Implement transform feedback emulation for hardware without native support (#5080)

* Implement transform feedback emulation for hardware without native support

* Stop doing some useless buffer updates and account for non-zero base instance

* Reduce redundant updates even more

* Update descriptor init logic to account for ResourceLayout

* Fix transform feedback and storage buffers not being updated in some cases

* Shader cache version bump

* PR feedback

* SetInstancedDrawVertexCount must be always called after UpdateState

* Minor typo

src/Ryujinx.Graphics.Shader/Constants.cs

@@ -10,5 +10,11 @@ namespace Ryujinx.Graphics.Shader
         public const int NvnBaseVertexByteOffset = 0x640;
         public const int NvnBaseInstanceByteOffset = 0x644;
         public const int NvnDrawIndexByteOffset = 0x648;
+
+        // Transform Feedback emulation.
+
+        public const int TfeInfoBinding = 0;
+        public const int TfeBufferBaseBinding = 1;
+        public const int TfeBuffersCount = 4;
     }
 }

src/Ryujinx.Graphics.Shader/IGpuAccessor.cs

@@ -367,6 +367,15 @@ namespace Ryujinx.Graphics.Shader
             return true;
         }
 
+        /// <summary>
+        /// Queries host GPU transform feedback support.
+        /// </summary>
+        /// <returns>True if the GPU and driver supports transform feedback, false otherwise</returns>
+        bool QueryHostSupportsTransformFeedback()
+        {
+            return true;
+        }
+
         /// <summary>
         /// Queries host support for writes to the viewport index from vertex or tessellation shader stages.
         /// </summary>

src/Ryujinx.Graphics.Shader/Translation/EmitterContext.cs

@@ -234,6 +234,45 @@ namespace Ryujinx.Graphics.Shader.Translation
 
         public void PrepareForVertexReturn()
         {
+            if (!Config.GpuAccessor.QueryHostSupportsTransformFeedback() && Config.GpuAccessor.QueryTransformFeedbackEnabled())
+            {
+                Operand vertexCount = this.Load(StorageKind.StorageBuffer, Constants.TfeInfoBinding, Const(1));
+
+                for (int tfbIndex = 0; tfbIndex < Constants.TfeBuffersCount; tfbIndex++)
+                {
+                    var locations = Config.GpuAccessor.QueryTransformFeedbackVaryingLocations(tfbIndex);
+                    var stride = Config.GpuAccessor.QueryTransformFeedbackStride(tfbIndex);
+
+                    Operand baseOffset = this.Load(StorageKind.StorageBuffer, Constants.TfeInfoBinding, Const(0), Const(tfbIndex));
+                    Operand baseVertex = this.Load(StorageKind.Input, IoVariable.BaseVertex);
+                    Operand baseInstance = this.Load(StorageKind.Input, IoVariable.BaseInstance);
+                    Operand vertexIndex = this.Load(StorageKind.Input, IoVariable.VertexIndex);
+                    Operand instanceIndex = this.Load(StorageKind.Input, IoVariable.InstanceIndex);
+
+                    Operand outputVertexOffset = this.ISubtract(vertexIndex, baseVertex);
+                    Operand outputInstanceOffset = this.ISubtract(instanceIndex, baseInstance);
+
+                    Operand outputBaseVertex = this.IMultiply(outputInstanceOffset, vertexCount);
+
+                    Operand vertexOffset = this.IMultiply(this.IAdd(outputBaseVertex, outputVertexOffset), Const(stride / 4));
+                    baseOffset = this.IAdd(baseOffset, vertexOffset);
+
+                    for (int j = 0; j < locations.Length; j++)
+                    {
+                        byte location = locations[j];
+                        if (location == 0xff)
+                        {
+                            continue;
+                        }
+
+                        Operand offset = this.IAdd(baseOffset, Const(j));
+                        Operand value = Instructions.AttributeMap.GenerateAttributeLoad(this, null, location * 4, isOutput: true, isPerPatch: false);
+
+                        this.Store(StorageKind.StorageBuffer, Constants.TfeBufferBaseBinding + tfbIndex, Const(0), offset, value);
+                    }
+                }
+            }
+
             if (Config.GpuAccessor.QueryViewportTransformDisable())
             {
                 Operand x = this.Load(StorageKind.Output, IoVariable.Position, null, Const(0));

src/Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs

@@ -132,6 +132,11 @@ namespace Ryujinx.Graphics.Shader.Translation
 
             _transformFeedbackDefinitions = new Dictionary<TransformFeedbackVariable, TransformFeedbackOutput>();
 
+            TransformFeedbackEnabled =
+                stage != ShaderStage.Compute &&
+                gpuAccessor.QueryTransformFeedbackEnabled() &&
+                gpuAccessor.QueryHostSupportsTransformFeedback();
+
             UsedInputAttributesPerPatch  = new HashSet<int>();
             UsedOutputAttributesPerPatch = new HashSet<int>();
 
@@ -139,6 +144,31 @@ namespace Ryujinx.Graphics.Shader.Translation
             _usedImages   = new Dictionary<TextureInfo, TextureMeta>();
 
             ResourceManager = new ResourceManager(stage, gpuAccessor, new ShaderProperties());
+
+            if (!gpuAccessor.QueryHostSupportsTransformFeedback() && gpuAccessor.QueryTransformFeedbackEnabled())
+            {
+                StructureType tfeInfoStruct = new StructureType(new StructureField[]
+                {
+                    new StructureField(AggregateType.Array | AggregateType.U32, "base_offset", 4),
+                    new StructureField(AggregateType.U32, "vertex_count")
+                });
+
+                BufferDefinition tfeInfoBuffer = new BufferDefinition(BufferLayout.Std430, 1, Constants.TfeInfoBinding, "tfe_info", tfeInfoStruct);
+
+                Properties.AddStorageBuffer(Constants.TfeInfoBinding, tfeInfoBuffer);
+
+                StructureType tfeDataStruct = new StructureType(new StructureField[]
+                {
+                    new StructureField(AggregateType.Array | AggregateType.U32, "data", 0)
+                });
+
+                for (int i = 0; i < Constants.TfeBuffersCount; i++)
+                {
+                    int binding = Constants.TfeBufferBaseBinding + i;
+                    BufferDefinition tfeDataBuffer = new BufferDefinition(BufferLayout.Std430, 1, binding, $"tfe_data{i}", tfeDataStruct);
+                    Properties.AddStorageBuffer(binding, tfeDataBuffer);
+                }
+            }
         }
 
         public ShaderConfig(
@@ -151,7 +181,6 @@ namespace Ryujinx.Graphics.Shader.Translation
             ThreadsPerInputPrimitive = 1;
             OutputTopology           = outputTopology;
             MaxOutputVertices        = maxOutputVertices;
-            TransformFeedbackEnabled = gpuAccessor.QueryTransformFeedbackEnabled();
         }
 
         public ShaderConfig(ShaderHeader header, IGpuAccessor gpuAccessor, TranslationOptions options) : this(header.Stage, gpuAccessor, options)
@@ -165,7 +194,6 @@ namespace Ryujinx.Graphics.Shader.Translation
             OmapTargets              = header.OmapTargets;
             OmapSampleMask           = header.OmapSampleMask;
             OmapDepth                = header.OmapDepth;
-            TransformFeedbackEnabled = gpuAccessor.QueryTransformFeedbackEnabled();
             LastInVertexPipeline     = header.Stage < ShaderStage.Fragment;
         }