daniel/Ryujinx
1
0
Fork 0
mirror of https://github.com/Ryujinx/Ryujinx.git synced 2025-06-28 22:40:47 -07:00
Code Issues Packages Projects Releases Wiki Activity

Move solution and projects to src

Browse Source
This commit is contained in:
TSR Berry
2023-04-08 01:22:00 +02:00
committed by Mary
parent cd124bda58
commit cee7121058
3466 changed files with 55 additions and 55 deletions
Download Patch File Download Diff File Expand all files Collapse all files

103
src/Ryujinx.Graphics.Vic/Image/BufferPool.cs Normal file
View File

@ -0,0 +1,103 @@
using System;
namespace Ryujinx.Graphics.Vic.Image
{
class BufferPool<T>
{
/// <summary>
/// Maximum number of buffers on the pool.
/// </summary>
private const int MaxBuffers = 4;
/// <summary>
/// Maximum size of a buffer that can be added on the pool.
/// If the required buffer is larger than this, it won't be
/// added to the pool to avoid long term high memory usage.
/// </summary>
private const int MaxBufferSize = 2048 * 2048;
private struct PoolItem
{
public bool InUse;
public T[] Buffer;
}
private readonly PoolItem[] _pool = new PoolItem[MaxBuffers];
/// <summary>
/// Rents a buffer with the exact size requested.
/// </summary>
/// <param name="length">Size of the buffer</param>
/// <param name="buffer">Span of the requested size</param>
/// <returns>The index of the buffer on the pool</returns>
public int Rent(int length, out Span<T> buffer)
{
int index = RentMinimum(length, out T[] bufferArray);
buffer = new Span<T>(bufferArray).Slice(0, length);
return index;
}
/// <summary>
/// Rents a buffer with a size greater than or equal to the requested size.
/// </summary>
/// <param name="length">Size of the buffer</param>
/// <param name="buffer">Array with a length greater than or equal to the requested length</param>
/// <returns>The index of the buffer on the pool</returns>
public int RentMinimum(int length, out T[] buffer)
{
if ((uint)length > MaxBufferSize)
{
buffer = new T[length];
return -1;
}
// Try to find a buffer that is larger or the same size of the requested one.
// This will avoid an allocation.
for (int i = 0; i < MaxBuffers; i++)
{
ref PoolItem item = ref _pool[i];
if (!item.InUse && item.Buffer != null && item.Buffer.Length >= length)
{
buffer = item.Buffer;
item.InUse = true;
return i;
}
}
buffer = new T[length];
// Try to add the new buffer to the pool.
// We try to find a slot that is not in use, and replace the buffer in it.
for (int i = 0; i < MaxBuffers; i++)
{
ref PoolItem item = ref _pool[i];
if (!item.InUse)
{
item.Buffer = buffer;
item.InUse = true;
return i;
}
}
return -1;
}
/// <summary>
/// Returns a buffer returned from <see cref="Rent(int)"/> to the pool.
/// </summary>
/// <param name="index">Index of the buffer on the pool</param>
public void Return(int index)
{
if (index < 0)
{
return;
}
_pool[index].InUse = false;
}
}
}

86
src/Ryujinx.Graphics.Vic/Image/InputSurface.cs Normal file
View File

@ -0,0 +1,86 @@
using System;
namespace Ryujinx.Graphics.Vic.Image
{
ref struct RentedBuffer
{
public static RentedBuffer Empty => new RentedBuffer(Span<byte>.Empty, -1);
public Span<byte> Data;
public int Index;
public RentedBuffer(Span<byte> data, int index)
{
Data = data;
Index = index;
}
public void Return(BufferPool<byte> pool)
{
if (Index != -1)
{
pool.Return(Index);
}
}
}
ref struct InputSurface
{
public ReadOnlySpan<byte> Buffer0;
public ReadOnlySpan<byte> Buffer1;
public ReadOnlySpan<byte> Buffer2;
public int Buffer0Index;
public int Buffer1Index;
public int Buffer2Index;
public int Width;
public int Height;
public int UvWidth;
public int UvHeight;
public void Initialize()
{
Buffer0Index = -1;
Buffer1Index = -1;
Buffer2Index = -1;
}
public void SetBuffer0(RentedBuffer buffer)
{
Buffer0 = buffer.Data;
Buffer0Index = buffer.Index;
}
public void SetBuffer1(RentedBuffer buffer)
{
Buffer1 = buffer.Data;
Buffer1Index = buffer.Index;
}
public void SetBuffer2(RentedBuffer buffer)
{
Buffer2 = buffer.Data;
Buffer2Index = buffer.Index;
}
public void Return(BufferPool<byte> pool)
{
if (Buffer0Index != -1)
{
pool.Return(Buffer0Index);
}
if (Buffer1Index != -1)
{
pool.Return(Buffer1Index);
}
if (Buffer2Index != -1)
{
pool.Return(Buffer2Index);
}
}
}
}

10
src/Ryujinx.Graphics.Vic/Image/Pixel.cs Normal file
View File

@ -0,0 +1,10 @@
namespace Ryujinx.Graphics.Vic.Image
{
struct Pixel
{
public ushort R;
public ushort G;
public ushort B;
public ushort A;
}
}

46
src/Ryujinx.Graphics.Vic/Image/Surface.cs Normal file
View File

@ -0,0 +1,46 @@
using System;
using System.Runtime.CompilerServices;
namespace Ryujinx.Graphics.Vic.Image
{
readonly struct Surface : IDisposable
{
private readonly int _bufferIndex;
private readonly BufferPool<Pixel> _pool;
public Pixel[] Data { get; }
public int Width { get; }
public int Height { get; }
public Surface(BufferPool<Pixel> pool, int width, int height)
{
_bufferIndex = pool.RentMinimum(width * height, out Pixel[] data);
_pool = pool;
Data = data;
Width = width;
Height = height;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public ushort GetR(int x, int y) => Data[y * Width + x].R;
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public ushort GetG(int x, int y) => Data[y * Width + x].G;
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public ushort GetB(int x, int y) => Data[y * Width + x].B;
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public ushort GetA(int x, int y) => Data[y * Width + x].A;
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public void SetR(int x, int y, ushort value) => Data[y * Width + x].R = value;
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public void SetG(int x, int y, ushort value) => Data[y * Width + x].G = value;
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public void SetB(int x, int y, ushort value) => Data[y * Width + x].B = value;
[MethodImpl(MethodImplOptions.AggressiveInlining)]
public void SetA(int x, int y, ushort value) => Data[y * Width + x].A = value;
public void Dispose() => _pool.Return(_bufferIndex);
}
}

