GPU: Pass SpanOrArray for Texture SetData to avoid copy (#3745)

* GPU: Pass SpanOrArray for Texture SetData to avoid copy

Texture data is often converted before upload, meaning that an array was allocated to perform the conversion into. However, the backend SetData methods were being passed a Span of that data, and the Multithreaded layer does `ToArray()` on it so that it can be stored for later! This method can't extract the original array, so it creates a copy.

This PR changes the type passed for textures to a new ref struct called SpanOrArray, which is backed by either a ReadOnlySpan or an array. The benefit here is that we can have a ToArray method that doesn't copy if it is originally backed by an array.

This will also avoid a copy when running the ASTC decoder.

On NieR this was taking 38% of texture upload time, which it does a _lot_ of when you move between areas, so there should be a 1.6x performance boost when strictly uploading textures. No doubt this will also improve texture streaming performance in UE4 games, and maybe a small reduction with video playback.

From the numbers, it's probably possible to improve the upload rate by a further 1.6x by performing layout conversion on GPU. I'm not sure if we could improve it further than that - multithreading conversion on CPU would probably result in memory bottleneck.

This doesn't extend to buffers, since we don't convert their data on the GPU emulator side.

* Remove implicit cast to array.

Ryujinx.Common/Memory/SpanOrArray.cs

@@ -0,0 +1,89 @@
+using System;
+
+namespace Ryujinx.Common.Memory
+{
+    /// <summary>
+    /// A struct that can represent both a Span and Array.
+    /// This is useful to keep the Array representation when possible to avoid copies.
+    /// </summary>
+    /// <typeparam name="T">Element Type</typeparam>
+    public ref struct SpanOrArray<T> where T : unmanaged
+    {
+        public readonly T[] Array;
+        public readonly ReadOnlySpan<T> Span;
+
+        /// <summary>
+        /// Create a new SpanOrArray from an array.
+        /// </summary>
+        /// <param name="array">Array to store</param>
+        public SpanOrArray(T[] array)
+        {
+            Array = array;
+            Span = ReadOnlySpan<T>.Empty;
+        }
+
+        /// <summary>
+        /// Create a new SpanOrArray from a readonly span.
+        /// </summary>
+        /// <param name="array">Span to store</param>
+        public SpanOrArray(ReadOnlySpan<T> span)
+        {
+            Array = null;
+            Span = span;
+        }
+
+        /// <summary>
+        /// Return the contained array, or convert the span if necessary.
+        /// </summary>
+        /// <returns>An array containing the data</returns>
+        public T[] ToArray()
+        {
+            return Array ?? Span.ToArray();
+        }
+
+        /// <summary>
+        /// Return a ReadOnlySpan from either the array or ReadOnlySpan.
+        /// </summary>
+        /// <returns>A ReadOnlySpan containing the data</returns>
+        public ReadOnlySpan<T> AsSpan()
+        {
+            return Array ?? Span;
+        }
+
+        /// <summary>
+        /// Cast an array to a SpanOrArray.
+        /// </summary>
+        /// <param name="array">Source array</param>
+        public static implicit operator SpanOrArray<T>(T[] array)
+        {
+            return new SpanOrArray<T>(array);
+        }
+
+        /// <summary>
+        /// Cast a ReadOnlySpan to a SpanOrArray.
+        /// </summary>
+        /// <param name="span">Source ReadOnlySpan</param>
+        public static implicit operator SpanOrArray<T>(ReadOnlySpan<T> span)
+        {
+            return new SpanOrArray<T>(span);
+        }
+
+        /// <summary>
+        /// Cast a Span to a SpanOrArray.
+        /// </summary>
+        /// <param name="span">Source Span</param>
+        public static implicit operator SpanOrArray<T>(Span<T> span)
+        {
+            return new SpanOrArray<T>(span);
+        }
+
+        /// <summary>
+        /// Cast a SpanOrArray to a ReadOnlySpan
+        /// </summary>
+        /// <param name="spanOrArray">Source SpanOrArray</param>
+        public static implicit operator ReadOnlySpan<T>(SpanOrArray<T> spanOrArray)
+        {
+            return spanOrArray.AsSpan();
+        }
+    }
+}

Ryujinx.Graphics.GAL/ITexture.cs

@@ -1,3 +1,4 @@
+using Ryujinx.Common.Memory;
 using System;
 
 namespace Ryujinx.Graphics.GAL
@@ -17,9 +18,9 @@ namespace Ryujinx.Graphics.GAL
         ReadOnlySpan<byte> GetData();
         ReadOnlySpan<byte> GetData(int layer, int level);
 
-        void SetData(ReadOnlySpan<byte> data);
+        void SetData(SpanOrArray<byte> data);
-        void SetData(ReadOnlySpan<byte> data, int layer, int level);
+        void SetData(SpanOrArray<byte> data, int layer, int level);
-        void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region);
+        void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region);
         void SetStorage(BufferRange buffer);
         void Release();
     }

