Force dynamic state update after rasterizer discard disable (#7007)

* Disallow concurrent fence waits on Adreno

* Ensure locks are released if exceptions are thrown

Directory.Packages.props

@@ -20,7 +20,7 @@
     <PackageVersion Include="LibHac" Version="0.19.0" />
     <PackageVersion Include="Microsoft.CodeAnalysis.Analyzers" Version="3.3.4" />
     <PackageVersion Include="Microsoft.CodeAnalysis.CSharp" Version="4.9.2" />
-    <PackageVersion Include="Microsoft.IdentityModel.JsonWebTokens" Version="7.6.0" />
+    <PackageVersion Include="Microsoft.IdentityModel.JsonWebTokens" Version="7.6.2" />
     <PackageVersion Include="Microsoft.NET.Test.Sdk" Version="17.9.0" />
     <PackageVersion Include="Microsoft.IO.RecyclableMemoryStream" Version="3.0.1" />
     <PackageVersion Include="MsgPack.Cli" Version="1.0.1" />

src/ARMeilleure/CodeGen/RegisterAllocators/LinearScanAllocator.cs

@@ -251,7 +251,20 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
                 }
             }
 
-            int selectedReg = GetHighestValueIndex(freePositions);
+            // If this is a copy destination variable, we prefer the register used for the copy source.
+            // If the register is available, then the copy can be eliminated later as both source
+            // and destination will use the same register.
+            int selectedReg;
+
+            if (current.TryGetCopySourceRegister(out int preferredReg) && freePositions[preferredReg] >= current.GetEnd())
+            {
+                selectedReg = preferredReg;
+            }
+            else
+            {
+                selectedReg = GetHighestValueIndex(freePositions);
+            }
+
             int selectedNextUse = freePositions[selectedReg];
 
             // Intervals starts and ends at odd positions, unless they span an entire
@@ -431,7 +444,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             }
         }
 
-        private static int GetHighestValueIndex(Span<int> span)
+        private static int GetHighestValueIndex(ReadOnlySpan<int> span)
         {
             int highest = int.MinValue;
 
@@ -798,12 +811,12 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             // The "visited" state is stored in the MSB of the local's value.
             const ulong VisitedMask = 1ul << 63;
 
-            bool IsVisited(Operand local)
+            static bool IsVisited(Operand local)
             {
                 return (local.GetValueUnsafe() & VisitedMask) != 0;
             }
 
-            void SetVisited(Operand local)
+            static void SetVisited(Operand local)
             {
                 local.GetValueUnsafe() |= VisitedMask;
             }
@@ -826,9 +839,25 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
                         {
                             dest.NumberLocal(_intervals.Count);
 
-                            _intervals.Add(new LiveInterval(dest));
+                            LiveInterval interval = new LiveInterval(dest);
+                            _intervals.Add(interval);
 
                             SetVisited(dest);
+
+                            // If this is a copy (or copy-like operation), set the copy source interval as well.
+                            // This is used for register preferencing later on, which allows the copy to be eliminated
+                            // in some cases.
+                            if (node.Instruction == Instruction.Copy || node.Instruction == Instruction.ZeroExtend32)
+                            {
+                                Operand source = node.GetSource(0);
+
+                                if (source.Kind == OperandKind.LocalVariable &&
+                                    source.GetLocalNumber() > 0 &&
+                                    (node.Instruction == Instruction.Copy || source.Type == OperandType.I32))
+                                {
+                                    interval.SetCopySource(_intervals[source.GetLocalNumber()]);
+                                }
+                            }
                         }
                     }
                 }

src/ARMeilleure/CodeGen/RegisterAllocators/LiveInterval.cs

@@ -19,6 +19,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             public LiveRange CurrRange;
 
             public LiveInterval Parent;
+            public LiveInterval CopySource;
 
             public UseList Uses;
             public LiveIntervalList Children;
@@ -37,6 +38,7 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
         private ref LiveRange CurrRange => ref _data->CurrRange;
         private ref LiveRange PrevRange => ref _data->PrevRange;
         private ref LiveInterval Parent => ref _data->Parent;
+        private ref LiveInterval CopySource => ref _data->CopySource;
         private ref UseList Uses => ref _data->Uses;
         private ref LiveIntervalList Children => ref _data->Children;
 
@@ -78,6 +80,25 @@ namespace ARMeilleure.CodeGen.RegisterAllocators
             Register = register;
         }
 
+        public void SetCopySource(LiveInterval copySource)
+        {
+            CopySource = copySource;
+        }
+
+        public bool TryGetCopySourceRegister(out int copySourceRegIndex)
+        {
+            if (CopySource._data != null)
+            {
+                copySourceRegIndex = CopySource.Register.Index;
+
+                return true;
+            }
+
+            copySourceRegIndex = 0;
+
+            return false;
+        }
+
         public void Reset()
         {
             PrevRange = default;

src/ARMeilleure/Translation/ControlFlowGraph.cs

@@ -11,7 +11,7 @@ namespace ARMeilleure.Translation
         private int[] _postOrderMap;
 
         public int LocalsCount { get; private set; }
-        public BasicBlock Entry { get; }
+        public BasicBlock Entry { get; private set; }
         public IntrusiveList<BasicBlock> Blocks { get; }
         public BasicBlock[] PostOrderBlocks => _postOrderBlocks;
         public int[] PostOrderMap => _postOrderMap;
@@ -34,6 +34,15 @@ namespace ARMeilleure.Translation
             return result;
         }
 
