Metal: Advanced Present (#6)

* Initial DrawTexture support & Advanced Present

* TODO: Get Scissors Working

* Chnage scissor state management

* Rebase problems…

* Rebase fixes again

* Update DrawTexture + Fix Topology

* Fix flipping

* Add clear action support

* Cleanup

src/Ryujinx.Graphics.Metal/Effects/IPostProcessingEffect.cs

@@ -0,0 +1,10 @@
+using System;
+
+namespace Ryujinx.Graphics.Metal.Effects
+{
+    internal interface IPostProcessingEffect : IDisposable
+    {
+        const int LocalGroupSize = 64;
+        Texture Run(Texture view, int width, int height);
+    }
+}

src/Ryujinx.Graphics.Metal/Effects/IScalingFilter.cs

@@ -0,0 +1,18 @@
+using Ryujinx.Graphics.GAL;
+using System;
+
+namespace Ryujinx.Graphics.Metal.Effects
+{
+    internal interface IScalingFilter : IDisposable
+    {
+        float Level { get; set; }
+        void Run(
+            Texture view,
+            Texture destinationTexture,
+            Format format,
+            int width,
+            int height,
+            Extents2D source,
+            Extents2D destination);
+    }
+}

src/Ryujinx.Graphics.Metal/EncoderState.cs

@@ -73,6 +73,9 @@ namespace Ryujinx.Graphics.Metal
         // Dirty flags
         public DirtyFlags Dirty = new();
 
+        // Only to be used for present
+        public bool ClearLoadAction = false;
+
         public EncoderState() { }
 
         public EncoderState Clone()

src/Ryujinx.Graphics.Metal/EncoderStateManager.cs

@@ -24,8 +24,6 @@ namespace Ryujinx.Graphics.Metal
         public readonly MTLIndexType IndexType => _currentState.IndexType;
         public readonly ulong IndexBufferOffset => _currentState.IndexBufferOffset;
         public readonly PrimitiveTopology Topology => _currentState.Topology;
-        public readonly Texture[] RenderTargets => _currentState.RenderTargets;
-        public readonly Texture DepthStencil => _currentState.DepthStencil;
 
         public EncoderStateManager(MTLDevice device, Pipeline pipeline)
         {
@@ -82,6 +80,11 @@ namespace Ryujinx.Graphics.Metal
             }
         }
 
+        public void SetClearLoadAction(bool clear)
+        {
+            _currentState.ClearLoadAction = clear;
+        }
+
         public MTLRenderCommandEncoder CreateRenderCommandEncoder()
         {
             // Initialise Pass & State
@@ -93,7 +96,7 @@ namespace Ryujinx.Graphics.Metal
                 {
                     var passAttachment = renderPassDescriptor.ColorAttachments.Object((ulong)i);
                     passAttachment.Texture = _currentState.RenderTargets[i].MTLTexture;
-                    passAttachment.LoadAction = MTLLoadAction.Load;
+                    passAttachment.LoadAction = _currentState.ClearLoadAction ? MTLLoadAction.Clear : MTLLoadAction.Load;
                     passAttachment.StoreAction = MTLStoreAction.Store;
                 }
             }
@@ -661,11 +664,18 @@ namespace Ryujinx.Graphics.Metal
             // TODO: Handle 'zero' buffers
             for (int i = 0; i < attribDescriptors.Length; i++)
             {
-                var attrib = vertexDescriptor.Attributes.Object((ulong)i);
+                if (!attribDescriptors[i].IsZero)
-                attrib.Format = attribDescriptors[i].Format.Convert();
+                {
-                indexMask |= 1u << attribDescriptors[i].BufferIndex;
+                    var attrib = vertexDescriptor.Attributes.Object((ulong)i);
-                attrib.BufferIndex = (ulong)attribDescriptors[i].BufferIndex;
+                    attrib.Format = attribDescriptors[i].Format.Convert();
-                attrib.Offset = (ulong)attribDescriptors[i].Offset;
+                    indexMask |= 1u << attribDescriptors[i].BufferIndex;
+                    attrib.BufferIndex = (ulong)attribDescriptors[i].BufferIndex;
+                    attrib.Offset = (ulong)attribDescriptors[i].Offset;
+                }
+                else
+                {
+                    // Logger.Warning?.PrintMsg(LogClass.Gpu, "Unhandled IsZero buffer!");
+                }
             }
 
