Add shader non-uniform indexing support (#91)

This PR marks ALL texture indexes as nonuniform to fix an issue with the paths in Tomodachi Life: Living the Dream on AMD cards. It should have a negligible impact on performance (and it should not have an impact at all on NVIDIA cards!)

It's caused by what is called 'implicit non-uniform sampler array indexing'. The idea is basically that some GPUs optimize texture lookups from indexed texture arrays, by assuming that you are never going to index different textures within a single workgroup. What this causes is that visual glitch where a subgroup is tasked with rendering a block of the screen, and in the boundaries some cores are indexing the wrong texture.

Co-authored-by: AsperTheDog <guillerman0000@gmail.com>
Reviewed-on: https://git.ryujinx.app/projects/Ryubing/pulls/91

src/Ryujinx.Graphics.GAL/Capabilities.cs

@@ -42,6 +42,7 @@ namespace Ryujinx.Graphics.GAL
         public readonly bool SupportsShaderBallot;
         public readonly bool SupportsShaderBarrierDivergence;
         public readonly bool SupportsShaderFloat64;
+        public readonly bool SupportsShaderNonUniformIndexing;
         public readonly bool SupportsTextureGatherOffsets;
         public readonly bool SupportsTextureShadowLod;
         public readonly bool SupportsVertexStoreAndAtomics;
@@ -110,6 +111,7 @@ namespace Ryujinx.Graphics.GAL
             bool supportsShaderBallot,
             bool supportsShaderBarrierDivergence,
             bool supportsShaderFloat64,
+            bool supportsShaderNonUniformIndexing,
             bool supportsTextureGatherOffsets,
             bool supportsTextureShadowLod,
             bool supportsVertexStoreAndAtomics,
@@ -172,6 +174,7 @@ namespace Ryujinx.Graphics.GAL
             SupportsShaderBallot = supportsShaderBallot;
             SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence;
             SupportsShaderFloat64 = supportsShaderFloat64;
+            SupportsShaderNonUniformIndexing = supportsShaderNonUniformIndexing;
             SupportsTextureGatherOffsets = supportsTextureGatherOffsets;
             SupportsTextureShadowLod = supportsTextureShadowLod;
             SupportsVertexStoreAndAtomics = supportsVertexStoreAndAtomics;

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 = 7353;
+        private const uint CodeGenVersion = 7354;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

src/Ryujinx.Graphics.Gpu/Shader/GpuAccessorBase.cs

@@ -231,6 +231,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
 
         public bool QueryHostSupportsShaderFloat64() => _context.Capabilities.SupportsShaderFloat64;
 
+        public bool QueryHostSupportsShaderNonUniformIndexing() => _context.Capabilities.SupportsShaderNonUniformIndexing;
+
         public bool QueryHostSupportsSnormBufferTextureFormat() => _context.Capabilities.SupportsSnormBufferTextureFormat;
 
         public bool QueryHostSupportsTextureGatherOffsets() => _context.Capabilities.SupportsTextureGatherOffsets;

src/Ryujinx.Graphics.OpenGL/OpenGLRenderer.cs

@@ -184,6 +184,7 @@ namespace Ryujinx.Graphics.OpenGL
                 supportsShaderBallot: HwCapabilities.SupportsShaderBallot,
                 supportsShaderBarrierDivergence: !(intelWindows || intelUnix),
                 supportsShaderFloat64: true,
+                supportsShaderNonUniformIndexing: false,
                 supportsTextureGatherOffsets: true,
                 supportsTextureShadowLod: HwCapabilities.SupportsTextureShadowLod,
                 supportsVertexStoreAndAtomics: true,

src/Ryujinx.Graphics.Shader/CodeGen/Spirv/Instructions.cs

@@ -587,6 +587,14 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
             return OperationResult.Invalid;
         }
 
