Eliminate redundant multiplications by gl_FragCoord.w on the shader (#4578)

* Eliminate redundant multiplications by gl_FragCoord.w on the shader

* Shader cache version bump

.github/workflows/build.yml

@@ -27,6 +27,7 @@ jobs:
   build:
     name: ${{ matrix.OS_NAME }} (${{ matrix.configuration }})
     runs-on: ${{ matrix.os }}
+    timeout-minutes: 35
     strategy:
       matrix:
         os: [ubuntu-latest, macOS-latest, windows-latest]
@@ -38,7 +39,7 @@ jobs:
           RELEASE_ZIP_OS_NAME: linux_x64
 
         - os: macOS-latest
-          OS_NAME: MacOS x64
+          OS_NAME: macOS x64
           DOTNET_RUNTIME_IDENTIFIER: osx-x64
           RELEASE_ZIP_OS_NAME: osx_x64
 
@@ -68,15 +69,15 @@ jobs:
 
       - name: Publish Ryujinx
         run: dotnet publish -c "${{ matrix.configuration }}" -r "${{ matrix.DOTNET_RUNTIME_IDENTIFIER }}" -o ./publish -p:Version="${{ env.RYUJINX_BASE_VERSION }}" -p:DebugType=embedded -p:SourceRevisionId="${{ steps.git_short_hash.outputs.result }}" -p:ExtraDefineConstants=DISABLE_UPDATER src/Ryujinx --self-contained true
-        if: github.event_name == 'pull_request'
+        if: github.event_name == 'pull_request' && matrix.os != 'macOS-latest'
 
       - name: Publish Ryujinx.Headless.SDL2
         run: dotnet publish -c "${{ matrix.configuration }}" -r "${{ matrix.DOTNET_RUNTIME_IDENTIFIER }}" -o ./publish_sdl2_headless -p:Version="${{ env.RYUJINX_BASE_VERSION }}" -p:DebugType=embedded -p:SourceRevisionId="${{ steps.git_short_hash.outputs.result }}" -p:ExtraDefineConstants=DISABLE_UPDATER src/Ryujinx.Headless.SDL2 --self-contained true
-        if: github.event_name == 'pull_request'
+        if: github.event_name == 'pull_request' && matrix.os != 'macOS-latest'
 
       - name: Publish Ryujinx.Ava
         run: dotnet publish -c "${{ matrix.configuration }}" -r "${{ matrix.DOTNET_RUNTIME_IDENTIFIER }}" -o ./publish_ava -p:Version="${{ env.RYUJINX_BASE_VERSION }}" -p:DebugType=embedded -p:SourceRevisionId="${{ steps.git_short_hash.outputs.result }}" -p:ExtraDefineConstants=DISABLE_UPDATER src/Ryujinx.Ava --self-contained true
-        if: github.event_name == 'pull_request'
+        if: github.event_name == 'pull_request' && matrix.os != 'macOS-latest'
 
       - name: Set executable bit
         run: |
@@ -90,25 +91,26 @@ jobs:
         with:
           name: ryujinx-${{ matrix.configuration }}-${{ env.RYUJINX_BASE_VERSION }}+${{ steps.git_short_hash.outputs.result }}-${{ matrix.RELEASE_ZIP_OS_NAME }}
           path: publish
-        if: github.event_name == 'pull_request'
+        if: github.event_name == 'pull_request' && matrix.os != 'macOS-latest'
 
       - name: Upload Ryujinx.Headless.SDL2 artifact
         uses: actions/upload-artifact@v3
         with:
           name: sdl2-ryujinx-headless-${{ matrix.configuration }}-${{ env.RYUJINX_BASE_VERSION }}+${{ steps.git_short_hash.outputs.result }}-${{ matrix.RELEASE_ZIP_OS_NAME }}
           path: publish_sdl2_headless
-        if: github.event_name == 'pull_request'
+        if: github.event_name == 'pull_request' && matrix.os != 'macOS-latest'
 
       - name: Upload Ryujinx.Ava artifact
         uses: actions/upload-artifact@v3
         with:
           name: ava-ryujinx-${{ matrix.configuration }}-${{ env.RYUJINX_BASE_VERSION }}+${{ steps.git_short_hash.outputs.result }}-${{ matrix.RELEASE_ZIP_OS_NAME }}
           path: publish_ava
-        if: github.event_name == 'pull_request'
+        if: github.event_name == 'pull_request' && matrix.os != 'macOS-latest'
 
   build_macos:
-    name: MacOS universal (${{ matrix.configuration }})
+    name: macOS Universal (${{ matrix.configuration }})
     runs-on: ubuntu-latest
+    timeout-minutes: 35
     strategy:
       matrix:
         configuration: [ Debug, Release ]