33
src/Ryujinx.Graphics.Vic/Image/SurfaceCommon.cs Normal file
View File

@ -0,0 +1,33 @@
using Ryujinx.Common;
using Ryujinx.Graphics.Texture;
namespace Ryujinx.Graphics.Vic.Image
{
static class SurfaceCommon
{
public static int GetPitch(int width, int bytesPerPixel)
{
return BitUtils.AlignUp(width * bytesPerPixel, 256);
}
public static int GetBlockLinearSize(int width, int height, int bytesPerPixel, int gobBlocksInY)
{
return SizeCalculator.GetBlockLinearTextureSize(width, height, 1, 1, 1, 1, 1, bytesPerPixel, gobBlocksInY, 1, 1).TotalSize;
}
public static ulong ExtendOffset(uint offset)
{
return (ulong)offset << 8;
}
public static ushort Upsample(byte value)
{
return (ushort)(value << 2);
}
public static byte Downsample(ushort value)
{
return (byte)(value >> 2);
}
}
}

495
src/Ryujinx.Graphics.Vic/Image/SurfaceReader.cs Normal file
View File

@ -0,0 +1,495 @@
using Ryujinx.Common.Logging;
using Ryujinx.Common.Memory;
using Ryujinx.Graphics.Texture;
using Ryujinx.Graphics.Vic.Types;
using System;
using System.Runtime.CompilerServices;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.Arm;
using System.Runtime.Intrinsics.X86;
using static Ryujinx.Graphics.Vic.Image.SurfaceCommon;
namespace Ryujinx.Graphics.Vic.Image
{
static class SurfaceReader
{
public static Surface Read(
ResourceManager rm,
ref SlotConfig config,
ref SlotSurfaceConfig surfaceConfig,
ref Array8<PlaneOffsets> offsets)
{
switch (surfaceConfig.SlotPixelFormat)
{
case PixelFormat.Y8___V8U8_N420: return ReadNv12(rm, ref config, ref surfaceConfig, ref offsets);
}
Logger.Error?.Print(LogClass.Vic, $"Unsupported pixel format \"{surfaceConfig.SlotPixelFormat}\".");
int lw = surfaceConfig.SlotLumaWidth + 1;
int lh = surfaceConfig.SlotLumaHeight + 1;
return new Surface(rm.SurfacePool, lw, lh);
}
private unsafe static Surface ReadNv12(
ResourceManager rm,
ref SlotConfig config,
ref SlotSurfaceConfig surfaceConfig,
ref Array8<PlaneOffsets> offsets)
{
InputSurface input = ReadSurface(rm, ref config, ref surfaceConfig, ref offsets, 1, 2);
int width = input.Width;
int height = input.Height;
int yStride = GetPitch(width, 1);
int uvStride = GetPitch(input.UvWidth, 2);
Surface output = new Surface(rm.SurfacePool, width, height);
if (Sse41.IsSupported)
{
Vector128<byte> shufMask = Vector128.Create(
(byte)0, (byte)2, (byte)3, (byte)1,
(byte)4, (byte)6, (byte)7, (byte)5,
(byte)8, (byte)10, (byte)11, (byte)9,
(byte)12, (byte)14, (byte)15, (byte)13);
Vector128<short> alphaMask = Vector128.Create(0xff << 24).AsInt16();
int yStrideGap = yStride - width;
int uvStrideGap = uvStride - input.UvWidth;
int widthTrunc = width & ~0xf;
fixed (Pixel* dstPtr = output.Data)
{
Pixel* op = dstPtr;
fixed (byte* src0Ptr = input.Buffer0, src1Ptr = input.Buffer1)
{
byte* i0p = src0Ptr;
for (int y = 0; y < height; y++)
{
byte* i1p = src1Ptr + (y >> 1) * uvStride;
int x = 0;
for (; x < widthTrunc; x += 16, i0p += 16, i1p += 16)
{
Vector128<short> ya0 = Sse41.ConvertToVector128Int16(i0p);
Vector128<short> ya1 = Sse41.ConvertToVector128Int16(i0p + 8);
Vector128<byte> uv = Sse2.LoadVector128(i1p);
Vector128<short> uv0 = Sse2.UnpackLow(uv.AsInt16(), uv.AsInt16());
Vector128<short> uv1 = Sse2.UnpackHigh(uv.AsInt16(), uv.AsInt16());
Vector128<short> rgba0 = Sse2.UnpackLow(ya0, uv0);
Vector128<short> rgba1 = Sse2.UnpackHigh(ya0, uv0);
Vector128<short> rgba2 = Sse2.UnpackLow(ya1, uv1);
Vector128<short> rgba3 = Sse2.UnpackHigh(ya1, uv1);
rgba0 = Ssse3.Shuffle(rgba0.AsByte(), shufMask).AsInt16();
rgba1 = Ssse3.Shuffle(rgba1.AsByte(), shufMask).AsInt16();
rgba2 = Ssse3.Shuffle(rgba2.AsByte(), shufMask).AsInt16();
rgba3 = Ssse3.Shuffle(rgba3.AsByte(), shufMask).AsInt16();
rgba0 = Sse2.Or(rgba0, alphaMask);
rgba1 = Sse2.Or(rgba1, alphaMask);
rgba2 = Sse2.Or(rgba2, alphaMask);
rgba3 = Sse2.Or(rgba3, alphaMask);
Vector128<short> rgba16_0 = Sse41.ConvertToVector128Int16(rgba0.AsByte());
Vector128<short> rgba16_1 = Sse41.ConvertToVector128Int16(HighToLow(rgba0.AsByte()));
Vector128<short> rgba16_2 = Sse41.ConvertToVector128Int16(rgba1.AsByte());
Vector128<short> rgba16_3 = Sse41.ConvertToVector128Int16(HighToLow(rgba1.AsByte()));
Vector128<short> rgba16_4 = Sse41.ConvertToVector128Int16(rgba2.AsByte());
Vector128<short> rgba16_5 = Sse41.ConvertToVector128Int16(HighToLow(rgba2.AsByte()));
Vector128<short> rgba16_6 = Sse41.ConvertToVector128Int16(rgba3.AsByte());
Vector128<short> rgba16_7 = Sse41.ConvertToVector128Int16(HighToLow(rgba3.AsByte()));
rgba16_0 = Sse2.ShiftLeftLogical(rgba16_0, 2);
rgba16_1 = Sse2.ShiftLeftLogical(rgba16_1, 2);
rgba16_2 = Sse2.ShiftLeftLogical(rgba16_2, 2);
rgba16_3 = Sse2.ShiftLeftLogical(rgba16_3, 2);
rgba16_4 = Sse2.ShiftLeftLogical(rgba16_4, 2);
rgba16_5 = Sse2.ShiftLeftLogical(rgba16_5, 2);
rgba16_6 = Sse2.ShiftLeftLogical(rgba16_6, 2);
rgba16_7 = Sse2.ShiftLeftLogical(rgba16_7, 2);
Sse2.Store((short*)(op + (uint)x + 0), rgba16_0);
Sse2.Store((short*)(op + (uint)x + 2), rgba16_1);
Sse2.Store((short*)(op + (uint)x + 4), rgba16_2);
