GPU: Don't sync/bind index buffer when it's not in use (#5526)

* GPU: Don't sync/bind index buffer when it's not in use

Sometimes draws don't use an index buffer. It's not necessary to check or upload data for the current index buffer binding as it won't be used.

This fixes Pokemon: Legends Arceus updating a stale index buffer for every draw during its TFB pass, which was all non-indexed draws.

This probably didn't cost much on normal PCs, but it had a large impact on MacOS, which the macos1 release build avoided by mirroring index buffers (the PR currently does not). Needs buffer mirrors still for the rest of the performance.

There are additional cases where index buffers are bound or checked with non-indexed draws on the backend, but this one was straightforward to fix and has the largest impact. Testing is welcome to ensure nothing weird broke.

* Fix case with _rebind

.github/workflows/build.yml

@@ -54,7 +54,11 @@ jobs:
         run: dotnet build -c "${{ matrix.configuration }}" -p:Version="${{ env.RYUJINX_BASE_VERSION }}" -p:SourceRevisionId="${{ steps.git_short_hash.outputs.result }}" -p:ExtraDefineConstants=DISABLE_UPDATER
 
       - name: Test
-        run: dotnet test --no-build -c "${{ matrix.configuration }}"
+        uses: TSRBerry/unstable-commands@v1
+        with:
+          commands: dotnet test --no-build -c "${{ matrix.configuration }}"
+          timeout-minutes: 10
+          retry-codes: 139
 
       - 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
@@ -141,4 +145,4 @@ jobs:
         with:
           name: ava-ryujinx-${{ matrix.configuration }}-${{ env.RYUJINX_BASE_VERSION }}+${{ steps.git_short_hash.outputs.result }}-macos_universal
           path: "publish_ava/*.tar.gz"
-        if: github.event_name == 'pull_request'
+        if: github.event_name == 'pull_request'

.github/workflows/checks.yml

@@ -40,23 +40,23 @@ jobs:
         run: |
           dotnet format whitespace --verify-no-changes --report ./whitespace-report.json -v d
 
+      # For some unknown reason this step sometimes fails with exit code 139 (segfault?),
+      # so in that case we'll try again (3 tries max).
       - name: Run dotnet format style
-        run: |
-          dotnet format style --severity info --verify-no-changes --report ./style-report.json -v d
+        uses: TSRBerry/unstable-commands@v1
+        with:
+          commands: dotnet format style --severity info --verify-no-changes --report ./style-report.json -v d
+          timeout-minutes: 5
+          retry-codes: 139
 
-      # For some reason this step sometimes fails with exit code 139 (segfault?),
-      # so should that be the case we'll try again (3 tries max).
+      # For some unknown reason this step sometimes fails with exit code 139 (segfault?),
+      # so in that case we'll try again (3 tries max).
       - name: Run dotnet format analyzers
-        run: |
-          attempt=0
-          exit_code=139
-          until [ $attempt -ge 3 ] || [ $exit_code -ne 139 ]; do
-            ((attempt+=1))
-            exit_code=0
-            echo "Attempt: ${attempt}/3"
-            dotnet format analyzers --severity info --verify-no-changes --report ./analyzers-report.json -v d || exit_code=$?
-          done
-          exit $exit_code
+        uses: TSRBerry/unstable-commands@v1
+        with:
+          commands: dotnet format analyzers --severity info --verify-no-changes --report ./analyzers-report.json -v d
+          timeout-minutes: 5
+          retry-codes: 139
 
       - name: Upload report
         if: failure()

Directory.Packages.props

@@ -12,7 +12,7 @@
     <PackageVersion Include="Avalonia.Svg.Skia" Version="0.10.18" />
     <PackageVersion Include="CommandLineParser" Version="2.9.1" />
     <PackageVersion Include="Concentus" Version="1.1.7" />
-    <PackageVersion Include="DiscordRichPresence" Version="1.1.3.18" />
+    <PackageVersion Include="DiscordRichPresence" Version="1.2.1.24" />
     <PackageVersion Include="DynamicData" Version="7.14.2" />
     <PackageVersion Include="FluentAvaloniaUI" Version="1.4.5" />
     <PackageVersion Include="GtkSharp.Dependencies" Version="1.1.1" />