.github/workflows/flatpak.yml

@@ -12,6 +12,7 @@ concurrency: flatpak-release
 
 jobs:
   release:
+    timeout-minutes: 35
     runs-on: ubuntu-latest
 
     env:

.github/workflows/nightly_pr_comment.yml

@@ -7,6 +7,7 @@ jobs:
   pr_comment:
     if: github.event.workflow_run.event == 'pull_request' && github.event.workflow_run.conclusion == 'success'
     runs-on: ubuntu-latest
+    timeout-minutes: 35
     steps:
       - uses: actions/github-script@v6
         with:
@@ -65,4 +66,4 @@ jobs:
             } else {
               core.info(`Creating a comment`);
               await github.rest.issues.createComment({repo, owner, issue_number, body});
-            }
+            }

.github/workflows/release.yml

@@ -46,6 +46,7 @@ jobs:
   release:
     name: Release ${{ matrix.OS_NAME }}
     runs-on: ${{ matrix.os }}
+    timeout-minutes: 35
     strategy:
       matrix:
         os: [ ubuntu-latest, windows-latest ]
@@ -143,13 +144,14 @@ jobs:
   macos_release:
     name: Release MacOS universal
     runs-on: ubuntu-latest
+    timeout-minutes: 35
     steps:
       - uses: actions/checkout@v3
 
       - uses: actions/setup-dotnet@v3
         with:
           global-json-file: global.json
-          
+
       - name: Setup LLVM 14
         run: |
           wget https://apt.llvm.org/llvm.sh
@@ -205,4 +207,4 @@ jobs:
     needs: release
     with:
       ryujinx_version: "1.1.${{ github.run_number }}"
-    secrets: inherit
+    secrets: inherit

Directory.Packages.props

@@ -13,7 +13,7 @@
     <PackageVersion Include="CommandLineParser" Version="2.9.1" />
     <PackageVersion Include="Concentus" Version="1.1.7" />
     <PackageVersion Include="DiscordRichPresence" Version="1.1.3.18" />
-    <PackageVersion Include="DynamicData" Version="7.13.5" />
+    <PackageVersion Include="DynamicData" Version="7.13.8" />
     <PackageVersion Include="FluentAvaloniaUI" Version="1.4.5" />
     <PackageVersion Include="GtkSharp.Dependencies" Version="1.1.1" />
     <PackageVersion Include="GtkSharp.Dependencies.osx" Version="0.0.5" />

distribution/macos/create_macos_build.sh

@@ -22,9 +22,9 @@ EXTRA_ARGS=$8
 
 if [ "$VERSION" == "1.1.0" ];
 then
-  RELEASE_TAR_FILE_NAME=Ryujinx-$CONFIGURATION-$VERSION+$SOURCE_REVISION_ID-macos_universal.app.tar
+  RELEASE_TAR_FILE_NAME=test-ava-ryujinx-$CONFIGURATION-$VERSION+$SOURCE_REVISION_ID-macos_universal.app.tar
 else
-  RELEASE_TAR_FILE_NAME=Ryujinx-$VERSION-macos_universal.app.tar
+  RELEASE_TAR_FILE_NAME=test-ava-ryujinx-$VERSION-macos_universal.app.tar
 fi
 
 ARM64_APP_BUNDLE="$TEMP_DIRECTORY/output_arm64/Ryujinx.app"

distribution/macos/updater.sh

@@ -36,4 +36,9 @@ sleep 1
 # Now replace and reopen.
 rm -rf "$INSTALL_DIRECTORY"
 mv "$NEW_APP_DIRECTORY" "$INSTALL_DIRECTORY"
-open -a "$INSTALL_DIRECTORY" --args "$APP_ARGUMENTS"
+
+if [ "$#" -le 3 ]; then
+    open -a "$INSTALL_DIRECTORY"
+else
+    open -a "$INSTALL_DIRECTORY" --args "$APP_ARGUMENTS"
+fi

src/Ryujinx.Audio.Backends.SDL2/SDL2HardwareDeviceDriver.cs

@@ -5,6 +5,7 @@ using Ryujinx.Memory;
 using Ryujinx.SDL2.Common;
 using System;
 using System.Collections.Concurrent;
+using System.Runtime.InteropServices;
 using System.Threading;
 
 using static Ryujinx.Audio.Integration.IHardwareDeviceDriver;
@@ -18,6 +19,13 @@ namespace Ryujinx.Audio.Backends.SDL2
         private readonly ManualResetEvent _pauseEvent;
         private readonly ConcurrentDictionary<SDL2HardwareDeviceSession, byte> _sessions;
 