+        public void UpdateEntry(BasicBlock newEntry)
+        {
+            newEntry.AddSuccessor(Entry);
+
+            Entry = newEntry;
+            Blocks.AddFirst(newEntry);
+            Update();
+        }
+
         public void Update()
         {
             RemoveUnreachableBlocks(Blocks);

src/ARMeilleure/Translation/PTC/Ptc.cs

@@ -29,7 +29,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 6634; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 6950; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";

src/ARMeilleure/Translation/RegisterUsage.cs

@@ -89,6 +89,17 @@ namespace ARMeilleure.Translation
 
         public static void RunPass(ControlFlowGraph cfg, ExecutionMode mode)
         {
+            if (cfg.Entry.Predecessors.Count != 0)
+            {
+                // We expect the entry block to have no predecessors.
+                // This is required because we have a implicit context load at the start of the function,
+                // but if there is a jump to the start of the function, the context load would trash the modified values.
+                // Here we insert a new entry block that will jump to the existing entry block.
+                BasicBlock newEntry = new BasicBlock(cfg.Blocks.Count);
+
+                cfg.UpdateEntry(newEntry);
+            }
+
             // Compute local register inputs and outputs used inside blocks.
             RegisterMask[] localInputs = new RegisterMask[cfg.Blocks.Count];
             RegisterMask[] localOutputs = new RegisterMask[cfg.Blocks.Count];
@@ -201,7 +212,7 @@ namespace ARMeilleure.Translation
 
                 // The only block without any predecessor should be the entry block.
                 // It always needs a context load as it is the first block to run.
-                if (block.Predecessors.Count == 0 || hasContextLoad)
+                if (block == cfg.Entry || hasContextLoad)
                 {
                     long vecMask = globalInputs[block.Index].VecMask;
                     long intMask = globalInputs[block.Index].IntMask;

src/Ryujinx.Common/Memory/MemoryOwner.cs

@@ -124,7 +124,7 @@ namespace Ryujinx.Common.Memory
 
             if (array is not null)
             {
-                ArrayPool<T>.Shared.Return(array);
+                ArrayPool<T>.Shared.Return(array, RuntimeHelpers.IsReferenceOrContainsReferences<T>());
             }
         }
 

src/Ryujinx.Common/Memory/SpanOwner.cs

@@ -108,7 +108,7 @@ namespace Ryujinx.Common.Memory
         [MethodImpl(MethodImplOptions.AggressiveInlining)]
         public void Dispose()
         {
-            ArrayPool<T>.Shared.Return(_array);
+            ArrayPool<T>.Shared.Return(_array, RuntimeHelpers.IsReferenceOrContainsReferences<T>());
         }
     }
 }

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 = 6852;
+        private const uint CodeGenVersion = 6921;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

src/Ryujinx.Graphics.Shader/Instructions/InstEmitPredicate.cs

@@ -24,7 +24,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
             if (op.BVal)
             {
-                context.Copy(dest, context.ConditionalSelect(res, ConstF(1), Const(0)));
+                context.Copy(dest, context.ConditionalSelect(res, ConstF(1), ConstF(0)));
             }
             else
             {

src/Ryujinx.Graphics.Shader/IntermediateRepresentation/Instruction.cs

@@ -156,6 +156,26 @@ namespace Ryujinx.Graphics.Shader.IntermediateRepresentation
             return false;
         }
 
+        public static bool IsComparison(this Instruction inst)
+        {
+            switch (inst & Instruction.Mask)
+            {
+                case Instruction.CompareEqual:
+                case Instruction.CompareGreater:
+                case Instruction.CompareGreaterOrEqual:
+                case Instruction.CompareGreaterOrEqualU32:
+                case Instruction.CompareGreaterU32:
+                case Instruction.CompareLess:
+                case Instruction.CompareLessOrEqual:
+                case Instruction.CompareLessOrEqualU32:
+                case Instruction.CompareLessU32:
+                case Instruction.CompareNotEqual:
+                    return true;
+            }
+
+            return false;
+        }
+
         public static bool IsTextureQuery(this Instruction inst)
         {
             inst &= Instruction.Mask;

src/Ryujinx.Graphics.Shader/Translation/Optimizations/BindlessElimination.cs

@@ -141,16 +141,16 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return true;
         }
 
-        private static bool IsBindlessAccessAllowed(Operand nvHandle)
+        private static bool IsBindlessAccessAllowed(Operand bindlessHandle)
         {
-            if (nvHandle.Type == OperandType.ConstantBuffer)
+            if (bindlessHandle.Type == OperandType.ConstantBuffer)
             {
                 // Bindless access with handles from constant buffer is allowed.
 
                 return true;
             }
 
-            if (nvHandle.AsgOp is not Operation handleOp ||
+            if (bindlessHandle.AsgOp is not Operation handleOp ||
                 handleOp.Inst != Instruction.Load ||
                 (handleOp.StorageKind != StorageKind.Input && handleOp.StorageKind != StorageKind.StorageBuffer))
             {
@@ -300,7 +300,7 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                         resourceManager,
                         gpuAccessor,
                         texOp,
-                        TextureHandle.PackOffsets(src0.GetCbufOffset(), ((src1.Value >> 20) & 0xfff), handleType),
+                        TextureHandle.PackOffsets(src0.GetCbufOffset(), (src1.Value >> 20) & 0xfff, handleType),
                         TextureHandle.PackSlots(src0.GetCbufSlot(), 0),
                         rewriteSamplerType,
                         isImage: false);

src/Ryujinx.Graphics.Shader/Translation/Optimizations/BindlessToArray.cs

@@ -126,7 +126,9 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                     continue;
                 }
 
