Initial transform feedback support (#1370)

* Initial transform feedback support

* Some nits and fixes

* Update ReportCounterType and Write method

* Can't change shader or TFB bindings while TFB is active

* Fix geometry shader input names with new naming

Ryujinx.Graphics.GAL/IPipeline.cs

@@ -7,6 +7,8 @@ public interface IPipeline
     {
         void Barrier();
 
+        void BeginTransformFeedback(PrimitiveTopology topology);
+
         void ClearRenderTargetColor(int index, uint componentMask, ColorF color);
 
         void ClearRenderTargetDepthStencil(
@@ -27,6 +29,8 @@ void DrawIndexed(
             int firstVertex,
             int firstInstance);
 
+        void EndTransformFeedback();
+
         void SetBlendState(int index, BlendDescriptor blend);
 
         void SetDepthBias(PolygonModeMask enables, float factor, float units, float clamp);
@@ -73,6 +77,7 @@ void DrawIndexed(
 
         void SetTexture(int index, ShaderStage stage, ITexture texture);
 
+        void SetTransformFeedbackBuffer(int index, BufferRange buffer);
         void SetUniformBuffer(int index, ShaderStage stage, BufferRange buffer);
 
         void SetUserClipDistance(int index, bool enableClip);

Ryujinx.Graphics.GAL/IRenderer.cs

@@ -13,7 +13,7 @@ public interface IRenderer : IDisposable
 
         BufferHandle CreateBuffer(int size);
 
-        IProgram CreateProgram(IShader[] shaders);
+        IProgram CreateProgram(IShader[] shaders, TransformFeedbackDescriptor[] transformFeedbackDescriptors);
 
         ISampler CreateSampler(SamplerCreateInfo info);
         ITexture CreateTexture(TextureCreateInfo info, float scale);

Ryujinx.Graphics.GAL/TransformFeedbackDescriptor.cs

@@ -0,0 +1,19 @@
+using System;
+
+namespace Ryujinx.Graphics.GAL
+{
+    public struct TransformFeedbackDescriptor
+    {
+        public int BufferIndex { get; }
+        public int Stride      { get; }
+
+        public byte[] VaryingLocations { get; }
+
+        public TransformFeedbackDescriptor(int bufferIndex, int stride, byte[] varyingLocations)
+        {
+            BufferIndex      = bufferIndex;
+            Stride           = stride;
+            VaryingLocations = varyingLocations ?? throw new ArgumentNullException(nameof(varyingLocations));
+        }
+    }
+}

Ryujinx.Graphics.Gpu/Constants.cs

@@ -35,6 +35,11 @@ static class Constants
         /// </remarks>
         public const int TotalGpStorageBuffers = 16;
 
+        /// <summary>
+        /// Maximum number of transform feedback buffers.
+        /// </summary>
+        public const int TotalTransformFeedbackBuffers = 4;
+
         /// <summary>
         /// Maximum number of render target color buffers.
         /// </summary>

Ryujinx.Graphics.Gpu/Engine/MethodReport.cs

@@ -61,8 +61,6 @@ private struct CounterData
         /// <param name="type">Counter to be written to memory</param>
         private void ReportCounter(GpuState state, ReportCounterType type)
         {
-            CounterData counterData = new CounterData();
-
             var rs = state.Get<SemaphoreState>(MethodOffset.ReportState);
 
             ulong gpuVa = rs.Address.Pack();
@@ -80,16 +78,14 @@ private void ReportCounter(GpuState state, ReportCounterType type)
 
             EventHandler<ulong> resultHandler = (object evt, ulong result) =>
             {
+                CounterData counterData = new CounterData();
+
                 counterData.Counter = result;
                 counterData.Timestamp = ticks;
 
-                Span<CounterData> counterDataSpan = MemoryMarshal.CreateSpan(ref counterData, 1);
-
-                Span<byte> data = MemoryMarshal.Cast<CounterData, byte>(counterDataSpan);
-
                 if (counter?.Invalid != true)
                 {
-                    _context.MemoryAccessor.Write(gpuVa, data);
+                    _context.MemoryAccessor.Write(gpuVa, counterData);
                 }
             };
 

