Add support for large sampler arrays on Vulkan (#6489)

* Add support for large sampler arrays on Vulkan

* Shader cache version bump

* Format whitespace

* Move DescriptorSetManager to PipelineLayoutCacheEntry to allow different pool sizes per layout

* Handle array textures with different types on the same buffer

* Somewhat better caching system

* Avoid useless buffer data modification checks

* Move redundant bindings update checking to the backend

* Fix an issue where texture arrays would get the same bindings across stages on Vulkan

* Backport some fixes from part 2

* Fix typo

* PR feedback

* Format whitespace

* Add some missing XML docs

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

@@ -17,6 +17,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
         public BufferDescriptor[][] ConstantBufferBindings { get; }
         public BufferDescriptor[][] StorageBufferBindings { get; }
 
+        public int[] TextureCounts { get; }
+
         public int MaxTextureBinding { get; }
         public int MaxImageBinding { get; }
 
@@ -34,6 +36,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
             ConstantBufferBindings = new BufferDescriptor[stageCount][];
             StorageBufferBindings = new BufferDescriptor[stageCount][];
 
+            TextureCounts = new int[stageCount];
+
             int maxTextureBinding = -1;
             int maxImageBinding = -1;
             int offset = isCompute ? 0 : 1;
@@ -59,13 +63,19 @@ namespace Ryujinx.Graphics.Gpu.Shader
                     var result = new TextureBindingInfo(
                         target,
                         descriptor.Binding,
+                        descriptor.ArrayLength,
                         descriptor.CbufSlot,
                         descriptor.HandleIndex,
                         descriptor.Flags);
 
-                    if (descriptor.Binding > maxTextureBinding)
+                    if (descriptor.ArrayLength <= 1)
                     {
-                        maxTextureBinding = descriptor.Binding;
+                        if (descriptor.Binding > maxTextureBinding)
+                        {
+                            maxTextureBinding = descriptor.Binding;
+                        }
+
+                        TextureCounts[i]++;
                     }
 
                     return result;
@@ -80,11 +90,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
                         target,
                         format,
                         descriptor.Binding,
+                        descriptor.ArrayLength,
                         descriptor.CbufSlot,
                         descriptor.HandleIndex,
                         descriptor.Flags);
 
-                    if (descriptor.Binding > maxImageBinding)
+                    if (descriptor.ArrayLength <= 1 && descriptor.Binding > maxImageBinding)
                     {
                         maxImageBinding = descriptor.Binding;
                     }

src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheGpuAccessor.cs

@@ -27,6 +27,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         /// <param name="cb1Data">The constant buffer 1 data of the shader</param>
         /// <param name="oldSpecState">Shader specialization state of the cached shader</param>
         /// <param name="newSpecState">Shader specialization state of the recompiled shader</param>
+        /// <param name="counts">Resource counts shared across all shader stages</param>
         /// <param name="stageIndex">Shader stage index</param>
         public DiskCacheGpuAccessor(
             GpuContext context,
@@ -108,6 +109,27 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             return _oldSpecState.GraphicsState.HasConstantBufferDrawParameters;
         }
 
+        /// <inheritdoc/>
+        public SamplerType QuerySamplerType(int handle, int cbufSlot)
+        {
+            _newSpecState.RecordTextureSamplerType(_stageIndex, handle, cbufSlot);
+            return _oldSpecState.GetTextureTarget(_stageIndex, handle, cbufSlot).ConvertSamplerType();
+        }
+
+        /// <inheritdoc/>
+        public int QueryTextureArrayLengthFromBuffer(int slot)
+        {
+            if (!_oldSpecState.TextureArrayFromBufferRegistered(_stageIndex, 0, slot))
+            {
+                throw new DiskCacheLoadException(DiskCacheLoadResult.MissingTextureArrayLength);
+            }
+
+            int arrayLength = _oldSpecState.GetTextureArrayFromBufferLength(_stageIndex, 0, slot);
+            _newSpecState.RegisterTextureArrayLengthFromBuffer(_stageIndex, 0, slot, arrayLength);
+
+            return arrayLength;
+        }
+
         /// <inheritdoc/>
         public TextureFormat QueryTextureFormat(int handle, int cbufSlot)
         {
@@ -116,13 +138,6 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             return ConvertToTextureFormat(format, formatSrgb);
         }
 
