Geometry shader emulation for macOS (#5551)

* Implement vertex and geometry shader conversion to compute

* Call InitializeReservedCounts for compute too

* PR feedback

* Set clip distance mask for geometry and tessellation shaders too

* Transform feedback emulation only for vertex

11 files changed, 317 insertions, 81 deletions

diff --git a/src/Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs b/src/Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs
index ff9c39a1..600c8a98 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/CachedShaderProgram.cs
@@ -15,6 +15,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
         public IProgram HostProgram { get; }
 
         /// <summary>
+        /// Optional vertex shader converted to compute.
+        /// </summary>
+        public ShaderAsCompute VertexAsCompute { get; }
+
+        /// <summary>
+        /// Optional geometry shader converted to compute.
+        /// </summary>
+        public ShaderAsCompute GeometryAsCompute { get; }
+
+        /// <summary>
         /// GPU state used to create this version of the shader.
         /// </summary>
         public ShaderSpecializationState SpecializationState { get; }
@@ -45,12 +55,25 @@ namespace Ryujinx.Graphics.Gpu.Shader
             Bindings = new CachedShaderBindings(shaders.Length == 1, shaders);
         }
 
+        public CachedShaderProgram(
+            IProgram hostProgram,
+            ShaderAsCompute vertexAsCompute,
+            ShaderAsCompute geometryAsCompute,
+            ShaderSpecializationState specializationState,
+            CachedShaderStage[] shaders) : this(hostProgram, specializationState, shaders)
+        {
+            VertexAsCompute = vertexAsCompute;
+            GeometryAsCompute = geometryAsCompute;
+        }
+
         /// <summary>
         /// Dispose of the host shader resources.
         /// </summary>
         public void Dispose()
         {
             HostProgram.Dispose();
+            VertexAsCompute?.HostProgram.Dispose();
+            GeometryAsCompute?.HostProgram.Dispose();
         }
     }
 }
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs
index 93d293f6..de6432bc 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs
@@ -35,7 +35,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             ShaderSpecializationState oldSpecState,
             ShaderSpecializationState newSpecState,
             ResourceCounts counts,
-            int stageIndex) : base(context, counts, stageIndex, oldSpecState.TransformFeedbackDescriptors != null)
+            int stageIndex) : base(context, counts, stageIndex)
         {
             _data = data;
             _cb1Data = cb1Data;
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs
index 08096bd5..99ef89ed 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheHostStorage.cs
@@ -22,7 +22,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         private const ushort FileFormatVersionMajor = 1;
         private const ushort FileFormatVersionMinor = 2;
         private const uint FileFormatVersionPacked = ((uint)FileFormatVersionMajor << 16) | FileFormatVersionMinor;
-        private const uint CodeGenVersion = 5609;
+        private const uint CodeGenVersion = 5551;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";
@@ -141,6 +141,21 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             public ShaderStage Stage;
 
             /// <summary>
+            /// Number of vertices that each output primitive has on a geometry shader.
+            /// </summary>
+            public byte GeometryVerticesPerPrimitive;
+
+            /// <summary>
+            /// Maximum number of vertices that a geometry shader may generate.
+            /// </summary>
+            public ushort GeometryMaxOutputVertices;
+
+            /// <summary>
+            /// Number of invocations per primitive on tessellation or geometry shaders.
+            /// </summary>
+            public ushort ThreadsPerInputPrimitive;
+
+            /// <summary>
             /// Indicates if the fragment shader accesses the fragment coordinate built-in variable.
             /// </summary>
             public bool UsesFragCoord;
@@ -783,9 +798,10 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                 sBuffers,
                 textures,
                 images,
-                ShaderIdentification.None,
-                0,
                 dataInfo.Stage,
+                dataInfo.GeometryVerticesPerPrimitive,
+                dataInfo.GeometryMaxOutputVertices,
+                dataInfo.ThreadsPerInputPrimitive,
                 dataInfo.UsesFragCoord,
                 dataInfo.UsesInstanceId,
                 dataInfo.UsesDrawParameters,
@@ -813,6 +829,9 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                 TexturesCount = (ushort)info.Textures.Count,
                 ImagesCount = (ushort)info.Images.Count,
                 Stage = info.Stage,
+                GeometryVerticesPerPrimitive = (byte)info.GeometryVerticesPerPrimitive,
+                GeometryMaxOutputVertices = (ushort)info.GeometryMaxOutputVertices,
+                ThreadsPerInputPrimitive = (ushort)info.ThreadsPerInputPrimitive,
                 UsesFragCoord = info.UsesFragCoord,
                 UsesInstanceId = info.UsesInstanceId,
                 UsesDrawParameters = info.UsesDrawParameters,
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/ParallelDiskCacheLoader.cs b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/ParallelDiskCacheLoader.cs
index 8c2108bf..153fc442 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/ParallelDiskCacheLoader.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/DiskCache/ParallelDiskCacheLoader.cs
@@ -595,6 +595,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
 