             for (int i = 0; i < bufferDescriptors.Length; i++)

src/Ryujinx.Graphics.Metal/HelperShader.cs

@@ -16,6 +16,8 @@ namespace Ryujinx.Graphics.Metal
         private readonly Pipeline _pipeline;
         private MTLDevice _device;
 
+        private readonly ISampler _samplerLinear;
+        private readonly ISampler _samplerNearest;
         private readonly IProgram _programColorBlit;
         private readonly List<IProgram> _programsColorClear = new();
         private readonly IProgram _programDepthStencilClear;
@@ -25,6 +27,9 @@ namespace Ryujinx.Graphics.Metal
             _device = device;
             _pipeline = pipeline;
 
+            _samplerNearest = new Sampler(_device, SamplerCreateInfo.Create(MinFilter.Nearest, MagFilter.Nearest));
+            _samplerLinear = new Sampler(_device, SamplerCreateInfo.Create(MinFilter.Linear, MagFilter.Linear));
+
             var blitSource = ReadMsl("Blit.metal");
             _programColorBlit = new Program(
             [
@@ -56,28 +61,140 @@ namespace Ryujinx.Graphics.Metal
             return EmbeddedResources.ReadAllText(string.Join('/', ShadersSourcePath, fileName));
         }
 
-        public void BlitColor(
+        public unsafe void BlitColor(
-            ITexture source,
+            ITexture src,
-            ITexture destination)
+            ITexture dst,
+            Extents2D srcRegion,
+            Extents2D dstRegion,
+            bool linearFilter)
         {
-            var sampler = _device.NewSamplerState(new MTLSamplerDescriptor
+            const int RegionBufferSize = 16;
+
+            var sampler = linearFilter ? _samplerLinear : _samplerNearest;
+
+            Span<float> region = stackalloc float[RegionBufferSize / sizeof(float)];
+
+            region[0] = srcRegion.X1 / src.Width;
+            region[1] = srcRegion.X2 / src.Width;
+            region[2] = srcRegion.Y1 / src.Height;
+            region[3] = srcRegion.Y2 / src.Height;
+
+            if (dstRegion.X1 > dstRegion.X2)
             {
-                MinFilter = MTLSamplerMinMagFilter.Nearest,
+                (region[0], region[1]) = (region[1], region[0]);
-                MagFilter = MTLSamplerMinMagFilter.Nearest,
+            }
-                MipFilter = MTLSamplerMipFilter.NotMipmapped
+
-            });
+            if (dstRegion.Y1 > dstRegion.Y2)
+            {
+                (region[2], region[3]) = (region[3], region[2]);
+            }
+
+            var rect = new Rectangle<float>(
+                MathF.Min(dstRegion.X1, dstRegion.X2),
+                MathF.Min(dstRegion.Y1, dstRegion.Y2),
+                MathF.Abs(dstRegion.X2 - dstRegion.X1),
+                MathF.Abs(dstRegion.Y2 - dstRegion.Y1));
+
+            Span<Viewport> viewports = stackalloc Viewport[1];
+
+            viewports[0] = new Viewport(
+                rect,
+                ViewportSwizzle.PositiveX,
+                ViewportSwizzle.PositiveY,
+                ViewportSwizzle.PositiveZ,
+                ViewportSwizzle.PositiveW,
+                0f,
+                1f);
+
+            int dstWidth = dst.Width;
+            int dstHeight = dst.Height;
 
             // Save current state
             _pipeline.SaveAndResetState();
 