-                if (texOp.GetSource(0).AsgOp is not Operation handleAsgOp)
+                Operand bindlessHandle = Utils.FindLastOperation(texOp.GetSource(0), block);
+
+                if (bindlessHandle.AsgOp is not Operation handleAsgOp)
                 {
                     continue;
                 }
@@ -137,8 +139,8 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
                 if (handleAsgOp.Inst == Instruction.BitwiseOr)
                 {
-                    Operand src0 = handleAsgOp.GetSource(0);
+                    Operand src0 = Utils.FindLastOperation(handleAsgOp.GetSource(0), block);
-                    Operand src1 = handleAsgOp.GetSource(1);
+                    Operand src1 = Utils.FindLastOperation(handleAsgOp.GetSource(1), block);
 
                     if (src0.Type == OperandType.ConstantBuffer && src1.AsgOp is Operation)
                     {

src/Ryujinx.Graphics.Shader/Translation/Optimizations/Optimizer.cs

@@ -152,18 +152,14 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
         {
             // If all phi sources are the same, we can propagate it and remove the phi.
 
-            Operand firstSrc = phi.GetSource(0);
+            if (!Utils.AreAllSourcesTheSameOperand(phi))
-
-            for (int index = 1; index < phi.SourcesCount; index++)
             {
-                if (!IsSameOperand(firstSrc, phi.GetSource(index)))
+                return false;
-                {
-                    return false;
-                }
             }
 
             // All sources are equal, we can propagate the value.
 
+            Operand firstSrc = phi.GetSource(0);
             Operand dest = phi.Dest;
 
             INode[] uses = dest.UseOps.ToArray();
@@ -182,17 +178,6 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return true;
         }
 
-        private static bool IsSameOperand(Operand x, Operand y)
-        {
-            if (x.Type != y.Type || x.Value != y.Value)
-            {
-                return false;
-            }
-
-            // TODO: Handle Load operations with the same storage and the same constant parameters.
-            return x.Type == OperandType.Constant || x.Type == OperandType.ConstantBuffer;
-        }
-
         private static bool PropagatePack(Operation packOp)
         {
             // Propagate pack source operands to uses by unpack

src/Ryujinx.Graphics.Shader/Translation/Optimizations/Simplification.cs

@@ -31,6 +31,10 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                     TryEliminateBitwiseOr(operation);
                     break;
 
+                case Instruction.CompareNotEqual:
+                    TryEliminateCompareNotEqual(operation);
+                    break;
+
                 case Instruction.ConditionalSelect:
                     TryEliminateConditionalSelect(operation);
                     break;
@@ -174,6 +178,32 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             }
         }
 
+        private static void TryEliminateCompareNotEqual(Operation operation)
+        {
+            // Comparison instruction returns 0 if the result is false, and -1 if true.
+            // Doing a not equal zero comparison on the result is redundant, so we can just copy the first result in this case.
+
+            Operand lhs = operation.GetSource(0);
+            Operand rhs = operation.GetSource(1);
+
+            if (lhs.Type == OperandType.Constant)
+            {
+                (lhs, rhs) = (rhs, lhs);
+            }
+
+            if (rhs.Type != OperandType.Constant || rhs.Value != 0)
+            {
+                return;
+            }
+
+            if (lhs.AsgOp is not Operation compareOp || !compareOp.Inst.IsComparison())
+            {
+                return;
+            }
+
+            operation.TurnIntoCopy(lhs);
+        }
+
         private static void TryEliminateConditionalSelect(Operation operation)
         {
             Operand cond = operation.GetSource(0);

src/Ryujinx.Graphics.Shader/Translation/Optimizations/Utils.cs

@@ -34,6 +34,50 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return elemIndexSrc.Type == OperandType.Constant && elemIndexSrc.Value == elemIndex;
         }
 
+        private static bool IsSameOperand(Operand x, Operand y)
+        {
+            if (x.Type != y.Type || x.Value != y.Value)
+            {
+                return false;
+            }
+
+            // TODO: Handle Load operations with the same storage and the same constant parameters.
+            return x == y || x.Type == OperandType.Constant || x.Type == OperandType.ConstantBuffer;
+        }
+
+        private static bool AreAllSourcesEqual(INode node, INode otherNode)
+        {
+            if (node.SourcesCount != otherNode.SourcesCount)
+            {
+                return false;
+            }
+
+            for (int index = 0; index < node.SourcesCount; index++)
+            {
+                if (!IsSameOperand(node.GetSource(index), otherNode.GetSource(index)))
+                {
+                    return false;
+                }
+            }
+
+            return true;
+        }
+
+        public static bool AreAllSourcesTheSameOperand(INode node)
+        {
+            Operand firstSrc = node.GetSource(0);
+
+            for (int index = 1; index < node.SourcesCount; index++)
+            {
+                if (!IsSameOperand(firstSrc, node.GetSource(index)))
+                {
+                    return false;
+                }
+            }
+
+            return true;
+        }
+
         private static Operation FindBranchSource(BasicBlock block)
         {
             foreach (BasicBlock sourceBlock in block.Predecessors)
@@ -55,6 +99,19 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return inst == Instruction.BranchIfFalse || inst == Instruction.BranchIfTrue;
         }
 
+        private static bool IsSameCondition(Operand currentCondition, Operand queryCondition)
+        {
+            if (currentCondition == queryCondition)
+            {
+                return true;
+            }
+
+            return currentCondition.AsgOp is Operation currentOperation &&
+                queryCondition.AsgOp is Operation queryOperation &&
+                currentOperation.Inst == queryOperation.Inst &&
+                AreAllSourcesEqual(currentOperation, queryOperation);
+        }
+
         private static bool BlockConditionsMatch(BasicBlock currentBlock, BasicBlock queryBlock)
         {
             // Check if all the conditions for the query block are satisfied by the current block.
@@ -70,10 +127,10 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 
             return currentBranch != null && queryBranch != null &&
                    currentBranch.Inst == queryBranch.Inst &&
-                   currentCondition == queryCondition;
+                   IsSameCondition(currentCondition, queryCondition);
         }
 
