Clone the state & flip viewport vertically (#16)

* implement texture get data

* reset all state before blit & clone state

* format

* support blit regions

* implement source region for blit

* replace bottom with top

* account for 0 size

* support image flipping

* revert presentation fixes & y flip

* revert

* flip viewport vertically

* switch face winding

* comment

* use SetBytes for texture clear

* implement missing compute builtins

* change storage and texture buffer alignment

* correct compute builtins

* don't use nullable for textures and samplers

* remove incorrect texture get data implementation

* Cleanup IntPtrs

---------

Co-authored-by: Isaac Marovitz <isaacryu@icloud.com>

src/Ryujinx.Graphics.Metal/Constants.cs

@@ -14,5 +14,7 @@ namespace Ryujinx.Graphics.Metal
         public const int MaxVertexAttributes = 16;
         // TODO: Check this value
         public const int MaxVertexLayouts = 16;
+        public const int MaxTextures = 31;
+        public const int MaxSamplers = 31;
     }
 }

src/Ryujinx.Graphics.Metal/EncoderState.cs

@@ -31,11 +31,11 @@ namespace Ryujinx.Graphics.Metal
         public MTLFunction? VertexFunction = null;
         public MTLFunction? FragmentFunction = null;
 
-        public Dictionary<ulong, MTLTexture> FragmentTextures = new();
+        public MTLTexture[] FragmentTextures = new MTLTexture[Constants.MaxTextures];
-        public Dictionary<ulong, MTLSamplerState> FragmentSamplers = new();
+        public MTLSamplerState[] FragmentSamplers = new MTLSamplerState[Constants.MaxSamplers];
 
-        public Dictionary<ulong, MTLTexture> VertexTextures = new();
+        public MTLTexture[] VertexTextures = new MTLTexture[Constants.MaxTextures];
-        public Dictionary<ulong, MTLSamplerState> VertexSamplers = new();
+        public MTLSamplerState[] VertexSamplers = new MTLSamplerState[Constants.MaxSamplers];
 
         public List<BufferInfo> UniformBuffers = [];
         public List<BufferInfo> StorageBuffers = [];
@@ -56,7 +56,7 @@ namespace Ryujinx.Graphics.Metal
 
         public PrimitiveTopology Topology = PrimitiveTopology.Triangles;
         public MTLCullMode CullMode = MTLCullMode.None;
-        public MTLWinding Winding = MTLWinding.Clockwise;
+        public MTLWinding Winding = MTLWinding.CounterClockwise;
 
         public MTLViewport[] Viewports = [];
         public MTLScissorRect[] Scissors = [];
@@ -64,7 +64,7 @@ namespace Ryujinx.Graphics.Metal
         // Changes to attachments take recreation!
         public Texture DepthStencil = default;
         public Texture[] RenderTargets = new Texture[Constants.MaxColorAttachments];
-        public Dictionary<int, BlendDescriptor> BlendDescriptors = new();
+        public BlendDescriptor?[] BlendDescriptors = new BlendDescriptor?[Constants.MaxColorAttachments];
         public ColorF BlendColor = new();
 
         public VertexBufferDescriptor[] VertexBuffers = [];
@@ -74,5 +74,20 @@ namespace Ryujinx.Graphics.Metal
         public DirtyFlags Dirty = new();
 
         public EncoderState() { }
+
+        public EncoderState Clone()
+        {
+            // Certain state (like viewport and scissor) doesn't need to be cloned, as it is always reacreated when assigned to
+            EncoderState clone = this;
+            clone.FragmentTextures = (MTLTexture[])FragmentTextures.Clone();
+            clone.FragmentSamplers = (MTLSamplerState[])FragmentSamplers.Clone();
+            clone.VertexTextures = (MTLTexture[])VertexTextures.Clone();
+            clone.VertexSamplers = (MTLSamplerState[])VertexSamplers.Clone();
+            clone.BlendDescriptors = (BlendDescriptor?[])BlendDescriptors.Clone();
+            clone.VertexBuffers = (VertexBufferDescriptor[])VertexBuffers.Clone();
+            clone.VertexAttribs = (VertexAttribDescriptor[])VertexAttribs.Clone();
+
+            return clone;
+        }
     }
 }

src/Ryujinx.Graphics.Metal/EncoderStateManager.cs

@@ -44,9 +44,16 @@ namespace Ryujinx.Graphics.Metal
             _depthStencilCache.Dispose();
         }
 
