Only use persistent buffers to flush on NVIDIA and Windows+AMD (#2489)

It seems like this method of flushing data is much slower on Mesa drivers, and slightly slower on Intel Windows. Have not tested Intel Mesa, but I'm assuming it is the same as AMD. This also adds vendor detection for AMD on Unix, which counted as "Unknown" before.
sxpisces · Jul 18, 2021 · 10e17ab · 10e17ab
1 parent b8ad676
commit 10e17ab
Show file tree

Hide file tree

Showing 4 changed files with 48 additions and 14 deletions.
diff --git a/Ryujinx.Graphics.OpenGL/HwCapabilities.cs b/Ryujinx.Graphics.OpenGL/HwCapabilities.cs
@@ -20,18 +20,24 @@ static class HwCapabilities
         public enum GpuVendor
         {
             Unknown,
-            Amd,
+            AmdWindows,
+            AmdUnix,
             IntelWindows,
             IntelUnix,
             Nvidia
         }
 
         private static readonly Lazy<GpuVendor> _gpuVendor = new Lazy<GpuVendor>(GetGpuVendor);
 
+        private static bool _isAMD   => _gpuVendor.Value == GpuVendor.AmdWindows || _gpuVendor.Value == GpuVendor.AmdUnix;
+        private static bool _isIntel => _gpuVendor.Value == GpuVendor.IntelWindows || _gpuVendor.Value == GpuVendor.IntelUnix;
+
         public static GpuVendor Vendor => _gpuVendor.Value;
 
         private static Lazy<float> _maxSupportedAnisotropy = new Lazy<float>(GL.GetFloat((GetPName)All.MaxTextureMaxAnisotropy));
 
+        public static bool UsePersistentBufferForFlush       => _gpuVendor.Value == GpuVendor.AmdWindows || _gpuVendor.Value == GpuVendor.Nvidia;
+
         public static bool SupportsAstcCompression           => _supportsAstcCompression.Value;
         public static bool SupportsImageLoadFormatted        => _supportsImageLoadFormatted.Value;
         public static bool SupportsParallelShaderCompile     => _supportsParallelShaderCompile.Value;
@@ -41,9 +47,9 @@ public enum GpuVendor
         public static bool SupportsTextureShadowLod          => _supportsTextureShadowLod.Value;
         public static bool SupportsViewportSwizzle           => _supportsViewportSwizzle.Value;
 
-        public static bool SupportsMismatchingViewFormat    => _gpuVendor.Value != GpuVendor.Amd && _gpuVendor.Value != GpuVendor.IntelWindows;
+        public static bool SupportsMismatchingViewFormat    => _gpuVendor.Value != GpuVendor.AmdWindows && _gpuVendor.Value != GpuVendor.IntelWindows;
         public static bool SupportsNonConstantTextureOffset => _gpuVendor.Value == GpuVendor.Nvidia;
-        public static bool RequiresSyncFlush                => _gpuVendor.Value == GpuVendor.Amd || _gpuVendor.Value == GpuVendor.IntelWindows || _gpuVendor.Value == GpuVendor.IntelUnix;
+        public static bool RequiresSyncFlush                => _gpuVendor.Value == GpuVendor.AmdWindows || _isIntel;
 
         public static int MaximumComputeSharedMemorySize => _maximumComputeSharedMemorySize.Value;
         public static int StorageBufferOffsetAlignment   => _storageBufferOffsetAlignment.Value;
@@ -86,7 +92,11 @@ private static GpuVendor GetGpuVendor()
             }
             else if (vendor == "ati technologies inc." || vendor == "advanced micro devices, inc.")
             {
-                return GpuVendor.Amd;
+                return GpuVendor.AmdWindows;
+            }
+            else if (vendor == "amd" || vendor == "x.org")
+            {
+                return GpuVendor.AmdUnix;
             }
             else
             {

diff --git a/Ryujinx.Graphics.OpenGL/Image/TextureView.cs b/Ryujinx.Graphics.OpenGL/Image/TextureView.cs
@@ -121,7 +121,31 @@ public void CopyTo(ITexture destination, Extents2D srcRegion, Extents2D dstRegio
 
         public byte[] GetData()
         {
-            return _renderer.PersistentBuffers.Default.GetTextureData(this);
+            int size = 0;
+
+            for (int level = 0; level < Info.Levels; level++)
+            {
+                size += Info.GetMipSize(level);
+            }
+
+            if (HwCapabilities.UsePersistentBufferForFlush)
+            {
+                return _renderer.PersistentBuffers.Default.GetTextureData(this, size);
+            }
+            else
+            {
+                byte[] data = new byte[size];
+
+                unsafe
+                {
+                    fixed (byte* ptr = data)
+                    {
+                        WriteTo((IntPtr)ptr);
+                    }
+                }
+
+                return data;
+            }
         }
 
         public void WriteToPbo(int offset, bool forceBgra)

diff --git a/Ryujinx.Graphics.OpenGL/PersistentBuffers.cs b/Ryujinx.Graphics.OpenGL/PersistentBuffers.cs
@@ -63,15 +63,8 @@ private void Sync()
             GL.DeleteSync(sync);
         }
 
-        public byte[] GetTextureData(TextureView view)
+        public byte[] GetTextureData(TextureView view, int size)
         {
-            int size = 0;
-
-            for (int level = 0; level < view.Info.Levels; level++)
-            {
-                size += view.Info.GetMipSize(level);
-            }
-
             EnsureBuffer(size);
 
             GL.BindBuffer(BufferTarget.PixelPackBuffer, _copyBufferHandle);

diff --git a/Ryujinx.Graphics.OpenGL/Renderer.cs b/Ryujinx.Graphics.OpenGL/Renderer.cs
@@ -93,7 +93,14 @@ public void DeleteBuffer(BufferHandle buffer)
 
         public byte[] GetBufferData(BufferHandle buffer, int offset, int size)
         {
-            return PersistentBuffers.Default.GetBufferData(buffer, offset, size);
+            if (HwCapabilities.UsePersistentBufferForFlush)
+            {
+                return PersistentBuffers.Default.GetBufferData(buffer, offset, size);
+            }
+            else
+            {
+                return Buffer.GetData(buffer, offset, size);
+            }
         }
 
         public Capabilities GetCapabilities()