Ryujinx.Graphics.Gpu/Engine/Methods.cs

@@ -40,6 +40,8 @@ partial class Methods
 
         private bool _forceShaderUpdate;
 
+        private bool _prevTfEnable;
+
         /// <summary>
         /// Creates a new instance of the GPU methods class.
         /// </summary>
@@ -124,6 +126,14 @@ public void RegisterCallbacksForFifo(GpuState state)
         /// <param name="state">Guest GPU state</param>
         private void UpdateState(GpuState state)
         {
+            bool tfEnable = state.Get<Boolean32>(MethodOffset.TfEnable);
+
+            if (!tfEnable && _prevTfEnable)
+            {
+                _context.Renderer.Pipeline.EndTransformFeedback();
+                _prevTfEnable = false;
+            }
+
             // Shaders must be the first one to be updated if modified, because
             // some of the other state depends on information from the currently
             // bound shaders.
@@ -134,6 +144,11 @@ private void UpdateState(GpuState state)
                 UpdateShaderState(state);
             }
 
+            if (state.QueryModified(MethodOffset.TfBufferState))
+            {
+                UpdateTfBufferState(state);
+            }
+
             if (state.QueryModified(MethodOffset.ClipDistanceEnable))
             {
                 UpdateUserClipState(state);
@@ -258,6 +273,12 @@ private void UpdateState(GpuState state)
             }
 
             CommitBindings();
+
+            if (tfEnable && !_prevTfEnable)
+            {
+                _context.Renderer.Pipeline.BeginTransformFeedback(PrimitiveType.Convert());
+                _prevTfEnable = true;
+            }
         }
 
         /// <summary>
@@ -318,7 +339,7 @@ private void UpdateStorageBuffers()
         /// </summary>
         /// <param name="state">Current GPU state</param>
         /// <param name="useControl">Use draw buffers information from render target control register</param>
-        /// <param name="singleUse">If this is not -1, it indicates that only the given indexed target will be used.</param> 
+        /// <param name="singleUse">If this is not -1, it indicates that only the given indexed target will be used.</param>
         private void UpdateRenderTargetState(GpuState state, bool useControl, int singleUse = -1)
         {
             var rtControl = state.Get<RtControl>(MethodOffset.RtControl);
@@ -1003,6 +1024,27 @@ private void UpdateShaderState(GpuState state)
             _context.Renderer.Pipeline.SetProgram(gs.HostProgram);
         }
 
+        /// <summary>
+        /// Updates transform feedback buffer state based on the guest GPU state.
+        /// </summary>
+        /// <param name="state">Current GPU state</param>
+        private void UpdateTfBufferState(GpuState state)
+        {
+            for (int index = 0; index < Constants.TotalTransformFeedbackBuffers; index++)
+            {
+                TfBufferState tfb = state.Get<TfBufferState>(MethodOffset.TfBufferState, index);
+
+                if (!tfb.Enable)
+                {
+                    BufferManager.SetTransformFeedbackBuffer(index, 0, 0);
+
+                    continue;
+                }
+
+                BufferManager.SetTransformFeedbackBuffer(index, tfb.Address.Pack(), (uint)tfb.Size);
+            }
+        }
+
         /// <summary>
         /// Updates user-defined clipping based on the guest GPU state.
         /// </summary>

Ryujinx.Graphics.Gpu/Memory/BufferManager.cs

@@ -24,8 +24,8 @@ class BufferManager
         private Buffer[] _bufferOverlaps;
 
         private IndexBuffer _indexBuffer;
-
         private VertexBuffer[] _vertexBuffers;
+        private BufferBounds[] _transformFeedbackBuffers;
 
         private class BuffersPerStage
         {
@@ -56,6 +56,7 @@ public void Bind(int index, ulong address, ulong size)
         private bool _indexBufferDirty;
         private bool _vertexBuffersDirty;
         private uint _vertexBuffersEnableMask;
+        private bool _transformFeedbackBuffersDirty;
 
         private bool _rebind;
 
@@ -73,6 +74,8 @@ public BufferManager(GpuContext context)
 
             _vertexBuffers = new VertexBuffer[Constants.TotalVertexBuffers];
 