+        private static void MarkNonUniform(CodeGenContext context, SpvInstruction inst)
+        {
+            if (context.HostCapabilities.SupportsShaderNonUniformIndexing)
+            {
+                context.Decorate(inst, Decoration.NonUniform);
+            }
+        }
+
         private static OperationResult GenerateImageAtomic(CodeGenContext context, AstOperation operation)
         {
             AstTextureOperation texOp = (AstTextureOperation)operation;
@@ -613,6 +621,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                 SpvInstruction textureIndex = Src(AggregateType.S32);
 
                 image = context.AccessChain(imagePointerType, image, textureIndex);
+                MarkNonUniform(context, image);
             }
 
             int coordsCount = texOp.Type.Dimensions;
@@ -683,15 +692,21 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
             ImageDeclaration declaration = context.Images[texOp.GetTextureSetAndBinding()];
             SpvInstruction image = declaration.Image;
+            bool isIndexed = declaration.IsIndexed;
 
-            if (declaration.IsIndexed)
+            if (isIndexed)
             {
                 SpvInstruction textureIndex = Src(AggregateType.S32);
 
                 image = context.AccessChain(declaration.ImagePointerType, image, textureIndex);
+                MarkNonUniform(context, image);
             }
 
             image = context.Load(declaration.ImageType, image);
+            if (isIndexed)
+            {
+                MarkNonUniform(context, image);
+            }
 
             int coordsCount = texOp.Type.Dimensions;
 
@@ -740,15 +755,21 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
 
             ImageDeclaration declaration = context.Images[texOp.GetTextureSetAndBinding()];
             SpvInstruction image = declaration.Image;
+            bool isIndexed = declaration.IsIndexed;
 
-            if (declaration.IsIndexed)
+            if (isIndexed)
             {
                 SpvInstruction textureIndex = Src(AggregateType.S32);
 
                 image = context.AccessChain(declaration.ImagePointerType, image, textureIndex);
+                MarkNonUniform(context, image);
             }
 
             image = context.Load(declaration.ImageType, image);
+            if (isIndexed)
+            {
+                MarkNonUniform(context, image);
+            }
 
             int coordsCount = texOp.Type.Dimensions;
 
@@ -1878,35 +1899,56 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
         private static SpvInstruction GenerateSampledImageLoad(CodeGenContext context, AstTextureOperation texOp, SamplerDeclaration declaration, ref int srcIndex)
         {
             SpvInstruction image = declaration.Image;
+            bool imageIndexed = declaration.IsIndexed;
 
-            if (declaration.IsIndexed)
+            if (imageIndexed)
             {
                 SpvInstruction textureIndex = context.Get(AggregateType.S32, texOp.GetSource(srcIndex++));
 
                 image = context.AccessChain(declaration.SampledImagePointerType, image, textureIndex);
+                MarkNonUniform(context, image);
             }
 
             if (texOp.IsSeparate)
             {
                 image = context.Load(declaration.ImageType, image);
+                if (imageIndexed)
+                {
+                    MarkNonUniform(context, image);
+                }
 
                 SamplerDeclaration samplerDeclaration = context.Samplers[texOp.GetSamplerSetAndBinding()];
 
                 SpvInstruction sampler = samplerDeclaration.Image;
+                bool samplerIndexed = samplerDeclaration.IsIndexed;
 
-                if (samplerDeclaration.IsIndexed)
+                if (samplerIndexed)
                 {
                     SpvInstruction samplerIndex = context.Get(AggregateType.S32, texOp.GetSource(srcIndex++));
 
                     sampler = context.AccessChain(samplerDeclaration.SampledImagePointerType, sampler, samplerIndex);
+                    MarkNonUniform(context, sampler);
                 }
 
                 sampler = context.Load(samplerDeclaration.ImageType, sampler);
+                if (samplerIndexed)
+                {
+                    MarkNonUniform(context, sampler);
+                }
+
                 image = context.SampledImage(declaration.SampledImageType, image, sampler);
+                if (imageIndexed || samplerIndexed)
+                {
+                    MarkNonUniform(context, image);
+                }
             }
             else
             {
                 image = context.Load(declaration.SampledImageType, image);
+                if (imageIndexed)
+                {
+                    MarkNonUniform(context, image);
+                }
             }
 
             return image;

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

@@ -60,6 +60,14 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                 context.AddCapability(Capability.Float64);
             }
 
