Ryujinx/Ryujinx.Graphics.Gpu/Engine/Twod/TwodClass.cs
riperiperi c9c65af59e
Perform unscaled 2d engine copy on CPU if source texture isn't in cache. (#3112)
* Initial implementation of fast 2d copy

TODO: Partial copy for mismatching region/size.

* WIP

* Cleanup

* Update Ryujinx.Graphics.Gpu/Engine/Twod/TwodClass.cs

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

Co-authored-by: gdkchan <gab.dark.100@gmail.com>
2022-02-22 11:21:29 -03:00

374 lines
15 KiB
C#

using Ryujinx.Common;
using Ryujinx.Graphics.Device;
using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Engine.Types;
using Ryujinx.Graphics.Gpu.Image;
using Ryujinx.Graphics.Texture;
using Ryujinx.Memory;
using System;
using System.Collections.Generic;
using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using System.Runtime.Intrinsics;
namespace Ryujinx.Graphics.Gpu.Engine.Twod
{
/// <summary>
/// Represents a 2D engine class.
/// </summary>
class TwodClass : IDeviceState
{
private readonly GpuChannel _channel;
private readonly DeviceState<TwodClassState> _state;
/// <summary>
/// Creates a new instance of the 2D engine class.
/// </summary>
/// <param name="channel">The channel that will make use of the engine</param>
public TwodClass(GpuChannel channel)
{
_channel = channel;
_state = new DeviceState<TwodClassState>(new Dictionary<string, RwCallback>
{
{ nameof(TwodClassState.PixelsFromMemorySrcY0Int), new RwCallback(PixelsFromMemorySrcY0Int, null) }
});
}
/// <summary>
/// Reads data from the class registers.
/// </summary>
/// <param name="offset">Register byte offset</param>
/// <returns>Data at the specified offset</returns>
public int Read(int offset) => _state.Read(offset);
/// <summary>
/// Writes data to the class registers.
/// </summary>
/// <param name="offset">Register byte offset</param>
/// <param name="data">Data to be written</param>
public void Write(int offset, int data) => _state.Write(offset, data);
/// <summary>
/// Determines if data is compatible between the source and destination texture.
/// The two textures must have the same size, layout, and bytes per pixel.
/// </summary>
/// <param name="lhs">Info for the first texture</param>
/// <param name="rhs">Info for the second texture</param>
/// <param name="lhsFormat">Format of the first texture</param>
/// <param name="rhsFormat">Format of the second texture</param>
/// <returns>True if the data is compatible, false otherwise</returns>
private bool IsDataCompatible(TwodTexture lhs, TwodTexture rhs, FormatInfo lhsFormat, FormatInfo rhsFormat)
{
if (lhsFormat.BytesPerPixel != rhsFormat.BytesPerPixel ||
lhs.Height != rhs.Height ||
lhs.Depth != rhs.Depth ||
lhs.LinearLayout != rhs.LinearLayout ||
lhs.MemoryLayout.Packed != rhs.MemoryLayout.Packed)
{
return false;
}
if (lhs.LinearLayout)
{
return lhs.Stride == rhs.Stride;
}
else
{
return lhs.Width == rhs.Width;
}
}
/// <summary>
/// Determine if the given region covers the full texture, also considering width alignment.
/// </summary>
/// <param name="texture">The texture to check</param>
/// <param name="formatInfo"></param>
/// <param name="x1">Region start x</param>
/// <param name="y1">Region start y</param>
/// <param name="x2">Region end x</param>
/// <param name="y2">Region end y</param>
/// <returns>True if the region covers the full texture, false otherwise</returns>
private bool IsCopyRegionComplete(TwodTexture texture, FormatInfo formatInfo, int x1, int y1, int x2, int y2)
{
if (x1 != 0 || y1 != 0 || y2 != texture.Height)
{
return false;
}
int width;
int widthAlignment;
if (texture.LinearLayout)
{
widthAlignment = 1;
width = texture.Stride / formatInfo.BytesPerPixel;
}
else
{
widthAlignment = Constants.GobAlignment / formatInfo.BytesPerPixel;
width = texture.Width;
}
return width == BitUtils.AlignUp(x2, widthAlignment);
}
/// <summary>
/// Performs a full data copy between two textures, reading and writing guest memory directly.
/// The textures must have a matching layout, size, and bytes per pixel.