-        /// <inheritdoc/>
-        public SamplerType QuerySamplerType(int handle, int cbufSlot)
-        {
-            _newSpecState.RecordTextureSamplerType(_stageIndex, handle, cbufSlot);
-            return _oldSpecState.GetTextureTarget(_stageIndex, handle, cbufSlot).ConvertSamplerType();
-        }
-
         /// <inheritdoc/>
         public bool QueryTextureCoordNormalized(int handle, int cbufSlot)
         {

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 = 6462;
+        private const uint CodeGenVersion = 6489;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

src/Ryujinx.Graphics.Gpu/Shader/DiskCache/DiskCacheLoadResult.cs

@@ -20,6 +20,11 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
         /// </summary>
         InvalidCb1DataLength,
 
+        /// <summary>
+        /// The cache is missing the length of a texture array used by the shader.
+        /// </summary>
+        MissingTextureArrayLength,
+
         /// <summary>
         /// The cache is missing the descriptor of a texture used by the shader.
         /// </summary>
@@ -60,6 +65,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                 DiskCacheLoadResult.Success => "No error.",
                 DiskCacheLoadResult.NoAccess => "Could not access the cache file.",
                 DiskCacheLoadResult.InvalidCb1DataLength => "Constant buffer 1 data length is too low.",
+                DiskCacheLoadResult.MissingTextureArrayLength => "Texture array length missing from the cache file.",
                 DiskCacheLoadResult.MissingTextureDescriptor => "Texture descriptor missing from the cache file.",
                 DiskCacheLoadResult.FileCorruptedGeneric => "The cache file is corrupted.",
                 DiskCacheLoadResult.FileCorruptedInvalidMagic => "Magic check failed, the cache file is corrupted.",

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

@@ -72,6 +72,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         public ReadOnlySpan<ulong> GetCode(ulong address, int minimumSize)
         {
             int size = Math.Max(minimumSize, 0x1000 - (int)(address & 0xfff));
+
             return MemoryMarshal.Cast<byte, ulong>(_channel.MemoryManager.GetSpan(address, size));
         }
 
@@ -119,6 +120,27 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return _state.GraphicsState.HasUnalignedStorageBuffer || _state.ComputeState.HasUnalignedStorageBuffer;
         }
 
+        /// <inheritdoc/>
+        public SamplerType QuerySamplerType(int handle, int cbufSlot)
+        {
+            _state.SpecializationState?.RecordTextureSamplerType(_stageIndex, handle, cbufSlot);
+            return GetTextureDescriptor(handle, cbufSlot).UnpackTextureTarget().ConvertSamplerType();
+        }
+
+        /// <inheritdoc/>
+        public int QueryTextureArrayLengthFromBuffer(int slot)
+        {
+            int size = _compute
+                ? _channel.BufferManager.GetComputeUniformBufferSize(slot)
+                : _channel.BufferManager.GetGraphicsUniformBufferSize(_stageIndex, slot);
+
+            int arrayLength = size / Constants.TextureHandleSizeInBytes;
+
+            _state.SpecializationState?.RegisterTextureArrayLengthFromBuffer(_stageIndex, 0, slot, arrayLength);
+
+            return arrayLength;
+        }
+
         //// <inheritdoc/>
         public TextureFormat QueryTextureFormat(int handle, int cbufSlot)
         {
@@ -127,13 +149,6 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return ConvertToTextureFormat(descriptor.UnpackFormat(), descriptor.UnpackSrgb());
         }
 
-        /// <inheritdoc/>
-        public SamplerType QuerySamplerType(int handle, int cbufSlot)
-        {
-            _state.SpecializationState?.RecordTextureSamplerType(_stageIndex, handle, cbufSlot);
-            return GetTextureDescriptor(handle, cbufSlot).UnpackTextureTarget().ConvertSamplerType();
-        }
-
         /// <inheritdoc/>
         public bool QueryTextureCoordNormalized(int handle, int cbufSlot)
         {

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

@@ -20,6 +20,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
         private int _reservedTextures;
         private int _reservedImages;
 
+        private int _staticTexturesCount;
+        private int _staticImagesCount;
+
         /// <summary>
         /// Creates a new GPU accessor.
         /// </summary>
@@ -48,7 +51,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
             _reservedImages = rrc.ReservedImages;
         }
 