             ResourceCounts counts = new();
 
+            DiskCacheGpuAccessor[] gpuAccessors = new DiskCacheGpuAccessor[Constants.ShaderStages];
             TranslatorContext[] translatorContexts = new TranslatorContext[Constants.ShaderStages + 1];
             TranslatorContext nextStage = null;
 
@@ -626,14 +627,22 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                         translatorContexts[0] = DecodeGraphicsShader(gpuAccessorA, api, DefaultFlags | TranslationFlags.VertexA, 0);
                     }
 
+                    gpuAccessors[stageIndex] = gpuAccessor;
                     translatorContexts[stageIndex + 1] = currentStage;
                     nextStage = currentStage;
                 }
             }
 
-            if (!_context.Capabilities.SupportsGeometryShader)
+            bool hasGeometryShader = translatorContexts[4] != null;
+            bool vertexHasStore = translatorContexts[1] != null && translatorContexts[1].HasStore;
+            bool geometryHasStore = hasGeometryShader && translatorContexts[4].HasStore;
+            bool vertexToCompute = ShouldConvertVertexToCompute(_context, vertexHasStore, geometryHasStore, hasGeometryShader);
+
+            // We don't support caching shader stages that have been converted to compute currently,
+            // so just eliminate them if they exist in the cache.
+            if (vertexToCompute)
             {
-                ShaderCache.TryRemoveGeometryStage(translatorContexts);
+                return;
             }
 
             CachedShaderStage[] shaders = new CachedShaderStage[guestShaders.Length];
@@ -647,6 +656,8 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
 
                 if (currentStage != null)
                 {
+                    gpuAccessors[stageIndex].InitializeReservedCounts(specState.TransformFeedbackDescriptors != null, vertexToCompute);
+
                     ShaderProgram program;
 
                     byte[] guestCode = guestShaders[stageIndex + 1].Value.Code;
@@ -701,6 +712,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             ResourceCounts counts = new();
             ShaderSpecializationState newSpecState = new(ref specState.ComputeState);
             DiskCacheGpuAccessor gpuAccessor = new(_context, shader.Code, shader.Cb1Data, specState, newSpecState, counts, 0);
+            gpuAccessor.InitializeReservedCounts(tfEnabled: false, vertexAsCompute: false);
 
             TranslatorContext translatorContext = DecodeComputeShader(gpuAccessor, _context.Capabilities.Api, 0);
 
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs b/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs
index b4f4a439..1d84d0e4 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessor.cs
@@ -25,11 +25,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="channel">GPU channel</param>
         /// <param name="state">Current GPU state</param>
         /// <param name="stageIndex">Graphics shader stage index (0 = Vertex, 4 = Fragment)</param>
-        public GpuAccessor(
-            GpuContext context,
-            GpuChannel channel,
-            GpuAccessorState state,
-            int stageIndex) : base(context, state.ResourceCounts, stageIndex, state.TransformFeedbackDescriptors != null)
+        public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state, int stageIndex) : base(context, state.ResourceCounts, stageIndex)
         {
             _isVulkan = context.Capabilities.Api == TargetApi.Vulkan;
             _channel = channel;
@@ -49,7 +45,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="context">GPU context</param>
         /// <param name="channel">GPU channel</param>
         /// <param name="state">Current GPU state</param>
-        public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state) : base(context, state.ResourceCounts, 0, false)
+        public GpuAccessor(GpuContext context, GpuChannel channel, GpuAccessorState state) : base(context, state.ResourceCounts, 0)
         {
             _channel = channel;
             _state = state;
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs b/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs
index 52193940..9d030cd6 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs
@@ -15,8 +15,10 @@ namespace Ryujinx.Graphics.Gpu.Shader
         private readonly ResourceCounts _resourceCounts;
         private readonly int _stageIndex;
 
-        private readonly int _reservedConstantBuffers;
-        private readonly int _reservedStorageBuffers;
+        private int _reservedConstantBuffers;
+        private int _reservedStorageBuffers;
+        private int _reservedTextures;
+        private int _reservedImages;
 