-        public readonly void SaveState()
+        public void SaveState()
         {
             _backStates.Push(_currentState);
+            _currentState = _currentState.Clone();
+        }
+
+        public void SaveAndResetState()
+        {
+            _backStates.Push(_currentState);
+            _currentState = new();
         }
 
         public void RestoreState()
@@ -65,6 +72,9 @@ namespace Ryujinx.Graphics.Metal
                 SetBuffers(renderCommandEncoder, _currentState.StorageBuffers, true);
                 SetCullMode(renderCommandEncoder);
                 SetFrontFace(renderCommandEncoder);
+
+                // Mark the other state as dirty
+                _currentState.Dirty.MarkAll();
             }
             else
             {
@@ -184,8 +194,9 @@ namespace Ryujinx.Graphics.Metal
                     pipelineAttachment.SourceRGBBlendFactor = MTLBlendFactor.SourceAlpha;
                     pipelineAttachment.DestinationRGBBlendFactor = MTLBlendFactor.OneMinusSourceAlpha;
 
-                    if (_currentState.BlendDescriptors.TryGetValue(i, out BlendDescriptor blendDescriptor))
+                    if (_currentState.BlendDescriptors[i] != null)
                     {
+                        var blendDescriptor = _currentState.BlendDescriptors[i].Value;
                         pipelineAttachment.SetBlendingEnabled(blendDescriptor.Enable);
                         pipelineAttachment.AlphaBlendOperation = blendDescriptor.AlphaOp.Convert();
                         pipelineAttachment.RgbBlendOperation = blendDescriptor.ColorOp.Convert();
@@ -469,12 +480,13 @@ namespace Ryujinx.Graphics.Metal
             for (int i = 0; i < viewports.Length; i++)
             {
                 var viewport = viewports[i];
+                // Y coordinate is inverted
                 _currentState.Viewports[i] = new MTLViewport
                 {
                     originX = viewport.Region.X,
-                    originY = viewport.Region.Y,
+                    originY = viewport.Region.Y + viewport.Region.Height,
                     width = viewport.Region.Width,
-                    height = viewport.Region.Height,
+                    height = -viewport.Region.Height,
                     znear = Clamp(viewport.DepthNear),
                     zfar = Clamp(viewport.DepthFar)
                 };
@@ -708,33 +720,41 @@ namespace Ryujinx.Graphics.Metal
             renderCommandEncoder.SetFrontFacingWinding(_currentState.Winding);
         }
 
-        private static void SetTextureAndSampler(MTLRenderCommandEncoder renderCommandEncoder, ShaderStage stage, Dictionary<ulong, MTLTexture> textures, Dictionary<ulong, MTLSamplerState> samplers)
+        private static void SetTextureAndSampler(MTLRenderCommandEncoder renderCommandEncoder, ShaderStage stage, MTLTexture[] textures, MTLSamplerState[] samplers)
         {
-            foreach (var texture in textures)
+            for (int i = 0; i < textures.Length; i++)
+            {
+                var texture = textures[i];
+                if (texture != IntPtr.Zero)
                 {
                     switch (stage)
                     {
                         case ShaderStage.Vertex:
-                        renderCommandEncoder.SetVertexTexture(texture.Value, texture.Key);
+                            renderCommandEncoder.SetVertexTexture(texture, (ulong)i);
                             break;
                         case ShaderStage.Fragment:
-                        renderCommandEncoder.SetFragmentTexture(texture.Value, texture.Key);
+                            renderCommandEncoder.SetFragmentTexture(texture, (ulong)i);
                             break;
                     }
                 }
+            }
 
-            foreach (var sampler in samplers)
+            for (int i = 0; i < samplers.Length; i++)
+            {
+                var sampler = samplers[i];
+                if (sampler != IntPtr.Zero)
                 {
                     switch (stage)
                     {
                         case ShaderStage.Vertex:
-                        renderCommandEncoder.SetVertexSamplerState(sampler.Value, sampler.Key);
+                            renderCommandEncoder.SetVertexSamplerState(sampler, (ulong)i);
                             break;
                         case ShaderStage.Fragment:
-                        renderCommandEncoder.SetFragmentSamplerState(sampler.Value, sampler.Key);
+                            renderCommandEncoder.SetFragmentSamplerState(sampler, (ulong)i);
                             break;
                     }
                 }
             }
         }
+    }
 }