+        private bool _supportSurroundConfiguration;
+
+        // TODO: Add this to SDL2-CS
+        // NOTE: We use a DllImport here because of marshaling issue for spec.
+        [DllImport("SDL2")]
+        private static extern int SDL_GetDefaultAudioInfo(IntPtr name, out SDL_AudioSpec spec, int isCapture);
+
         public SDL2HardwareDeviceDriver()
         {
             _updateRequiredEvent = new ManualResetEvent(false);
@@ -25,6 +33,20 @@ namespace Ryujinx.Audio.Backends.SDL2
             _sessions = new ConcurrentDictionary<SDL2HardwareDeviceSession, byte>();
 
             SDL2Driver.Instance.Initialize();
+
+            int res = SDL_GetDefaultAudioInfo(IntPtr.Zero, out var spec, 0);
+
+            if (res != 0)
+            {
+                Logger.Error?.Print(LogClass.Application,
+                    $"SDL_GetDefaultAudioInfo failed with error \"{SDL_GetError()}\"");
+
+                _supportSurroundConfiguration = true;
+            }
+            else
+            {
+                _supportSurroundConfiguration = spec.channels == 6;
+            }
         }
 
         public static bool IsSupported => IsSupportedInternal();
@@ -164,6 +186,11 @@ namespace Ryujinx.Audio.Backends.SDL2
 
         public bool SupportsChannelCount(uint channelCount)
         {
+            if (channelCount == 6)
+            {
+                return _supportSurroundConfiguration;
+            }
+
             return true;
         }
 

src/Ryujinx.Audio/Renderer/Device/VirtualDevice.cs

@@ -45,7 +45,7 @@ namespace Ryujinx.Audio.Renderer.Device
         /// <param name="name">The name of the <see cref="VirtualDevice"/>.</param>
         /// <param name="channelCount">The count of channels supported by the <see cref="VirtualDevice"/>.</param>
         /// <param name="isExternalOutput">Indicate if the <see cref="VirtualDevice"/> is provided by an external interface.</param>
-        private VirtualDevice(string name, uint channelCount, bool isExternalOutput)
+        public VirtualDevice(string name, uint channelCount, bool isExternalOutput)
         {
             Name = name;
             ChannelCount = channelCount;

src/Ryujinx.Audio/Renderer/Device/VirtualDeviceSessionRegistry.cs

@@ -1,3 +1,4 @@
+using Ryujinx.Audio.Integration;
 using System.Collections.Generic;
 
 namespace Ryujinx.Audio.Renderer.Device
@@ -22,7 +23,23 @@ namespace Ryujinx.Audio.Renderer.Device
         /// The current active <see cref="VirtualDevice"/>.
         /// </summary>
         // TODO: make this configurable
-        public VirtualDevice ActiveDevice = VirtualDevice.Devices[2];
+        public VirtualDevice ActiveDevice { get; }
+
+        public VirtualDeviceSessionRegistry(IHardwareDeviceDriver driver)
+        {
+            uint channelCount;
+
+            if (driver.GetRealDeviceDriver().SupportsChannelCount(6))
+            {
+                channelCount = 6;
+            }
+            else
+            {
+                channelCount = 2;
+            }
+
+            ActiveDevice = new VirtualDevice("AudioTvOutput", channelCount, false);
+        }
 
         /// <summary>
         /// Get the associated <see cref="T:VirtualDeviceSession[]"/> from an AppletResourceId.

src/Ryujinx.Ava/AppHost.cs

@@ -86,7 +86,7 @@ namespace Ryujinx.Ava
         private KeyboardHotkeyState         _prevHotkeyState;
 
         private long _lastCursorMoveTime;
-        private bool _isCursorInRenderer;
+        private bool _isCursorInRenderer = true;
 
         private bool _isStopped;
         private bool _isActive;
@@ -160,7 +160,9 @@ namespace Ryujinx.Ava
 
             ConfigurationState.Instance.HideCursor.Event += HideCursorState_Changed;
 
-            _topLevel.PointerMoved += TopLevel_PointerMoved;
+            _topLevel.PointerMoved += TopLevel_PointerEnterOrMoved;
+            _topLevel.PointerEnter += TopLevel_PointerEnterOrMoved;
+            _topLevel.PointerLeave += TopLevel_PointerLeave;
 
             if (OperatingSystem.IsWindows())
             {
@@ -183,7 +185,7 @@ namespace Ryujinx.Ava
             _gpuCancellationTokenSource = new CancellationTokenSource();
         }
 