             _pipeline.SetProgram(_programColorBlit);
-            // Viewport and scissor needs to be set before render pass begin so as not to bind the old ones
+            _pipeline.SetViewports(viewports);
-            _pipeline.SetViewports([]);
+            _pipeline.SetScissors(stackalloc Rectangle<int>[] { new Rectangle<int>(0, 0, dstWidth, dstHeight) });
-            _pipeline.SetScissors([]);
+            _pipeline.SetRenderTargets([dst], null);
-            _pipeline.SetRenderTargets([destination], null);
+            _pipeline.SetClearLoadAction(true);
-            _pipeline.SetTextureAndSampler(ShaderStage.Fragment, 0, source, new Sampler(sampler));
+            _pipeline.SetTextureAndSampler(ShaderStage.Fragment, 0, src, sampler);
-            _pipeline.SetPrimitiveTopology(PrimitiveTopology.Triangles);
+            _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
-            _pipeline.Draw(6, 1, 0, 0);
+
+            fixed (float* ptr = region)
+            {
+                _pipeline.GetOrCreateRenderEncoder().SetVertexBytes((IntPtr)ptr, RegionBufferSize, 0);
+            }
+
+            _pipeline.Draw(4, 1, 0, 0);
+
+            // Restore previous state
+            _pipeline.RestoreState();
+        }
+
+        public unsafe void DrawTexture(
+            ITexture src,
+            ISampler srcSampler,
+            Extents2DF srcRegion,
+            Extents2DF dstRegion)
+        {
+            const int RegionBufferSize = 16;
+
+            Span<float> region = stackalloc float[RegionBufferSize / sizeof(float)];
+
+            region[0] = srcRegion.X1 / src.Width;
+            region[1] = srcRegion.X2 / src.Width;
+            region[2] = srcRegion.Y1 / src.Height;
+            region[3] = srcRegion.Y2 / src.Height;
+
+            if (dstRegion.X1 > dstRegion.X2)
+            {
+                (region[0], region[1]) = (region[1], region[0]);
+            }
+
+            if (dstRegion.Y1 > dstRegion.Y2)
+            {
+                (region[2], region[3]) = (region[3], region[2]);
+            }
+
+            Span<Viewport> viewports = stackalloc Viewport[1];
+            Span<Rectangle<int>> scissors = stackalloc Rectangle<int>[1];
+
+            var rect = new Rectangle<float>(
+                MathF.Min(dstRegion.X1, dstRegion.X2),
+                MathF.Min(dstRegion.Y1, dstRegion.Y2),
+                MathF.Abs(dstRegion.X2 - dstRegion.X1),
+                MathF.Abs(dstRegion.Y2 - dstRegion.Y1));
+
+            viewports[0] = new Viewport(
+                rect,
+                ViewportSwizzle.PositiveX,
+                ViewportSwizzle.PositiveY,
+                ViewportSwizzle.PositiveZ,
+                ViewportSwizzle.PositiveW,
+                0f,
+                1f);
+
+            scissors[0] = new Rectangle<int>(0, 0, 0xFFFF, 0xFFFF);
+
+            // Save current state
+            _pipeline.SaveState();
+
+            _pipeline.SetProgram(_programColorBlit);
+            _pipeline.SetViewports(viewports);
+            _pipeline.SetScissors(scissors);
+            _pipeline.SetTextureAndSampler(ShaderStage.Fragment, 0, src, srcSampler);
+            _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
+            _pipeline.SetFaceCulling(false, Face.FrontAndBack);
+            // For some reason this results in a SIGSEGV
+            // _pipeline.SetStencilTest(CreateStencilTestDescriptor(false));
+            _pipeline.SetDepthTest(new DepthTestDescriptor(false, false, CompareOp.Always));
+
+            fixed (float* ptr = region)
+            {
+                _pipeline.GetOrCreateRenderEncoder().SetVertexBytes((IntPtr)ptr, RegionBufferSize, 0);
+            }
+
+            _pipeline.Draw(4, 1, 0, 0);
 