Sse2.Store((short*)(op + (uint)x + 6), rgba16_3);
Sse2.Store((short*)(op + (uint)x + 8), rgba16_4);
Sse2.Store((short*)(op + (uint)x + 10), rgba16_5);
Sse2.Store((short*)(op + (uint)x + 12), rgba16_6);
Sse2.Store((short*)(op + (uint)x + 14), rgba16_7);
}
for (; x < width; x++, i1p += (x & 1) * 2)
{
Pixel* px = op + (uint)x;
px->R = Upsample(*i0p++);
px->G = Upsample(*i1p);
px->B = Upsample(*(i1p + 1));
px->A = 0x3ff;
}
op += width;
i0p += yStrideGap;
i1p += uvStrideGap;
}
}
}
}
else if (AdvSimd.Arm64.IsSupported)
{
Vector128<int> alphaMask = Vector128.Create(0xffu << 24).AsInt32();
int yStrideGap = yStride - width;
int uvStrideGap = uvStride - input.UvWidth;
int widthTrunc = width & ~0xf;
fixed (Pixel* dstPtr = output.Data)
{
Pixel* op = dstPtr;
fixed (byte* src0Ptr = input.Buffer0, src1Ptr = input.Buffer1)
{
byte* i0p = src0Ptr;
for (int y = 0; y < height; y++)
{
byte* i1p = src1Ptr + (y >> 1) * uvStride;
int x = 0;
for (; x < widthTrunc; x += 16, i0p += 16, i1p += 16)
{
Vector128<byte> ya = AdvSimd.LoadVector128(i0p);
Vector128<byte> uv = AdvSimd.LoadVector128(i1p);
Vector128<short> ya0 = AdvSimd.ZeroExtendWideningLower(ya.GetLower()).AsInt16();
Vector128<short> ya1 = AdvSimd.ZeroExtendWideningUpper(ya).AsInt16();
Vector128<short> uv0 = AdvSimd.Arm64.ZipLow(uv.AsInt16(), uv.AsInt16());
Vector128<short> uv1 = AdvSimd.Arm64.ZipHigh(uv.AsInt16(), uv.AsInt16());
ya0 = AdvSimd.ShiftLeftLogical(ya0, 8);
ya1 = AdvSimd.ShiftLeftLogical(ya1, 8);
Vector128<short> rgba0 = AdvSimd.Arm64.ZipLow(ya0, uv0);
Vector128<short> rgba1 = AdvSimd.Arm64.ZipHigh(ya0, uv0);
Vector128<short> rgba2 = AdvSimd.Arm64.ZipLow(ya1, uv1);
Vector128<short> rgba3 = AdvSimd.Arm64.ZipHigh(ya1, uv1);
rgba0 = AdvSimd.ShiftRightLogicalAdd(alphaMask, rgba0.AsInt32(), 8).AsInt16();
rgba1 = AdvSimd.ShiftRightLogicalAdd(alphaMask, rgba1.AsInt32(), 8).AsInt16();
rgba2 = AdvSimd.ShiftRightLogicalAdd(alphaMask, rgba2.AsInt32(), 8).AsInt16();
rgba3 = AdvSimd.ShiftRightLogicalAdd(alphaMask, rgba3.AsInt32(), 8).AsInt16();
Vector128<short> rgba16_0 = AdvSimd.ZeroExtendWideningLower(rgba0.AsByte().GetLower()).AsInt16();
Vector128<short> rgba16_1 = AdvSimd.ZeroExtendWideningUpper(rgba0.AsByte()).AsInt16();
Vector128<short> rgba16_2 = AdvSimd.ZeroExtendWideningLower(rgba1.AsByte().GetLower()).AsInt16();
Vector128<short> rgba16_3 = AdvSimd.ZeroExtendWideningUpper(rgba1.AsByte()).AsInt16();
Vector128<short> rgba16_4 = AdvSimd.ZeroExtendWideningLower(rgba2.AsByte().GetLower()).AsInt16();
Vector128<short> rgba16_5 = AdvSimd.ZeroExtendWideningUpper(rgba2.AsByte()).AsInt16();
Vector128<short> rgba16_6 = AdvSimd.ZeroExtendWideningLower(rgba3.AsByte().GetLower()).AsInt16();
Vector128<short> rgba16_7 = AdvSimd.ZeroExtendWideningUpper(rgba3.AsByte()).AsInt16();
rgba16_0 = AdvSimd.ShiftLeftLogical(rgba16_0, 2);
rgba16_1 = AdvSimd.ShiftLeftLogical(rgba16_1, 2);
rgba16_2 = AdvSimd.ShiftLeftLogical(rgba16_2, 2);
rgba16_3 = AdvSimd.ShiftLeftLogical(rgba16_3, 2);
rgba16_4 = AdvSimd.ShiftLeftLogical(rgba16_4, 2);
rgba16_5 = AdvSimd.ShiftLeftLogical(rgba16_5, 2);
rgba16_6 = AdvSimd.ShiftLeftLogical(rgba16_6, 2);
rgba16_7 = AdvSimd.ShiftLeftLogical(rgba16_7, 2);
AdvSimd.Store((short*)(op + (uint)x + 0), rgba16_0);
AdvSimd.Store((short*)(op + (uint)x + 2), rgba16_1);
AdvSimd.Store((short*)(op + (uint)x + 4), rgba16_2);
AdvSimd.Store((short*)(op + (uint)x + 6), rgba16_3);
AdvSimd.Store((short*)(op + (uint)x + 8), rgba16_4);
AdvSimd.Store((short*)(op + (uint)x + 10), rgba16_5);
AdvSimd.Store((short*)(op + (uint)x + 12), rgba16_6);
AdvSimd.Store((short*)(op + (uint)x + 14), rgba16_7);
}
for (; x < width; x++, i1p += (x & 1) * 2)
{
Pixel* px = op + (uint)x;
px->R = Upsample(*i0p++);
px->G = Upsample(*i1p);
px->B = Upsample(*(i1p + 1));
px->A = 0x3ff;
}
op += width;
i0p += yStrideGap;
i1p += uvStrideGap;
}
}
}
}
else
{
for (int y = 0; y < height; y++)
{
int uvBase = (y >> 1) * uvStride;
for (int x = 0; x < width; x++)
{
output.SetR(x, y, Upsample(input.Buffer0[y * yStride + x]));
int uvOffs = uvBase + (x & ~1);
output.SetG(x, y, Upsample(input.Buffer1[uvOffs]));
output.SetB(x, y, Upsample(input.Buffer1[uvOffs + 1]));
output.SetA(x, y, 0x3ff);
}
}
}
input.Return(rm.BufferPool);
return output;
}
[MethodImpl(MethodImplOptions.AggressiveInlining)]
private static Vector128<byte> HighToLow(Vector128<byte> value)
{
return Sse.MoveHighToLow(value.AsSingle(), value.AsSingle()).AsByte();
}
private static InputSurface ReadSurface(
ResourceManager rm,
ref SlotConfig config,
ref SlotSurfaceConfig surfaceConfig,
ref Array8<PlaneOffsets> offsets,
int bytesPerPixel,
int planes)
{
InputSurface surface = new InputSurface();
surface.Initialize();
int gobBlocksInY = 1 << surfaceConfig.SlotBlkHeight;
bool linear = surfaceConfig.SlotBlkKind == 0;
int lw = surfaceConfig.SlotLumaWidth + 1;
int lh = surfaceConfig.SlotLumaHeight + 1;
int cw = surfaceConfig.SlotChromaWidth + 1;
int ch = surfaceConfig.SlotChromaHeight + 1;