-        private void TopLevel_PointerMoved(object sender, PointerEventArgs e)
+        private void TopLevel_PointerEnterOrMoved(object sender, PointerEventArgs e)
         {
             if (sender is MainWindow window)
             {
@@ -201,6 +203,12 @@ namespace Ryujinx.Ava
                 }
             }
         }
+
+        private void TopLevel_PointerLeave(object sender, PointerEventArgs e)
+        {
+            _isCursorInRenderer = false;
+        }
+
         private void UpdateScalingFilterLevel(object sender, ReactiveEventArgs<int> e)
         {
             _renderer.Window?.SetScalingFilter((Graphics.GAL.ScalingFilter)ConfigurationState.Instance.Graphics.ScalingFilter.Value);
@@ -446,7 +454,9 @@ namespace Ryujinx.Ava
             ConfigurationState.Instance.Graphics.ScalingFilterLevel.Event  -= UpdateScalingFilterLevel;
             ConfigurationState.Instance.Graphics.AntiAliasing.Event        -= UpdateAntiAliasing;
 
-            _topLevel.PointerMoved -= TopLevel_PointerMoved;
+            _topLevel.PointerMoved -= TopLevel_PointerEnterOrMoved;
+            _topLevel.PointerEnter -= TopLevel_PointerEnterOrMoved;
+            _topLevel.PointerLeave -= TopLevel_PointerLeave;
 
             _gpuCancellationTokenSource.Cancel();
             _gpuCancellationTokenSource.Dispose();

src/Ryujinx.Ava/Common/ApplicationHelper.cs

@@ -193,7 +193,7 @@ namespace Ryujinx.Ava.Common
                             if (nca.Header.ContentType == NcaContentType.Program)
                             {
                                 int dataIndex = Nca.GetSectionIndexFromType(NcaSectionType.Data, NcaContentType.Program);
-                                if (nca.Header.GetFsHeader(dataIndex).IsPatchSection())
+                                if (nca.SectionExists(NcaSectionType.Data) && nca.Header.GetFsHeader(dataIndex).IsPatchSection())
                                 {
                                     patchNca = nca;
                                 }

src/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs

@@ -23,6 +23,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         public const int PrimitiveRestartStateIndex = 12;
         public const int RenderTargetStateIndex = 27;
 
+        private const ulong MaxUnknownStorageSize = 0x100000;
+
         private readonly GpuContext _context;
         private readonly GpuChannel _channel;
         private readonly DeviceStateWithShadow<ThreedClassState> _state;
@@ -356,7 +358,19 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
                     SbDescriptor sbDescriptor = _channel.MemoryManager.Physical.Read<SbDescriptor>(sbDescAddress);
 
-                    _channel.BufferManager.SetGraphicsStorageBuffer(stage, sb.Slot, sbDescriptor.PackAddress(), (uint)sbDescriptor.Size, sb.Flags);
+                    uint size;
+                    if (sb.SbCbSlot == 0)
+                    {
+                        // Only trust the SbDescriptor size if it comes from slot 0.
+                        size = (uint)sbDescriptor.Size;
+                    }
+                    else
+                    {
+                        // TODO: Use full mapped size and somehow speed up buffer sync.
+                        size = (uint)_channel.MemoryManager.GetMappedSize(sbDescriptor.PackAddress(), MaxUnknownStorageSize);
+                    }
+
+                    _channel.BufferManager.SetGraphicsStorageBuffer(stage, sb.Slot, sbDescriptor.PackAddress(), size, sb.Flags);
                 }
             }
         }

src/Ryujinx.Graphics.Gpu/GpuContext.cs

@@ -99,6 +99,7 @@ namespace Ryujinx.Graphics.Gpu
         private bool _pendingSync;
 
         private long _modifiedSequence;
+        private ulong _firstTimestamp;
 
         /// <summary>
         /// Creates a new instance of the GPU emulation context.
@@ -123,6 +124,8 @@ namespace Ryujinx.Graphics.Gpu
             DeferredActions = new Queue<Action>();
 
             PhysicalMemoryRegistry = new ConcurrentDictionary<ulong, PhysicalMemory>();
+
+            _firstTimestamp = ConvertNanosecondsToTicks((ulong)PerformanceCounter.ElapsedNanoseconds);
         }
 
         /// <summary>
@@ -217,7 +220,8 @@ namespace Ryujinx.Graphics.Gpu
         /// <returns>The current GPU timestamp</returns>
         public ulong GetTimestamp()
         {
-            ulong ticks = ConvertNanosecondsToTicks((ulong)PerformanceCounter.ElapsedNanoseconds);
+            // Guest timestamp will start at 0, instead of host value. 
+            ulong ticks = ConvertNanosecondsToTicks((ulong)PerformanceCounter.ElapsedNanoseconds) - _firstTimestamp;
 
             if (GraphicsConfig.FastGpuTime)
             {

src/Ryujinx.Graphics.Gpu/Memory/MemoryManager.cs

@@ -637,6 +637,33 @@ namespace Ryujinx.Graphics.Gpu.Memory
             return UnpackPaFromPte(pte) + (va & PageMask);
         }
 