Ryujinx.Graphics.GAL/Multithreading/Resources/ThreadedTexture.cs

@@ -1,4 +1,5 @@
-using Ryujinx.Graphics.GAL.Multithreading.Commands.Texture;
+using Ryujinx.Common.Memory;
+using Ryujinx.Graphics.GAL.Multithreading.Commands.Texture;
 using Ryujinx.Graphics.GAL.Multithreading.Model;
 using System;
 
@@ -107,19 +108,19 @@ namespace Ryujinx.Graphics.GAL.Multithreading.Resources
             }
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
             _renderer.New<TextureSetDataCommand>().Set(Ref(this), Ref(data.ToArray()));
             _renderer.QueueCommand();
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             _renderer.New<TextureSetDataSliceCommand>().Set(Ref(this), Ref(data.ToArray()), layer, level);
             _renderer.QueueCommand();
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
             _renderer.New<TextureSetDataSliceRegionCommand>().Set(Ref(this), Ref(data.ToArray()), layer, level, region);
             _renderer.QueueCommand();

Ryujinx.Graphics.Gpu/Engine/Dma/DmaClass.cs

@@ -229,7 +229,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Dma
 
                     if (target != null)
                     {
-                        ReadOnlySpan<byte> data;
+                        byte[] data;
                         if (srcLinear)
                         {
                             data = LayoutConverter.ConvertLinearStridedToLinear(

Ryujinx.Graphics.Gpu/Image/Texture.cs

@@ -1,5 +1,6 @@
 using Ryujinx.Common;
 using Ryujinx.Common.Logging;
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Texture;
@@ -720,9 +721,9 @@ namespace Ryujinx.Graphics.Gpu.Image
                 }
             }
 
-            data = ConvertToHostCompatibleFormat(data);
+            SpanOrArray<byte> result = ConvertToHostCompatibleFormat(data);
 
-            HostTexture.SetData(data);
+            HostTexture.SetData(result);
 
             _hasData = true;
         }
@@ -731,7 +732,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// Uploads new texture data to the host GPU.
         /// </summary>
         /// <param name="data">New data</param>
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
             BlacklistScale();
 
@@ -750,7 +751,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="data">New data</param>
         /// <param name="layer">Target layer</param>
         /// <param name="level">Target level</param>
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             BlacklistScale();
 
@@ -786,7 +787,7 @@ namespace Ryujinx.Graphics.Gpu.Image
         /// <param name="level">Mip level to convert</param>
         /// <param name="single">True to convert a single slice</param>
         /// <returns>Converted data</returns>
-        public ReadOnlySpan<byte> ConvertToHostCompatibleFormat(ReadOnlySpan<byte> data, int level = 0, bool single = false)
+        public SpanOrArray<byte> ConvertToHostCompatibleFormat(ReadOnlySpan<byte> data, int level = 0, bool single = false)
         {
             int width = Info.Width;
             int height = Info.Height;
@@ -799,9 +800,11 @@ namespace Ryujinx.Graphics.Gpu.Image
             height = Math.Max(height >> level, 1);
             depth = Math.Max(depth >> level, 1);
 
+            SpanOrArray<byte> result;
+
             if (Info.IsLinear)
             {
-                data = LayoutConverter.ConvertLinearStridedToLinear(
+                result = LayoutConverter.ConvertLinearStridedToLinear(
                     width,
                     height,
                     Info.FormatInfo.BlockWidth,
@@ -813,7 +816,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             }
             else
             {
-                data = LayoutConverter.ConvertBlockLinearToLinear(
+                result = LayoutConverter.ConvertBlockLinearToLinear(
                     width,
                     height,
                     depth,
@@ -836,7 +839,7 @@ namespace Ryujinx.Graphics.Gpu.Image
             if (!_context.Capabilities.SupportsAstcCompression && Format.IsAstc())
             {
                 if (!AstcDecoder.TryDecodeToRgba8P(
-                    data.ToArray(),
+                    result.ToArray(),
                     Info.FormatInfo.BlockWidth,
                     Info.FormatInfo.BlockHeight,
                     width,
@@ -856,11 +859,11 @@ namespace Ryujinx.Graphics.Gpu.Image
                     decoded = BCnEncoder.EncodeBC7(decoded, width, height, depth, levels, layers);
                 }
 
