New NVDEC and VIC implementation (#1384)

* Initial NVDEC and VIC implementation

* Update FFmpeg.AutoGen to 4.3.0

* Add nvdec dependencies for Windows

* Unify some VP9 structures

* Rename VP9 structure fields

* Improvements to Video API

* XML docs for Common.Memory

* Remove now unused or redundant overloads from MemoryAccessor

* NVDEC UV surface read/write scalar paths

* Add FIXME comments about hacky things/stuff that will need to be fixed in the future

* Cleaned up VP9 memory allocation

* Remove some debug logs

* Rename some VP9 structs

* Remove unused struct

* No need to compile Ryujinx.Graphics.Host1x with unsafe anymore

* Name AsyncWorkQueue threads to make debugging easier

* Make Vp9PictureInfo a ref struct

* LayoutConverter no longer needs the depth argument (broken by rebase)

* Pooling of VP9 buffers, plus fix a memory leak on VP9

* Really wish VS could rename projects properly...

* Address feedback

* Remove using

* Catch OperationCanceledException

* Add licensing informations

* Add THIRDPARTY.md to release too

Co-authored-by: Thog <me@thog.eu>
This commit is contained in:
gdkchan 2020-07-12 00:07:01 -03:00 committed by GitHub
parent 38b26cf424
commit 4d02a2d2c0
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
202 changed files with 20563 additions and 2567 deletions

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 * 1280;
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;
}
}
}

View file

@ -0,0 +1,17 @@
using System;
namespace Ryujinx.Graphics.Vic.Image
{
ref struct InputSurface
{
public ReadOnlySpan<byte> Buffer0;
public ReadOnlySpan<byte> Buffer1;
public ReadOnlySpan<byte> Buffer2;
public int Width;
public int Height;
public int UvWidth;
public int UvHeight;
}
}

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;
}
}

View file

@ -0,0 +1,46 @@
using System;
using System.Runtime.CompilerServices;
namespace Ryujinx.Graphics.Vic.Image
{
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);
}
}

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);
}
}
}

View file

@ -0,0 +1,253 @@
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.Gpu.Memory;
using Ryujinx.Graphics.Texture;
using Ryujinx.Graphics.Vic.Types;
using System;
using System.Runtime.CompilerServices;
using System.Runtime.Intrinsics;
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 SlotSurfaceConfig config, ref PlaneOffsets offsets)
{
switch (config.SlotPixelFormat)
{
case PixelFormat.Y8___V8U8_N420: return ReadNv12(rm, ref config, ref offsets);
}
Logger.PrintError(LogClass.Vic, $"Unsupported pixel format \"{config.SlotPixelFormat}\".");
int lw = config.SlotLumaWidth + 1;
int lh = config.SlotLumaHeight + 1;
return new Surface(rm.SurfacePool, lw, lh);
}
private unsafe static Surface ReadNv12(ResourceManager rm, ref SlotSurfaceConfig config, ref PlaneOffsets offsets)
{
InputSurface input = ReadSurface(rm.Gmm, ref config, 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(0xffUL << 48).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();
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.Or(rgba16_0, alphaMask);
rgba16_1 = Sse2.Or(rgba16_1, alphaMask);
rgba16_2 = Sse2.Or(rgba16_2, alphaMask);
rgba16_3 = Sse2.Or(rgba16_3, alphaMask);
rgba16_4 = Sse2.Or(rgba16_4, alphaMask);
rgba16_5 = Sse2.Or(rgba16_5, alphaMask);
rgba16_6 = Sse2.Or(rgba16_6, alphaMask);
rgba16_7 = Sse2.Or(rgba16_7, alphaMask);
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
{
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);
}
}
}
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(
MemoryManager gmm,
ref SlotSurfaceConfig config,
ref PlaneOffsets offsets,
int bytesPerPixel,
int planes)
{
InputSurface surface = new InputSurface();
int gobBlocksInY = 1 << config.SlotBlkHeight;
bool linear = config.SlotBlkKind == 0;
int lw = config.SlotLumaWidth + 1;
int lh = config.SlotLumaHeight + 1;
int cw = config.SlotChromaWidth + 1;
int ch = config.SlotChromaHeight + 1;
surface.Width = lw;
surface.Height = lh;
surface.UvWidth = cw;
surface.UvHeight = ch;
if (planes > 0)
{
surface.Buffer0 = ReadBuffer(gmm, offsets.LumaOffset, linear, lw, lh, bytesPerPixel, gobBlocksInY);
}
if (planes > 1)
{
surface.Buffer1 = ReadBuffer(gmm, offsets.ChromaUOffset, linear, cw, ch, planes == 2 ? 2 : 1, gobBlocksInY);
}
if (planes > 2)
{
surface.Buffer2 = ReadBuffer(gmm, offsets.ChromaVOffset, linear, cw, ch, 1, gobBlocksInY);
}
return surface;
}
private static ReadOnlySpan<byte> ReadBuffer(
MemoryManager gmm,
uint offset,
bool linear,
int width,
int height,
int bytesPerPixel,
int gobBlocksInY)
{
int stride = GetPitch(width, bytesPerPixel);
if (linear)
{
return gmm.GetSpan(ExtendOffset(offset), stride * height);
}
return ReadBuffer(gmm, offset, width, height, stride, bytesPerPixel, gobBlocksInY);
}
private static ReadOnlySpan<byte> ReadBuffer(
MemoryManager gmm,
uint offset,
int width,
int height,
int dstStride,
int bytesPerPixel,
int gobBlocksInY)
{
int inSize = GetBlockLinearSize(width, height, bytesPerPixel, gobBlocksInY);
ReadOnlySpan<byte> src = gmm.GetSpan(ExtendOffset(offset), inSize);
Span<byte> dst = new byte[dstStride * height];
LayoutConverter.ConvertBlockLinearToLinear(dst, width, height, dstStride, bytesPerPixel, gobBlocksInY, src);
return dst;
}
}
}

View file

@ -0,0 +1,361 @@
using Ryujinx.Common.Logging;
using Ryujinx.Graphics.Texture;
using Ryujinx.Graphics.Vic.Types;
using System;
using System.Runtime.Intrinsics;
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:
WriteA8B8G8R8(rm, input, ref config, ref offsets);
break;
case PixelFormat.Y8___V8U8_N420:
WriteNv12(rm, input, ref config, ref offsets);
break;
default:
Logger.PrintError(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
{
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 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
{
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
{
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.Write(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.Write(ExtendOffset(offset), dst);
rm.BufferPool.Return(dstIndex);
}
}
}