0
0
Fork 0
mirror of https://github.com/ryujinx-mirror/ryujinx.git synced 2024-12-23 06:05:45 +00:00

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>
This commit is contained in:
riperiperi 2022-02-22 14:21:29 +00:00 committed by GitHub
parent dc063eac83
commit c9c65af59e
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 220 additions and 7 deletions

View file

@ -1,11 +1,15 @@
using Ryujinx.Graphics.Device; using Ryujinx.Common;
using Ryujinx.Graphics.Device;
using Ryujinx.Graphics.GAL; using Ryujinx.Graphics.GAL;
using Ryujinx.Graphics.Gpu.Engine.Types; using Ryujinx.Graphics.Gpu.Engine.Types;
using Ryujinx.Graphics.Gpu.Image; using Ryujinx.Graphics.Gpu.Image;
using Ryujinx.Graphics.Texture; using Ryujinx.Graphics.Texture;
using Ryujinx.Memory;
using System; using System;
using System.Collections.Generic; using System.Collections.Generic;
using System.Runtime.CompilerServices; using System.Runtime.CompilerServices;
using System.Runtime.InteropServices;
using System.Runtime.Intrinsics;
namespace Ryujinx.Graphics.Gpu.Engine.Twod namespace Ryujinx.Graphics.Gpu.Engine.Twod
{ {
@ -44,6 +48,180 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
/// <param name="data">Data to be written</param> /// <param name="data">Data to be written</param>
public void Write(int offset, int data) => _state.Write(offset, data); 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> /// <summary>
/// Performs the blit operation, triggered by the register write. /// Performs the blit operation, triggered by the register write.
/// </summary> /// </summary>
@ -114,16 +292,31 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
srcX1 = 0; 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( var srcTexture = memoryManager.Physical.TextureCache.FindOrCreateTexture(
memoryManager, memoryManager,
srcCopyTexture, srcCopyTexture,
offset, offset,
srcCopyTextureFormat, srcCopyTextureFormat,
!canDirectCopy,
false, false,
srcHint); srcHint);
if (srcTexture == null) if (srcTexture == null)
{ {
if (canDirectCopy)
{
// Directly copy the data on CPU.
UnscaledFullCopy(srcCopyTexture, dstCopyTexture, srcX2, srcY2, srcCopyTextureFormat.BytesPerPixel);
}
return; return;
} }
@ -132,7 +325,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
// When the source texture that was found has a depth format, // When the source texture that was found has a depth format,
// we must enforce the target texture also 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. // as copies between depth and color formats are not allowed.
FormatInfo dstCopyTextureFormat;
if (srcTexture.Format.IsDepthOrStencil()) if (srcTexture.Format.IsDepthOrStencil())
{ {
@ -148,6 +340,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Twod
dstCopyTexture, dstCopyTexture,
0, 0,
dstCopyTextureFormat, dstCopyTextureFormat,
true,
srcTexture.ScaleMode == TextureScaleMode.Scaled, srcTexture.ScaleMode == TextureScaleMode.Scaled,
dstHint); dstHint);

View file

@ -28,6 +28,14 @@ namespace Ryujinx.Graphics.Gpu
/// </summary> /// </summary>
public static bool FastGpuTime = true; public static bool FastGpuTime = true;
/// <summary>
/// Enables or disables fast 2d engine texture copies entirely on CPU when possible.
/// Reduces stuttering and # of textures in games that copy textures around for streaming,
/// as textures will not need to be created for the copy, and the data does not need to be
/// flushed from GPU.
/// </summary>
public static bool Fast2DCopy = true;
/// <summary> /// <summary>
/// Enables or disables the Just-in-Time compiler for GPU Macro code. /// Enables or disables the Just-in-Time compiler for GPU Macro code.
/// </summary> /// </summary>

View file

@ -194,6 +194,7 @@ namespace Ryujinx.Graphics.Gpu.Image
TwodTexture copyTexture, TwodTexture copyTexture,
ulong offset, ulong offset,
FormatInfo formatInfo, FormatInfo formatInfo,
bool shouldCreate,
bool preferScaling = true, bool preferScaling = true,
Size? sizeHint = null) Size? sizeHint = null)
{ {
@ -234,6 +235,11 @@ namespace Ryujinx.Graphics.Gpu.Image
flags |= TextureSearchFlags.WithUpscale; flags |= TextureSearchFlags.WithUpscale;
} }
if (!shouldCreate)
{
flags |= TextureSearchFlags.NoCreate;
}
Texture texture = FindOrCreateTexture(memoryManager, flags, info, 0, sizeHint); Texture texture = FindOrCreateTexture(memoryManager, flags, info, 0, sizeHint);
texture?.SynchronizeMemory(); texture?.SynchronizeMemory();
@ -480,6 +486,10 @@ namespace Ryujinx.Graphics.Gpu.Image
return texture; return texture;
} }
else if (flags.HasFlag(TextureSearchFlags.NoCreate))
{
return null;
}
// Calculate texture sizes, used to find all overlapping textures. // Calculate texture sizes, used to find all overlapping textures.
SizeInfo sizeInfo = info.CalculateSizeInfo(layerSize); SizeInfo sizeInfo = info.CalculateSizeInfo(layerSize);

View file

@ -12,6 +12,7 @@ namespace Ryujinx.Graphics.Gpu.Image
Strict = 1 << 0, Strict = 1 << 0,
ForSampler = 1 << 1, ForSampler = 1 << 1,
ForCopy = 1 << 2, ForCopy = 1 << 2,
WithUpscale = 1 << 3 WithUpscale = 1 << 3,
NoCreate = 1 << 4
} }
} }

View file

@ -154,14 +154,15 @@ namespace Ryujinx.Graphics.Gpu.Memory
/// <summary> /// <summary>
/// Gets a writable region from GPU mapped memory. /// Gets a writable region from GPU mapped memory.
/// </summary> /// </summary>
/// <param name="address">Start address of the range</param> /// <param name="va">Start address of the range</param>
/// <param name="size">Size in bytes to be range</param> /// <param name="size">Size in bytes to be range</param>
/// <param name="tracked">True if write tracking is triggered on the span</param>
/// <returns>A writable region with the data at the specified memory location</returns> /// <returns>A writable region with the data at the specified memory location</returns>
public WritableRegion GetWritableRegion(ulong va, int size) public WritableRegion GetWritableRegion(ulong va, int size, bool tracked = false)
{ {
if (IsContiguous(va, size)) if (IsContiguous(va, size))
{ {
return Physical.GetWritableRegion(Translate(va), size); return Physical.GetWritableRegion(Translate(va), size, tracked);
} }
else else
{ {
@ -169,7 +170,7 @@ namespace Ryujinx.Graphics.Gpu.Memory
GetSpan(va, size).CopyTo(memory.Span); GetSpan(va, size).CopyTo(memory.Span);
return new WritableRegion(this, va, memory); return new WritableRegion(this, va, memory, tracked);
} }
} }