-        public static Operand FindLastOperation(Operand source, BasicBlock block)
+        public static Operand FindLastOperation(Operand source, BasicBlock block, bool recurse = true)
         {
             if (source.AsgOp is PhiNode phiNode)
             {
@@ -84,10 +141,23 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                 for (int i = phiNode.SourcesCount - 1; i >= 0; i--)
                 {
                     BasicBlock phiBlock = phiNode.GetBlock(i);
+                    Operand phiSource = phiNode.GetSource(i);
 
                     if (BlockConditionsMatch(block, phiBlock))
                     {
-                        return phiNode.GetSource(i);
+                        return phiSource;
+                    }
+                    else if (recurse && phiSource.AsgOp is PhiNode)
+                    {
+                        // Phi source is another phi.
+                        // Let's check if that phi has a block that matches our condition.
+
+                        Operand match = FindLastOperation(phiSource, block, false);
+
+                        if (match != phiSource)
+                        {
+                            return match;
+                        }
                     }
                 }
             }

src/Ryujinx.Graphics.Vulkan/BackgroundResources.cs

@@ -29,7 +29,14 @@ namespace Ryujinx.Graphics.Vulkan
 
                 lock (queueLock)
                 {
-                    _pool = new CommandBufferPool(_gd.Api, _device, queue, queueLock, _gd.QueueFamilyIndex, isLight: true);
+                    _pool = new CommandBufferPool(
+                        _gd.Api,
+                        _device,
+                        queue,
+                        queueLock,
+                        _gd.QueueFamilyIndex,
+                        _gd.IsConcurrentFenceWaitUnsupported,
+                        isLight: true);
                 }
             }
 

src/Ryujinx.Graphics.Vulkan/BufferManager.cs

@@ -103,12 +103,19 @@ namespace Ryujinx.Graphics.Vulkan
                 usage |= BufferUsageFlags.IndirectBufferBit;
             }
 
+            var externalMemoryBuffer = new ExternalMemoryBufferCreateInfo
+            {
+                SType = StructureType.ExternalMemoryBufferCreateInfo,
+                HandleTypes = ExternalMemoryHandleTypeFlags.HostAllocationBitExt,
+            };
+
             var bufferCreateInfo = new BufferCreateInfo
             {
                 SType = StructureType.BufferCreateInfo,
                 Size = (ulong)size,
                 Usage = usage,
                 SharingMode = SharingMode.Exclusive,
+                PNext = &externalMemoryBuffer,
             };
 
             gd.Api.CreateBuffer(_device, in bufferCreateInfo, null, out var buffer).ThrowOnError();

src/Ryujinx.Graphics.Vulkan/CommandBufferPool.cs

@@ -18,6 +18,7 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly Device _device;
         private readonly Queue _queue;
         private readonly object _queueLock;
+        private readonly bool _concurrentFenceWaitUnsupported;
         private readonly CommandPool _pool;
         private readonly Thread _owner;
 
@@ -61,12 +62,20 @@ namespace Ryujinx.Graphics.Vulkan
         private int _queuedCount;
         private int _inUseCount;
 
-        public unsafe CommandBufferPool(Vk api, Device device, Queue queue, object queueLock, uint queueFamilyIndex, bool isLight = false)
+        public unsafe CommandBufferPool(
+            Vk api,
+            Device device,
+            Queue queue,
+            object queueLock,
+            uint queueFamilyIndex,
+            bool concurrentFenceWaitUnsupported,
+            bool isLight = false)
         {
             _api = api;
             _device = device;
             _queue = queue;
             _queueLock = queueLock;
+            _concurrentFenceWaitUnsupported = concurrentFenceWaitUnsupported;
             _owner = Thread.CurrentThread;
 
             var commandPoolCreateInfo = new CommandPoolCreateInfo
@@ -357,7 +366,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             if (refreshFence)
             {
-                entry.Fence = new FenceHolder(_api, _device);
+                entry.Fence = new FenceHolder(_api, _device, _concurrentFenceWaitUnsupported);
             }
             else
             {

src/Ryujinx.Graphics.Vulkan/DescriptorSetUpdater.cs

@@ -73,7 +73,6 @@ namespace Ryujinx.Graphics.Vulkan
 
         private readonly VulkanRenderer _gd;
         private readonly Device _device;
-        private readonly PipelineBase _pipeline;
         private ShaderCollection _program;
 
         private readonly BufferRef[] _uniformBufferRefs;
@@ -125,11 +124,10 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly TextureView _dummyTexture;
         private readonly SamplerHolder _dummySampler;
 
-        public DescriptorSetUpdater(VulkanRenderer gd, Device device, PipelineBase pipeline)
+        public DescriptorSetUpdater(VulkanRenderer gd, Device device)
         {
             _gd = gd;
             _device = device;
-            _pipeline = pipeline;
 