         /// <summary>
         /// Creates a new GPU accessor.
@@ -24,15 +26,26 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="context">GPU context</param>
         /// <param name="resourceCounts">Counter of GPU resources used by the shader</param>
         /// <param name="stageIndex">Index of the shader stage, 0 for compute</param>
-        /// <param name="tfEnabled">Indicates if the current graphics shader is used with transform feedback enabled</param>
-        public GpuAccessorBase(GpuContext context, ResourceCounts resourceCounts, int stageIndex, bool tfEnabled)
+        public GpuAccessorBase(GpuContext context, ResourceCounts resourceCounts, int stageIndex)
         {
             _context = context;
             _resourceCounts = resourceCounts;
             _stageIndex = stageIndex;
+        }
+
+        /// <summary>
+        /// Initializes counts for bindings that will be reserved for emulator use.
+        /// </summary>
+        /// <param name="tfEnabled">Indicates if the current graphics shader is used with transform feedback enabled</param>
+        /// <param name="vertexAsCompute">Indicates that the vertex shader will be emulated on a compute shader</param>
+        public void InitializeReservedCounts(bool tfEnabled, bool vertexAsCompute)
+        {
+            ResourceReservationCounts rrc = new(!_context.Capabilities.SupportsTransformFeedback && tfEnabled, vertexAsCompute);
 
-            _reservedConstantBuffers = 1; // For the support buffer.
-            _reservedStorageBuffers = !context.Capabilities.SupportsTransformFeedback && tfEnabled ? 5 : 0;
+            _reservedConstantBuffers = rrc.ReservedConstantBuffers;
+            _reservedStorageBuffers = rrc.ReservedStorageBuffers;
+            _reservedTextures = rrc.ReservedTextures;
+            _reservedImages = rrc.ReservedImages;
         }
 
         public int QueryBindingConstantBuffer(int index)
@@ -69,6 +82,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
         public int QueryBindingTexture(int index, bool isBuffer)
         {
+            int binding;
+
             if (_context.Capabilities.Api == TargetApi.Vulkan)
             {
                 if (isBuffer)
@@ -76,16 +91,20 @@ namespace Ryujinx.Graphics.Gpu.Shader
                     index += (int)_context.Capabilities.MaximumTexturesPerStage;
                 }
 
-                return GetBindingFromIndex(index, _context.Capabilities.MaximumTexturesPerStage * 2, "Texture");
+                binding = GetBindingFromIndex(index, _context.Capabilities.MaximumTexturesPerStage * 2, "Texture");
             }
             else
             {
-                return _resourceCounts.TexturesCount++;
+                binding = _resourceCounts.TexturesCount++;
             }
+
+            return binding + _reservedTextures;
         }
 
         public int QueryBindingImage(int index, bool isBuffer)
         {
+            int binding;
+
             if (_context.Capabilities.Api == TargetApi.Vulkan)
             {
                 if (isBuffer)
@@ -93,12 +112,14 @@ namespace Ryujinx.Graphics.Gpu.Shader
                     index += (int)_context.Capabilities.MaximumImagesPerStage;
                 }
 
-                return GetBindingFromIndex(index, _context.Capabilities.MaximumImagesPerStage * 2, "Image");
+                binding = GetBindingFromIndex(index, _context.Capabilities.MaximumImagesPerStage * 2, "Image");
             }
             else
             {
-                return _resourceCounts.ImagesCount++;
+                binding = _resourceCounts.ImagesCount++;
             }
+
+            return binding + _reservedImages;
         }
 
         private int GetBindingFromIndex(int index, uint maxPerStage, string resourceName)
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderAsCompute.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderAsCompute.cs
new file mode 100644
index 00000000..71540a13
--- /dev/null
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderAsCompute.cs
@@ -0,0 +1,20 @@
+using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Shader;
+using Ryujinx.Graphics.Shader.Translation;
+
+namespace Ryujinx.Graphics.Gpu.Shader
+{
+    class ShaderAsCompute
+    {
+        public IProgram HostProgram { get; }
+        public ShaderProgramInfo Info { get; }
+        public ResourceReservations Reservations { get; }
+
+        public ShaderAsCompute(IProgram hostProgram, ShaderProgramInfo info, ResourceReservations reservations)
+        {
+            HostProgram = hostProgram;
+            Info = info;
+            Reservations = reservations;
+        }
+    }
+}
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs
index 97d7a720..38be262a 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderCache.cs
@@ -215,9 +215,10 @@ namespace Ryujinx.Graphics.Gpu.Shader
             ShaderSpecializationState specState = new(ref computeState);
             GpuAccessorState gpuAccessorState = new(poolState, computeState, default, specState);
             GpuAccessor gpuAccessor = new(_context, channel, gpuAccessorState);
+            gpuAccessor.InitializeReservedCounts(tfEnabled: false, vertexAsCompute: false);
 