src/Ryujinx.Graphics.Metal/EnumConversion.cs

@@ -93,10 +93,11 @@ namespace Ryujinx.Graphics.Metal
 
         public static MTLWinding Convert(this FrontFace frontFace)
         {
+            // The viewport is flipped vertically, therefore we need to switch the winding order as well
             return frontFace switch
             {
-                FrontFace.Clockwise => MTLWinding.Clockwise,
+                FrontFace.Clockwise => MTLWinding.CounterClockwise,
-                FrontFace.CounterClockwise => MTLWinding.CounterClockwise,
+                FrontFace.CounterClockwise => MTLWinding.Clockwise,
                 _ => LogInvalidAndReturn(frontFace, nameof(FrontFace), MTLWinding.Clockwise)
             };
         }

src/Ryujinx.Graphics.Metal/HelperShader.cs

@@ -2,11 +2,9 @@ using Ryujinx.Common;
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Shader;
 using Ryujinx.Graphics.Shader.Translation;
-using SharpMetal.Foundation;
 using SharpMetal.Metal;
 using System;
 using System.Collections.Generic;
-using System.Runtime.CompilerServices;
 using System.Runtime.Versioning;
 
 namespace Ryujinx.Graphics.Metal
@@ -70,11 +68,9 @@ namespace Ryujinx.Graphics.Metal
             });
 
             // Save current state
-            _pipeline.SaveState();
+            _pipeline.SaveAndResetState();
 
             _pipeline.SetProgram(_programColorBlit);
-            _pipeline.SetFaceCulling(false, Face.Front);
-            _pipeline.SetDepthTest(new DepthTestDescriptor(false, false, CompareOp.Always));
             // Viewport and scissor needs to be set before render pass begin so as not to bind the old ones
             _pipeline.SetViewports([]);
             _pipeline.SetScissors([]);
@@ -93,29 +89,21 @@ namespace Ryujinx.Graphics.Metal
         {
             const int ClearColorBufferSize = 16;
 
-            var buffer = _device.NewBuffer(ClearColorBufferSize, MTLResourceOptions.ResourceStorageModeManaged);
-            var span = new Span<float>(buffer.Contents.ToPointer(), ClearColorBufferSize);
-            clearColor.CopyTo(span);
-
-            buffer.DidModifyRange(new NSRange
-            {
-                location = 0,
-                length = ClearColorBufferSize
-            });
-
-            var handle = buffer.NativePtr;
-            var range = new BufferRange(Unsafe.As<IntPtr, BufferHandle>(ref handle), 0, ClearColorBufferSize);
-
             // Save current state
             _pipeline.SaveState();
 
-            _pipeline.SetUniformBuffers([new BufferAssignment(0, range)]);
-
             _pipeline.SetProgram(_programsColorClear[index]);
+            _pipeline.SetBlendState(index, new BlendDescriptor(false, new ColorF(0f, 0f, 0f, 1f), BlendOp.Add, BlendFactor.One, BlendFactor.Zero, BlendOp.Add, BlendFactor.One, BlendFactor.Zero));
             _pipeline.SetFaceCulling(false, Face.Front);
             _pipeline.SetDepthTest(new DepthTestDescriptor(false, false, CompareOp.Always));
             // _pipeline.SetRenderTargetColorMasks([componentMask]);
             _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
+
+            fixed (float* ptr = clearColor)
+            {
+                _pipeline.GetOrCreateRenderEncoder().SetFragmentBytes((IntPtr)ptr, ClearColorBufferSize, 0);
+            }
+
             _pipeline.Draw(4, 1, 0, 0);
 
             // Restore previous state
@@ -123,37 +111,25 @@ namespace Ryujinx.Graphics.Metal
         }
 
         public unsafe void ClearDepthStencil(
-            ReadOnlySpan<float> depthValue,
+            float depthValue,
             bool depthMask,
             int stencilValue,
             int stencilMask)
         {
-            const int ClearColorBufferSize = 16;
+            const int ClearDepthBufferSize = 4;
 
-            var buffer = _device.NewBuffer(ClearColorBufferSize, MTLResourceOptions.ResourceStorageModeManaged);
+            IntPtr ptr = new(&depthValue);
-            var span = new Span<float>(buffer.Contents.ToPointer(), ClearColorBufferSize);
-            depthValue.CopyTo(span);
-
-            buffer.DidModifyRange(new NSRange
-            {
-                location = 0,
-                length = ClearColorBufferSize
-            });
-
-            var handle = buffer.NativePtr;
-            var range = new BufferRange(Unsafe.As<IntPtr, BufferHandle>(ref handle), 0, ClearColorBufferSize);
 
             // Save current state
             _pipeline.SaveState();
 