src/ARMeilleure/Decoders/OpCodeTable.cs

@@ -330,6 +330,7 @@ namespace ARMeilleure.Decoders
             SetA64("011111100x110000110010xxxxxxxxxx", InstName.Fmaxnmp_S,       InstEmit.Fmaxnmp_S,       OpCodeSimd.Create);
             SetA64("0>1011100<1xxxxx110001xxxxxxxxxx", InstName.Fmaxnmp_V,       InstEmit.Fmaxnmp_V,       OpCodeSimdReg.Create);
             SetA64("0110111000110000110010xxxxxxxxxx", InstName.Fmaxnmv_V,       InstEmit.Fmaxnmv_V,       OpCodeSimd.Create);
+            SetA64("011111100x110000111110xxxxxxxxxx", InstName.Fmaxp_S,         InstEmit.Fmaxp_S,         OpCodeSimd.Create);
             SetA64("0>1011100<1xxxxx111101xxxxxxxxxx", InstName.Fmaxp_V,         InstEmit.Fmaxp_V,         OpCodeSimdReg.Create);
             SetA64("0110111000110000111110xxxxxxxxxx", InstName.Fmaxv_V,         InstEmit.Fmaxv_V,         OpCodeSimd.Create);
             SetA64("000111100x1xxxxx010110xxxxxxxxxx", InstName.Fmin_S,          InstEmit.Fmin_S,          OpCodeSimdReg.Create);
@@ -339,6 +340,7 @@ namespace ARMeilleure.Decoders
             SetA64("011111101x110000110010xxxxxxxxxx", InstName.Fminnmp_S,       InstEmit.Fminnmp_S,       OpCodeSimd.Create);
             SetA64("0>1011101<1xxxxx110001xxxxxxxxxx", InstName.Fminnmp_V,       InstEmit.Fminnmp_V,       OpCodeSimdReg.Create);
             SetA64("0110111010110000110010xxxxxxxxxx", InstName.Fminnmv_V,       InstEmit.Fminnmv_V,       OpCodeSimd.Create);
+            SetA64("011111101x110000111110xxxxxxxxxx", InstName.Fminp_S,         InstEmit.Fminp_S,         OpCodeSimd.Create);
             SetA64("0>1011101<1xxxxx111101xxxxxxxxxx", InstName.Fminp_V,         InstEmit.Fminp_V,         OpCodeSimdReg.Create);
             SetA64("0110111010110000111110xxxxxxxxxx", InstName.Fminv_V,         InstEmit.Fminv_V,         OpCodeSimd.Create);
             SetA64("010111111xxxxxxx0001x0xxxxxxxxxx", InstName.Fmla_Se,         InstEmit.Fmla_Se,         OpCodeSimdRegElemF.Create);

src/ARMeilleure/Instructions/InstEmitSimdArithmetic.cs

@@ -883,6 +883,31 @@ namespace ARMeilleure.Instructions
             }
         }
 
+        public static void Fmaxp_S(ArmEmitterContext context)
+        {
+            if (Optimizations.UseAdvSimd)
+            {
+                InstEmitSimdHelperArm64.EmitScalarUnaryOpF(context, Intrinsic.Arm64FmaxpS);
+            }
+            else if (Optimizations.FastFP && Optimizations.UseSse41)
+            {
+                EmitSse2ScalarPairwiseOpF(context, (op1, op2) =>
+                {
+                    return EmitSse41ProcessNaNsOpF(context, (op1, op2) =>
+                    {
+                        return EmitSse2VectorMaxMinOpF(context, op1, op2, isMax: true);
+                    }, scalar: true, op1, op2);
+                });
+            }
+            else
+            {
+                EmitScalarPairwiseOpF(context, (op1, op2) =>
+                {
+                    return EmitSoftFloatCall(context, nameof(SoftFloat32.FPMax), op1, op2);
+                });
+            }
+        }
+
         public static void Fmaxp_V(ArmEmitterContext context)
         {
             if (Optimizations.UseAdvSimd)
@@ -1081,6 +1106,31 @@ namespace ARMeilleure.Instructions
             }
         }
 