             TranslatorContext translatorContext = DecodeComputeShader(gpuAccessor, _context.Capabilities.Api, gpuVa);
-            TranslatedShader translatedShader = TranslateShader(_dumper, channel, translatorContext, cachedGuestCode);
+            TranslatedShader translatedShader = TranslateShader(_dumper, channel, translatorContext, cachedGuestCode, asCompute: false);
 
             ShaderSource[] shaderSourcesArray = new ShaderSource[] { CreateShaderSource(translatedShader.Program) };
             ShaderInfo info = ShaderInfoBuilder.BuildForCompute(_context, translatedShader.Program.Info);
@@ -321,6 +322,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
             ReadOnlySpan<ulong> addressesSpan = addresses.AsSpan();
 
+            GpuAccessor[] gpuAccessors = new GpuAccessor[Constants.ShaderStages];
             TranslatorContext[] translatorContexts = new TranslatorContext[Constants.ShaderStages + 1];
             TranslatorContext nextStage = null;
 
@@ -345,22 +347,31 @@ namespace Ryujinx.Graphics.Gpu.Shader
                         translatorContexts[0] = DecodeGraphicsShader(gpuAccessor, api, DefaultFlags | TranslationFlags.VertexA, addresses.VertexA);
                     }
 
+                    gpuAccessors[stageIndex] = gpuAccessor;
                     translatorContexts[stageIndex + 1] = currentStage;
                     nextStage = currentStage;
                 }
             }
 
-            if (!_context.Capabilities.SupportsGeometryShader)
-            {
-                TryRemoveGeometryStage(translatorContexts);
-            }
+            bool hasGeometryShader = translatorContexts[4] != null;
+            bool vertexHasStore = translatorContexts[1] != null && translatorContexts[1].HasStore;
+            bool geometryHasStore = hasGeometryShader && translatorContexts[4].HasStore;
+            bool vertexToCompute = ShouldConvertVertexToCompute(_context, vertexHasStore, geometryHasStore, hasGeometryShader);
+            bool geometryToCompute = ShouldConvertGeometryToCompute(_context, geometryHasStore);
 
             CachedShaderStage[] shaders = new CachedShaderStage[Constants.ShaderStages + 1];
             List<ShaderSource> shaderSources = new();
 
             TranslatorContext previousStage = null;
+            ShaderInfoBuilder infoBuilder = new(_context, transformFeedbackDescriptors != null, vertexToCompute);
+
+            if (geometryToCompute && translatorContexts[4] != null)
+            {
+                translatorContexts[4].SetVertexOutputMapForGeometryAsCompute(translatorContexts[1]);
+            }
 
-            ShaderInfoBuilder infoBuilder = new(_context, transformFeedbackDescriptors != null);
+            ShaderAsCompute vertexAsCompute = null;
+            ShaderAsCompute geometryAsCompute = null;
 
             for (int stageIndex = 0; stageIndex < Constants.ShaderStages; stageIndex++)
             {
@@ -368,8 +379,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
                 if (currentStage != null)
                 {
+                    gpuAccessors[stageIndex].InitializeReservedCounts(transformFeedbackDescriptors != null, vertexToCompute);
+
                     ShaderProgram program;
 
+                    bool asCompute = (stageIndex == 0 && vertexToCompute) || (stageIndex == 3 && geometryToCompute);
+
                     if (stageIndex == 0 && translatorContexts[0] != null)
                     {
                         TranslatedShaderVertexPair translatedShader = TranslateShader(
@@ -378,7 +393,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
                             currentStage,
                             translatorContexts[0],
                             cachedGuestCode.VertexACode,
-                            cachedGuestCode.VertexBCode);
+                            cachedGuestCode.VertexBCode,
+                            asCompute);
 
                         shaders[0] = translatedShader.VertexA;
                         shaders[1] = translatedShader.VertexB;
@@ -388,12 +404,31 @@ namespace Ryujinx.Graphics.Gpu.Shader
                     {
                         byte[] code = cachedGuestCode.GetByIndex(stageIndex);
 
-                        TranslatedShader translatedShader = TranslateShader(_dumper, channel, currentStage, code);
+                        TranslatedShader translatedShader = TranslateShader(_dumper, channel, currentStage, code, asCompute);
 
                         shaders[stageIndex + 1] = translatedShader.Shader;
                         program = translatedShader.Program;
                     }
 
+                    if (asCompute)
+                    {
+                        bool tfEnabled = transformFeedbackDescriptors != null;
+
+                        if (stageIndex == 0)
+                        {
+                            vertexAsCompute = CreateHostVertexAsComputeProgram(program, currentStage, tfEnabled);
+
+                            TranslatorContext lastInVertexPipeline = geometryToCompute ? translatorContexts[4] ?? currentStage : currentStage;
+
+                            program = lastInVertexPipeline.GenerateVertexPassthroughForCompute();
+                        }
+                        else
+                        {
+                            geometryAsCompute = CreateHostVertexAsComputeProgram(program, currentStage, tfEnabled);
+                            program = null;
+                        }
+                    }
+
                     if (program != null)
                     {
                         shaderSources.Add(CreateShaderSource(program));
@@ -418,46 +453,81 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
             IProgram hostProgram = _context.Renderer.CreateProgram(shaderSourcesArray, info);
 