-        public int QueryBindingConstantBuffer(int index)
+        public int CreateConstantBufferBinding(int index)
         {
             int binding;
 
@@ -64,7 +67,39 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return binding + _reservedConstantBuffers;
         }
 
-        public int QueryBindingStorageBuffer(int index)
+        public int CreateImageBinding(int count, bool isBuffer)
+        {
+            int binding;
+
+            if (_context.Capabilities.Api == TargetApi.Vulkan)
+            {
+                if (count == 1)
+                {
+                    int index = _staticImagesCount++;
+
+                    if (isBuffer)
+                    {
+                        index += (int)_context.Capabilities.MaximumImagesPerStage;
+                    }
+
+                    binding = GetBindingFromIndex(index, _context.Capabilities.MaximumImagesPerStage * 2, "Image");
+                }
+                else
+                {
+                    binding = (int)GetDynamicBaseIndexDual(_context.Capabilities.MaximumImagesPerStage) + _resourceCounts.ImagesCount++;
+                }
+            }
+            else
+            {
+                binding = _resourceCounts.ImagesCount;
+
+                _resourceCounts.ImagesCount += count;
+            }
+
+            return binding + _reservedImages;
+        }
+
+        public int CreateStorageBufferBinding(int index)
         {
             int binding;
 
@@ -80,48 +115,38 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return binding + _reservedStorageBuffers;
         }
 
-        public int QueryBindingTexture(int index, bool isBuffer)
+        public int CreateTextureBinding(int count, bool isBuffer)
         {
             int binding;
 
             if (_context.Capabilities.Api == TargetApi.Vulkan)
             {
-                if (isBuffer)
+                if (count == 1)
                 {
-                    index += (int)_context.Capabilities.MaximumTexturesPerStage;
-                }
+                    int index = _staticTexturesCount++;
 
-                binding = GetBindingFromIndex(index, _context.Capabilities.MaximumTexturesPerStage * 2, "Texture");
+                    if (isBuffer)
+                    {
+                        index += (int)_context.Capabilities.MaximumTexturesPerStage;
+                    }
+
+                    binding = GetBindingFromIndex(index, _context.Capabilities.MaximumTexturesPerStage * 2, "Texture");
+                }
+                else
+                {
+                    binding = (int)GetDynamicBaseIndexDual(_context.Capabilities.MaximumTexturesPerStage) + _resourceCounts.TexturesCount++;
+                }
             }
             else
             {
-                binding = _resourceCounts.TexturesCount++;
+                binding = _resourceCounts.TexturesCount;
+
+                _resourceCounts.TexturesCount += count;
             }
 
             return binding + _reservedTextures;
         }
 
-        public int QueryBindingImage(int index, bool isBuffer)
-        {
-            int binding;
-
-            if (_context.Capabilities.Api == TargetApi.Vulkan)
-            {
-                if (isBuffer)
-                {
-                    index += (int)_context.Capabilities.MaximumImagesPerStage;
-                }
-
-                binding = GetBindingFromIndex(index, _context.Capabilities.MaximumImagesPerStage * 2, "Image");
-            }
-            else
-            {
-                binding = _resourceCounts.ImagesCount++;
-            }
-
-            return binding + _reservedImages;
-        }
-
         private int GetBindingFromIndex(int index, uint maxPerStage, string resourceName)
         {
             if ((uint)index >= maxPerStage)
@@ -148,6 +173,16 @@ namespace Ryujinx.Graphics.Gpu.Shader
             };
         }
 
+        private static uint GetDynamicBaseIndexDual(uint maxPerStage)
+        {
+            return GetDynamicBaseIndex(maxPerStage) * 2;
+        }
+
+        private static uint GetDynamicBaseIndex(uint maxPerStage)
+        {
+            return maxPerStage * Constants.ShaderStages;
+        }
+
         public int QueryHostGatherBiasPrecision() => _context.Capabilities.GatherBiasPrecision;
 
         public bool QueryHostReducedPrecision() => _context.Capabilities.ReduceShaderPrecision;

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

@@ -132,6 +132,9 @@ namespace Ryujinx.Graphics.Gpu.Shader
             AddDualDescriptor(stages, ResourceType.TextureAndSampler, ResourceType.BufferTexture, TextureSetIndex, textureBinding, texturesPerStage);
             AddDualDescriptor(stages, ResourceType.Image, ResourceType.BufferImage, ImageSetIndex, imageBinding, imagesPerStage);
 