             // Some of the bindings counts needs to be multiplied by 2 because we have buffer and
             // regular textures/images interleaved on the same descriptor set.
@@ -684,7 +682,14 @@ namespace Ryujinx.Graphics.Vulkan
 
             if (_dirty.HasFlag(DirtyFlags.Texture))
             {
-                UpdateAndBind(cbs, program, PipelineBase.TextureSetIndex, pbp);
+                if (program.UpdateTexturesWithoutTemplate)
+                {
+                    UpdateAndBindTexturesWithoutTemplate(cbs, program, pbp);
+                }
+                else
+                {
+                    UpdateAndBind(cbs, program, PipelineBase.TextureSetIndex, pbp);
+                }
             }
 
             if (_dirty.HasFlag(DirtyFlags.Image))
@@ -918,31 +923,84 @@ namespace Ryujinx.Graphics.Vulkan
             _gd.Api.CmdBindDescriptorSets(cbs.CommandBuffer, pbp, _program.PipelineLayout, (uint)setIndex, 1, sets, 0, ReadOnlySpan<uint>.Empty);
         }
 
-        private unsafe void UpdateBuffers(
+        private void UpdateAndBindTexturesWithoutTemplate(CommandBufferScoped cbs, ShaderCollection program, PipelineBindPoint pbp)
-            CommandBufferScoped cbs,
-            PipelineBindPoint pbp,
-            int baseBinding,
-            ReadOnlySpan<DescriptorBufferInfo> bufferInfo,
-            DescriptorType type)
         {
-            if (bufferInfo.Length == 0)
+            int setIndex = PipelineBase.TextureSetIndex;
+            var bindingSegments = program.BindingSegments[setIndex];
+
+            if (bindingSegments.Length == 0)
             {
                 return;
             }
 
-            fixed (DescriptorBufferInfo* pBufferInfo = bufferInfo)
+            if (_updateDescriptorCacheCbIndex)
             {
-                var writeDescriptorSet = new WriteDescriptorSet
+                _updateDescriptorCacheCbIndex = false;
-                {
+                program.UpdateDescriptorCacheCommandBufferIndex(cbs.CommandBufferIndex);
-                    SType = StructureType.WriteDescriptorSet,
-                    DstBinding = (uint)baseBinding,
-                    DescriptorType = type,
-                    DescriptorCount = (uint)bufferInfo.Length,
-                    PBufferInfo = pBufferInfo,
-                };
-
-                _gd.PushDescriptorApi.CmdPushDescriptorSet(cbs.CommandBuffer, pbp, _program.PipelineLayout, 0, 1, &writeDescriptorSet);
             }
+
+            var dsc = program.GetNewDescriptorSetCollection(setIndex, out _).Get(cbs);
+
+            foreach (ResourceBindingSegment segment in bindingSegments)
+            {
+                int binding = segment.Binding;
+                int count = segment.Count;
+
+                if (!segment.IsArray)
+                {
+                    if (segment.Type != ResourceType.BufferTexture)
+                    {
+                        Span<DescriptorImageInfo> textures = _textures;
+
+                        for (int i = 0; i < count; i++)
+                        {
+                            ref var texture = ref textures[i];
+                            ref var refs = ref _textureRefs[binding + i];
+
+                            texture.ImageView = refs.View?.Get(cbs).Value ?? default;
+                            texture.Sampler = refs.Sampler?.Get(cbs).Value ?? default;
+
+                            if (texture.ImageView.Handle == 0)
+                            {
+                                texture.ImageView = _dummyTexture.GetImageView().Get(cbs).Value;
+                            }
+
+                            if (texture.Sampler.Handle == 0)
+                            {
+                                texture.Sampler = _dummySampler.GetSampler().Get(cbs).Value;
+                            }
+                        }
+
+                        dsc.UpdateImages(0, binding, textures[..count], DescriptorType.CombinedImageSampler);
+                    }
+                    else
+                    {
+                        Span<BufferView> bufferTextures = _bufferTextures;
+
+                        for (int i = 0; i < count; i++)
+                        {
+                            bufferTextures[i] = _bufferTextureRefs[binding + i]?.GetBufferView(cbs, false) ?? default;
+                        }
+
+                        dsc.UpdateBufferImages(0, binding, bufferTextures[..count], DescriptorType.UniformTexelBuffer);
+                    }
+                }
+                else
+                {
+                    if (segment.Type != ResourceType.BufferTexture)
+                    {
+                        dsc.UpdateImages(0, binding, _textureArrayRefs[binding].Array.GetImageInfos(_gd, cbs, _dummyTexture, _dummySampler), DescriptorType.CombinedImageSampler);
+                    }
+                    else
+                    {
+                        dsc.UpdateBufferImages(0, binding, _textureArrayRefs[binding].Array.GetBufferViews(cbs), DescriptorType.UniformTexelBuffer);
+                    }
+                }
+            }
+
+            var sets = dsc.GetSets();
+
+            _gd.Api.CmdBindDescriptorSets(cbs.CommandBuffer, pbp, _program.PipelineLayout, (uint)setIndex, 1, sets, 0, ReadOnlySpan<uint>.Empty);
         }
 
         [MethodImpl(MethodImplOptions.AggressiveInlining)]

src/Ryujinx.Graphics.Vulkan/FenceHolder.cs

@@ -10,12 +10,15 @@ namespace Ryujinx.Graphics.Vulkan
         private readonly Device _device;
         private Fence _fence;
         private int _referenceCount;
+        private int _lock;
+        private readonly bool _concurrentWaitUnsupported;
         private bool _disposed;
 
-        public unsafe FenceHolder(Vk api, Device device)
+        public unsafe FenceHolder(Vk api, Device device, bool concurrentWaitUnsupported)
         {
             _api = api;
             _device = device;
+            _concurrentWaitUnsupported = concurrentWaitUnsupported;
 
             var fenceCreateInfo = new FenceCreateInfo
             {
@@ -47,6 +50,11 @@ namespace Ryujinx.Graphics.Vulkan
             }
             while (Interlocked.CompareExchange(ref _referenceCount, lastValue + 1, lastValue) != lastValue);
 
