diff options
| author | TSR Berry <20988865+TSRBerry@users.noreply.github.com> | 2023-04-08 01:22:00 +0200 |
|---|---|---|
| committer | Mary <thog@protonmail.com> | 2023-04-27 23:51:14 +0200 |
| commit | cee712105850ac3385cd0091a923438167433f9f (patch) | |
| tree | 4a5274b21d8b7f938c0d0ce18736d3f2993b11b1 /src/Ryujinx.Graphics.Nvdec/Image/SurfaceWriter.cs | |
| parent | cd124bda587ef09668a971fa1cac1c3f0cfc9f21 (diff) | |
Move solution and projects to src
Diffstat (limited to 'src/Ryujinx.Graphics.Nvdec/Image/SurfaceWriter.cs')
| -rw-r--r-- | src/Ryujinx.Graphics.Nvdec/Image/SurfaceWriter.cs | 175 |
1 files changed, 175 insertions, 0 deletions
diff --git a/src/Ryujinx.Graphics.Nvdec/Image/SurfaceWriter.cs b/src/Ryujinx.Graphics.Nvdec/Image/SurfaceWriter.cs new file mode 100644 index 00000000..cc5c251b --- /dev/null +++ b/src/Ryujinx.Graphics.Nvdec/Image/SurfaceWriter.cs @@ -0,0 +1,175 @@ +using Ryujinx.Common; +using Ryujinx.Graphics.Gpu.Memory; +using Ryujinx.Graphics.Texture; +using Ryujinx.Graphics.Video; +using System; +using System.Runtime.Intrinsics; +using System.Runtime.Intrinsics.X86; +using static Ryujinx.Graphics.Nvdec.Image.SurfaceCommon; +using static Ryujinx.Graphics.Nvdec.MemoryExtensions; + +namespace Ryujinx.Graphics.Nvdec.Image +{ + static class SurfaceWriter + { + public static void Write(MemoryManager gmm, ISurface surface, uint lumaOffset, uint chromaOffset) + { + int lumaSize = GetBlockLinearSize(surface.Width, surface.Height, 1); + + using var luma = gmm.GetWritableRegion(ExtendOffset(lumaOffset), lumaSize); + + WriteLuma( + luma.Memory.Span, + surface.YPlane.AsSpan(), + surface.Stride, + surface.Width, + surface.Height); + + int chromaSize = GetBlockLinearSize(surface.UvWidth, surface.UvHeight, 2); + + using var chroma = gmm.GetWritableRegion(ExtendOffset(chromaOffset), chromaSize); + + WriteChroma( + chroma.Memory.Span, + surface.UPlane.AsSpan(), + surface.VPlane.AsSpan(), + surface.UvStride, + surface.UvWidth, + surface.UvHeight); + } + + public static void WriteInterlaced( + MemoryManager gmm, + ISurface surface, + uint lumaTopOffset, + uint chromaTopOffset, + uint lumaBottomOffset, + uint chromaBottomOffset) + { + int lumaSize = GetBlockLinearSize(surface.Width, surface.Height / 2, 1); + + using var lumaTop = gmm.GetWritableRegion(ExtendOffset(lumaTopOffset), lumaSize); + using var lumaBottom = gmm.GetWritableRegion(ExtendOffset(lumaBottomOffset), lumaSize); + + WriteLuma( + lumaTop.Memory.Span, + surface.YPlane.AsSpan(), + surface.Stride * 2, + surface.Width, + surface.Height / 2); + + WriteLuma( + lumaBottom.Memory.Span, + surface.YPlane.AsSpan().Slice(surface.Stride), + surface.Stride * 2, + surface.Width, + surface.Height / 2); + + int chromaSize = GetBlockLinearSize(surface.UvWidth, surface.UvHeight / 2, 2); + + using var chromaTop = gmm.GetWritableRegion(ExtendOffset(chromaTopOffset), chromaSize); + using var chromaBottom = gmm.GetWritableRegion(ExtendOffset(chromaBottomOffset), chromaSize); + + WriteChroma( + chromaTop.Memory.Span, + surface.UPlane.AsSpan(), + surface.VPlane.AsSpan(), + surface.UvStride * 2, + surface.UvWidth, + surface.UvHeight / 2); + + WriteChroma( + chromaBottom.Memory.Span, + surface.UPlane.AsSpan().Slice(surface.UvStride), + surface.VPlane.AsSpan().Slice(surface.UvStride), + surface.UvStride * 2, + surface.UvWidth, + surface.UvHeight / 2); + } + + private static void WriteLuma(Span<byte> dst, ReadOnlySpan<byte> src, int srcStride, int width, int height) + { + LayoutConverter.ConvertLinearToBlockLinear(dst, width, height, srcStride, 1, 2, src); + } + + private unsafe static void WriteChroma( + Span<byte> dst, + ReadOnlySpan<byte> srcU, + ReadOnlySpan<byte> srcV, + int srcStride, + int width, + int height) + { + OffsetCalculator calc = new OffsetCalculator(width, height, 0, false, 2, 2); + + if (Sse2.IsSupported) + { + int strideTrunc64 = BitUtils.AlignDown(width * 2, 64); + + int inStrideGap = srcStride - width; + + fixed (byte* outputPtr = dst, srcUPtr = srcU, srcVPtr = srcV) + { + byte* inUPtr = srcUPtr; + byte* inVPtr = srcVPtr; + + for (int y = 0; y < height; y++) + { + calc.SetY(y); + + for (int x = 0; x < strideTrunc64; x += 64, inUPtr += 32, inVPtr += 32) + { + byte* offset = outputPtr + calc.GetOffsetWithLineOffset64(x); + byte* offset2 = offset + 0x20; + byte* offset3 = offset + 0x100; + byte* offset4 = offset + 0x120; + + Vector128<byte> value = *(Vector128<byte>*)inUPtr; + Vector128<byte> value2 = *(Vector128<byte>*)inVPtr; + Vector128<byte> value3 = *(Vector128<byte>*)(inUPtr + 16); + Vector128<byte> value4 = *(Vector128<byte>*)(inVPtr + 16); + + Vector128<byte> uv0 = Sse2.UnpackLow(value, value2); + Vector128<byte> uv1 = Sse2.UnpackHigh(value, value2); + Vector128<byte> uv2 = Sse2.UnpackLow(value3, value4); + Vector128<byte> uv3 = Sse2.UnpackHigh(value3, value4); + + *(Vector128<byte>*)offset = uv0; + *(Vector128<byte>*)offset2 = uv1; + *(Vector128<byte>*)offset3 = uv2; + *(Vector128<byte>*)offset4 = uv3; + } + + for (int x = strideTrunc64 / 2; x < width; x++, inUPtr++, inVPtr++) + { + byte* offset = outputPtr + calc.GetOffset(x); + + *offset = *inUPtr; + *(offset + 1) = *inVPtr; + } + + inUPtr += inStrideGap; + inVPtr += inStrideGap; + } + } + } + else + { + for (int y = 0; y < height; y++) + { + int srcBaseOffset = y * srcStride; + + calc.SetY(y); + + for (int x = 0; x < width; x++) + { + int dstOffset = calc.GetOffset(x); + + dst[dstOffset + 0] = srcU[srcBaseOffset + x]; + dst[dstOffset + 1] = srcV[srcBaseOffset + x]; + } + } + } + } + } +} |