// Interlaced inputs have double the height when deinterlaced.
int heightShift = config.FrameFormat.IsField() ? 1 : 0;
surface.Width = lw;
surface.Height = lh << heightShift;
surface.UvWidth = cw;
surface.UvHeight = ch << heightShift;
if (planes > 0)
{
surface.SetBuffer0(ReadBuffer(rm, ref config, ref offsets, linear, 0, lw, lh, bytesPerPixel, gobBlocksInY));
}
if (planes > 1)
{
surface.SetBuffer1(ReadBuffer(rm, ref config, ref offsets, linear, 1, cw, ch, planes == 2 ? 2 : 1, gobBlocksInY));
}
if (planes > 2)
{
surface.SetBuffer2(ReadBuffer(rm, ref config, ref offsets, linear, 2, cw, ch, 1, gobBlocksInY));
}
return surface;
}
private static RentedBuffer ReadBuffer(
ResourceManager rm,
scoped ref SlotConfig config,
scoped ref Array8<PlaneOffsets> offsets,
bool linear,
int plane,
int width,
int height,
int bytesPerPixel,
int gobBlocksInY)
{
FrameFormat frameFormat = config.FrameFormat;
bool isLuma = plane == 0;
bool isField = frameFormat.IsField();
bool isTopField = frameFormat.IsTopField(isLuma);
int stride = GetPitch(width, bytesPerPixel);
uint offset = GetOffset(ref offsets[0], plane);
int dstStart = 0;
int dstStride = stride;
if (isField)
{
dstStart = isTopField ? 0 : stride;
dstStride = stride * 2;
}
RentedBuffer buffer;
if (linear)
{
buffer = ReadBufferLinear(rm, offset, width, height, dstStart, dstStride, bytesPerPixel);
}
else
{
buffer = ReadBufferBlockLinear(rm, offset, width, height, dstStart, dstStride, bytesPerPixel, gobBlocksInY);
}
if (isField || frameFormat.IsInterlaced())
{
RentedBuffer prevBuffer = RentedBuffer.Empty;
RentedBuffer nextBuffer = RentedBuffer.Empty;
if (config.PrevFieldEnable)
{
prevBuffer = ReadBufferNoDeinterlace(rm, ref offsets[1], linear, plane, width, height, bytesPerPixel, gobBlocksInY);
}
if (config.NextFieldEnable)
{
nextBuffer = ReadBufferNoDeinterlace(rm, ref offsets[2], linear, plane, width, height, bytesPerPixel, gobBlocksInY);
}
int w = width * bytesPerPixel;
switch (config.DeinterlaceMode)
{
case DeinterlaceMode.Weave:
Scaler.DeinterlaceWeave(buffer.Data, prevBuffer.Data, w, stride, isTopField);
break;
case DeinterlaceMode.BobField:
Scaler.DeinterlaceBob(buffer.Data, w, stride, isTopField);
break;
case DeinterlaceMode.Bob:
bool isCurrentTop = isLuma ? config.IsEven : config.ChromaEven;
Scaler.DeinterlaceBob(buffer.Data, w, stride, isCurrentTop ^ frameFormat.IsInterlacedBottomFirst());
break;
case DeinterlaceMode.NewBob:
case DeinterlaceMode.Disi1:
Scaler.DeinterlaceMotionAdaptive(buffer.Data, prevBuffer.Data, nextBuffer.Data, w, stride, isTopField);
break;
case DeinterlaceMode.WeaveLumaBobFieldChroma:
if (isLuma)
{
Scaler.DeinterlaceWeave(buffer.Data, prevBuffer.Data, w, stride, isTopField);
}
else
{
Scaler.DeinterlaceBob(buffer.Data, w, stride, isTopField);
}
break;
default:
Logger.Error?.Print(LogClass.Vic, $"Unsupported deinterlace mode \"{config.DeinterlaceMode}\".");
break;
}
prevBuffer.Return(rm.BufferPool);
nextBuffer.Return(rm.BufferPool);
}
return buffer;
}
private static uint GetOffset(ref PlaneOffsets offsets, int plane)
{
return plane switch
{
0 => offsets.LumaOffset,
1 => offsets.ChromaUOffset,
2 => offsets.ChromaVOffset,
_ => throw new ArgumentOutOfRangeException(nameof(plane))
};
}
private static RentedBuffer ReadBufferNoDeinterlace(
ResourceManager rm,
ref PlaneOffsets offsets,
bool linear,
int plane,
int width,
int height,
int bytesPerPixel,
int gobBlocksInY)
{
int stride = GetPitch(width, bytesPerPixel);
uint offset = GetOffset(ref offsets, plane);
if (linear)
{
return ReadBufferLinear(rm, offset, width, height, 0, stride, bytesPerPixel);
}
return ReadBufferBlockLinear(rm, offset, width, height, 0, stride, bytesPerPixel, gobBlocksInY);
}
private static RentedBuffer ReadBufferLinear(
ResourceManager rm,
uint offset,
int width,
int height,
int dstStart,
int dstStride,
int bytesPerPixel)
{
int srcStride = GetPitch(width, bytesPerPixel);
int inSize = srcStride * height;
ReadOnlySpan<byte> src = rm.Gmm.GetSpan(ExtendOffset(offset), inSize);
int outSize = dstStride * height;
int bufferIndex = rm.BufferPool.RentMinimum(outSize, out byte[] buffer);
Span<byte> dst = buffer;
dst = dst.Slice(0, outSize);
for (int y = 0; y < height; y++)
{
src.Slice(y * srcStride, srcStride).CopyTo(dst.Slice(dstStart + y * dstStride, srcStride));
}
return new RentedBuffer(dst, bufferIndex);
}
private static RentedBuffer ReadBufferBlockLinear(
ResourceManager rm,
uint offset,
int width,
int height,
int dstStart,
int dstStride,
int bytesPerPixel,
int gobBlocksInY)
{
int inSize = GetBlockLinearSize(width, height, bytesPerPixel, gobBlocksInY);
ReadOnlySpan<byte> src = rm.Gmm.GetSpan(ExtendOffset(offset), inSize);
int outSize = dstStride * height;
int bufferIndex = rm.BufferPool.RentMinimum(outSize, out byte[] buffer);
Span<byte> dst = buffer;
dst = dst.Slice(0, outSize);
LayoutConverter.ConvertBlockLinearToLinear(dst.Slice(dstStart), width, height, dstStride, bytesPerPixel, gobBlocksInY, src);
return new RentedBuffer(dst, bufferIndex);
}
}
}