-            _pipeline.SetUniformBuffers([new BufferAssignment(0, range)]);
-
             _pipeline.SetProgram(_programDepthStencilClear);
             _pipeline.SetFaceCulling(false, Face.Front);
             _pipeline.SetDepthTest(new DepthTestDescriptor(false, false, CompareOp.Always));
             _pipeline.SetPrimitiveTopology(PrimitiveTopology.TriangleStrip);
             _pipeline.SetDepthTest(new DepthTestDescriptor(true, depthMask, CompareOp.Always));
             // _pipeline.SetStencilTest(CreateStencilTestDescriptor(stencilMask != 0, stencilValue, 0xFF, stencilMask));
+            _pipeline.GetOrCreateRenderEncoder().SetFragmentBytes(ptr, ClearDepthBufferSize, 0);
             _pipeline.Draw(4, 1, 0, 0);
 
             // Restore previous state

src/Ryujinx.Graphics.Metal/MetalRenderer.cs

@@ -186,8 +186,8 @@ namespace Ryujinx.Graphics.Metal
                 maximumComputeSharedMemorySize: (int)_device.MaxThreadgroupMemoryLength,
                 maximumSupportedAnisotropy: 0,
                 shaderSubgroupSize: 256,
-                storageBufferOffsetAlignment: 0,
+                storageBufferOffsetAlignment: 16,
-                textureBufferOffsetAlignment: 0,
+                textureBufferOffsetAlignment: 16,
                 gatherBiasPrecision: 0
             );
         }

src/Ryujinx.Graphics.Metal/Pipeline.cs

@@ -51,6 +51,11 @@ namespace Ryujinx.Graphics.Metal
             _encoderStateManager.SaveState();
         }
 
+        public void SaveAndResetState()
+        {
+            _encoderStateManager.SaveAndResetState();
+        }
+
         public void RestoreState()
         {
             _encoderStateManager.RestoreState();
@@ -242,7 +247,7 @@ namespace Ryujinx.Graphics.Metal
 
         public void ClearRenderTargetDepthStencil(int layer, int layerCount, float depthValue, bool depthMask, int stencilValue, int stencilMask)
         {
-            _helperShader.ClearDepthStencil([depthValue], depthMask, stencilValue, stencilMask);
+            _helperShader.ClearDepthStencil(depthValue, depthMask, stencilValue, stencilMask);
         }
 
         public void CommandBufferBarrier()

src/Ryujinx.Graphics.Shader/CodeGen/Msl/Instructions/IoMap.cs

@@ -19,18 +19,21 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Msl.Instructions
             {
                 IoVariable.BaseInstance => ("base_instance", AggregateType.S32),
                 IoVariable.BaseVertex => ("base_vertex", AggregateType.S32),
+                IoVariable.CtaId => ("threadgroup_position_in_grid", AggregateType.Vector3 | AggregateType.U32),
                 IoVariable.ClipDistance => ("clip_distance", AggregateType.Array | AggregateType.FP32),
                 IoVariable.FragmentOutputColor => ($"out.color{location}", AggregateType.Vector4 | AggregateType.FP32),
                 IoVariable.FragmentOutputDepth => ("out.depth", AggregateType.FP32),
                 IoVariable.FrontFacing => ("front_facing", AggregateType.Bool),
+                IoVariable.GlobalId => ("thread_position_in_grid", AggregateType.Vector3 | AggregateType.U32),
                 IoVariable.InstanceId => ("instance_id", AggregateType.S32),
+                IoVariable.InvocationId => ("INVOCATION_ID", AggregateType.S32),
                 IoVariable.PointCoord => ("point_coord", AggregateType.Vector2),
                 IoVariable.PointSize => ("out.point_size", AggregateType.FP32),
                 IoVariable.Position => ("out.position", AggregateType.Vector4 | AggregateType.FP32),
                 IoVariable.PrimitiveId => ("primitive_id", AggregateType.S32),
                 IoVariable.UserDefined => GetUserDefinedVariableName(definitions, location, component, isOutput, isPerPatch),
+                IoVariable.ThreadId => ("thread_position_in_threadgroup", AggregateType.Vector3 | AggregateType.U32),
                 IoVariable.VertexId => ("vertex_id", AggregateType.S32),
-                IoVariable.GlobalId => ("global_id", AggregateType.Vector3 | AggregateType.U32),
                 // gl_VertexIndex does not have a direct equivalent in MSL
                 IoVariable.VertexIndex => ("vertex_index", AggregateType.U32),
                 IoVariable.ViewportIndex => ("viewport_array_index", AggregateType.S32),