+        /// <summary>
+        /// Translates a GPU virtual address and returns the number of bytes that are mapped after it.
+        /// </summary>
+        /// <param name="va">GPU virtual address to be translated</param>
+        /// <param name="maxSize">Maximum size in bytes to scan</param>
+        /// <returns>Number of bytes, 0 if unmapped</returns>
+        public ulong GetMappedSize(ulong va, ulong maxSize)
+        {
+            if (!ValidateAddress(va))
+            {
+                return 0;
+            }
+
+            ulong startVa = va;
+            ulong endVa = va + maxSize;
+
+            ulong pte = GetPte(va);
+
+            while (pte != PteUnmapped && va < endVa)
+            {
+                va += PageSize - (va & PageMask);
+                pte = GetPte(va);
+            }
+
+            return Math.Min(maxSize, va - startVa);
+        }
+
         /// <summary>
         /// Gets the kind of a given memory page.
         /// This might indicate the type of resource that can be allocated on the page, and also texture tiling.

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 = 4892;
+        private const uint CodeGenVersion = 4578;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

src/Ryujinx.Graphics.OpenGL/Image/TextureView.cs

@@ -306,6 +306,8 @@ namespace Ryujinx.Graphics.OpenGL.Image
             int offset = WriteToPbo2D(range.Offset, layer, level);
 
             Debug.Assert(offset == 0);
+
+            GL.BindBuffer(BufferTarget.PixelPackBuffer, 0);
         }
 
         public void WriteToPbo(int offset, bool forceBgra)

src/Ryujinx.Graphics.OpenGL/Window.cs

@@ -93,7 +93,7 @@ namespace Ryujinx.Graphics.OpenGL
                     oldView.Dispose();
                 }
             }
-            
+
             GL.BindFramebuffer(FramebufferTarget.DrawFramebuffer, drawFramebuffer);
             GL.BindFramebuffer(FramebufferTarget.ReadFramebuffer, readFramebuffer);
 

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

@@ -20,6 +20,12 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
                 GlobalToStorage.RunPass(blocks[blkIndex], config, ref sbUseMask, ref ubeUseMask);
                 BindlessToIndexed.RunPass(blocks[blkIndex], config);
                 BindlessElimination.RunPass(blocks[blkIndex], config);
+
+                // FragmentCoord only exists on fragment shaders, so we don't need to check other stages.
+                if (config.Stage == ShaderStage.Fragment)
+                {
+                    EliminateMultiplyByFragmentCoordW(blocks[blkIndex]);
+                }
             }
 
             config.SetAccessibleBufferMasks(sbUseMask, ubeUseMask);
@@ -281,6 +287,75 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
             return modified;
         }
 
+        private static void EliminateMultiplyByFragmentCoordW(BasicBlock block)
+        {
+            foreach (INode node in block.Operations)
+            {
+                if (node is Operation operation)
+                {
+                    EliminateMultiplyByFragmentCoordW(operation);
+                }
+            }
+        }
+
+        private static void EliminateMultiplyByFragmentCoordW(Operation operation)
+        {
+            // We're looking for the pattern:
+            //  y = x * gl_FragCoord.w
+            //  v = y * (1.0 / gl_FragCoord.w)
+            // Then we transform it into:
+            //  v = x
+            // This pattern is common on fragment shaders due to the way how perspective correction is done.
+
+            // We are expecting a multiplication by the reciprocal of gl_FragCoord.w.
+            if (operation.Inst != (Instruction.FP32 | Instruction.Multiply))
+            {
+                return;
+            }
+
+            Operand lhs = operation.GetSource(0);
+            Operand rhs = operation.GetSource(1);
+
+            // Check LHS of the the main multiplication operation. We expect an input being multiplied by gl_FragCoord.w.
+            if (!(lhs.AsgOp is Operation attrMulOp) || attrMulOp.Inst != (Instruction.FP32 | Instruction.Multiply))
+            {
+                return;
+            }
+
+            Operand attrMulLhs = attrMulOp.GetSource(0);
+            Operand attrMulRhs = attrMulOp.GetSource(1);
+
+            // LHS should be any input, RHS should be exactly gl_FragCoord.w.
+            if (!Utils.IsInputLoad(attrMulLhs.AsgOp) || !Utils.IsInputLoad(attrMulRhs.AsgOp, IoVariable.FragmentCoord, 3))
+            {
+                return;
+            }
+
+            // RHS of the main multiplication should be a reciprocal operation (1.0 / x).
+            if (!(rhs.AsgOp is Operation reciprocalOp) || reciprocalOp.Inst != (Instruction.FP32 | Instruction.Divide))
+            {
+                return;
+            }
+
+            Operand reciprocalLhs = reciprocalOp.GetSource(0);
+            Operand reciprocalRhs = reciprocalOp.GetSource(1);
+
+            // Check if the divisor is a constant equal to 1.0.
+            if (reciprocalLhs.Type != OperandType.Constant || reciprocalLhs.AsFloat() != 1.0f)
+            {
+                return;
+            }
+
+            // Check if the dividend is gl_FragCoord.w.
+            if (!Utils.IsInputLoad(reciprocalRhs.AsgOp, IoVariable.FragmentCoord, 3))
+            {
+                return;
+            }
+
+            // If everything matches, we can replace the operation with the input load result.
+            operation.TurnIntoCopy(attrMulLhs);
+        }
+
         private static void RemoveNode(BasicBlock block, LinkedListNode<INode> llNode)
         {
             // Remove a node from the nodes list, and also remove itself

