GPU: Pre-emptively flush textures that are flushed often (to imported memory when available) (#4711)

* WIP texture pre-flush

Improve performance of TextureView GetData to buffer

Fix copy/sync ordering

Fix minor bug

Make this actually work

WIP host mapping stuff

* Fix usage flags

* message

* Cleanup 1

* Fix rebase

* Fix

* Improve pre-flush rules

* Fix pre-flush

* A lot of cleanup

* Use the host memory bits

* Select the correct memory type

* Cleanup TextureGroupHandle

* Missing comment

* Remove debugging logs

* Revert BufferHandle _value access modifier

* One interrupt action at a time.

* Support D32S8 to D24S8 conversion, safeguards

* Interrupt cannot happen in sync handle's lock

Waitable needs to be checked twice now, but this should stop it from deadlocking.

* Remove unused using

* Address some feedback

* Address feedback

* Address more feedback

* Address more feedback

* Improve sync rules

Should allow for faster sync in some cases.

src/Ryujinx.Graphics.OpenGL/Buffer.cs

@@ -42,6 +42,20 @@ namespace Ryujinx.Graphics.OpenGL
             return Handle.FromInt32<BufferHandle>(handle);
         }
 
+        public static BufferHandle CreatePersistent(int size)
+        {
+            int handle = GL.GenBuffer();
+
+            GL.BindBuffer(BufferTarget.CopyWriteBuffer, handle);
+            GL.BufferStorage(BufferTarget.CopyWriteBuffer, size, IntPtr.Zero,
+                BufferStorageFlags.MapPersistentBit |
+                BufferStorageFlags.MapCoherentBit |
+                BufferStorageFlags.ClientStorageBit |
+                BufferStorageFlags.MapReadBit);
+
+            return Handle.FromInt32<BufferHandle>(handle);
+        }
+
         public static void Copy(BufferHandle source, BufferHandle destination, int srcOffset, int dstOffset, int size)
         {
             GL.BindBuffer(BufferTarget.CopyReadBuffer, source.ToInt32());
@@ -60,7 +74,11 @@ namespace Ryujinx.Graphics.OpenGL
             // Data in the persistent buffer and host array is guaranteed to be available
             // until the next time the host thread requests data.
 
-            if (HwCapabilities.UsePersistentBufferForFlush)
+            if (renderer.PersistentBuffers.TryGet(buffer, out IntPtr ptr))
+            {
+                return new PinnedSpan<byte>(IntPtr.Add(ptr, offset).ToPointer(), size);
+            }
+            else if (HwCapabilities.UsePersistentBufferForFlush)
             {
                 return PinnedSpan<byte>.UnsafeFromSpan(renderer.PersistentBuffers.Default.GetBufferData(buffer, offset, size));
             }

src/Ryujinx.Graphics.OpenGL/Image/TextureBuffer.cs

@@ -49,6 +49,11 @@ namespace Ryujinx.Graphics.OpenGL.Image
             return GetData();
         }
 
+        public void CopyTo(BufferRange range, int layer, int level, int stride)
+        {
+            throw new NotImplementedException();
+        }
+
         public void SetData(SpanOrArray<byte> data)
         {
             var dataSpan = data.AsSpan();

src/Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -3,6 +3,7 @@ using Ryujinx.Common;
 using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using System;
+using System.Diagnostics;
 
 namespace Ryujinx.Graphics.OpenGL.Image
 {
@@ -287,6 +288,26 @@ namespace Ryujinx.Graphics.OpenGL.Image
             }
         }
 