+            if (_concurrentWaitUnsupported)
+            {
+                AcquireLock();
+            }
+
             fence = _fence;
             return true;
         }
@@ -57,6 +65,16 @@ namespace Ryujinx.Graphics.Vulkan
             return _fence;
         }
 
+        public void PutLock()
+        {
+            Put();
+
+            if (_concurrentWaitUnsupported)
+            {
+                ReleaseLock();
+            }
+        }
+
         public void Put()
         {
             if (Interlocked.Decrement(ref _referenceCount) == 0)
@@ -66,24 +84,67 @@ namespace Ryujinx.Graphics.Vulkan
             }
         }
 
+        private void AcquireLock()
+        {
+            while (!TryAcquireLock())
+            {
+                Thread.SpinWait(32);
+            }
+        }
+
+        private bool TryAcquireLock()
+        {
+            return Interlocked.Exchange(ref _lock, 1) == 0;
+        }
+
+        private void ReleaseLock()
+        {
+            Interlocked.Exchange(ref _lock, 0);
+        }
+
         public void Wait()
         {
-            Span<Fence> fences = stackalloc Fence[]
+            if (_concurrentWaitUnsupported)
             {
-                _fence,
+                AcquireLock();
-            };
 
-            FenceHelper.WaitAllIndefinitely(_api, _device, fences);
+                try
+                {
+                    FenceHelper.WaitAllIndefinitely(_api, _device, stackalloc Fence[] { _fence });
+                }
+                finally
+                {
+                    ReleaseLock();
+                }
+            }
+            else
+            {
+                FenceHelper.WaitAllIndefinitely(_api, _device, stackalloc Fence[] { _fence });
+            }
         }
 
         public bool IsSignaled()
         {
-            Span<Fence> fences = stackalloc Fence[]
+            if (_concurrentWaitUnsupported)
             {
-                _fence,
+                if (!TryAcquireLock())
-            };
+                {
+                    return false;
+                }
 
-            return FenceHelper.AllSignaled(_api, _device, fences);
+                try
+                {
+                    return FenceHelper.AllSignaled(_api, _device, stackalloc Fence[] { _fence });
+                }
+                finally
+                {
+                    ReleaseLock();
+                }
+            }
+            else
+            {
+                return FenceHelper.AllSignaled(_api, _device, stackalloc Fence[] { _fence });
+            }
         }
 
         public void Dispose()

src/Ryujinx.Graphics.Vulkan/MultiFenceHolder.cs

@@ -196,18 +196,23 @@ namespace Ryujinx.Graphics.Vulkan
 
             bool signaled = true;
 
-            if (hasTimeout)
+            try
             {
-                signaled = FenceHelper.AllSignaled(api, device, fences[..fenceCount], timeout);
+                if (hasTimeout)
+                {
+                    signaled = FenceHelper.AllSignaled(api, device, fences[..fenceCount], timeout);
+                }
+                else
+                {
+                    FenceHelper.WaitAllIndefinitely(api, device, fences[..fenceCount]);
+                }
             }
-            else
+            finally
             {
-                FenceHelper.WaitAllIndefinitely(api, device, fences[..fenceCount]);
+                for (int i = 0; i < fenceCount; i++)
-            }
+                {
-
+                    fenceHolders[i].PutLock();
-            for (int i = 0; i < fenceCount; i++)
+                }
-            {
-                fenceHolders[i].Put();
             }
 
             return signaled;

src/Ryujinx.Graphics.Vulkan/PipelineBase.cs

@@ -105,7 +105,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             gd.Api.CreatePipelineCache(device, pipelineCacheCreateInfo, null, out PipelineCache).ThrowOnError();
 
-            _descriptorSetUpdater = new DescriptorSetUpdater(gd, device, this);
+            _descriptorSetUpdater = new DescriptorSetUpdater(gd, device);
             _vertexBufferUpdater = new VertexBufferUpdater(gd);
 
             _transformFeedbackBuffers = new BufferState[Constants.MaxTransformFeedbackBuffers];
@@ -1020,6 +1020,13 @@ namespace Ryujinx.Graphics.Vulkan
         {
             _newState.RasterizerDiscardEnable = discard;
             SignalStateChange();
+
+            if (!discard && Gd.Vendor == Vendor.Qualcomm)
+            {
+                // On Adreno, enabling rasterizer discard somehow corrupts the viewport state.
+                // Force it to be updated on next use to work around this bug.
+                DynamicState.ForceAllDirty();
+            }
         }
 
         public void SetRenderTargetColorMasks(ReadOnlySpan<uint> componentMask)

src/Ryujinx.Graphics.Vulkan/ShaderCollection.cs

@@ -23,6 +23,8 @@ namespace Ryujinx.Graphics.Vulkan
         public bool IsCompute { get; }
         public bool HasTessellationControlShader => (Stages & (1u << 3)) != 0;
 
+        public bool UpdateTexturesWithoutTemplate { get; }
+
         public uint Stages { get; }
 
         public ResourceBindingSegment[][] ClearSegments { get; }
@@ -127,9 +129,12 @@ namespace Ryujinx.Graphics.Vulkan
             Stages = stages;
 
             ClearSegments = BuildClearSegments(sets);
-            BindingSegments = BuildBindingSegments(resourceLayout.SetUsages);
+            BindingSegments = BuildBindingSegments(resourceLayout.SetUsages, out bool usesBufferTextures);
             Templates = BuildTemplates(usePushDescriptors);
 