-            gpShaders = new CachedShaderProgram(hostProgram, specState, shaders);
+            gpShaders = new(hostProgram, vertexAsCompute, geometryAsCompute, specState, shaders);
 
             _graphicsShaderCache.Add(gpShaders);
-            EnqueueProgramToSave(gpShaders, hostProgram, shaderSourcesArray);
+
+            // We don't currently support caching shaders that have been converted to compute.
+            if (vertexAsCompute == null)
+            {
+                EnqueueProgramToSave(gpShaders, hostProgram, shaderSourcesArray);
+            }
+
             _gpPrograms[addresses] = gpShaders;
 
             return gpShaders;
         }
 
         /// <summary>
-        /// Tries to eliminate the geometry stage from the array of translator contexts.
+        /// Checks if a vertex shader should be converted to a compute shader due to it making use of
+        /// features that are not supported on the host.
         /// </summary>
-        /// <param name="translatorContexts">Array of translator contexts</param>
-        public static void TryRemoveGeometryStage(TranslatorContext[] translatorContexts)
+        /// <param name="context">GPU context of the shader</param>
+        /// <param name="vertexHasStore">Whether the vertex shader has image or storage buffer store operations</param>
+        /// <param name="geometryHasStore">Whether the geometry shader has image or storage buffer store operations, if one exists</param>
+        /// <param name="hasGeometryShader">Whether a geometry shader exists</param>
+        /// <returns>True if the vertex shader should be converted to compute, false otherwise</returns>
+        public static bool ShouldConvertVertexToCompute(GpuContext context, bool vertexHasStore, bool geometryHasStore, bool hasGeometryShader)
         {
-            if (translatorContexts[4] != null)
+            // If the host does not support store operations on vertex,
+            // we need to emulate it on a compute shader.
+            if (!context.Capabilities.SupportsVertexStoreAndAtomics && vertexHasStore)
             {
-                // We have a geometry shader, but geometry shaders are not supported.
-                // Try to eliminate the geometry shader.
+                return true;
+            }
 
-                ShaderProgramInfo info = translatorContexts[4].Translate().Info;
+            // If any stage after the vertex stage is converted to compute,
+            // we need to convert vertex to compute too.
+            return hasGeometryShader && ShouldConvertGeometryToCompute(context, geometryHasStore);
+        }
 
-                if (info.Identification == ShaderIdentification.GeometryLayerPassthrough)
-                {
-                    // We managed to identify that this geometry shader is only used to set the output Layer value,
-                    // we can set the Layer on the previous stage instead (usually the vertex stage) and eliminate it.
+        /// <summary>
+        /// Checks if a geometry shader should be converted to a compute shader due to it making use of
+        /// features that are not supported on the host.
+        /// </summary>
+        /// <param name="context">GPU context of the shader</param>
+        /// <param name="geometryHasStore">Whether the geometry shader has image or storage buffer store operations, if one exists</param>
+        /// <returns>True if the geometry shader should be converted to compute, false otherwise</returns>
+        public static bool ShouldConvertGeometryToCompute(GpuContext context, bool geometryHasStore)
+        {
+            return (!context.Capabilities.SupportsVertexStoreAndAtomics && geometryHasStore) ||
+                   !context.Capabilities.SupportsGeometryShader;
+        }
 
-                    for (int i = 3; i >= 1; i--)
-                    {
-                        if (translatorContexts[i] != null)
-                        {
-                            translatorContexts[i].SetGeometryShaderLayerInputAttribute(info.GpLayerInputAttribute);
-                            translatorContexts[i].SetLastInVertexPipeline();
-                            break;
-                        }
-                    }
+        /// <summary>
+        /// Checks if it might be necessary for any vertex, tessellation or geometry shader to be converted to compute,
+        /// based on the supported host features.
+        /// </summary>
+        /// <param name="capabilities">Host capabilities</param>
+        /// <returns>True if the possibility of a shader being converted to compute exists, false otherwise</returns>
+        public static bool MayConvertVtgToCompute(ref Capabilities capabilities)
+        {
+            return !capabilities.SupportsVertexStoreAndAtomics || !capabilities.SupportsGeometryShader;
+        }
 
