Geometry shader passthrough emulation

Ryujinx.Graphics.GAL/Capabilities.cs

@@ -16,6 +16,7 @@ namespace Ryujinx.Graphics.GAL
         public readonly bool SupportsR4G4Format;
         public readonly bool SupportsFragmentShaderInterlock;
         public readonly bool SupportsFragmentShaderOrderingIntel;
+        public readonly bool SupportsGeometryShaderPassthrough;
         public readonly bool SupportsImageLoadFormatted;
         public readonly bool SupportsMismatchingViewFormat;
         public readonly bool SupportsNonConstantTextureOffset;
@@ -44,6 +45,7 @@ namespace Ryujinx.Graphics.GAL
             bool supportsR4G4Format,
             bool supportsFragmentShaderInterlock,
             bool supportsFragmentShaderOrderingIntel,
+            bool supportsGeometryShaderPassthrough,
             bool supportsImageLoadFormatted,
             bool supportsMismatchingViewFormat,
             bool supportsNonConstantTextureOffset,
@@ -69,6 +71,7 @@ namespace Ryujinx.Graphics.GAL
             SupportsR4G4Format = supportsR4G4Format;
             SupportsFragmentShaderInterlock = supportsFragmentShaderInterlock;
             SupportsFragmentShaderOrderingIntel = supportsFragmentShaderOrderingIntel;
+            SupportsGeometryShaderPassthrough = supportsGeometryShaderPassthrough;
             SupportsImageLoadFormatted = supportsImageLoadFormatted;
             SupportsMismatchingViewFormat = supportsMismatchingViewFormat;
             SupportsNonConstantTextureOffset = supportsNonConstantTextureOffset;

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 = 1;
+        private const uint CodeGenVersion = 2;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs

@@ -105,64 +105,37 @@ namespace Ryujinx.Graphics.Gpu.Shader
             };
         }
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host about the presence of the FrontFacing built-in variable bug.
-        /// </summary>
-        /// <returns>True if the bug is present on the host device used, false otherwise</returns>
         public bool QueryHostHasFrontFacingBug() => _context.Capabilities.HasFrontFacingBug;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host about the presence of the vector indexing bug.
-        /// </summary>
-        /// <returns>True if the bug is present on the host device used, false otherwise</returns>
         public bool QueryHostHasVectorIndexingBug() => _context.Capabilities.HasVectorIndexingBug;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host storage buffer alignment required.
-        /// </summary>
-        /// <returns>Host storage buffer alignment in bytes</returns>
         public int QueryHostStorageBufferOffsetAlignment() => _context.Capabilities.StorageBufferOffsetAlignment;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host support for texture formats with BGRA component order (such as BGRA8).
-        /// </summary>
-        /// <returns>True if BGRA formats are supported, false otherwise</returns>
         public bool QueryHostSupportsBgraFormat() => _context.Capabilities.SupportsBgraFormat;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host support for fragment shader ordering critical sections on the shader code.
-        /// </summary>
-        /// <returns>True if fragment shader interlock is supported, false otherwise</returns>
         public bool QueryHostSupportsFragmentShaderInterlock() => _context.Capabilities.SupportsFragmentShaderInterlock;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host support for fragment shader ordering scoped critical sections on the shader code.
-        /// </summary>
-        /// <returns>True if fragment shader ordering is supported, false otherwise</returns>
         public bool QueryHostSupportsFragmentShaderOrderingIntel() => _context.Capabilities.SupportsFragmentShaderOrderingIntel;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host support for readable images without a explicit format declaration on the shader.
+        public bool QueryHostSupportsGeometryShaderPassthrough() => _context.Capabilities.SupportsGeometryShaderPassthrough;
-        /// </summary>
+
-        /// <returns>True if formatted image load is supported, false otherwise</returns>
+        /// <inheritdoc/>
         public bool QueryHostSupportsImageLoadFormatted() => _context.Capabilities.SupportsImageLoadFormatted;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host GPU non-constant texture offset support.
-        /// </summary>
-        /// <returns>True if the GPU and driver supports non-constant texture offsets, false otherwise</returns>
         public bool QueryHostSupportsNonConstantTextureOffset() => _context.Capabilities.SupportsNonConstantTextureOffset;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host GPU shader ballot support.
-        /// </summary>
-        /// <returns>True if the GPU and driver supports shader ballot, false otherwise</returns>
         public bool QueryHostSupportsShaderBallot() => _context.Capabilities.SupportsShaderBallot;
 
-        /// <summary>
+        /// <inheritdoc/>
-        /// Queries host GPU texture shadow LOD support.
-        /// </summary>
-        /// <returns>True if the GPU and driver supports texture shadow LOD, false otherwise</returns>
         public bool QueryHostSupportsTextureShadowLod() => _context.Capabilities.SupportsTextureShadowLod;
 