             // Restore previous state
             _pipeline.RestoreState();
@@ -169,6 +286,8 @@ namespace Ryujinx.Graphics.Metal
             }
             _programDepthStencilClear.Dispose();
             _pipeline.Dispose();
+            _samplerLinear.Dispose();
+            _samplerNearest.Dispose();
         }
     }
 }

src/Ryujinx.Graphics.Metal/Pipeline.cs

@@ -61,6 +61,11 @@ namespace Ryujinx.Graphics.Metal
             _encoderStateManager.RestoreState();
         }
 
+        public void SetClearLoadAction(bool clear)
+        {
+            _encoderStateManager.SetClearLoadAction(clear);
+        }
+
         public MTLRenderCommandEncoder GetOrCreateRenderEncoder()
         {
             MTLRenderCommandEncoder renderCommandEncoder;
@@ -167,22 +172,17 @@ namespace Ryujinx.Graphics.Metal
             return computeCommandEncoder;
         }
 
-        public void Present(CAMetalDrawable drawable, ITexture texture)
+        public void Present(CAMetalDrawable drawable, Texture src, Extents2D srcRegion, Extents2D dstRegion, bool isLinear)
         {
-            if (texture is not Texture tex)
-            {
-                return;
-            }
-
             EndCurrentPass();
 
             SaveState();
 
             // TODO: Clean this up
             var textureInfo = new TextureCreateInfo((int)drawable.Texture.Width, (int)drawable.Texture.Height, (int)drawable.Texture.Depth, (int)drawable.Texture.MipmapLevelCount, (int)drawable.Texture.SampleCount, 0, 0, 0, Format.B8G8R8A8Unorm, 0, Target.Texture2D, SwizzleComponent.Red, SwizzleComponent.Green, SwizzleComponent.Blue, SwizzleComponent.Alpha);
-            var dest = new Texture(_device, this, textureInfo, drawable.Texture, 0, 0);
+            var dst = new Texture(_device, this, textureInfo, drawable.Texture, 0, 0);
 
-            _helperShader.BlitColor(tex, dest);
+            _helperShader.BlitColor(src, dst, srcRegion, dstRegion, isLinear);
 
             EndCurrentPass();
 
@@ -194,7 +194,7 @@ namespace Ryujinx.Graphics.Metal
             RestoreState();
 
             // Cleanup
-            dest.Dispose();
+            dst.Dispose();
         }
 
         public void Barrier()
@@ -338,9 +338,7 @@ namespace Ryujinx.Graphics.Metal
 
         public void DrawTexture(ITexture texture, ISampler sampler, Extents2DF srcRegion, Extents2DF dstRegion)
         {
-            // var renderCommandEncoder = GetOrCreateRenderEncoder();
+            _helperShader.DrawTexture(texture, sampler, srcRegion, dstRegion);
-
-            Logger.Warning?.Print(LogClass.Gpu, "Not Implemented!");
         }
 
         public void SetAlphaTest(bool enable, float reference, CompareOp op)

src/Ryujinx.Graphics.Metal/Shaders/Blit.metal

@@ -2,32 +2,23 @@
 
 using namespace metal;
 
-// ------------------
-// Simple Blit Shader
-// ------------------
-
-constant float2 quadVertices[] = {
-    float2(-1, -1),
-    float2(-1,  1),
-    float2( 1,  1),
-    float2(-1, -1),
-    float2( 1,  1),
-    float2( 1, -1)
-};
-
 struct CopyVertexOut {
     float4 position [[position]];
     float2 uv;
 };
 
-vertex CopyVertexOut vertexMain(unsigned short vid [[vertex_id]]) {
+vertex CopyVertexOut vertexMain(uint vid [[vertex_id]],
-    float2 position = quadVertices[vid];
+                                const device float* texCoord [[buffer(0)]]) {
-
     CopyVertexOut out;
 