-                    translatorContexts[4] = null;
-                }
-            }
+        /// <summary>
+        /// Creates a compute shader from a vertex, tessellation or geometry shader that has been converted to compute.
+        /// </summary>
+        /// <param name="program">Shader program</param>
+        /// <param name="context">Translation context of the shader</param>
+        /// <param name="tfEnabled">Whether transform feedback is enabled</param>
+        /// <returns>Compute shader</returns>
+        private ShaderAsCompute CreateHostVertexAsComputeProgram(ShaderProgram program, TranslatorContext context, bool tfEnabled)
+        {
+            ShaderSource source = new(program.Code, program.BinaryCode, ShaderStage.Compute, program.Language);
+            ShaderInfo info = ShaderInfoBuilder.BuildForVertexAsCompute(_context, program.Info, tfEnabled);
+
+            return new(_context.Renderer.CreateProgram(new[] { source }, info), program.Info, context.GetResourceReservations());
         }
 
         /// <summary>
@@ -573,9 +643,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 }
             }
 
+            bool vertexAsCompute = gpShaders.VertexAsCompute != null;
             bool usesDrawParameters = gpShaders.Shaders[1]?.Info.UsesDrawParameters ?? false;
 
-            return gpShaders.SpecializationState.MatchesGraphics(channel, ref poolState, ref graphicsState, usesDrawParameters, true);
+            return gpShaders.SpecializationState.MatchesGraphics(
+                channel,
+                ref poolState,
+                ref graphicsState,
+                vertexAsCompute,
+                usesDrawParameters,
+                checkTextures: true);
         }
 
         /// <summary>
@@ -636,6 +713,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="vertexA">Optional translator context of the shader that should be combined</param>
         /// <param name="codeA">Optional Maxwell binary code of the Vertex A shader, if present</param>
         /// <param name="codeB">Optional Maxwell binary code of the Vertex B or current stage shader, if present on cache</param>
+        /// <param name="asCompute">Indicates that the vertex shader should be converted to a compute shader</param>
         /// <returns>Compiled graphics shader code</returns>
         private static TranslatedShaderVertexPair TranslateShader(
             ShaderDumper dumper,
@@ -643,7 +721,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
             TranslatorContext currentStage,
             TranslatorContext vertexA,
             byte[] codeA,
-            byte[] codeB)
+            byte[] codeB,
+            bool asCompute)
         {
             ulong cb1DataAddress = channel.BufferManager.GetGraphicsUniformBufferAddress(0, 1);
 
@@ -663,7 +742,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 pathsB = dumper.Dump(codeB, compute: false);
             }
 
-            ShaderProgram program = currentStage.Translate(vertexA);
+            ShaderProgram program = currentStage.Translate(vertexA, asCompute);
 
             pathsB.Prepend(program);
             pathsA.Prepend(program);
@@ -681,8 +760,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="channel">GPU channel using the shader</param>
         /// <param name="context">Translator context of the stage to be translated</param>
         /// <param name="code">Optional Maxwell binary code of the current stage shader, if present on cache</param>
+        /// <param name="asCompute">Indicates that the vertex shader should be converted to a compute shader</param>
         /// <returns>Compiled graphics shader code</returns>
-        private static TranslatedShader TranslateShader(ShaderDumper dumper, GpuChannel channel, TranslatorContext context, byte[] code)
+        private static TranslatedShader TranslateShader(ShaderDumper dumper, GpuChannel channel, TranslatorContext context, byte[] code, bool asCompute)
         {
             var memoryManager = channel.MemoryManager;
 
@@ -694,7 +774,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
             code ??= memoryManager.GetSpan(context.Address, context.Size).ToArray();
 
             ShaderDumpPaths paths = dumper?.Dump(code, context.Stage == ShaderStage.Compute) ?? default;
-            ShaderProgram program = context.Translate();
+            ShaderProgram program = context.Translate(asCompute);
 
             paths.Prepend(program);
 
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs
index af1e1ee3..bea916a6 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderInfoBuilder.cs
@@ -33,6 +33,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
         private readonly int _reservedConstantBuffers;
         private readonly int _reservedStorageBuffers;
+        private readonly int _reservedTextures;
+        private readonly int _reservedImages;
 
         private readonly List<ResourceDescriptor>[] _resourceDescriptors;
         private readonly List<ResourceUsage>[] _resourceUsages;
@@ -42,7 +44,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         /// <param name="context">GPU context that owns the shaders that will be added to the builder</param>
         /// <param name="tfEnabled">Indicates if the graphics shader is used with transform feedback enabled</param>
-        public ShaderInfoBuilder(GpuContext context, bool tfEnabled)
+        /// <param name="vertexAsCompute">Indicates that the vertex shader will be emulated on a compute shader</param>
+        public ShaderInfoBuilder(GpuContext context, bool tfEnabled, bool vertexAsCompute = false)
         {
             _context = context;
 