+            AddArrayDescriptors(info.Textures, stages, TextureSetIndex, isImage: false);
+            AddArrayDescriptors(info.Images, stages, TextureSetIndex, isImage: true);
+
             AddUsage(info.CBuffers, stages, UniformSetIndex, isStorage: false);
             AddUsage(info.SBuffers, stages, StorageSetIndex, isStorage: true);
             AddUsage(info.Textures, stages, TextureSetIndex, isImage: false);
@@ -169,6 +172,30 @@ namespace Ryujinx.Graphics.Gpu.Shader
             AddDescriptor(stages, type2, setIndex, binding + count, count);
         }
 
+        /// <summary>
+        /// Adds all array descriptors (those with an array length greater than one).
+        /// </summary>
+        /// <param name="textures">Textures to be added</param>
+        /// <param name="stages">Stages where the textures are used</param>
+        /// <param name="setIndex">Descriptor set index where the textures will be bound</param>
+        /// <param name="isImage">True for images, false for textures</param>
+        private void AddArrayDescriptors(IEnumerable<TextureDescriptor> textures, ResourceStages stages, int setIndex, bool isImage)
+        {
+            foreach (TextureDescriptor texture in textures)
+            {
+                if (texture.ArrayLength > 1)
+                {
+                    bool isBuffer = (texture.Type & SamplerType.Mask) == SamplerType.TextureBuffer;
+
+                    ResourceType type = isBuffer
+                        ? (isImage ? ResourceType.BufferImage : ResourceType.BufferTexture)
+                        : (isImage ? ResourceType.Image : ResourceType.TextureAndSampler);
+
+                    _resourceDescriptors[setIndex].Add(new ResourceDescriptor(texture.Binding, texture.ArrayLength, type, stages));
+                }
+            }
+        }
+
         /// <summary>
         /// Adds buffer usage information to the list of usages.
         /// </summary>
@@ -181,7 +208,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         {
             for (int index = 0; index < count; index++)
             {
-                _resourceUsages[setIndex].Add(new ResourceUsage(binding + index, type, stages));
+                _resourceUsages[setIndex].Add(new ResourceUsage(binding + index, 1, type, stages));
             }
         }
 
@@ -198,6 +225,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
             {
                 _resourceUsages[setIndex].Add(new ResourceUsage(
                     buffer.Binding,
+                    1,
                     isStorage ? ResourceType.StorageBuffer : ResourceType.UniformBuffer,
                     stages));
             }
@@ -220,10 +248,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
                     ? (isImage ? ResourceType.BufferImage : ResourceType.BufferTexture)
                     : (isImage ? ResourceType.Image : ResourceType.TextureAndSampler);
 
-                _resourceUsages[setIndex].Add(new ResourceUsage(
-                    texture.Binding,
-                    type,
-                    stages));
+                _resourceUsages[setIndex].Add(new ResourceUsage(texture.Binding, texture.ArrayLength, type, stages));
             }
         }
 

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

@@ -30,6 +30,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         {
             PrimitiveTopology = 1 << 1,
             TransformFeedback = 1 << 3,
+            TextureArrayFromBuffer = 1 << 4,
         }
 
         private QueriedStateFlags _queriedState;
@@ -153,6 +154,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         }
 
         private readonly Dictionary<TextureKey, Box<TextureSpecializationState>> _textureSpecialization;
+        private readonly Dictionary<TextureKey, int> _textureArraySpecialization;
         private KeyValuePair<TextureKey, Box<TextureSpecializationState>>[] _allTextures;
         private Box<TextureSpecializationState>[][] _textureByBinding;
         private Box<TextureSpecializationState>[][] _imageByBinding;
@@ -163,6 +165,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
         private ShaderSpecializationState()
         {
             _textureSpecialization = new Dictionary<TextureKey, Box<TextureSpecializationState>>();
+            _textureArraySpecialization = new Dictionary<TextureKey, int>();
         }
 
         /// <summary>
@@ -323,6 +326,19 @@ namespace Ryujinx.Graphics.Gpu.Shader
             state.Value.CoordNormalized = coordNormalized;
         }
 