-                data = decoded;
+                result = decoded;
             }
             else if (!_context.Capabilities.SupportsR4G4Format && Format == Format.R4G4Unorm)
             {
-                data = PixelConverter.ConvertR4G4ToR4G4B4A4(data);
+                result = PixelConverter.ConvertR4G4ToR4G4B4A4(result);
             }
             else if (!TextureCompatibility.HostSupportsBcFormat(Format, Target, _context.Capabilities))
             {
@@ -868,36 +871,36 @@ namespace Ryujinx.Graphics.Gpu.Image
                 {
                     case Format.Bc1RgbaSrgb:
                     case Format.Bc1RgbaUnorm:
-                        data = BCnDecoder.DecodeBC1(data, width, height, depth, levels, layers);
+                        result = BCnDecoder.DecodeBC1(result, width, height, depth, levels, layers);
                         break;
                     case Format.Bc2Srgb:
                     case Format.Bc2Unorm:
-                        data = BCnDecoder.DecodeBC2(data, width, height, depth, levels, layers);
+                        result = BCnDecoder.DecodeBC2(result, width, height, depth, levels, layers);
                         break;
                     case Format.Bc3Srgb:
                     case Format.Bc3Unorm:
-                        data = BCnDecoder.DecodeBC3(data, width, height, depth, levels, layers);
+                        result = BCnDecoder.DecodeBC3(result, width, height, depth, levels, layers);
                         break;
                     case Format.Bc4Snorm:
                     case Format.Bc4Unorm:
-                        data = BCnDecoder.DecodeBC4(data, width, height, depth, levels, layers, Format == Format.Bc4Snorm);
+                        result = BCnDecoder.DecodeBC4(result, width, height, depth, levels, layers, Format == Format.Bc4Snorm);
                         break;
                     case Format.Bc5Snorm:
                     case Format.Bc5Unorm:
-                        data = BCnDecoder.DecodeBC5(data, width, height, depth, levels, layers, Format == Format.Bc5Snorm);
+                        result = BCnDecoder.DecodeBC5(result, width, height, depth, levels, layers, Format == Format.Bc5Snorm);
                         break;
                     case Format.Bc6HSfloat:
                     case Format.Bc6HUfloat:
-                        data = BCnDecoder.DecodeBC6(data, width, height, depth, levels, layers, Format == Format.Bc6HSfloat);
+                        result = BCnDecoder.DecodeBC6(result, width, height, depth, levels, layers, Format == Format.Bc6HSfloat);
                         break;
                     case Format.Bc7Srgb:
                     case Format.Bc7Unorm:
-                        data = BCnDecoder.DecodeBC7(data, width, height, depth, levels, layers);
+                        result = BCnDecoder.DecodeBC7(result, width, height, depth, levels, layers);
                         break;
                 }
             }
 
-            return data;
+            return result;
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Image/TextureGroup.cs

@@ -1,4 +1,5 @@
-using Ryujinx.Cpu.Tracking;
+using Ryujinx.Common.Memory;
+using Ryujinx.Cpu.Tracking;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Texture;
@@ -348,9 +349,9 @@ namespace Ryujinx.Graphics.Gpu.Image
 
                             ReadOnlySpan<byte> data = _physicalMemory.GetSpan(Storage.Range.GetSlice((ulong)offset, (ulong)size));
 
-                            data = Storage.ConvertToHostCompatibleFormat(data, info.BaseLevel, true);
+                            SpanOrArray<byte> result = Storage.ConvertToHostCompatibleFormat(data, info.BaseLevel, true);
 
-                            Storage.SetData(data, info.BaseLayer, info.BaseLevel);
+                            Storage.SetData(result, info.BaseLayer, info.BaseLevel);
 
                             offsetIndex++;
                         }

Ryujinx.Graphics.OpenGL/Image/TextureBuffer.cs

@@ -1,4 +1,5 @@
 using OpenTK.Graphics.OpenGL;
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using System;
 
@@ -48,17 +49,19 @@ namespace Ryujinx.Graphics.OpenGL.Image
             return GetData();
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
-            Buffer.SetData(_buffer, _bufferOffset, data.Slice(0, Math.Min(data.Length, _bufferSize)));
+            var dataSpan = data.AsSpan();
+
+            Buffer.SetData(_buffer, _bufferOffset, dataSpan.Slice(0, Math.Min(dataSpan.Length, _bufferSize)));
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             throw new NotSupportedException();
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
             throw new NotSupportedException();
         }

Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -1,5 +1,6 @@
 using OpenTK.Graphics.OpenGL;
 using Ryujinx.Common;
+using Ryujinx.Common.Memory;
 using Ryujinx.Graphics.GAL;
 using System;
 