         /// <summary>

Ryujinx.Graphics.OpenGL/HwCapabilities.cs

@@ -9,6 +9,7 @@ namespace Ryujinx.Graphics.OpenGL
         private static readonly Lazy<bool> _supportsDrawTexture               = new Lazy<bool>(() => HasExtension("GL_NV_draw_texture"));
         private static readonly Lazy<bool> _supportsFragmentShaderInterlock   = new Lazy<bool>(() => HasExtension("GL_ARB_fragment_shader_interlock"));
         private static readonly Lazy<bool> _supportsFragmentShaderOrdering    = new Lazy<bool>(() => HasExtension("GL_INTEL_fragment_shader_ordering"));
+        private static readonly Lazy<bool> _supportsGeometryShaderPassthrough = new Lazy<bool>(() => HasExtension("GL_NV_geometry_shader_passthrough"));
         private static readonly Lazy<bool> _supportsImageLoadFormatted        = new Lazy<bool>(() => HasExtension("GL_EXT_shader_image_load_formatted"));
         private static readonly Lazy<bool> _supportsIndirectParameters        = new Lazy<bool>(() => HasExtension("GL_ARB_indirect_parameters"));
         private static readonly Lazy<bool> _supportsParallelShaderCompile     = new Lazy<bool>(() => HasExtension("GL_ARB_parallel_shader_compile"));
@@ -47,6 +48,7 @@ namespace Ryujinx.Graphics.OpenGL
         public static bool SupportsDrawTexture               => _supportsDrawTexture.Value;
         public static bool SupportsFragmentShaderInterlock   => _supportsFragmentShaderInterlock.Value;
         public static bool SupportsFragmentShaderOrdering    => _supportsFragmentShaderOrdering.Value;
+        public static bool SupportsGeometryShaderPassthrough => _supportsGeometryShaderPassthrough.Value;
         public static bool SupportsImageLoadFormatted        => _supportsImageLoadFormatted.Value;
         public static bool SupportsIndirectParameters        => _supportsIndirectParameters.Value;
         public static bool SupportsParallelShaderCompile     => _supportsParallelShaderCompile.Value;

Ryujinx.Graphics.OpenGL/Renderer.cs

@@ -110,6 +110,7 @@ namespace Ryujinx.Graphics.OpenGL
                 supportsR4G4Format: false,
                 supportsFragmentShaderInterlock: HwCapabilities.SupportsFragmentShaderInterlock,
                 supportsFragmentShaderOrderingIntel: HwCapabilities.SupportsFragmentShaderOrdering,
+                supportsGeometryShaderPassthrough: HwCapabilities.SupportsGeometryShaderPassthrough,
                 supportsImageLoadFormatted: HwCapabilities.SupportsImageLoadFormatted,
                 supportsMismatchingViewFormat: HwCapabilities.SupportsMismatchingViewFormat,
                 supportsNonConstantTextureOffset: HwCapabilities.SupportsNonConstantTextureOffset,

Ryujinx.Graphics.Shader/CodeGen/Glsl/Declarations.cs

@@ -48,7 +48,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 context.AppendLine("#extension GL_ARB_shader_viewport_layer_array : enable");
             }
 
-            if (context.Config.GpPassthrough)
+            if (context.Config.GpPassthrough && context.Config.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough())
             {
                 context.AppendLine("#extension GL_NV_geometry_shader_passthrough : enable");
             }
@@ -127,11 +127,12 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
             {
                 if (context.Config.Stage == ShaderStage.Geometry)
                 {
-                    string inPrimitive = context.Config.GpuAccessor.QueryPrimitiveTopology().ToGlslString();
+                    InputTopology inputTopology = context.Config.GpuAccessor.QueryPrimitiveTopology();
+                    string inPrimitive = inputTopology.ToGlslString();
 
                     context.AppendLine($"layout ({inPrimitive}) in;");
 
-                    if (context.Config.GpPassthrough)
+                    if (context.Config.GpPassthrough && context.Config.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough())
                     {
                         context.AppendLine($"layout (passthrough) in gl_PerVertex");
                         context.EnterScope();
@@ -144,7 +145,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                     {
                         string outPrimitive = context.Config.OutputTopology.ToGlslString();
 
-                        int maxOutputVertices = context.Config.MaxOutputVertices;
+                        int maxOutputVertices = context.Config.GpPassthrough
+                            ? inputTopology.ToInputVertices()
+                            : context.Config.MaxOutputVertices;
 