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

@@ -4,6 +4,35 @@ namespace Ryujinx.Graphics.Shader.Translation.Optimizations
 {
     static class Utils
     {
+        public static bool IsInputLoad(INode node)
+        {
+            return (node is Operation operation) &&
+                   operation.Inst == Instruction.Load &&
+                   operation.StorageKind == StorageKind.Input;
+        }
+
+        public static bool IsInputLoad(INode node, IoVariable ioVariable, int elemIndex)
+        {
+            if (!(node is Operation operation) ||
+                operation.Inst != Instruction.Load ||
+                operation.StorageKind != StorageKind.Input ||
+                operation.SourcesCount != 2)
+            {
+                return false;
+            }
+
+            Operand ioVariableSrc = operation.GetSource(0);
+
+            if (ioVariableSrc.Type != OperandType.Constant || (IoVariable)ioVariableSrc.Value != ioVariable)
+            {
+                return false;
+            }
+
+            Operand elemIndexSrc = operation.GetSource(1);
+
+            return elemIndexSrc.Type == OperandType.Constant && elemIndexSrc.Value == elemIndex;
+        }
+
         private static Operation FindBranchSource(BasicBlock block)
         {
             foreach (BasicBlock sourceBlock in block.Predecessors)

src/Ryujinx.Graphics.Vulkan/BufferHolder.cs

@@ -56,6 +56,7 @@ namespace Ryujinx.Graphics.Vulkan
         private int _writeCount;
         private int _flushCount;
         private int _flushTemp;
+        private int _lastFlushWrite = -1;
 
         private ReaderWriterLock _flushLock;
         private FenceHolder _flushFence;
@@ -200,14 +201,21 @@ namespace Ryujinx.Graphics.Vulkan
         {
             if (_baseType == BufferAllocationType.Auto)
             {
-                if (_writeCount >= WriteCountThreshold || _setCount >= SetCountThreshold || _flushCount >= FlushCountThreshold)
+                // When flushed, wait for a bit more info to make a decision.
+                bool wasFlushed = _flushTemp > 0;
+                int multiplier = wasFlushed ? 2 : 0;
+                if (_writeCount >= (WriteCountThreshold << multiplier) || _setCount >= (SetCountThreshold << multiplier) || _flushCount >= (FlushCountThreshold << multiplier))
                 {
                     if (_flushCount > 0 || _flushTemp-- > 0)
                     {
                         // Buffers that flush should ideally be mapped in host address space for easy copies.
                         // If the buffer is large it will do better on GPU memory, as there will be more writes than data flushes (typically individual pages).
                         // If it is small, then it's likely most of the buffer will be flushed so we want it on host memory, as access is cached.
-                        DesiredType = Size > DeviceLocalSizeThreshold ? BufferAllocationType.DeviceLocalMapped : BufferAllocationType.HostMapped;
+
+                        bool hostMappingSensitive = _gd.Vendor == Vendor.Nvidia;
+                        bool deviceLocalMapped = Size > DeviceLocalSizeThreshold || (wasFlushed && _writeCount > _flushCount * 10 && hostMappingSensitive) || _currentType == BufferAllocationType.DeviceLocalMapped;
+
+                        DesiredType = deviceLocalMapped ? BufferAllocationType.DeviceLocalMapped : BufferAllocationType.HostMapped;
 