+        public static void Fminp_S(ArmEmitterContext context)
+        {
+            if (Optimizations.UseAdvSimd)
+            {
+                InstEmitSimdHelperArm64.EmitScalarUnaryOpF(context, Intrinsic.Arm64FminpS);
+            }
+            else if (Optimizations.FastFP && Optimizations.UseSse41)
+            {
+                EmitSse2ScalarPairwiseOpF(context, (op1, op2) =>
+                {
+                    return EmitSse41ProcessNaNsOpF(context, (op1, op2) =>
+                    {
+                        return EmitSse2VectorMaxMinOpF(context, op1, op2, isMax: false);
+                    }, scalar: true, op1, op2);
+                });
+            }
+            else
+            {
+                EmitScalarPairwiseOpF(context, (op1, op2) =>
+                {
+                    return EmitSoftFloatCall(context, nameof(SoftFloat32.FPMin), op1, op2);
+                });
+            }
+        }
+
         public static void Fminp_V(ArmEmitterContext context)
         {
             if (Optimizations.UseAdvSimd)

src/ARMeilleure/Instructions/InstName.cs

@@ -228,6 +228,7 @@ namespace ARMeilleure.Instructions
         Fmaxnmp_S,
         Fmaxnmp_V,
         Fmaxnmv_V,
+        Fmaxp_S,
         Fmaxp_V,
         Fmaxv_V,
         Fmin_S,
@@ -237,6 +238,7 @@ namespace ARMeilleure.Instructions
         Fminnmp_S,
         Fminnmp_V,
         Fminnmv_V,
+        Fminp_S,
         Fminp_V,
         Fminv_V,
         Fmla_Se,

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 = 5343; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 5502; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";

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

@@ -342,5 +342,19 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 Signal();
             }
         }
+
+        /// <summary>
+        /// Sets the Y negate enabled state.
+        /// </summary>
+        /// <param name="enabled">True if Y negate of the fragment coordinates is enabled</param>
+        public void SetYNegateEnabled(bool enabled)
+        {
+            if (enabled != _graphics.YNegateEnabled)
+            {
+                _graphics.YNegateEnabled = enabled;
+
+                Signal();
+            }
+        }
     }
 }

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

@@ -37,6 +37,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
         private ProgramPipelineState _pipeline;
 
+        private bool _fsReadsFragCoord;
         private bool _vsUsesDrawParameters;
         private bool _vtgWritesRtLayer;
         private byte _vsClipDistancesWritten;
@@ -330,7 +331,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 UpdateShaderState();
             }
 
-            _channel.BufferManager.CommitGraphicsBindings();
+            _channel.BufferManager.CommitGraphicsBindings(_drawState.DrawIndexed);
         }
 
         /// <summary>
@@ -692,12 +693,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             var face = _state.State.FaceState;
 
             bool disableTransform = _state.State.ViewportTransformEnable == 0;
+            bool yNegate = yControl.HasFlag(YControl.NegateY);
 
             UpdateFrontFace(yControl, face.FrontFace);
             UpdateDepthMode();
 
-            bool flipY = yControl.HasFlag(YControl.NegateY);
-
             Span<Viewport> viewports = stackalloc Viewport[Constants.TotalViewports];
 
             for (int index = 0; index < Constants.TotalViewports; index++)
@@ -719,7 +719,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 float scaleX = MathF.Abs(transform.ScaleX);
                 float scaleY = transform.ScaleY;
 
-                if (flipY)
+                if (yNegate)
                 {
                     scaleY = -scaleY;
                 }
@@ -771,8 +771,17 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 _channel.TextureManager.RenderTargetScale,
                 disableTransform);
 
+            // Viewport size is only used on the shader when YNegate is enabled,
+            // and if the fragment shader accesses gl_FragCoord,
+            // so there's no need to update it in other cases.
+            if (yNegate && _fsReadsFragCoord)
+            {
+                UpdateSupportBufferViewportSize();
+            }
+
             _currentSpecState.SetViewportTransformDisable(disableTransform);
             _currentSpecState.SetDepthMode(GetDepthMode() == DepthMode.MinusOneToOne);
+            _currentSpecState.SetYNegateEnabled(yNegate);
         }
 
         /// <summary>
@@ -1415,9 +1424,41 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 _currentProgramInfo[stageIndex] = info;
             }
 