667
src/Ryujinx.Graphics.Vic/Image/SurfaceWriter.cs Normal file
View File

@ -0,0 +1,667 @@
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.Texture;
using Ryujinx.Graphics.Vic.Types;
using System;
using System.Runtime.Intrinsics;
using System.Runtime.Intrinsics.Arm;
using System.Runtime.Intrinsics.X86;
using static Ryujinx.Graphics.Vic.Image.SurfaceCommon;
namespace Ryujinx.Graphics.Vic.Image
{
class SurfaceWriter
{
public static void Write(ResourceManager rm, Surface input, ref OutputSurfaceConfig config, ref PlaneOffsets offsets)
{
switch (config.OutPixelFormat)
{
case PixelFormat.A8B8G8R8:
case PixelFormat.X8B8G8R8:
WriteA8B8G8R8(rm, input, ref config, ref offsets);
break;
case PixelFormat.A8R8G8B8:
WriteA8R8G8B8(rm, input, ref config, ref offsets);
break;
case PixelFormat.Y8___V8U8_N420:
WriteNv12(rm, input, ref config, ref offsets);
break;
default:
Logger.Error?.Print(LogClass.Vic, $"Unsupported pixel format \"{config.OutPixelFormat}\".");
break;
}
}
private unsafe static void WriteA8B8G8R8(ResourceManager rm, Surface input, ref OutputSurfaceConfig config, ref PlaneOffsets offsets)
{
int width = input.Width;
int height = input.Height;
int stride = GetPitch(width, 4);
int dstIndex = rm.BufferPool.Rent(height * stride, out Span<byte> dst);
if (Sse2.IsSupported)
{
int widthTrunc = width & ~7;
int strideGap = stride - width * 4;
fixed (Pixel* srcPtr = input.Data)
{
Pixel* ip = srcPtr;
fixed (byte* dstPtr = dst)
{
byte* op = dstPtr;
for (int y = 0; y < height; y++, ip += input.Width)
{
int x = 0;
for (; x < widthTrunc; x += 8)
{
Vector128<ushort> pixel12 = Sse2.LoadVector128((ushort*)(ip + (uint)x));
Vector128<ushort> pixel34 = Sse2.LoadVector128((ushort*)(ip + (uint)x + 2));
Vector128<ushort> pixel56 = Sse2.LoadVector128((ushort*)(ip + (uint)x + 4));
Vector128<ushort> pixel78 = Sse2.LoadVector128((ushort*)(ip + (uint)x + 6));
pixel12 = Sse2.ShiftRightLogical(pixel12, 2);
pixel34 = Sse2.ShiftRightLogical(pixel34, 2);
pixel56 = Sse2.ShiftRightLogical(pixel56, 2);
pixel78 = Sse2.ShiftRightLogical(pixel78, 2);
Vector128<byte> pixel1234 = Sse2.PackUnsignedSaturate(pixel12.AsInt16(), pixel34.AsInt16());
Vector128<byte> pixel5678 = Sse2.PackUnsignedSaturate(pixel56.AsInt16(), pixel78.AsInt16());
Sse2.Store(op + 0x00, pixel1234);
Sse2.Store(op + 0x10, pixel5678);
op += 0x20;
}
for (; x < width; x++)
{
Pixel* px = ip + (uint)x;
*(op + 0) = Downsample(px->R);
*(op + 1) = Downsample(px->G);
*(op + 2) = Downsample(px->B);
*(op + 3) = Downsample(px->A);
op += 4;
}
op += strideGap;
}
}
}
}
else if (AdvSimd.IsSupported)
{
int widthTrunc = width & ~7;
int strideGap = stride - width * 4;
fixed (Pixel* srcPtr = input.Data)
{
Pixel* ip = srcPtr;
fixed (byte* dstPtr = dst)
{
byte* op = dstPtr;
for (int y = 0; y < height; y++, ip += input.Width)
{
int x = 0;
for (; x < widthTrunc; x += 8)
{
Vector128<ushort> pixel12 = AdvSimd.LoadVector128((ushort*)(ip + (uint)x));
Vector128<ushort> pixel34 = AdvSimd.LoadVector128((ushort*)(ip + (uint)x + 2));
Vector128<ushort> pixel56 = AdvSimd.LoadVector128((ushort*)(ip + (uint)x + 4));
Vector128<ushort> pixel78 = AdvSimd.LoadVector128((ushort*)(ip + (uint)x + 6));
pixel12 = AdvSimd.ShiftRightLogical(pixel12, 2);
pixel34 = AdvSimd.ShiftRightLogical(pixel34, 2);
pixel56 = AdvSimd.ShiftRightLogical(pixel56, 2);
pixel78 = AdvSimd.ShiftRightLogical(pixel78, 2);
Vector64<byte> lower12 = AdvSimd.ExtractNarrowingLower(pixel12.AsUInt16());
Vector64<byte> lower56 = AdvSimd.ExtractNarrowingLower(pixel56.AsUInt16());
Vector128<byte> pixel1234 = AdvSimd.ExtractNarrowingUpper(lower12, pixel34.AsUInt16());
Vector128<byte> pixel5678 = AdvSimd.ExtractNarrowingUpper(lower56, pixel78.AsUInt16());
AdvSimd.Store(op + 0x00, pixel1234);
AdvSimd.Store(op + 0x10, pixel5678);
op += 0x20;
}
for (; x < width; x++)
{
Pixel* px = ip + (uint)x;
*(op + 0) = Downsample(px->R);
*(op + 1) = Downsample(px->G);
*(op + 2) = Downsample(px->B);
*(op + 3) = Downsample(px->A);
op += 4;
}
op += strideGap;
}
}
}
}
else
{
for (int y = 0; y < height; y++)
{
int baseOffs = y * stride;
for (int x = 0; x < width; x++)
{
int offs = baseOffs + x * 4;
dst[offs + 0] = Downsample(input.GetR(x, y));
dst[offs + 1] = Downsample(input.GetG(x, y));
dst[offs + 2] = Downsample(input.GetB(x, y));
dst[offs + 3] = Downsample(input.GetA(x, y));
}
}
}
bool outLinear = config.OutBlkKind == 0;
int gobBlocksInY = 1 << config.OutBlkHeight;
WriteBuffer(rm, dst, offsets.LumaOffset, outLinear, width, height, 4, gobBlocksInY);