+            // Updating buffer texture bindings using template updates crashes the Adreno driver on Windows.
+            UpdateTexturesWithoutTemplate = gd.Vendor == Vendor.Qualcomm && usesBufferTextures;
+
             _compileTask = Task.CompletedTask;
             _firstBackgroundUse = false;
         }
@@ -280,8 +285,10 @@ namespace Ryujinx.Graphics.Vulkan
             return segments;
         }
 
-        private static ResourceBindingSegment[][] BuildBindingSegments(ReadOnlyCollection<ResourceUsageCollection> setUsages)
+        private static ResourceBindingSegment[][] BuildBindingSegments(ReadOnlyCollection<ResourceUsageCollection> setUsages, out bool usesBufferTextures)
         {
+            usesBufferTextures = false;
+
             ResourceBindingSegment[][] segments = new ResourceBindingSegment[setUsages.Count][];
 
             for (int setIndex = 0; setIndex < setUsages.Count; setIndex++)
@@ -295,6 +302,11 @@ namespace Ryujinx.Graphics.Vulkan
                 {
                     ResourceUsage usage = setUsages[setIndex].Usages[index];
 
+                    if (usage.Type == ResourceType.BufferTexture)
+                    {
+                        usesBufferTextures = true;
+                    }
+
                     if (currentUsage.Binding + currentCount != usage.Binding ||
                         currentUsage.Type != usage.Type ||
                         currentUsage.Stages != usage.Stages ||

src/Ryujinx.Graphics.Vulkan/TextureStorage.cs

@@ -80,7 +80,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             var usage = GetImageUsage(info.Format, info.Target, gd.Capabilities.SupportsShaderStorageImageMultisample);
 
-            var flags = ImageCreateFlags.CreateMutableFormatBit;
+            var flags = ImageCreateFlags.CreateMutableFormatBit | ImageCreateFlags.CreateExtendedUsageBit;
 
             // This flag causes mipmapped texture arrays to break on AMD GCN, so for that copy dependencies are forced for aliasing as cube.
             bool isCube = info.Target == Target.Cubemap || info.Target == Target.CubemapArray;

src/Ryujinx.Graphics.Vulkan/TextureView.cs

@@ -100,7 +100,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             unsafe Auto<DisposableImageView> CreateImageView(ComponentMapping cm, ImageSubresourceRange sr, ImageViewType viewType, ImageUsageFlags usageFlags)
             {
-                var usage = new ImageViewUsageCreateInfo
+                var imageViewUsage = new ImageViewUsageCreateInfo
                 {
                     SType = StructureType.ImageViewUsageCreateInfo,
                     Usage = usageFlags,
@@ -114,7 +114,7 @@ namespace Ryujinx.Graphics.Vulkan
                     Format = format,
                     Components = cm,
                     SubresourceRange = sr,
-                    PNext = &usage,
+                    PNext = &imageViewUsage,
                 };
 
                 gd.Api.CreateImageView(device, imageCreateInfo, null, out var imageView).ThrowOnError();
@@ -123,7 +123,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             ImageUsageFlags shaderUsage = ImageUsageFlags.SampledBit;
 
-            if (info.Format.IsImageCompatible())
+            if (info.Format.IsImageCompatible() && (_gd.Capabilities.SupportsShaderStorageImageMultisample || !info.Target.IsMultisample()))
             {
                 shaderUsage |= ImageUsageFlags.StorageBit;
             }
@@ -154,7 +154,7 @@ namespace Ryujinx.Graphics.Vulkan
                 }
                 else
                 {
-                    subresourceRange = new ImageSubresourceRange(aspectFlags, (uint)firstLevel, levels, (uint)firstLayer, (uint)info.Depth);
+                    subresourceRange = new ImageSubresourceRange(aspectFlags, (uint)firstLevel, 1, (uint)firstLayer, (uint)info.Depth);
 
                     _imageView2dArray = CreateImageView(identityComponentMapping, subresourceRange, ImageViewType.Type2DArray, usage);
                 }

src/Ryujinx.Graphics.Vulkan/VulkanInitialization.cs

@@ -42,6 +42,8 @@ namespace Ryujinx.Graphics.Vulkan
             "VK_EXT_depth_clip_control",
             "VK_KHR_portability_subset", // As per spec, we should enable this if present.
             "VK_EXT_4444_formats",
+            "VK_KHR_8bit_storage",
+            "VK_KHR_maintenance2",
         };
 
         private static readonly string[] _requiredExtensions = {
@@ -355,6 +357,14 @@ namespace Ryujinx.Graphics.Vulkan
                 features2.PNext = &supportedFeaturesDepthClipControl;
             }
 
+            PhysicalDeviceVulkan12Features supportedPhysicalDeviceVulkan12Features = new()
+            {
+                SType = StructureType.PhysicalDeviceVulkan12Features,
+                PNext = features2.PNext,
+            };
+
+            features2.PNext = &supportedPhysicalDeviceVulkan12Features;
+
             api.GetPhysicalDeviceFeatures2(physicalDevice.PhysicalDevice, &features2);
 
             var supportedFeatures = features2.Features;
@@ -382,6 +392,7 @@ namespace Ryujinx.Graphics.Vulkan
                 TessellationShader = supportedFeatures.TessellationShader,
                 VertexPipelineStoresAndAtomics = supportedFeatures.VertexPipelineStoresAndAtomics,
                 RobustBufferAccess = useRobustBufferAccess,
+                SampleRateShading = supportedFeatures.SampleRateShading,
             };
 