/// </summary>
/// <param name="src">The source texture</param>
/// <param name="dst">The destination texture</param>
/// <param name="w">Copy width</param>
/// <param name="h">Copy height</param>
/// <param name="bpp">Bytes per pixel</param>
private void UnscaledFullCopy(TwodTexture src, TwodTexture dst, int w, int h, int bpp)
{
var srcCalculator = new OffsetCalculator(
w,
h,
src.Stride,
src.LinearLayout,
src.MemoryLayout.UnpackGobBlocksInY(),
src.MemoryLayout.UnpackGobBlocksInZ(),
bpp);
(int _, int srcSize) = srcCalculator.GetRectangleRange(0, 0, w, h);
var memoryManager = _channel.MemoryManager;
ulong srcGpuVa = src.Address.Pack();
ulong dstGpuVa = dst.Address.Pack();
ReadOnlySpan<byte> srcSpan = memoryManager.GetSpan(srcGpuVa, srcSize, true);
int width;
int height = src.Height;
if (src.LinearLayout)
{
width = src.Stride / bpp;
}
else
{
width = src.Width;
}
// If the copy is not equal to the width and height of the texture, we will need to copy partially.
// It's worth noting that it has already been established that the src and dst are the same size.
if (w == width && h == height)
{
memoryManager.Write(dstGpuVa, srcSpan);
}
else
{
using WritableRegion dstRegion = memoryManager.GetWritableRegion(dstGpuVa, srcSize, true);
Span<byte> dstSpan = dstRegion.Memory.Span;
if (src.LinearLayout)
{
int stride = src.Stride;
int offset = 0;
int lineSize = width * bpp;
for (int y = 0; y < height; y++)
{
srcSpan.Slice(offset, lineSize).CopyTo(dstSpan.Slice(offset));
offset += stride;
}
}
else
{
// Copy with the block linear layout in mind.
// Recreate the offset calculate with bpp 1 for copy.
int stride = w * bpp;
srcCalculator = new OffsetCalculator(
stride,
h,
0,
false,
src.MemoryLayout.UnpackGobBlocksInY(),
src.MemoryLayout.UnpackGobBlocksInZ(),
1);
int strideTrunc = BitUtils.AlignDown(stride, 16);
ReadOnlySpan<Vector128<byte>> srcVec = MemoryMarshal.Cast<byte, Vector128<byte>>(srcSpan);
Span<Vector128<byte>> dstVec = MemoryMarshal.Cast<byte, Vector128<byte>>(dstSpan);
for (int y = 0; y < h; y++)
{
int x = 0;
srcCalculator.SetY(y);
for (; x < strideTrunc; x += 16)
{
int offset = srcCalculator.GetOffset(x) >> 4;
dstVec[offset] = srcVec[offset];
}
for (; x < stride; x++)
{
int offset = srcCalculator.GetOffset(x);
dstSpan[offset] = srcSpan[offset];
}
}
}
}
}
/// <summary>
/// Performs the blit operation, triggered by the register write.
/// </summary>
/// <param name="argument">Method call argument</param>
private void PixelsFromMemorySrcY0Int(int argument)
{
var memoryManager = _channel.MemoryManager;
var dstCopyTexture = Unsafe.As<uint, TwodTexture>(ref _state.State.SetDstFormat);
var srcCopyTexture = Unsafe.As<uint, TwodTexture>(ref _state.State.SetSrcFormat);
long srcX = ((long)_state.State.SetPixelsFromMemorySrcX0Int << 32) | (long)(ulong)_state.State.SetPixelsFromMemorySrcX0Frac;
long srcY = ((long)_state.State.PixelsFromMemorySrcY0Int << 32) | (long)(ulong)_state.State.SetPixelsFromMemorySrcY0Frac;
long duDx = ((long)_state.State.SetPixelsFromMemoryDuDxInt << 32) | (long)(ulong)_state.State.SetPixelsFromMemoryDuDxFrac;
long dvDy = ((long)_state.State.SetPixelsFromMemoryDvDyInt << 32) | (long)(ulong)_state.State.SetPixelsFromMemoryDvDyFrac;
bool originCorner = _state.State.SetPixelsFromMemorySampleModeOrigin == SetPixelsFromMemorySampleModeOrigin.Corner;
if (originCorner)
{
// If the origin is corner, it is assumed that the guest API
// is manually centering the origin by adding a offset to the
// source region X/Y coordinates.
// Here we attempt to remove such offset to ensure we have the correct region.
// The offset is calculated as FactorXY / 2.0, where FactorXY = SrcXY / DstXY,
// so we do the same here by dividing the fixed point value by 2, while
// throwing away the fractional part to avoid rounding errors.
srcX -= (duDx >> 33) << 32;
srcY -= (dvDy >> 33) << 32;
}
int srcX1 = (int)(srcX >> 32);
int srcY1 = (int)(srcY >> 32);