+        public void CopyTo(BufferRange range, int layer, int level, int stride)
+        {
+            if (stride != 0 && stride != BitUtils.AlignUp(Info.Width * Info.BytesPerPixel, 4))
+            {
+                throw new NotSupportedException("Stride conversion for texture copy to buffer not supported.");
+            }
+
+            GL.BindBuffer(BufferTarget.PixelPackBuffer, range.Handle.ToInt32());
+
+            FormatInfo format = FormatTable.GetFormatInfo(Info.Format);
+            if (format.PixelFormat == PixelFormat.DepthStencil)
+            {
+                throw new InvalidOperationException("DepthStencil copy to buffer is not supported for layer/level > 0.");
+            }
+
+            int offset = WriteToPbo2D(range.Offset, layer, level);
+
+            Debug.Assert(offset == 0);
+        }
+
         public void WriteToPbo(int offset, bool forceBgra)
         {
             WriteTo(IntPtr.Zero + offset, forceBgra);

src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs

@@ -58,10 +58,31 @@ namespace Ryujinx.Graphics.OpenGL
         }
 
         public BufferHandle CreateBuffer(int size, BufferHandle storageHint)
+        {
+            return CreateBuffer(size, GAL.BufferAccess.Default);
+        }
+
+        public BufferHandle CreateBuffer(int size, GAL.BufferAccess access)
         {
             BufferCount++;
 
-            return Buffer.Create(size);
+            if (access == GAL.BufferAccess.FlushPersistent)
+            {
+                BufferHandle handle = Buffer.CreatePersistent(size);
+
+                PersistentBuffers.Map(handle, size);
+
+                return handle;
+            }
+            else
+            {
+                return Buffer.Create(size);
+            }
+        }
+
+        public BufferHandle CreateBuffer(nint pointer, int size)
+        {
+            throw new NotSupportedException();
         }
 
         public IProgram CreateProgram(ShaderSource[] shaders, ShaderInfo info)
@@ -88,6 +109,8 @@ namespace Ryujinx.Graphics.OpenGL
 
         public void DeleteBuffer(BufferHandle buffer)
         {
+            PersistentBuffers.Unmap(buffer);
+
             Buffer.Delete(buffer);
         }
 
@@ -272,5 +295,10 @@ namespace Ryujinx.Graphics.OpenGL
         {
             ScreenCaptured?.Invoke(this, bitmap);
         }
+
+        public bool PrepareHostMapping(nint address, ulong size)
+        {
+            return false;
+        }
     }
 }

src/Ryujinx.Graphics.OpenGL/PersistentBuffers.cs

@@ -1,8 +1,9 @@
-using OpenTK.Graphics.OpenGL;
+using OpenTK.Graphics.OpenGL;
 using Ryujinx.Common.Logging;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.OpenGL.Image;
 using System;
+using System.Collections.Generic;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 
@@ -13,6 +14,8 @@ namespace Ryujinx.Graphics.OpenGL
         private PersistentBuffer _main = new PersistentBuffer();
         private PersistentBuffer _background = new PersistentBuffer();
 
+        private Dictionary<BufferHandle, IntPtr> _maps = new Dictionary<BufferHandle, IntPtr>();
+
         public PersistentBuffer Default => BackgroundContextWorker.InBackground ? _background : _main;
 
         public void Dispose()
@@ -20,6 +23,30 @@ namespace Ryujinx.Graphics.OpenGL
             _main?.Dispose();
             _background?.Dispose();
         }
+
+        public void Map(BufferHandle handle, int size)
+        {
+            GL.BindBuffer(BufferTarget.CopyWriteBuffer, handle.ToInt32());
+            IntPtr ptr = GL.MapBufferRange(BufferTarget.CopyWriteBuffer, IntPtr.Zero, size, BufferAccessMask.MapReadBit | BufferAccessMask.MapPersistentBit);
+
+            _maps[handle] = ptr;
+        }
+
+        public void Unmap(BufferHandle handle)
+        {
+            if (_maps.ContainsKey(handle))
+            {
+                GL.BindBuffer(BufferTarget.CopyWriteBuffer, handle.ToInt32());
+                GL.UnmapBuffer(BufferTarget.CopyWriteBuffer);
+
+                _maps.Remove(handle);
+            }
+        }
+
+        public bool TryGet(BufferHandle handle, out IntPtr ptr)
+        {
+            return _maps.TryGetValue(handle, out ptr);
+        }
     }
 
     class PersistentBuffer : IDisposable