@@ -60,27 +63,34 @@ namespace Ryujinx.Graphics.Gpu.Shader
             AddDescriptor(SupportBufferStages, ResourceType.UniformBuffer, UniformSetIndex, 0, 1);
             AddUsage(SupportBufferStages, ResourceType.UniformBuffer, ResourceAccess.Read, UniformSetIndex, 0, 1);
 
-            _reservedConstantBuffers = 1; // For the support buffer.
+            ResourceReservationCounts rrc = new(!context.Capabilities.SupportsTransformFeedback && tfEnabled, vertexAsCompute);
 
-            if (!context.Capabilities.SupportsTransformFeedback && tfEnabled)
-            {
-                _reservedStorageBuffers = 5;
+            _reservedConstantBuffers = rrc.ReservedConstantBuffers;
+            _reservedStorageBuffers = rrc.ReservedStorageBuffers;
+            _reservedTextures = rrc.ReservedTextures;
+            _reservedImages = rrc.ReservedImages;
 
-                AddDescriptor(VtgStages, ResourceType.StorageBuffer, StorageSetIndex, 0, 5);
-                AddUsage(VtgStages, ResourceType.StorageBuffer, ResourceAccess.Read, StorageSetIndex, 0, 1);
-                AddUsage(VtgStages, ResourceType.StorageBuffer, ResourceAccess.Write, StorageSetIndex, 1, 4);
-            }
-            else
-            {
-                _reservedStorageBuffers = 0;
-            }
+            // TODO: Handle that better? Maybe we should only set the binding that are really needed on each shader.
+            ResourceStages stages = vertexAsCompute ? ResourceStages.Compute | ResourceStages.Vertex : VtgStages;
+
+            PopulateDescriptorAndUsages(stages, ResourceType.UniformBuffer, ResourceAccess.Read, UniformSetIndex, 1, rrc.ReservedConstantBuffers - 1);
+            PopulateDescriptorAndUsages(stages, ResourceType.StorageBuffer, ResourceAccess.ReadWrite, StorageSetIndex, 0, rrc.ReservedStorageBuffers);
+            PopulateDescriptorAndUsages(stages, ResourceType.BufferTexture, ResourceAccess.Read, TextureSetIndex, 0, rrc.ReservedTextures);
+            PopulateDescriptorAndUsages(stages, ResourceType.BufferImage, ResourceAccess.ReadWrite, ImageSetIndex, 0, rrc.ReservedImages);
+        }
+
+        private void PopulateDescriptorAndUsages(ResourceStages stages, ResourceType type, ResourceAccess access, int setIndex, int start, int count)
+        {
+            AddDescriptor(stages, type, setIndex, start, count);
+            AddUsage(stages, type, access, setIndex, start, count);
         }
 
         /// <summary>
         /// Adds information from a given shader stage.
         /// </summary>
         /// <param name="info">Shader stage information</param>
-        public void AddStageInfo(ShaderProgramInfo info)
+        /// <param name="vertexAsCompute">True if the shader stage has been converted into a compute shader</param>
+        public void AddStageInfo(ShaderProgramInfo info, bool vertexAsCompute = false)
         {
             if (info.Stage == ShaderStage.Fragment)
             {
@@ -96,7 +106,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 _ => 0,
             });
 
-            ResourceStages stages = info.Stage switch
+            ResourceStages stages = vertexAsCompute ? ResourceStages.Compute : info.Stage switch
             {
                 ShaderStage.Compute => ResourceStages.Compute,
                 ShaderStage.Vertex => ResourceStages.Vertex,
@@ -114,8 +124,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
             int uniformBinding = _reservedConstantBuffers + stageIndex * uniformsPerStage;
             int storageBinding = _reservedStorageBuffers + stageIndex * storagesPerStage;
-            int textureBinding = stageIndex * texturesPerStage * 2;
-            int imageBinding = stageIndex * imagesPerStage * 2;
+            int textureBinding = _reservedTextures + stageIndex * texturesPerStage * 2;
+            int imageBinding = _reservedImages + stageIndex * imagesPerStage * 2;
 