+            if (gs.Shaders[5]?.Info.UsesFragCoord == true)
+            {
+                // Make sure we update the viewport size on the support buffer if it will be consumed on the new shader.
+
+                if (!_fsReadsFragCoord && _state.State.YControl.HasFlag(YControl.NegateY))
+                {
+                    UpdateSupportBufferViewportSize();
+                }
+
+                _fsReadsFragCoord = true;
+            }
+            else
+            {
+                _fsReadsFragCoord = false;
+            }
+
             _context.Renderer.Pipeline.SetProgram(gs.HostProgram);
         }
 
+        /// <summary>
+        /// Updates the viewport size on the support buffer for fragment shader access.
+        /// </summary>
+        private void UpdateSupportBufferViewportSize()
+        {
+            ref var transform = ref _state.State.ViewportTransform[0];
+
+            float scaleX = MathF.Abs(transform.ScaleX);
+            float scaleY = transform.ScaleY;
+
+            float width = scaleX * 2;
+            float height = scaleY * 2;
+
+            _context.SupportBufferUpdater.SetViewportSize(width, MathF.Abs(height));
+        }
+
         /// <summary>
         /// Updates bindings consumed by the shader on the texture and buffer managers.
         /// </summary>

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

@@ -515,24 +515,32 @@ namespace Ryujinx.Graphics.Gpu.Memory
         /// Ensures that the graphics engine bindings are visible to the host GPU.
         /// Note: this actually performs the binding using the host graphics API.
         /// </summary>
-        public void CommitGraphicsBindings()
+        /// <param name="indexed">True if the index buffer is in use</param>
+        public void CommitGraphicsBindings(bool indexed)
         {
             var bufferCache = _channel.MemoryManager.Physical.BufferCache;
 
-            if (_indexBufferDirty || _rebind)
+            if (indexed)
             {
-                _indexBufferDirty = false;
-
-                if (_indexBuffer.Address != 0)
+                if (_indexBufferDirty || _rebind)
                 {
-                    BufferRange buffer = bufferCache.GetBufferRange(_indexBuffer.Address, _indexBuffer.Size);
+                    _indexBufferDirty = false;
 
-                    _context.Renderer.Pipeline.SetIndexBuffer(buffer, _indexBuffer.Type);
+                    if (_indexBuffer.Address != 0)
+                    {
+                        BufferRange buffer = bufferCache.GetBufferRange(_indexBuffer.Address, _indexBuffer.Size);
+
+                        _context.Renderer.Pipeline.SetIndexBuffer(buffer, _indexBuffer.Type);
+                    }
+                }
+                else if (_indexBuffer.Address != 0)
+                {
+                    bufferCache.SynchronizeBufferRange(_indexBuffer.Address, _indexBuffer.Size);
                 }
             }
-            else if (_indexBuffer.Address != 0)
+            else if (_rebind)
             {
-                bufferCache.SynchronizeBufferRange(_indexBuffer.Address, _indexBuffer.Size);
+                _indexBufferDirty = true;
             }
 
             uint vbEnableMask = _vertexBuffersEnableMask;

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

@@ -112,6 +112,17 @@ namespace Ryujinx.Graphics.Gpu.Memory
             MarkDirty(SupportBuffer.ViewportInverseOffset, SupportBuffer.FieldSize);
         }
 
+        /// <summary>
+        /// Updates the viewport size vector.
+        /// </summary>
+        /// <param name="data">Viewport size vector</param>
+        private void UpdateViewportSize(Vector4<float> data)
+        {
+            _data.ViewportSize = data;
+
+            MarkDirty(SupportBuffer.ViewportSizeOffset, SupportBuffer.FieldSize);
+        }
+
         /// <summary>
         /// Sets the scale of all output render targets (they should all have the same scale).
         /// </summary>
@@ -192,6 +203,25 @@ namespace Ryujinx.Graphics.Gpu.Memory
             }
         }
 
+        /// <summary>
+        /// Sets the viewport size, used to invert the fragment coordinates Y value.
+        /// </summary>
+        /// <param name="viewportWidth">Value used as viewport width</param>
+        /// <param name="viewportHeight">Value used as viewport height</param>
+        public void SetViewportSize(float viewportWidth, float viewportHeight)
+        {
+            if (_data.ViewportSize.X != viewportWidth || _data.ViewportSize.Y != viewportHeight)
+            {
+                UpdateViewportSize(new Vector4<float>
+                {
+                    X = viewportWidth,
+                    Y = viewportHeight,
+                    Z = 1,
+                    W = 0
+                });
+            }
+        }
+
         /// <summary>
         /// Submits all pending buffer updates to the GPU.
         /// </summary>