rm.BufferPool.Return(dstIndex);
}
private unsafe static void WriteA8R8G8B8(ResourceManager rm, Surface input, ref OutputSurfaceConfig config, ref PlaneOffsets offsets)
{
int width = input.Width;
int height = input.Height;
int stride = GetPitch(width, 4);
int dstIndex = rm.BufferPool.Rent(height * stride, out Span<byte> dst);
if (Ssse3.IsSupported)
{
Vector128<byte> shuffleMask = Vector128.Create(
(byte)2, (byte)1, (byte)0, (byte)3,
(byte)6, (byte)5, (byte)4, (byte)7,
(byte)10, (byte)9, (byte)8, (byte)11,
(byte)14, (byte)13, (byte)12, (byte)15);
int widthTrunc = width & ~7;
int strideGap = stride - width * 4;
fixed (Pixel* srcPtr = input.Data)
{
Pixel* ip = srcPtr;
fixed (byte* dstPtr = dst)
{
byte* op = dstPtr;
for (int y = 0; y < height; y++, ip += input.Width)
{
int x = 0;
for (; x < widthTrunc; x += 8)
{
Vector128<ushort> pixel12 = Sse2.LoadVector128((ushort*)(ip + (uint)x));
Vector128<ushort> pixel34 = Sse2.LoadVector128((ushort*)(ip + (uint)x + 2));
Vector128<ushort> pixel56 = Sse2.LoadVector128((ushort*)(ip + (uint)x + 4));
Vector128<ushort> pixel78 = Sse2.LoadVector128((ushort*)(ip + (uint)x + 6));
pixel12 = Sse2.ShiftRightLogical(pixel12, 2);
pixel34 = Sse2.ShiftRightLogical(pixel34, 2);
pixel56 = Sse2.ShiftRightLogical(pixel56, 2);
pixel78 = Sse2.ShiftRightLogical(pixel78, 2);
Vector128<byte> pixel1234 = Sse2.PackUnsignedSaturate(pixel12.AsInt16(), pixel34.AsInt16());
Vector128<byte> pixel5678 = Sse2.PackUnsignedSaturate(pixel56.AsInt16(), pixel78.AsInt16());
pixel1234 = Ssse3.Shuffle(pixel1234, shuffleMask);
pixel5678 = Ssse3.Shuffle(pixel5678, shuffleMask);
Sse2.Store(op + 0x00, pixel1234);
Sse2.Store(op + 0x10, pixel5678);
op += 0x20;
}
for (; x < width; x++)
{
Pixel* px = ip + (uint)x;
*(op + 0) = Downsample(px->B);
*(op + 1) = Downsample(px->G);
*(op + 2) = Downsample(px->R);
*(op + 3) = Downsample(px->A);
op += 4;
}
op += strideGap;
}
}
}
}
else
{
for (int y = 0; y < height; y++)
{
int baseOffs = y * stride;
for (int x = 0; x < width; x++)
{
int offs = baseOffs + x * 4;
dst[offs + 0] = Downsample(input.GetB(x, y));
dst[offs + 1] = Downsample(input.GetG(x, y));
dst[offs + 2] = Downsample(input.GetR(x, y));
dst[offs + 3] = Downsample(input.GetA(x, y));
}
}
}
bool outLinear = config.OutBlkKind == 0;
int gobBlocksInY = 1 << config.OutBlkHeight;
WriteBuffer(rm, dst, offsets.LumaOffset, outLinear, width, height, 4, gobBlocksInY);
rm.BufferPool.Return(dstIndex);
}
private unsafe static void WriteNv12(ResourceManager rm, Surface input, ref OutputSurfaceConfig config, ref PlaneOffsets offsets)
{
int gobBlocksInY = 1 << config.OutBlkHeight;
bool outLinear = config.OutBlkKind == 0;
int width = Math.Min(config.OutLumaWidth + 1, input.Width);
int height = Math.Min(config.OutLumaHeight + 1, input.Height);
int yStride = GetPitch(config.OutLumaWidth + 1, 1);
int dstYIndex = rm.BufferPool.Rent((config.OutLumaHeight + 1) * yStride, out Span<byte> dstY);
if (Sse41.IsSupported)
{
Vector128<ushort> mask = Vector128.Create(0xffffUL).AsUInt16();
int widthTrunc = width & ~0xf;
int strideGap = yStride - width;
fixed (Pixel* srcPtr = input.Data)
{
Pixel* ip = srcPtr;
fixed (byte* dstPtr = dstY)
{
byte* op = dstPtr;
for (int y = 0; y < height; y++, ip += input.Width)
{
int x = 0;
for (; x < widthTrunc; x += 16)
{
byte* baseOffset = (byte*)(ip + (ulong)(uint)x);
Vector128<ushort> pixelp1 = Sse2.LoadVector128((ushort*)baseOffset);
Vector128<ushort> pixelp2 = Sse2.LoadVector128((ushort*)(baseOffset + 0x10));
Vector128<ushort> pixelp3 = Sse2.LoadVector128((ushort*)(baseOffset + 0x20));
Vector128<ushort> pixelp4 = Sse2.LoadVector128((ushort*)(baseOffset + 0x30));
Vector128<ushort> pixelp5 = Sse2.LoadVector128((ushort*)(baseOffset + 0x40));
Vector128<ushort> pixelp6 = Sse2.LoadVector128((ushort*)(baseOffset + 0x50));
Vector128<ushort> pixelp7 = Sse2.LoadVector128((ushort*)(baseOffset + 0x60));
Vector128<ushort> pixelp8 = Sse2.LoadVector128((ushort*)(baseOffset + 0x70));
pixelp1 = Sse2.And(pixelp1, mask);
pixelp2 = Sse2.And(pixelp2, mask);
pixelp3 = Sse2.And(pixelp3, mask);
pixelp4 = Sse2.And(pixelp4, mask);
pixelp5 = Sse2.And(pixelp5, mask);
pixelp6 = Sse2.And(pixelp6, mask);
pixelp7 = Sse2.And(pixelp7, mask);
pixelp8 = Sse2.And(pixelp8, mask);
Vector128<ushort> pixelq1 = Sse41.PackUnsignedSaturate(pixelp1.AsInt32(), pixelp2.AsInt32());
Vector128<ushort> pixelq2 = Sse41.PackUnsignedSaturate(pixelp3.AsInt32(), pixelp4.AsInt32());
Vector128<ushort> pixelq3 = Sse41.PackUnsignedSaturate(pixelp5.AsInt32(), pixelp6.AsInt32());