             AddDescriptor(stages, ResourceType.UniformBuffer, UniformSetIndex, uniformBinding, uniformsPerStage);
             AddDescriptor(stages, ResourceType.StorageBuffer, StorageSetIndex, storageBinding, storagesPerStage);
@@ -285,11 +295,28 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <returns>Shader information</returns>
         public static ShaderInfo BuildForCompute(GpuContext context, ShaderProgramInfo info, bool fromCache = false)
         {
-            ShaderInfoBuilder builder = new(context, tfEnabled: false);
+            ShaderInfoBuilder builder = new(context, tfEnabled: false, vertexAsCompute: false);
 
             builder.AddStageInfo(info);
 
             return builder.Build(null, fromCache);
         }
+
+        /// <summary>
+        /// Builds shader information for a vertex or geometry shader thas was converted to compute shader.
+        /// </summary>
+        /// <param name="context">GPU context that owns the shader</param>
+        /// <param name="info">Compute shader information</param>
+        /// <param name="tfEnabled">Indicates if the graphics shader is used with transform feedback enabled</param>
+        /// <param name="fromCache">True if the compute shader comes from a disk cache, false otherwise</param>
+        /// <returns>Shader information</returns>
+        public static ShaderInfo BuildForVertexAsCompute(GpuContext context, ShaderProgramInfo info, bool tfEnabled, bool fromCache = false)
+        {
+            ShaderInfoBuilder builder = new(context, tfEnabled, vertexAsCompute: true);
+
+            builder.AddStageInfo(info, vertexAsCompute: true);
+
+            return builder.Build(null, fromCache);
+        }
     }
 }
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs
index e57e1df1..3c2f0b9b 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationList.cs
@@ -35,9 +35,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
         {
             foreach (var entry in _entries)
             {
+                bool vertexAsCompute = entry.VertexAsCompute != null;
                 bool usesDrawParameters = entry.Shaders[1]?.Info.UsesDrawParameters ?? false;
 
-                if (entry.SpecializationState.MatchesGraphics(channel, ref poolState, ref graphicsState, usesDrawParameters, true))
+                if (entry.SpecializationState.MatchesGraphics(
+                    channel,
+                    ref poolState,
+                    ref graphicsState,
+                    vertexAsCompute,
+                    usesDrawParameters,
+                    checkTextures: true))
                 {
                     program = entry;
                     return true;
diff --git a/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs b/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs
index fcd95375..a41f761b 100644
--- a/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs
+++ b/src/Ryujinx.Graphics.Gpu/Shader/ShaderSpecializationState.cs
@@ -457,6 +457,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="channel">GPU channel</param>
         /// <param name="poolState">Texture pool state</param>
         /// <param name="graphicsState">Graphics state</param>
+        /// <param name="vertexAsCompute">Indicates that the vertex shader has been converted into a compute shader</param>
         /// <param name="usesDrawParameters">Indicates whether the vertex shader accesses draw parameters</param>
         /// <param name="checkTextures">Indicates whether texture descriptors should be checked</param>
         /// <returns>True if the state matches, false otherwise</returns>
@@ -464,6 +465,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
             GpuChannel channel,
             ref GpuChannelPoolState poolState,
             ref GpuChannelGraphicsState graphicsState,
+            bool vertexAsCompute,
             bool usesDrawParameters,
             bool checkTextures)
         {
@@ -497,9 +499,25 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 return false;
             }
 
-            if (!graphicsState.AttributeTypes.AsSpan().SequenceEqual(GraphicsState.AttributeTypes.AsSpan()))
+            if (ShaderCache.MayConvertVtgToCompute(ref channel.Capabilities) && !vertexAsCompute)
             {
-                return false;
+                for (int index = 0; index < graphicsState.AttributeTypes.Length; index++)
+                {
+                    AttributeType lType = FilterAttributeType(channel, graphicsState.AttributeTypes[index]);
+                    AttributeType rType = FilterAttributeType(channel, GraphicsState.AttributeTypes[index]);
+
+                    if (lType != rType)
+                    {
+                        return false;
+                    }
+                }
+            }
+            else
+            {
+                if (!graphicsState.AttributeTypes.AsSpan().SequenceEqual(GraphicsState.AttributeTypes.AsSpan()))
+                {
+                    return false;
+                }
             }
 
             if (usesDrawParameters && graphicsState.HasConstantBufferDrawParameters != GraphicsState.HasConstantBufferDrawParameters)
@@ -530,6 +548,19 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return Matches(channel, ref poolState, checkTextures, isCompute: false);
         }
 
+        private static AttributeType FilterAttributeType(GpuChannel channel, AttributeType type)
+        {
+            type &= ~(AttributeType.Packed | AttributeType.PackedRgb10A2Signed);
+
+            if (channel.Capabilities.SupportsScaledVertexFormats &&
+                (type == AttributeType.Sscaled || type == AttributeType.Uscaled))
+            {
+                type = AttributeType.Float;
+            }
+
+            return type;
+        }
+
         /// <summary>
         /// Checks if the recorded state matches the current GPU compute engine state.
         /// </summary>