+            _transformFeedbackBuffers = new BufferBounds[Constants.TotalTransformFeedbackBuffers];
+
             _cpStorageBuffers = new BuffersPerStage(Constants.TotalCpStorageBuffers);
             _cpUniformBuffers = new BuffersPerStage(Constants.TotalCpUniformBuffers);
 
@@ -144,6 +147,16 @@ public void SetVertexBuffer(int index, ulong gpuVa, ulong size, int stride, int
             }
         }
 
+        public void SetTransformFeedbackBuffer(int index, ulong gpuVa, ulong size)
+        {
+            ulong address = TranslateAndCreateBuffer(gpuVa, size);
+
+            _transformFeedbackBuffers[index].Address = address;
+            _transformFeedbackBuffers[index].Size    = size;
+
+            _transformFeedbackBuffersDirty = true;
+        }
+
         /// <summary>
         /// Sets a storage buffer on the compute pipeline.
         /// Storage buffers can be read and written to on shaders.
@@ -522,6 +535,41 @@ public void CommitGraphicsBindings()
                 }
             }
 
+            if (_transformFeedbackBuffersDirty)
+            {
+                _transformFeedbackBuffersDirty = false;
+
+                for (int index = 0; index < Constants.TotalTransformFeedbackBuffers; index++)
+                {
+                    BufferBounds tfb = _transformFeedbackBuffers[index];
+
+                    if (tfb.Address == 0)
+                    {
+                        _context.Renderer.Pipeline.SetTransformFeedbackBuffer(index, new BufferRange(BufferHandle.Null, 0, 0));
+
+                        continue;
+                    }
+
+                    BufferRange buffer = GetBufferRange(tfb.Address, tfb.Size);
+
+                    _context.Renderer.Pipeline.SetTransformFeedbackBuffer(index, buffer);
+                }
+            }
+            else
+            {
+                for (int index = 0; index < Constants.TotalTransformFeedbackBuffers; index++)
+                {
+                    BufferBounds tfb = _transformFeedbackBuffers[index];
+
+                    if (tfb.Address == 0)
+                    {
+                        continue;
+                    }
+
+                    SynchronizeBufferRange(tfb.Address, tfb.Size);
+                }
+            }
+
             if (_gpStorageBuffersDirty || _rebind)
             {
                 _gpStorageBuffersDirty = false;

Ryujinx.Graphics.Gpu/Memory/MemoryAccessor.cs

@@ -63,11 +63,11 @@ public T Read<T>(ulong gpuVa) where T : unmanaged
         /// </summary>
         /// <param name="gpuVa">GPU virtual address to write the value into</param>
         /// <param name="value">The value to be written</param>
-        public void Write(ulong gpuVa, int value)
+        public void Write<T>(ulong gpuVa, T value) where T : unmanaged
         {
             ulong processVa = _context.MemoryManager.Translate(gpuVa);
 
-            _context.PhysicalMemory.Write(processVa, BitConverter.GetBytes(value));
+            _context.PhysicalMemory.Write(processVa, MemoryMarshal.Cast<T, byte>(MemoryMarshal.CreateSpan(ref value, 1)));
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs

@@ -82,7 +82,7 @@ public ShaderBundle GetComputeShader(
 
             shader.HostShader = _context.Renderer.CompileShader(shader.Program);
 
-            IProgram hostProgram = _context.Renderer.CreateProgram(new IShader[] { shader.HostShader });
+            IProgram hostProgram = _context.Renderer.CreateProgram(new IShader[] { shader.HostShader }, null);
 
             ShaderBundle cpShader = new ShaderBundle(hostProgram, shader);
 
@@ -150,14 +150,16 @@ public ShaderBundle GetGraphicsShader(GpuState state, ShaderAddresses addresses)
                     continue;
                 }
 
+                var tfd = GetTransformFeedbackDescriptors(state);
+
                 IShader hostShader = _context.Renderer.CompileShader(program);
 
                 shaders[stage].HostShader = hostShader;
 
                 hostShaders.Add(hostShader);
             }
 
-            IProgram hostProgram = _context.Renderer.CreateProgram(hostShaders.ToArray());
+            IProgram hostProgram = _context.Renderer.CreateProgram(hostShaders.ToArray(), GetTransformFeedbackDescriptors(state));
 
             ShaderBundle gpShaders = new ShaderBundle(hostProgram, shaders);
 