                         context.AppendLine($"layout ({outPrimitive}, max_vertices = {maxOutputVertices}) out;");
                     }
@@ -563,7 +566,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 };
             }
 
-            string pass = (context.Config.PassthroughAttributes & (1 << attr)) != 0 ? "passthrough, " : string.Empty;
+            bool passthrough = (context.Config.PassthroughAttributes & (1 << attr)) != 0;
+            string pass = passthrough && context.Config.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough() ? "passthrough, " : string.Empty;
             string name = $"{DefaultNames.IAttributePrefix}{attr}";
 
             if (context.Config.TransformFeedbackEnabled && context.Config.Stage != ShaderStage.Vertex)

Ryujinx.Graphics.Shader/CodeGen/Spirv/Declarations.cs

@@ -425,7 +425,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                         var spvType = context.TypePointer(StorageClass.Input, attrType);
                         var spvVar = context.Variable(spvType, StorageClass.Input);
 
-                        if (context.Config.PassthroughAttributes != 0)
+                        if (context.Config.PassthroughAttributes != 0 && context.Config.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough())
                         {
                             context.Decorate(spvVar, Decoration.PassthroughNV);
                         }
@@ -534,7 +534,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             {
                 attrType = context.TypeArray(attrType, context.Constant(context.TypeU32(), (LiteralInteger)context.InputVertices));
 
-                if (context.Config.GpPassthrough)
+                if (context.Config.GpPassthrough && context.Config.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough())
                 {
                     builtInPassthrough = true;
                 }
@@ -581,7 +581,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
                 if (!isOutAttr)
                 {
-                    if (!perPatch && (context.Config.PassthroughAttributes & (1 << location)) != 0)
+                    if (!perPatch &&
+                        (context.Config.PassthroughAttributes & (1 << location)) != 0 &&
+                        context.Config.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough())
                     {
                         context.Decorate(spvVar, Decoration.PassthroughNV);
                     }
@@ -646,7 +648,8 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             }
             else
             {
-                if ((context.Config.PassthroughAttributes & (1 << location)) != 0)
+                if ((context.Config.PassthroughAttributes & (1 << location)) != 0 &&
+                    context.Config.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough())
                 {
                     context.Decorate(spvVar, Decoration.PassthroughNV);
                 }

Ryujinx.Graphics.Shader/CodeGen/Spirv/SpirvGenerator.cs

@@ -70,7 +70,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             {
                 context.AddCapability(Capability.Geometry);
 
-                if (config.GpPassthrough)
+                if (config.GpPassthrough && context.Config.GpuAccessor.QueryHostSupportsGeometryShaderPassthrough())
                 {
                     context.AddExtension("SPV_NV_geometry_shader_passthrough");
                     context.AddCapability(Capability.GeometryShaderPassthroughNV);
@@ -230,7 +230,9 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                         _ => throw new InvalidOperationException($"Invalid output topology \"{context.Config.OutputTopology}\".")
                     });
 
-                    context.AddExecutionMode(spvFunc, ExecutionMode.OutputVertices, (SpvLiteralInteger)context.Config.MaxOutputVertices);
+                    int maxOutputVertices = context.Config.GpPassthrough ? context.InputVertices : context.Config.MaxOutputVertices;
+
+                    context.AddExecutionMode(spvFunc, ExecutionMode.OutputVertices, (SpvLiteralInteger)maxOutputVertices);
                 }
                 else if (context.Config.Stage == ShaderStage.Fragment)
                 {

Ryujinx.Graphics.Shader/IGpuAccessor.cs

@@ -197,6 +197,15 @@ namespace Ryujinx.Graphics.Shader
             return false;
         }
 
+        /// <summary>
+        /// Queries host GPU geometry shader passthrough support.
+        /// </summary>
+        /// <returns>True if the GPU and driver supports geometry shader passthrough, false otherwise</returns>
+        bool QueryHostSupportsGeometryShaderPassthrough()
+        {
+            return true;
+        }
+
         /// <summary>
         /// Queries host support for readable images without a explicit format declaration on the shader.
         /// </summary>

Ryujinx.Graphics.Shader/Translation/AttributeInfo.cs

@@ -91,7 +91,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                 AggregateType elemType;
 
-                if (!isOutAttr)
+                if (config.Stage == ShaderStage.Vertex && !isOutAttr)
                 {
                     elemType = config.GpuAccessor.QueryAttributeType(location) switch
                     {