@@ -317,32 +318,36 @@ namespace Ryujinx.Graphics.OpenGL.Image
             }
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
+            var dataSpan = data.AsSpan();
+
             if (Format == Format.S8UintD24Unorm)
             {
-                data = FormatConverter.ConvertS8D24ToD24S8(data);
+                dataSpan = FormatConverter.ConvertS8D24ToD24S8(dataSpan);
             }
 
             unsafe
             {
-                fixed (byte* ptr = data)
+                fixed (byte* ptr = dataSpan)
                 {
-                    ReadFrom((IntPtr)ptr, data.Length);
+                    ReadFrom((IntPtr)ptr, dataSpan.Length);
                 }
             }
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
+            var dataSpan = data.AsSpan();
+
             if (Format == Format.S8UintD24Unorm)
             {
-                data = FormatConverter.ConvertS8D24ToD24S8(data);
+                dataSpan = FormatConverter.ConvertS8D24ToD24S8(dataSpan);
             }
 
             unsafe
             {
-                fixed (byte* ptr = data)
+                fixed (byte* ptr = dataSpan)
                 {
                     int width = Math.Max(Info.Width >> level, 1);
                     int height = Math.Max(Info.Height >> level, 1);
@@ -352,11 +357,13 @@ namespace Ryujinx.Graphics.OpenGL.Image
             }
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
+            var dataSpan = data.AsSpan();
+
             if (Format == Format.S8UintD24Unorm)
             {
-                data = FormatConverter.ConvertS8D24ToD24S8(data);
+                dataSpan = FormatConverter.ConvertS8D24ToD24S8(dataSpan);
             }
 
             int wInBlocks = BitUtils.DivRoundUp(region.Width, Info.BlockWidth);
@@ -364,7 +371,7 @@ namespace Ryujinx.Graphics.OpenGL.Image
 
             unsafe
             {
-                fixed (byte* ptr = data)
+                fixed (byte* ptr = dataSpan)
                 {
                     ReadFrom2D(
                         (IntPtr)ptr,

Ryujinx.Graphics.Texture/LayoutConverter.cs

@@ -93,7 +93,7 @@ namespace Ryujinx.Graphics.Texture
             };
         }
 
-        public static Span<byte> ConvertBlockLinearToLinear(
+        public static byte[] ConvertBlockLinearToLinear(
             int width,
             int height,
             int depth,
@@ -119,7 +119,7 @@ namespace Ryujinx.Graphics.Texture
                 blockHeight,
                 bytesPerPixel);
 
-            Span<byte> output = new byte[outSize];
+            byte[] output = new byte[outSize];
 
             int outOffs = 0;
 
@@ -246,7 +246,7 @@ namespace Ryujinx.Graphics.Texture
             return output;
         }
 
-        public static Span<byte> ConvertLinearStridedToLinear(
+        public static byte[] ConvertLinearStridedToLinear(
             int width,
             int height,
             int blockWidth,
@@ -262,14 +262,15 @@ namespace Ryujinx.Graphics.Texture
             int outStride = BitUtils.AlignUp(w * bytesPerPixel, HostStrideAlignment);
             lineSize = Math.Min(lineSize, outStride);
 
-            Span<byte> output = new byte[h * outStride];
+            byte[] output = new byte[h * outStride];
+            Span<byte> outSpan = output;
 
             int outOffs = 0;
             int inOffs = 0;
 
             for (int y = 0; y < h; y++)
             {
-                data.Slice(inOffs, lineSize).CopyTo(output.Slice(outOffs, lineSize));
+                data.Slice(inOffs, lineSize).CopyTo(outSpan.Slice(outOffs, lineSize));
 
                 inOffs += stride;
                 outOffs += outStride;

Ryujinx.Graphics.Vulkan/TextureBuffer.cs

@@ -1,4 +1,5 @@
-using Ryujinx.Graphics.GAL;
+using Ryujinx.Common.Memory;
+using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
@@ -90,17 +91,17 @@ namespace Ryujinx.Graphics.Vulkan
             _bufferView = null;
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
             _gd.SetBufferData(_bufferHandle, _offset, data);
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             throw new NotSupportedException();
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
             throw new NotSupportedException();
         }

Ryujinx.Graphics.Vulkan/TextureView.cs

@@ -1,4 +1,5 @@
-using Ryujinx.Graphics.GAL;
+using Ryujinx.Common.Memory;
+using Ryujinx.Graphics.GAL;
 using Silk.NET.Vulkan;
 using System;
 using System.Collections.Generic;
@@ -873,17 +874,17 @@ namespace Ryujinx.Graphics.Vulkan
             return GetDataFromBuffer(result, size, result);
         }
 
-        public void SetData(ReadOnlySpan<byte> data)
+        public void SetData(SpanOrArray<byte> data)
         {
             SetData(data, 0, 0, Info.GetLayers(), Info.Levels, singleSlice: false);
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level)
+        public void SetData(SpanOrArray<byte> data, int layer, int level)
         {
             SetData(data, layer, level, 1, 1, singleSlice: true);
         }
 
-        public void SetData(ReadOnlySpan<byte> data, int layer, int level, Rectangle<int> region)
+        public void SetData(SpanOrArray<byte> data, int layer, int level, Rectangle<int> region)
         {
             SetData(data, layer, level, 1, 1, singleSlice: true, region);
         }