                         // It's harder for a buffer that is flushed to revert to another type of mapping.
                         if (_flushCount > 0)
@@ -215,17 +223,18 @@ namespace Ryujinx.Graphics.Vulkan
                             _flushTemp = 1000;
                         }
                     }
-                    else if (_writeCount >= WriteCountThreshold)
+                    else if (_writeCount >= (WriteCountThreshold << multiplier))
                     {
                         // Buffers that are written often should ideally be in the device local heap. (Storage buffers)
                         DesiredType = BufferAllocationType.DeviceLocal;
                     }
-                    else if (_setCount > SetCountThreshold)
+                    else if (_setCount > (SetCountThreshold << multiplier))
                     {
                         // Buffers that have their data set often should ideally be host mapped. (Constant buffers)
                         DesiredType = BufferAllocationType.HostMapped;
                     }
 
+                    _lastFlushWrite = -1;
                     _flushCount = 0;
                     _writeCount = 0;
                     _setCount = 0;
@@ -418,7 +427,12 @@ namespace Ryujinx.Graphics.Vulkan
 
             WaitForFlushFence();
 
-            _flushCount++;
+            if (_lastFlushWrite != _writeCount)
+            {
+                // If it's on the same page as the last flush, ignore it.
+                _lastFlushWrite = _writeCount;
+                _flushCount++;
+            }
 
             Span<byte> result;
 

src/Ryujinx.Graphics.Vulkan/DescriptorSetUpdater.cs

@@ -228,7 +228,12 @@ namespace Ryujinx.Graphics.Vulkan
             SignalDirty(DirtyFlags.Storage);
         }
 
-        public void SetTextureAndSampler(CommandBufferScoped cbs, ShaderStage stage, int binding, ITexture texture, ISampler sampler)
+        public void SetTextureAndSampler(
+            CommandBufferScoped cbs,
+            ShaderStage stage,
+            int binding,
+            ITexture texture,
+            ISampler sampler)
         {
             if (texture is TextureBuffer textureBuffer)
             {
@@ -251,6 +256,28 @@ namespace Ryujinx.Graphics.Vulkan
             SignalDirty(DirtyFlags.Texture);
         }
 
+        public void SetTextureAndSamplerIdentitySwizzle(
+            CommandBufferScoped cbs,
+            ShaderStage stage,
+            int binding,
+            ITexture texture,
+            ISampler sampler)
+        {
+            if (texture is TextureView view)
+            {
+                view.Storage.InsertWriteToReadBarrier(cbs, AccessFlags.ShaderReadBit, stage.ConvertToPipelineStageFlags());
+
+                _textureRefs[binding] = view.GetIdentityImageView();
+                _samplerRefs[binding] = ((SamplerHolder)sampler)?.GetSampler();
+
+                SignalDirty(DirtyFlags.Texture);
+            }
+            else
+            {
+                SetTextureAndSampler(cbs, stage, binding, texture, sampler);
+            }
+        }
+
         public void SetUniformBuffers(CommandBuffer commandBuffer, ReadOnlySpan<BufferAssignment> buffers)
         {
             for (int i = 0; i < buffers.Length; i++)

src/Ryujinx.Graphics.Vulkan/HelperShader.cs

@@ -415,7 +415,7 @@ namespace Ryujinx.Graphics.Vulkan
 
             var sampler = linearFilter ? _samplerLinear : _samplerNearest;
 
-            _pipeline.SetTextureAndSampler(ShaderStage.Fragment, 0, src, sampler);
+            _pipeline.SetTextureAndSamplerIdentitySwizzle(ShaderStage.Fragment, 0, src, sampler);
 
             Span<float> region = stackalloc float[RegionBufferSize / sizeof(float)];
 
@@ -625,7 +625,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         private void BlitDepthStencilDraw(TextureView src, bool isDepth)
         {
-            _pipeline.SetTextureAndSampler(ShaderStage.Fragment, 0, src, _samplerNearest);
+            _pipeline.SetTextureAndSamplerIdentitySwizzle(ShaderStage.Fragment, 0, src, _samplerNearest);
 
             if (isDepth)
             {
@@ -1037,7 +1037,7 @@ namespace Ryujinx.Graphics.Vulkan
                     var srcView = Create2DLayerView(src, srcLayer + z, srcLevel + l, srcFormat);
                     var dstView = Create2DLayerView(dst, dstLayer + z, dstLevel + l);
 
-                    _pipeline.SetTextureAndSampler(ShaderStage.Compute, 0, srcView, null);
+                    _pipeline.SetTextureAndSamplerIdentitySwizzle(ShaderStage.Compute, 0, srcView, null);
                     _pipeline.SetImage(0, dstView, dstFormat);
 
                     int dispatchX = (Math.Min(srcView.Info.Width, dstView.Info.Width) + 31) / 32;
@@ -1177,7 +1177,7 @@ namespace Ryujinx.Graphics.Vulkan
                     var srcView = Create2DLayerView(src, srcLayer + z, 0, format);
                     var dstView = Create2DLayerView(dst, dstLayer + z, 0);
 