Vector128<ushort> pixelq4 = Sse41.PackUnsignedSaturate(pixelp7.AsInt32(), pixelp8.AsInt32());
pixelq1 = Sse41.PackUnsignedSaturate(pixelq1.AsInt32(), pixelq2.AsInt32());
pixelq2 = Sse41.PackUnsignedSaturate(pixelq3.AsInt32(), pixelq4.AsInt32());
pixelq1 = Sse2.ShiftRightLogical(pixelq1, 2);
pixelq2 = Sse2.ShiftRightLogical(pixelq2, 2);
Vector128<byte> pixel = Sse2.PackUnsignedSaturate(pixelq1.AsInt16(), pixelq2.AsInt16());
Sse2.Store(op, pixel);
op += 0x10;
}
for (; x < width; x++)
{
Pixel* px = ip + (uint)x;
*op++ = Downsample(px->R);
}
op += strideGap;
}
}
}
}
else if (AdvSimd.IsSupported)
{
Vector128<ushort> mask = Vector128.Create(0xffffUL).AsUInt16();
int widthTrunc = width & ~0xf;
int strideGap = yStride - width;
fixed (Pixel* srcPtr = input.Data)
{
Pixel* ip = srcPtr;
fixed (byte* dstPtr = dstY)
{
byte* op = dstPtr;
for (int y = 0; y < height; y++, ip += input.Width)
{
int x = 0;
for (; x < widthTrunc; x += 16)
{
byte* baseOffset = (byte*)(ip + (ulong)(uint)x);
Vector128<ushort> pixelp1 = AdvSimd.LoadVector128((ushort*)baseOffset);
Vector128<ushort> pixelp2 = AdvSimd.LoadVector128((ushort*)(baseOffset + 0x10));
Vector128<ushort> pixelp3 = AdvSimd.LoadVector128((ushort*)(baseOffset + 0x20));
Vector128<ushort> pixelp4 = AdvSimd.LoadVector128((ushort*)(baseOffset + 0x30));
Vector128<ushort> pixelp5 = AdvSimd.LoadVector128((ushort*)(baseOffset + 0x40));
Vector128<ushort> pixelp6 = AdvSimd.LoadVector128((ushort*)(baseOffset + 0x50));
Vector128<ushort> pixelp7 = AdvSimd.LoadVector128((ushort*)(baseOffset + 0x60));
Vector128<ushort> pixelp8 = AdvSimd.LoadVector128((ushort*)(baseOffset + 0x70));
pixelp1 = AdvSimd.And(pixelp1, mask);
pixelp2 = AdvSimd.And(pixelp2, mask);
pixelp3 = AdvSimd.And(pixelp3, mask);
pixelp4 = AdvSimd.And(pixelp4, mask);
pixelp5 = AdvSimd.And(pixelp5, mask);
pixelp6 = AdvSimd.And(pixelp6, mask);
pixelp7 = AdvSimd.And(pixelp7, mask);
pixelp8 = AdvSimd.And(pixelp8, mask);
Vector64<ushort> lowerp1 = AdvSimd.ExtractNarrowingLower(pixelp1.AsUInt32());
Vector64<ushort> lowerp3 = AdvSimd.ExtractNarrowingLower(pixelp3.AsUInt32());
Vector64<ushort> lowerp5 = AdvSimd.ExtractNarrowingLower(pixelp5.AsUInt32());
Vector64<ushort> lowerp7 = AdvSimd.ExtractNarrowingLower(pixelp7.AsUInt32());
Vector128<ushort> pixelq1 = AdvSimd.ExtractNarrowingUpper(lowerp1, pixelp2.AsUInt32());
Vector128<ushort> pixelq2 = AdvSimd.ExtractNarrowingUpper(lowerp3, pixelp4.AsUInt32());
Vector128<ushort> pixelq3 = AdvSimd.ExtractNarrowingUpper(lowerp5, pixelp6.AsUInt32());
Vector128<ushort> pixelq4 = AdvSimd.ExtractNarrowingUpper(lowerp7, pixelp8.AsUInt32());
Vector64<ushort> lowerq1 = AdvSimd.ExtractNarrowingLower(pixelq1.AsUInt32());
Vector64<ushort> lowerq3 = AdvSimd.ExtractNarrowingLower(pixelq3.AsUInt32());
pixelq1 = AdvSimd.ExtractNarrowingUpper(lowerq1, pixelq2.AsUInt32());
pixelq2 = AdvSimd.ExtractNarrowingUpper(lowerq3, pixelq4.AsUInt32());
pixelq1 = AdvSimd.ShiftRightLogical(pixelq1, 2);
pixelq2 = AdvSimd.ShiftRightLogical(pixelq2, 2);
Vector64<byte> pixelLower = AdvSimd.ExtractNarrowingLower(pixelq1.AsUInt16());
Vector128<byte> pixel = AdvSimd.ExtractNarrowingUpper(pixelLower, pixelq2.AsUInt16());
AdvSimd.Store(op, pixel);
op += 0x10;
}
for (; x < width; x++)
{
Pixel* px = ip + (uint)x;
*op++ = Downsample(px->R);
}
op += strideGap;
}
}
}
}
else
{
for (int y = 0; y < height; y++)
{
for (int x = 0; x < width; x++)
{
dstY[y * yStride + x] = Downsample(input.GetR(x, y));
}
}
}
WriteBuffer(
rm,
dstY,
offsets.LumaOffset,
outLinear,
config.OutLumaWidth + 1,
config.OutLumaHeight + 1,
1,
gobBlocksInY);
rm.BufferPool.Return(dstYIndex);
int uvWidth = Math.Min(config.OutChromaWidth + 1, (width + 1) >> 1);
int uvHeight = Math.Min(config.OutChromaHeight + 1, (height + 1) >> 1);
int uvStride = GetPitch(config.OutChromaWidth + 1, 2);
int dstUvIndex = rm.BufferPool.Rent((config.OutChromaHeight + 1) * uvStride, out Span<byte> dstUv);
if (Sse2.IsSupported)
{
int widthTrunc = uvWidth & ~7;
int strideGap = uvStride - uvWidth * 2;
fixed (Pixel* srcPtr = input.Data)
{
Pixel* ip = srcPtr;
fixed (byte* dstPtr = dstUv)
{
byte* op = dstPtr;
for (int y = 0; y < uvHeight; y++, ip += input.Width * 2)
{
int x = 0;
for (; x < widthTrunc; x += 8)
{
byte* baseOffset = (byte*)ip + (ulong)(uint)x * 16;
Vector128<uint> pixel1 = Sse2.LoadScalarVector128((uint*)(baseOffset + 0x02));
Vector128<uint> pixel2 = Sse2.LoadScalarVector128((uint*)(baseOffset + 0x12));
Vector128<uint> pixel3 = Sse2.LoadScalarVector128((uint*)(baseOffset + 0x22));