             void* pExtendedFeatures = null;
@@ -451,9 +462,11 @@ namespace Ryujinx.Graphics.Vulkan
             {
                 SType = StructureType.PhysicalDeviceVulkan12Features,
                 PNext = pExtendedFeatures,
-                DescriptorIndexing = physicalDevice.IsDeviceExtensionPresent("VK_EXT_descriptor_indexing"),
+                DescriptorIndexing = supportedPhysicalDeviceVulkan12Features.DescriptorIndexing,
-                DrawIndirectCount = physicalDevice.IsDeviceExtensionPresent(KhrDrawIndirectCount.ExtensionName),
+                DrawIndirectCount = supportedPhysicalDeviceVulkan12Features.DrawIndirectCount,
-                UniformBufferStandardLayout = physicalDevice.IsDeviceExtensionPresent("VK_KHR_uniform_buffer_standard_layout"),
+                UniformBufferStandardLayout = supportedPhysicalDeviceVulkan12Features.UniformBufferStandardLayout,
+                UniformAndStorageBuffer8BitAccess = supportedPhysicalDeviceVulkan12Features.UniformAndStorageBuffer8BitAccess,
+                StorageBuffer8BitAccess = supportedPhysicalDeviceVulkan12Features.StorageBuffer8BitAccess,
             };
 
             pExtendedFeatures = &featuresVk12;

src/Ryujinx.Graphics.Vulkan/VulkanRenderer.cs

@@ -90,6 +90,8 @@ namespace Ryujinx.Graphics.Vulkan
         internal bool IsMoltenVk { get; private set; }
         internal bool IsTBDR { get; private set; }
         internal bool IsSharedMemory { get; private set; }
+        internal bool IsConcurrentFenceWaitUnsupported { get; private set; }
+
         public string GpuVendor { get; private set; }
         public string GpuDriver { get; private set; }
         public string GpuRenderer { get; private set; }
@@ -323,6 +325,8 @@ namespace Ryujinx.Graphics.Vulkan
                 Vendor == Vendor.Broadcom ||
                 Vendor == Vendor.ImgTec;
 
+            IsConcurrentFenceWaitUnsupported = Vendor == Vendor.Qualcomm;
+
             GpuVendor = VendorUtils.GetNameFromId(properties.VendorID);
             GpuDriver = hasDriverProperties && !OperatingSystem.IsMacOS() ?
                 VendorUtils.GetFriendlyDriverName(driverProperties.DriverID) : GpuVendor; // Fallback to vendor name if driver is unavailable or on MacOS where vendor is preferred.
@@ -411,7 +415,7 @@ namespace Ryujinx.Graphics.Vulkan
             Api.TryGetDeviceExtension(_instance.Instance, _device, out ExtExternalMemoryHost hostMemoryApi);
             HostMemoryAllocator = new HostMemoryAllocator(MemoryAllocator, Api, hostMemoryApi, _device);
 
-            CommandBufferPool = new CommandBufferPool(Api, _device, Queue, QueueLock, queueFamilyIndex);
+            CommandBufferPool = new CommandBufferPool(Api, _device, Queue, QueueLock, queueFamilyIndex, IsConcurrentFenceWaitUnsupported);
 
             PipelineLayoutCache = new PipelineLayoutCache();
 

src/Ryujinx.Graphics.Vulkan/Window.cs

@@ -623,7 +623,8 @@ namespace Ryujinx.Graphics.Vulkan
 
         public override void SetSize(int width, int height)
         {
-            // Not needed as we can get the size from the surface.
+            // We don't need to use width and height as we can get the size from the surface.
+            _swapchainIsDirty = true;
         }
 
         public override void ChangeVSyncMode(bool vsyncEnabled)

src/Ryujinx.HLE/HOS/Kernel/SupervisorCall/Syscall.cs

@@ -616,7 +616,7 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
                 }
             }
 
-            ArrayPool<KSynchronizationObject>.Shared.Return(syncObjsArray);
+            ArrayPool<KSynchronizationObject>.Shared.Return(syncObjsArray, true);
 
             return result;
         }
@@ -1546,8 +1546,8 @@ namespace Ryujinx.HLE.HOS.Kernel.SupervisorCall
 #pragma warning disable CA1822 // Mark member as static
         public Result SetProcessMemoryPermission(
             int handle,
-            [PointerSized] ulong src,
+            ulong src,
-            [PointerSized] ulong size,
+            ulong size,
             KMemoryPermission permission)
         {
             if (!PageAligned(src))

src/Ryujinx.HLE/HOS/Kernel/Threading/KSynchronization.cs

@@ -104,7 +104,7 @@ namespace Ryujinx.HLE.HOS.Kernel.Threading
                     }
                 }
 
-                ArrayPool<LinkedListNode<KThread>>.Shared.Return(syncNodesArray);
+                ArrayPool<LinkedListNode<KThread>>.Shared.Return(syncNodesArray, true);
             }
 
             _context.CriticalSection.Leave();

src/Ryujinx.UI.Common/DiscordIntegrationModule.cs

@@ -104,8 +104,13 @@ namespace Ryujinx.UI.Common
             // Find the length to trim the string to guarantee we have space for the trailing ellipsis.
             int trimLimit = byteLimit - Encoding.UTF8.GetByteCount(Ellipsis);
 
-            // Basic trim to best case scenario of 1 byte characters.
+            // Make sure the string is long enough to perform the basic trim.
-            input = input[..trimLimit];
+            // Amount of bytes != Length of the string
+            if (input.Length > trimLimit)
+            {
+                // Basic trim to best case scenario of 1 byte characters.
+                input = input[..trimLimit];
+            }
 
             while (Encoding.UTF8.GetByteCount(input) > trimLimit)
             {