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

@@ -247,6 +247,12 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             return _oldSpecState.GraphicsState.ViewportTransformDisable;
         }
 
+        /// <inheritdoc/>
+        public bool QueryYNegateEnabled()
+        {
+            return _oldSpecState.GraphicsState.YNegateEnabled;
+        }
+
         /// <inheritdoc/>
         public void RegisterTexture(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 = 5266;
+        private const uint CodeGenVersion = 4675;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";
@@ -140,6 +140,11 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
             /// </summary>
             public ShaderStage Stage;
 
+            /// <summary>
+            /// Indicates if the fragment shader accesses the fragment coordinate built-in variable.
+            /// </summary>
+            public bool UsesFragCoord;
+
             /// <summary>
             /// Indicates if the shader accesses the Instance ID built-in variable.
             /// </summary>
@@ -781,6 +786,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                 ShaderIdentification.None,
                 0,
                 dataInfo.Stage,
+                dataInfo.UsesFragCoord,
                 dataInfo.UsesInstanceId,
                 dataInfo.UsesDrawParameters,
                 dataInfo.UsesRtLayer,
@@ -807,6 +813,7 @@ namespace Ryujinx.Graphics.Gpu.Shader.DiskCache
                 TexturesCount = (ushort)info.Textures.Count,
                 ImagesCount = (ushort)info.Images.Count,
                 Stage = info.Stage,
+                UsesFragCoord = info.UsesFragCoord,
                 UsesInstanceId = info.UsesInstanceId,
                 UsesDrawParameters = info.UsesDrawParameters,
                 UsesRtLayer = info.UsesRtLayer,

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

@@ -113,6 +113,13 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return _state.GraphicsState.AttributeTypes[location];
         }
 
+        /// <inheritdoc/>
+        public bool QueryEarlyZForce()
+        {
+            _state.SpecializationState?.RecordEarlyZForce();
+            return _state.GraphicsState.EarlyZForce;
+        }
+
         /// <inheritdoc/>
         public AttributeType QueryFragmentOutputType(int location)
         {
@@ -275,19 +282,18 @@ namespace Ryujinx.Graphics.Gpu.Shader
             return _state.TransformFeedbackDescriptors[bufferIndex].Stride;
         }
 
-        /// <inheritdoc/>
-        public bool QueryEarlyZForce()
-        {
-            _state.SpecializationState?.RecordEarlyZForce();
-            return _state.GraphicsState.EarlyZForce;
-        }
-
         /// <inheritdoc/>
         public bool QueryViewportTransformDisable()
         {
             return _state.GraphicsState.ViewportTransformDisable;
         }
 
+        /// <inheritdoc/>
+        public bool QueryYNegateEnabled()
+        {
+            return _state.GraphicsState.YNegateEnabled;
+        }
+
         /// <inheritdoc/>
         public void RegisterTexture(int handle, int cbufSlot)
         {

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

@@ -97,6 +97,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// </summary>
         public bool DualSourceBlendEnable;
 
+        /// <summary>
+        /// Indicates whether Y negate of the fragment coordinates is enabled.
+        /// </summary>
+        public bool YNegateEnabled;
+
         /// <summary>
         /// Creates a new GPU graphics state.
         /// </summary>
@@ -116,7 +121,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
         /// <param name="hasConstantBufferDrawParameters">Indicates that the draw is writing the base vertex, base instance and draw index to Constant Buffer 0</param>
         /// <param name="hasUnalignedStorageBuffer">Indicates that any storage buffer use is unaligned</param>
         /// <param name="fragmentOutputTypes">Type of the fragment shader outputs</param>
-        /// <param name="dualSourceBlendEnable">Type of the vertex attributes consumed by the shader</param>
+        /// <param name="dualSourceBlendEnable">Indicates whether dual source blend is enabled</param>
+        /// <param name="yNegateEnabled">Indicates whether Y negate of the fragment coordinates is enabled</param>
         public GpuChannelGraphicsState(
             bool earlyZForce,
             PrimitiveTopology topology,
@@ -134,7 +140,8 @@ namespace Ryujinx.Graphics.Gpu.Shader
             bool hasConstantBufferDrawParameters,
             bool hasUnalignedStorageBuffer,
             ref Array8<AttributeType> fragmentOutputTypes,
-            bool dualSourceBlendEnable)
+            bool dualSourceBlendEnable,
+            bool yNegateEnabled)
         {
             EarlyZForce = earlyZForce;
             Topology = topology;
@@ -153,6 +160,7 @@ namespace Ryujinx.Graphics.Gpu.Shader
             HasUnalignedStorageBuffer = hasUnalignedStorageBuffer;
             FragmentOutputTypes = fragmentOutputTypes;
             DualSourceBlendEnable = dualSourceBlendEnable;
+            YNegateEnabled = yNegateEnabled;
         }
     }
 }