-    out.position = float4(position, 0, 1);
+    int low = vid & 1;
-    out.position.y = -out.position.y;
+    int high = vid >> 1;
-    out.uv = position * 0.5f + 0.5f;
+    out.uv.x = texCoord[low];
+    out.uv.y = texCoord[2 + high];
+    out.position.x = (float(low) - 0.5f) * 2.0f;
+    out.position.y = (float(high) - 0.5f) * 2.0f;
+    out.position.z = 0.0f;
+    out.position.w = 1.0f;
 
     return out;
 }

src/Ryujinx.Graphics.Metal/Window.cs

@@ -1,5 +1,6 @@
 using Ryujinx.Common.Logging;
 using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Metal.Effects;
 using SharpMetal.ObjectiveCCore;
 using SharpMetal.QuartzCore;
 using System;
@@ -10,51 +11,196 @@ namespace Ryujinx.Graphics.Metal
     [SupportedOSPlatform("macos")]
     class Window : IWindow, IDisposable
     {
+        public bool ScreenCaptureRequested { get; set; }
+
         private readonly MetalRenderer _renderer;
         private readonly CAMetalLayer _metalLayer;
 
+        private int _width;
+        private int _height;
+        private bool _vsyncEnabled;
+        private AntiAliasing _currentAntiAliasing;
+        private bool _updateEffect;
+        private IPostProcessingEffect _effect;
+        private IScalingFilter _scalingFilter;
+        private bool _isLinear;
+        private float _scalingFilterLevel;
+        private bool _updateScalingFilter;
+        private ScalingFilter _currentScalingFilter;
+        private bool _colorSpacePassthroughEnabled;
+
         public Window(MetalRenderer renderer, CAMetalLayer metalLayer)
         {
             _renderer = renderer;
             _metalLayer = metalLayer;
         }
 
-        // TODO: Handle ImageCrop
         public void Present(ITexture texture, ImageCrop crop, Action swapBuffersCallback)
         {
             if (_renderer.Pipeline is Pipeline pipeline && texture is Texture tex)
             {
                 var drawable = new CAMetalDrawable(ObjectiveC.IntPtr_objc_msgSend(_metalLayer, "nextDrawable"));
-                pipeline.Present(drawable, tex);
+
+                _width = (int)drawable.Texture.Width;
+                _height = (int)drawable.Texture.Height;
+
+                UpdateEffect();
+
+                if (_effect != null)
+                {
+                    // TODO: Run Effects
+                    // view = _effect.Run()
+                }
+
+                int srcX0, srcX1, srcY0, srcY1;
+
+                if (crop.Left == 0 && crop.Right == 0)
+                {
+                    srcX0 = 0;
+                    srcX1 = tex.Width;
+                }
+                else
+                {
+                    srcX0 = crop.Left;
+                    srcX1 = crop.Right;
+                }
+
+                if (crop.Top == 0 && crop.Bottom == 0)
+                {
+                    srcY0 = 0;
+                    srcY1 = tex.Height;
+                }
+                else
+                {
+                    srcY0 = crop.Top;
+                    srcY1 = crop.Bottom;
+                }
+
+                if (ScreenCaptureRequested)
+                {
+                    // TODO: Support screen captures
+
+                    ScreenCaptureRequested = false;
+                }
+
+                float ratioX = crop.IsStretched ? 1.0f : MathF.Min(1.0f, _height * crop.AspectRatioX / (_width * crop.AspectRatioY));
+                float ratioY = crop.IsStretched ? 1.0f : MathF.Min(1.0f, _width * crop.AspectRatioY / (_height * crop.AspectRatioX));
+
+                int dstWidth = (int)(_width * ratioX);
+                int dstHeight = (int)(_height * ratioY);
+
+                int dstPaddingX = (_width - dstWidth) / 2;
+                int dstPaddingY = (_height - dstHeight) / 2;
+
+                int dstX0 = crop.FlipX ? _width - dstPaddingX : dstPaddingX;
+                int dstX1 = crop.FlipX ? dstPaddingX : _width - dstPaddingX;
+
+                int dstY0 = crop.FlipY ? _height - dstPaddingY : dstPaddingY;
+                int dstY1 = crop.FlipY ? dstPaddingY : _height - dstPaddingY;
+
+                if (_scalingFilter != null)
+                {
+                    // TODO: Run scaling filter
+                }
+
+                pipeline.Present(
+                    drawable,
+                    tex,
+                    new Extents2D(srcX0, srcY0, srcX1, srcY1),
+                    new Extents2D(dstX0, dstY0, dstX1, dstY1),
+                    _isLinear);
             }
         }
 