int srcX2 = srcX1 + (int)((duDx * _state.State.SetPixelsFromMemoryDstWidth + uint.MaxValue) >> 32);
int srcY2 = srcY1 + (int)((dvDy * _state.State.SetPixelsFromMemoryDstHeight + uint.MaxValue) >> 32);
int dstX1 = (int)_state.State.SetPixelsFromMemoryDstX0;
int dstY1 = (int)_state.State.SetPixelsFromMemoryDstY0;
int dstX2 = dstX1 + (int)_state.State.SetPixelsFromMemoryDstWidth;
int dstY2 = dstY1 + (int)_state.State.SetPixelsFromMemoryDstHeight;
// The source and destination textures should at least be as big as the region being requested.
// The hints will only resize within alignment constraints, so out of bound copies won't resize in most cases.
var srcHint = new Size(srcX2, srcY2, 1);
var dstHint = new Size(dstX2, dstY2, 1);
var srcCopyTextureFormat = srcCopyTexture.Format.Convert();
int srcWidthAligned = srcCopyTexture.Stride / srcCopyTextureFormat.BytesPerPixel;
ulong offset = 0;
// For an out of bounds copy, we must ensure that the copy wraps to the next line,
// so for a copy from a 64x64 texture, in the region [32, 96[, there are 32 pixels that are
// outside the bounds of the texture. We fill the destination with the first 32 pixels
// of the next line on the source texture.
// This can be done by simply adding an offset to the texture address, so that the initial
// gap is skipped and the copy is inside bounds again.
// This is required by the proprietary guest OpenGL driver.
if (srcCopyTexture.LinearLayout && srcCopyTexture.Width == srcX2 && srcX2 > srcWidthAligned && srcX1 > 0)
{
offset = (ulong)(srcX1 * srcCopyTextureFormat.BytesPerPixel);
srcCopyTexture.Width -= srcX1;
srcX2 -= srcX1;
srcX1 = 0;
}
FormatInfo dstCopyTextureFormat = dstCopyTexture.Format.Convert();
bool canDirectCopy = GraphicsConfig.Fast2DCopy &&
srcX2 == dstX2 && srcY2 == dstY2 &&
IsDataCompatible(srcCopyTexture, dstCopyTexture, srcCopyTextureFormat, dstCopyTextureFormat) &&
IsCopyRegionComplete(srcCopyTexture, srcCopyTextureFormat, srcX1, srcY1, srcX2, srcY2) &&
IsCopyRegionComplete(dstCopyTexture, dstCopyTextureFormat, dstX1, dstY1, dstX2, dstY2);
var srcTexture = memoryManager.Physical.TextureCache.FindOrCreateTexture(
memoryManager,
srcCopyTexture,
offset,
srcCopyTextureFormat,
!canDirectCopy,
false,
srcHint);
if (srcTexture == null)
{
if (canDirectCopy)
{
// Directly copy the data on CPU.
UnscaledFullCopy(srcCopyTexture, dstCopyTexture, srcX2, srcY2, srcCopyTextureFormat.BytesPerPixel);
}
return;
}
memoryManager.Physical.TextureCache.Lift(srcTexture);
// When the source texture that was found has a depth format,
// we must enforce the target texture also has a depth format,
// as copies between depth and color formats are not allowed.
if (srcTexture.Format.IsDepthOrStencil())
{
dstCopyTextureFormat = srcTexture.Info.FormatInfo;
}
else
{
dstCopyTextureFormat = dstCopyTexture.Format.Convert();
}
var dstTexture = memoryManager.Physical.TextureCache.FindOrCreateTexture(
memoryManager,
dstCopyTexture,
0,
dstCopyTextureFormat,
true,
srcTexture.ScaleMode == TextureScaleMode.Scaled,
dstHint);
if (dstTexture == null)
{
return;
}
float scale = srcTexture.ScaleFactor;
float dstScale = dstTexture.ScaleFactor;
Extents2D srcRegion = new Extents2D(
(int)Math.Ceiling(scale * (srcX1 / srcTexture.Info.SamplesInX)),
(int)Math.Ceiling(scale * (srcY1 / srcTexture.Info.SamplesInY)),
(int)Math.Ceiling(scale * (srcX2 / srcTexture.Info.SamplesInX)),
(int)Math.Ceiling(scale * (srcY2 / srcTexture.Info.SamplesInY)));
Extents2D dstRegion = new Extents2D(
(int)Math.Ceiling(dstScale * (dstX1 / dstTexture.Info.SamplesInX)),
(int)Math.Ceiling(dstScale * (dstY1 / dstTexture.Info.SamplesInY)),
(int)Math.Ceiling(dstScale * (dstX2 / dstTexture.Info.SamplesInX)),
(int)Math.Ceiling(dstScale * (dstY2 / dstTexture.Info.SamplesInY)));
bool linearFilter = _state.State.SetPixelsFromMemorySampleModeFilter == SetPixelsFromMemorySampleModeFilter.Bilinear;
srcTexture.HostTexture.CopyTo(dstTexture.HostTexture, srcRegion, dstRegion, linearFilter);
dstTexture.SignalModified();
}
}
}