12 13 * * ... + // 20 21 30 31 22 22 32 33 * * ... + + // Convert both to 16 bit. + Vector128 src0_16b = Sse2.UnpackLow(srcLow0.AsByte(), Vector128.Zero); + Vector128 src1_16b = Sse2.UnpackLow(srcLow1.AsByte(), Vector128.Zero); + Vector128 ref0_16b = Sse2.UnpackLow(refLow0.AsByte(), Vector128.Zero); + Vector128 ref1_16b = Sse2.UnpackLow(refLow1.AsByte(), Vector128.Zero); + + // Compute the difference. + Vector128 row01 = Sse2.Subtract(src0_16b.AsInt16(), ref0_16b.AsInt16()); + Vector128 row23 = Sse2.Subtract(src1_16b.AsInt16(), ref1_16b.AsInt16()); + + // First pass. + FTransformPass1SSE2(row01, row23, out Vector128 v01, out Vector128 v32); + + // Second pass. + FTransformPass2SSE2(v01, v32, output); } else #endif