Ryujinx.Graphics.Shader/Translation/EmitterContext.cs

@@ -2,6 +2,7 @@ using Ryujinx.Graphics.Shader.Decoders;
 using Ryujinx.Graphics.Shader.IntermediateRepresentation;
 using System.Collections.Generic;
 using System.Diagnostics;
+using System.Numerics;
 using System.Runtime.CompilerServices;
 
 using static Ryujinx.Graphics.Shader.IntermediateRepresentation.OperandHelper;
@@ -232,6 +233,44 @@ namespace Ryujinx.Graphics.Shader.Translation
             {
                 PrepareForVertexReturn();
             }
+            else if (Config.Stage == ShaderStage.Geometry)
+            {
+                void WriteOutput(int index, int primIndex)
+                {
+                    Operand x = this.LoadAttribute(Const(index), Const(0), Const(primIndex));
+                    Operand y = this.LoadAttribute(Const(index + 4), Const(0), Const(primIndex));
+                    Operand z = this.LoadAttribute(Const(index + 8), Const(0), Const(primIndex));
+                    Operand w = this.LoadAttribute(Const(index + 12), Const(0), Const(primIndex));
+
+                    this.Copy(Attribute(index), x);
+                    this.Copy(Attribute(index + 4), y);
+                    this.Copy(Attribute(index + 8), z);
+                    this.Copy(Attribute(index + 12), w);
+                }
+
+                if (Config.GpPassthrough)
+                {
+                    int inputVertices = Config.GpuAccessor.QueryPrimitiveTopology().ToInputVertices();
+
+                    for (int primIndex = 0; primIndex < inputVertices; primIndex++)
+                    {
+                        WriteOutput(AttributeConsts.PositionX, primIndex);
+
+                        int passthroughAttributes = Config.PassthroughAttributes;
+                        while (passthroughAttributes != 0)
+                        {
+                            int index = BitOperations.TrailingZeroCount(passthroughAttributes);
+                            WriteOutput(AttributeConsts.UserAttributeBase + index * 16, primIndex);
+                            Config.SetOutputUserAttribute(index, perPatch: false);
+                            passthroughAttributes &= ~(1 << index);
+                        }
+
+                        this.EmitVertex();
+                    }
+
+                    this.EndPrimitive();
+                }
+            }
             else if (Config.Stage == ShaderStage.Fragment)
             {
                 bool supportsBgra = Config.GpuAccessor.QueryHostSupportsBgraFormat();

Ryujinx.Graphics.Shader/Translation/ShaderConfig.cs

@@ -15,6 +15,7 @@ namespace Ryujinx.Graphics.Shader.Translation
         public ShaderStage Stage { get; }
 
         public bool GpPassthrough { get; }
+        public bool GpPassthroughWithHostSupport => GpPassthrough && GpuAccessor.QueryHostSupportsGeometryShaderPassthrough();
         public bool LastInVertexPipeline { get; private set; }
 
         public int ThreadsPerInputPrimitive { get; }

Ryujinx.Graphics.Vulkan/VulkanGraphicsDevice.cs

@@ -38,6 +38,7 @@ namespace Ryujinx.Graphics.Vulkan
         internal bool SupportsCustomBorderColor { get; private set; }
         internal bool SupportsIndirectParameters { get; private set; }
         internal bool SupportsFragmentShaderInterlock { get; private set; }
+        internal bool SupportsGeometryShaderPassthrough { get; private set; }
         internal bool SupportsSubgroupSizeControl { get; private set; }
 
         internal uint QueueFamilyIndex { get; private set; }
@@ -123,6 +124,7 @@ namespace Ryujinx.Graphics.Vulkan
             SupportsCustomBorderColor = supportedExtensions.Contains("VK_EXT_custom_border_color");
             SupportsIndirectParameters = supportedExtensions.Contains(KhrDrawIndirectCount.ExtensionName);
             SupportsFragmentShaderInterlock = supportedExtensions.Contains("VK_EXT_fragment_shader_interlock");
+            SupportsGeometryShaderPassthrough = supportedExtensions.Contains("VK_NV_geometry_shader_passthrough");
             SupportsSubgroupSizeControl = supportedExtensions.Contains("VK_EXT_subgroup_size_control");
 
             if (api.TryGetDeviceExtension(_instance, _device, out KhrSwapchain swapchainApi))
@@ -311,6 +313,7 @@ namespace Ryujinx.Graphics.Vulkan
                 supportsR4G4Format: false,
                 supportsFragmentShaderInterlock: SupportsFragmentShaderInterlock,
                 supportsFragmentShaderOrderingIntel: false,
+                supportsGeometryShaderPassthrough: SupportsGeometryShaderPassthrough,
                 supportsImageLoadFormatted: features.ShaderStorageImageReadWithoutFormat,
                 supportsMismatchingViewFormat: true,
                 supportsNonConstantTextureOffset: false,