+        /// <summary>
+        /// Indicates that the coordinate normalization state of a given texture was used during the shader translation process.
+        /// </summary>
+        /// <param name="stageIndex">Shader stage where the texture is used</param>
+        /// <param name="handle">Offset in words of the texture handle on the texture buffer</param>
+        /// <param name="cbufSlot">Slot of the texture buffer constant buffer</param>
+        /// <param name="length">Number of elements in the texture array</param>
+        public void RegisterTextureArrayLengthFromBuffer(int stageIndex, int handle, int cbufSlot, int length)
+        {
+            _textureArraySpecialization[new TextureKey(stageIndex, handle, cbufSlot)] = length;
+            _queriedState |= QueriedStateFlags.TextureArrayFromBuffer;
+        }
+
         /// <summary>
         /// Indicates that the format of a given texture was used during the shader translation process.
         /// </summary>
@@ -379,6 +395,17 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return GetTextureSpecState(stageIndex, handle, cbufSlot) != null;
         }
 
+        /// <summary>
+        /// Checks if a given texture array (from constant buffer) was registerd on this specialization state.
+        /// </summary>
+        /// <param name="stageIndex">Shader stage where the texture is used</param>
+        /// <param name="handle">Offset in words of the texture handle on the texture buffer</param>
+        /// <param name="cbufSlot">Slot of the texture buffer constant buffer</param>
+        public bool TextureArrayFromBufferRegistered(int stageIndex, int handle, int cbufSlot)
+        {
+            return _textureArraySpecialization.ContainsKey(new TextureKey(stageIndex, handle, cbufSlot));
+        }
+
         /// <summary>
         /// Gets the recorded format of a given texture.
         /// </summary>
@@ -413,6 +440,17 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return GetTextureSpecState(stageIndex, handle, cbufSlot).Value.CoordNormalized;
         }
 
+        /// <summary>
+        /// Gets the recorded length of a given texture array (from constant buffer).
+        /// </summary>
+        /// <param name="stageIndex">Shader stage where the texture is used</param>
+        /// <param name="handle">Offset in words of the texture handle on the texture buffer</param>
+        /// <param name="cbufSlot">Slot of the texture buffer constant buffer</param>
+        public int GetTextureArrayFromBufferLength(int stageIndex, int handle, int cbufSlot)
+        {
+            return _textureArraySpecialization[new TextureKey(stageIndex, handle, cbufSlot)];
+        }
+
         /// <summary>
         /// Gets texture specialization state for a given texture, or create a new one if not present.
         /// </summary>
@@ -548,6 +586,12 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return Matches(channel, ref poolState, checkTextures, isCompute: false);
         }
 
+        /// <summary>
+        /// Converts special vertex attribute groups to their generic equivalents, for comparison purposes.
+        /// </summary>
+        /// <param name="channel">GPU channel</param>
+        /// <param name="type">Vertex attribute type</param>
+        /// <returns>Filtered attribute</returns>
         private static AttributeType FilterAttributeType(GpuChannel channel, AttributeType type)
         {
             type &= ~(AttributeType.Packed | AttributeType.PackedRgb10A2Signed);
@@ -838,6 +882,22 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 specState._textureSpecialization[textureKey] = textureState;
             }
 
+            if (specState._queriedState.HasFlag(QueriedStateFlags.TextureArrayFromBuffer))
+            {
+                dataReader.Read(ref count);
+
+                for (int index = 0; index < count; index++)
+                {
+                    TextureKey textureKey = default;
+                    int length = 0;
+
+                    dataReader.ReadWithMagicAndSize(ref textureKey, TexkMagic);
+                    dataReader.Read(ref length);
+
+                    specState._textureArraySpecialization[textureKey] = length;
+                }
+            }
+
             return specState;
         }
 
@@ -902,6 +962,21 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 dataWriter.WriteWithMagicAndSize(ref textureKey, TexkMagic);
                 dataWriter.WriteWithMagicAndSize(ref textureState.Value, TexsMagic);
             }
+
+            if (_queriedState.HasFlag(QueriedStateFlags.TextureArrayFromBuffer))
+            {
+                count = (ushort)_textureArraySpecialization.Count;
+                dataWriter.Write(ref count);
+
+                foreach (var kv in _textureArraySpecialization)
+                {
+                    var textureKey = kv.Key;
+                    var length = kv.Value;
+
+                    dataWriter.WriteWithMagicAndSize(ref textureKey, TexkMagic);
+                    dataWriter.Write(ref length);
+                }
+            }
         }
     }
 }