+            if (parameters.HostCapabilities.SupportsShaderNonUniformIndexing)
+            {
+                context.AddExtension("SPV_EXT_descriptor_indexing");
+                context.AddCapability(Capability.ShaderNonUniform);
+                context.AddCapability(Capability.SampledImageArrayNonUniformIndexing);
+                context.AddCapability(Capability.StorageImageArrayNonUniformIndexing);
+            }
+
             if (parameters.Definitions.TransformFeedbackEnabled && parameters.Definitions.LastInVertexPipeline)
             {
                 context.AddCapability(Capability.TransformFeedback);

src/Ryujinx.Graphics.Shader/IGpuAccessor.cs

@@ -336,6 +336,10 @@ namespace Ryujinx.Graphics.Shader
         {
             return true;
         }
+        bool QueryHostSupportsShaderNonUniformIndexing()
+        {
+            return false;
+        }
 
         /// <summary>
         /// Queries host GPU support for signed normalized buffer texture formats.

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

@@ -9,6 +9,7 @@ namespace Ryujinx.Graphics.Shader.Translation
         public readonly bool SupportsShaderBallot;
         public readonly bool SupportsShaderBarrierDivergence;
         public readonly bool SupportsShaderFloat64;
+        public readonly bool SupportsShaderNonUniformIndexing;
         public readonly bool SupportsTextureShadowLod;
         public readonly bool SupportsViewportMask;
 
@@ -20,6 +21,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             bool supportsShaderBallot,
             bool supportsShaderBarrierDivergence,
             bool supportsShaderFloat64,
+            bool supportsShaderNonUniformIndexing,
             bool supportsTextureShadowLod,
             bool supportsViewportMask)
         {
@@ -30,6 +32,7 @@ namespace Ryujinx.Graphics.Shader.Translation
             SupportsShaderBallot = supportsShaderBallot;
             SupportsShaderBarrierDivergence = supportsShaderBarrierDivergence;
             SupportsShaderFloat64 = supportsShaderFloat64;
+            SupportsShaderNonUniformIndexing = supportsShaderNonUniformIndexing;
             SupportsTextureShadowLod = supportsTextureShadowLod;
             SupportsViewportMask = supportsViewportMask;
         }

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

@@ -364,6 +364,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 GpuAccessor.QueryHostSupportsShaderBallot(),
                 GpuAccessor.QueryHostSupportsShaderBarrierDivergence(),
                 GpuAccessor.QueryHostSupportsShaderFloat64(),
+                GpuAccessor.QueryHostSupportsShaderNonUniformIndexing(),
                 GpuAccessor.QueryHostSupportsTextureShadowLod(),
                 GpuAccessor.QueryHostSupportsViewportMask());
 

src/Ryujinx.Graphics.Vulkan/VulkanInitialization.cs

@@ -494,6 +494,8 @@ namespace Ryujinx.Graphics.Vulkan
                 UniformBufferStandardLayout = supportedPhysicalDeviceVulkan12Features.UniformBufferStandardLayout,
                 UniformAndStorageBuffer8BitAccess = supportedPhysicalDeviceVulkan12Features.UniformAndStorageBuffer8BitAccess,
                 StorageBuffer8BitAccess = supportedPhysicalDeviceVulkan12Features.StorageBuffer8BitAccess,
+                ShaderSampledImageArrayNonUniformIndexing = supportedPhysicalDeviceVulkan12Features.ShaderSampledImageArrayNonUniformIndexing,
+                ShaderStorageImageArrayNonUniformIndexing = supportedPhysicalDeviceVulkan12Features.ShaderStorageImageArrayNonUniformIndexing,
             };
 
             pExtendedFeatures = &featuresVk12;

src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs

@@ -775,6 +775,9 @@ namespace Ryujinx.Graphics.Vulkan
                 supportsShaderBallot: false,
                 supportsShaderBarrierDivergence: Vendor != Vendor.Intel,
                 supportsShaderFloat64: Capabilities.SupportsShaderFloat64,
+                supportsShaderNonUniformIndexing:
+                    featuresVk12.ShaderSampledImageArrayNonUniformIndexing &&
+                    featuresVk12.ShaderStorageImageArrayNonUniformIndexing,
                 supportsTextureGatherOffsets: features2.Features.ShaderImageGatherExtended && !IsMoltenVk,
                 supportsTextureShadowLod: false,
                 supportsVertexStoreAndAtomics: features2.Features.VertexPipelineStoresAndAtomics,