-                    _pipeline.SetTextureAndSampler(ShaderStage.Compute, 0, srcView, null);
+                    _pipeline.SetTextureAndSamplerIdentitySwizzle(ShaderStage.Compute, 0, srcView, null);
                     _pipeline.SetImage(0, dstView, format);
 
                     _pipeline.DispatchCompute(dispatchX, dispatchY, 1);
@@ -1313,7 +1313,7 @@ namespace Ryujinx.Graphics.Vulkan
                     var srcView = Create2DLayerView(src, srcLayer + z, 0, format);
                     var dstView = Create2DLayerView(dst, dstLayer + z, 0);
 
-                    _pipeline.SetTextureAndSampler(ShaderStage.Fragment, 0, srcView, null);
+                    _pipeline.SetTextureAndSamplerIdentitySwizzle(ShaderStage.Fragment, 0, srcView, null);
                     _pipeline.SetRenderTarget(
                         ((TextureView)dstView).GetView(format).GetImageViewForAttachment(),
                         (uint)dst.Width,
@@ -1384,7 +1384,7 @@ namespace Ryujinx.Graphics.Vulkan
 
         private void CopyMSAspectDraw(TextureView src, bool fromMS, bool isDepth)
         {
-            _pipeline.SetTextureAndSampler(ShaderStage.Fragment, 0, src, _samplerNearest);
+            _pipeline.SetTextureAndSamplerIdentitySwizzle(ShaderStage.Fragment, 0, src, _samplerNearest);
 
             if (isDepth)
             {

src/Ryujinx.Graphics.Vulkan/PipelineBase.cs

@@ -1098,6 +1098,11 @@ namespace Ryujinx.Graphics.Vulkan
             _descriptorSetUpdater.SetTextureAndSampler(Cbs, stage, binding, texture, sampler);
         }
 
+        public void SetTextureAndSamplerIdentitySwizzle(ShaderStage stage, int binding, ITexture texture, ISampler sampler)
+        {
+            _descriptorSetUpdater.SetTextureAndSamplerIdentitySwizzle(Cbs, stage, binding, texture, sampler);
+        }
+
         public void SetTransformFeedbackBuffers(ReadOnlySpan<BufferRange> buffers)
         {
             PauseTransformFeedbackInternal();

src/Ryujinx.HLE/HOS/Horizon.cs

@@ -261,7 +261,7 @@ namespace Ryujinx.HLE.HOS
             AudioInputManager = new AudioInputManager();
             AudioRendererManager = new AudioRendererManager(tickSource);
             AudioRendererManager.SetVolume(Device.Configuration.AudioVolume);
-            AudioDeviceSessionRegistry = new VirtualDeviceSessionRegistry();
+            AudioDeviceSessionRegistry = new VirtualDeviceSessionRegistry(Device.AudioDeviceDriver);
 
             IWritableEvent[] audioOutputRegisterBufferEvents = new IWritableEvent[Constants.AudioOutSessionCountMax];
 

src/Ryujinx/Ui/RendererWidgetBase.cs

@@ -321,27 +321,27 @@ namespace Ryujinx.Ui
 
             _toggleDockedMode = toggleDockedMode;
 
-            if (ConfigurationState.Instance.Hid.EnableMouse.Value)
+            if (_isMouseInClient)
             {
-                if (_isMouseInClient)
+                if (ConfigurationState.Instance.Hid.EnableMouse.Value)
                 {
                     Window.Cursor = _invisibleCursor;
                 }
-            }
+                else
-            else
-            {
-                switch (_hideCursorMode)
                 {
-                    case HideCursorMode.OnIdle:
+                    switch (_hideCursorMode)
-                        long cursorMoveDelta = Stopwatch.GetTimestamp() - _lastCursorMoveTime;
+                    {
-                        Window.Cursor = (cursorMoveDelta >= CursorHideIdleTime * Stopwatch.Frequency) ? _invisibleCursor : null;
+                        case HideCursorMode.OnIdle:
-                        break;
+                            long cursorMoveDelta = Stopwatch.GetTimestamp() - _lastCursorMoveTime;
-                    case HideCursorMode.Always:
+                            Window.Cursor = (cursorMoveDelta >= CursorHideIdleTime * Stopwatch.Frequency) ? _invisibleCursor : null;
-                        Window.Cursor = _invisibleCursor;
+                            break;
-                        break;
+                        case HideCursorMode.Always:
-                    case HideCursorMode.Never:
+                            Window.Cursor = _invisibleCursor;
-                        Window.Cursor = null;
+                            break;
-                        break;
+                        case HideCursorMode.Never:
+                            Window.Cursor = null;
+                            break;
+                    }
                 }
             }
         }