@@ -173,6 +175,36 @@ public ShaderBundle GetGraphicsShader(GpuState state, ShaderAddresses addresses)
             return gpShaders;
         }
 
+        /// <summary>
+        /// Gets transform feedback state from the current GPU state.
+        /// </summary>
+        /// <param name="state">Current GPU state</param>
+        /// <returns>Four transform feedback descriptors for the enabled TFBs, or null if TFB is disabled</returns>
+        private TransformFeedbackDescriptor[] GetTransformFeedbackDescriptors(GpuState state)
+        {
+            bool tfEnable = state.Get<Boolean32>(MethodOffset.TfEnable);
+
+            if (!tfEnable)
+            {
+                return null;
+            }
+
+            TransformFeedbackDescriptor[] descs = new TransformFeedbackDescriptor[Constants.TotalTransformFeedbackBuffers];
+
+            for (int i = 0; i < Constants.TotalTransformFeedbackBuffers; i++)
+            {
+                var tf = state.Get<TfState>(MethodOffset.TfState, i);
+
+                int length = (int)Math.Min((uint)tf.VaryingsCount, 0x80);
+
+                var varyingLocations = state.GetSpan(MethodOffset.TfVaryingLocations + i * 0x80, length).ToArray();
+
+                descs[i] = new TransformFeedbackDescriptor(tf.BufferIndex, tf.Stride, varyingLocations);
+            }
+
+            return descs;
+        }
+
         /// <summary>
         /// Checks if compute shader code in memory is equal to the cached shader.
         /// </summary>

Ryujinx.Graphics.Gpu/State/GpuState.cs

@@ -346,7 +346,7 @@ public bool QueryModified(
         /// <param name="offset">Register offset</param>
         /// <param name="index">Index for indexed data</param>
         /// <returns>The data at the specified location</returns>
-        public T Get<T>(MethodOffset offset, int index) where T : struct
+        public T Get<T>(MethodOffset offset, int index) where T : unmanaged
         {
             Register register = _registers[(int)offset];
 
@@ -364,19 +364,30 @@ public T Get<T>(MethodOffset offset, int index) where T : struct
         /// <typeparam name="T">Type of the data</typeparam>
         /// <param name="offset">Register offset</param>
         /// <returns>The data at the specified location</returns>
-        public T Get<T>(MethodOffset offset) where T : struct
+        public T Get<T>(MethodOffset offset) where T : unmanaged
         {
             return MemoryMarshal.Cast<int, T>(_memory.AsSpan().Slice((int)offset))[0];
         }
 
+        /// <summary>
+        /// Gets a span of the data at a given register offset.
+        /// </summary>
+        /// <param name="offset">Register offset</param>
+        /// <param name="length">Length of the data in bytes</param>
+        /// <returns>The data at the specified location</returns>
+        public Span<byte> GetSpan(MethodOffset offset, int length)
+        {
+            return MemoryMarshal.Cast<int, byte>(_memory.AsSpan().Slice((int)offset)).Slice(0, length);
+        }
+
         /// <summary>
         /// Sets indexed data to a given register offset.
         /// </summary>
         /// <typeparam name="T">Type of the data</typeparam>
         /// <param name="offset">Register offset</param>
         /// <param name="index">Index for indexed data</param>
         /// <param name="data">The data to set</param>
-        public void Set<T>(MethodOffset offset, int index, T data) where T : struct
+        public void Set<T>(MethodOffset offset, int index, T data) where T : unmanaged
         {
             Register register = _registers[(int)offset];
 
@@ -394,7 +405,7 @@ public void Set<T>(MethodOffset offset, int index, T data) where T : struct
         /// <typeparam name="T">Type of the data</typeparam>
         /// <param name="offset">Register offset</param>
         /// <param name="data">The data to set</param>
-        public void Set<T>(MethodOffset offset, T data) where T : struct
+        public void Set<T>(MethodOffset offset, T data) where T : unmanaged
         {
             ReadOnlySpan<int> intSpan = MemoryMarshal.Cast<T, int>(MemoryMarshal.CreateReadOnlySpan(ref data, 1));
             intSpan.CopyTo(_memory.AsSpan().Slice((int)offset, intSpan.Length));