Vector128<uint> pixel4 = Sse2.LoadScalarVector128((uint*)(baseOffset + 0x32));
Vector128<uint> pixel5 = Sse2.LoadScalarVector128((uint*)(baseOffset + 0x42));
Vector128<uint> pixel6 = Sse2.LoadScalarVector128((uint*)(baseOffset + 0x52));
Vector128<uint> pixel7 = Sse2.LoadScalarVector128((uint*)(baseOffset + 0x62));
Vector128<uint> pixel8 = Sse2.LoadScalarVector128((uint*)(baseOffset + 0x72));
Vector128<uint> pixel12 = Sse2.UnpackLow(pixel1, pixel2);
Vector128<uint> pixel34 = Sse2.UnpackLow(pixel3, pixel4);
Vector128<uint> pixel56 = Sse2.UnpackLow(pixel5, pixel6);
Vector128<uint> pixel78 = Sse2.UnpackLow(pixel7, pixel8);
Vector128<ulong> pixel1234 = Sse2.UnpackLow(pixel12.AsUInt64(), pixel34.AsUInt64());
Vector128<ulong> pixel5678 = Sse2.UnpackLow(pixel56.AsUInt64(), pixel78.AsUInt64());
pixel1234 = Sse2.ShiftRightLogical(pixel1234, 2);
pixel5678 = Sse2.ShiftRightLogical(pixel5678, 2);
Vector128<byte> pixel = Sse2.PackUnsignedSaturate(pixel1234.AsInt16(), pixel5678.AsInt16());
Sse2.Store(op, pixel);
op += 0x10;
}
for (; x < uvWidth; x++)
{
Pixel* px = ip + (uint)(x << 1);
*op++ = Downsample(px->G);
*op++ = Downsample(px->B);
}
op += strideGap;
}
}
}
}
else if (AdvSimd.Arm64.IsSupported)
{
int widthTrunc = uvWidth & ~7;
int strideGap = uvStride - uvWidth * 2;
fixed (Pixel* srcPtr = input.Data)
{
Pixel* ip = srcPtr;
fixed (byte* dstPtr = dstUv)
{
byte* op = dstPtr;
for (int y = 0; y < uvHeight; y++, ip += input.Width * 2)
{
int x = 0;
for (; x < widthTrunc; x += 8)
{
byte* baseOffset = (byte*)ip + (ulong)(uint)x * 16;
Vector128<uint> pixel1 = AdvSimd.LoadAndReplicateToVector128((uint*)(baseOffset + 0x02));
Vector128<uint> pixel2 = AdvSimd.LoadAndReplicateToVector128((uint*)(baseOffset + 0x12));
Vector128<uint> pixel3 = AdvSimd.LoadAndReplicateToVector128((uint*)(baseOffset + 0x22));
Vector128<uint> pixel4 = AdvSimd.LoadAndReplicateToVector128((uint*)(baseOffset + 0x32));
Vector128<uint> pixel5 = AdvSimd.LoadAndReplicateToVector128((uint*)(baseOffset + 0x42));
Vector128<uint> pixel6 = AdvSimd.LoadAndReplicateToVector128((uint*)(baseOffset + 0x52));
Vector128<uint> pixel7 = AdvSimd.LoadAndReplicateToVector128((uint*)(baseOffset + 0x62));
Vector128<uint> pixel8 = AdvSimd.LoadAndReplicateToVector128((uint*)(baseOffset + 0x72));
Vector128<uint> pixel12 = AdvSimd.Arm64.ZipLow(pixel1, pixel2);
Vector128<uint> pixel34 = AdvSimd.Arm64.ZipLow(pixel3, pixel4);
Vector128<uint> pixel56 = AdvSimd.Arm64.ZipLow(pixel5, pixel6);
Vector128<uint> pixel78 = AdvSimd.Arm64.ZipLow(pixel7, pixel8);
Vector128<ulong> pixel1234 = AdvSimd.Arm64.ZipLow(pixel12.AsUInt64(), pixel34.AsUInt64());
Vector128<ulong> pixel5678 = AdvSimd.Arm64.ZipLow(pixel56.AsUInt64(), pixel78.AsUInt64());
pixel1234 = AdvSimd.ShiftRightLogical(pixel1234, 2);
pixel5678 = AdvSimd.ShiftRightLogical(pixel5678, 2);
Vector64<byte> pixelLower = AdvSimd.ExtractNarrowingLower(pixel1234.AsUInt16());
Vector128<byte> pixel = AdvSimd.ExtractNarrowingUpper(pixelLower, pixel5678.AsUInt16());
AdvSimd.Store(op, pixel);
op += 0x10;
}
for (; x < uvWidth; x++)
{
Pixel* px = ip + (uint)(x << 1);
*op++ = Downsample(px->G);
*op++ = Downsample(px->B);
}
op += strideGap;
}
}
}
}
else
{
for (int y = 0; y < uvHeight; y++)
{
for (int x = 0; x < uvWidth; x++)
{
int xx = x << 1;
int yy = y << 1;
int uvOffs = y * uvStride + xx;
dstUv[uvOffs + 0] = Downsample(input.GetG(xx, yy));
dstUv[uvOffs + 1] = Downsample(input.GetB(xx, yy));
}
}
}
WriteBuffer(
rm,
dstUv,
offsets.ChromaUOffset,
outLinear,
config.OutChromaWidth + 1,
config.OutChromaHeight + 1, 2,
gobBlocksInY);
rm.BufferPool.Return(dstUvIndex);
}
private static void WriteBuffer(
ResourceManager rm,
ReadOnlySpan<byte> src,
uint offset,
bool linear,
int width,
int height,
int bytesPerPixel,
int gobBlocksInY)
{
if (linear)
{
rm.Gmm.WriteMapped(ExtendOffset(offset), src);
return;
}
WriteBuffer(rm, src, offset, width, height, bytesPerPixel, gobBlocksInY);
}
private static void WriteBuffer(
ResourceManager rm,
ReadOnlySpan<byte> src,
uint offset,
int width,
int height,
int bytesPerPixel,
int gobBlocksInY)
{
int outSize = GetBlockLinearSize(width, height, bytesPerPixel, gobBlocksInY);
int dstStride = GetPitch(width, bytesPerPixel);
int dstIndex = rm.BufferPool.Rent(outSize, out Span<byte> dst);
LayoutConverter.ConvertLinearToBlockLinear(dst, width, height, dstStride, bytesPerPixel, gobBlocksInY, src);
rm.Gmm.WriteMapped(ExtendOffset(offset), dst);
rm.BufferPool.Return(dstIndex);
}
}
}