         public void SetSize(int width, int height)
         {
-            Logger.Warning?.Print(LogClass.Gpu, "Not Implemented!");
+            // Ignore
         }
 
         public void ChangeVSyncMode(bool vsyncEnabled)
         {
-            Logger.Warning?.Print(LogClass.Gpu, "Not Implemented!");
+            _vsyncEnabled = vsyncEnabled;
         }
 
-        public void SetAntiAliasing(AntiAliasing antialiasing)
+        public void SetAntiAliasing(AntiAliasing effect)
         {
-            Logger.Warning?.Print(LogClass.Gpu, "Not Implemented!");
+            if (_currentAntiAliasing == effect && _effect != null)
+            {
+                return;
+            }
+
+            _currentAntiAliasing = effect;
+
+            _updateEffect = true;
         }
 
         public void SetScalingFilter(ScalingFilter type)
         {
-            Logger.Warning?.Print(LogClass.Gpu, "Not Implemented!");
+            if (_currentScalingFilter == type && _effect != null)
+            {
+                return;
+            }
+
+            _currentScalingFilter = type;
+
+            _updateScalingFilter = true;
         }
 
         public void SetScalingFilterLevel(float level)
         {
-            Logger.Warning?.Print(LogClass.Gpu, "Not Implemented!");
+            _scalingFilterLevel = level;
+            _updateScalingFilter = true;
         }
 
-        public void SetColorSpacePassthrough(bool colorSpacePassThroughEnabled) { }
+        public void SetColorSpacePassthrough(bool colorSpacePassThroughEnabled)
+        {
+            _colorSpacePassthroughEnabled = colorSpacePassThroughEnabled;
+        }
+
+        private void UpdateEffect()
+        {
+            if (_updateEffect)
+            {
+                _updateEffect = false;
+
+                switch (_currentAntiAliasing)
+                {
+                    case AntiAliasing.Fxaa:
+                        _effect?.Dispose();
+                        Logger.Warning?.PrintMsg(LogClass.Gpu, "FXAA not implemented for Metal backend!");
+                        break;
+                    case AntiAliasing.None:
+                        _effect?.Dispose();
+                        _effect = null;
+                        break;
+                    case AntiAliasing.SmaaLow:
+                    case AntiAliasing.SmaaMedium:
+                    case AntiAliasing.SmaaHigh:
+                    case AntiAliasing.SmaaUltra:
+                        var quality = _currentAntiAliasing - AntiAliasing.SmaaLow;
+                        Logger.Warning?.PrintMsg(LogClass.Gpu, "SMAA not implemented for Metal backend!");
+                        break;
+                }
+            }
+
+            if (_updateScalingFilter)
+            {
+                _updateScalingFilter = false;
+
+                switch (_currentScalingFilter)
+                {
+                    case ScalingFilter.Bilinear:
+                    case ScalingFilter.Nearest:
+                        _scalingFilter?.Dispose();
+                        _scalingFilter = null;
+                        _isLinear = _currentScalingFilter == ScalingFilter.Bilinear;
+                        break;
+                    case ScalingFilter.Fsr:
+                        Logger.Warning?.PrintMsg(LogClass.Gpu, "FSR not implemented for Metal backend!");
+                        break;
+                }
+            }
+        }
 
         public void Dispose()
         {