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

@@ -540,6 +540,11 @@ namespace Ryujinx.Graphics.Gpu.Shader
                 return false;
             }
 
+            if (graphicsState.YNegateEnabled != GraphicsState.YNegateEnabled)
+            {
+                return false;
+            }
+
             return Matches(channel, ref poolState, checkTextures, isCompute: false);
         }
 

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

@@ -188,10 +188,19 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Glsl
                 context.AppendLine();
             }
 
-            if (context.Config.Stage == ShaderStage.Fragment && context.Config.GpuAccessor.QueryEarlyZForce())
+            if (context.Config.Stage == ShaderStage.Fragment)
             {
-                context.AppendLine("layout(early_fragment_tests) in;");
-                context.AppendLine();
+                if (context.Config.GpuAccessor.QueryEarlyZForce())
+                {
+                    context.AppendLine("layout (early_fragment_tests) in;");
+                    context.AppendLine();
+                }
+
+                if (context.Config.Properties.OriginUpperLeft)
+                {
+                    context.AppendLine("layout (origin_upper_left) in vec4 gl_FragCoord;");
+                    context.AppendLine();
+                }
             }
 
             if ((info.HelperFunctionsMask & HelperFunctionsMask.MultiplyHighS32) != 0)

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

@@ -251,7 +251,7 @@ namespace Ryujinx.Graphics.Shader.CodeGen.Spirv
                 }
                 else if (context.Config.Stage == ShaderStage.Fragment)
                 {
-                    context.AddExecutionMode(spvFunc, context.Config.Options.TargetApi == TargetApi.Vulkan
+                    context.AddExecutionMode(spvFunc, context.Config.Properties.OriginUpperLeft
                         ? ExecutionMode.OriginUpperLeft
                         : ExecutionMode.OriginLowerLeft);
 

src/Ryujinx.Graphics.Shader/IGpuAccessor.cs

@@ -178,6 +178,15 @@ namespace Ryujinx.Graphics.Shader
             return 0;
         }
 
+        /// <summary>
+        /// Queries if host state forces early depth testing.
+        /// </summary>
+        /// <returns>True if early depth testing is forced</returns>
+        bool QueryEarlyZForce()
+        {
+            return false;
+        }
+
         /// <summary>
         /// Queries whenever the current draw has written the base vertex and base instance into Constant Buffer 0.
         /// </summary>
@@ -534,19 +543,19 @@ namespace Ryujinx.Graphics.Shader
         }
 
         /// <summary>
-        /// Queries if host state forces early depth testing.
+        /// Queries if host state disables the viewport transform.
         /// </summary>
-        /// <returns>True if early depth testing is forced</returns>
-        bool QueryEarlyZForce()
+        /// <returns>True if the viewport transform is disabled</returns>
+        bool QueryViewportTransformDisable()
         {
             return false;
         }
 
         /// <summary>
-        /// Queries if host state disables the viewport transform.
+        /// Queries Y negate enable state.
         /// </summary>
-        /// <returns>True if the viewport transform is disabled</returns>
-        bool QueryViewportTransformDisable()
+        /// <returns>True if Y negate of the fragment coordinates is enabled, false otherwise</returns>
+        bool QueryYNegateEnabled()
         {
             return false;
         }

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

@@ -161,6 +161,18 @@ namespace Ryujinx.Graphics.Shader.Instructions
                     // FragCoord X/Y must be divided by the render target scale, if resolution scaling is active,
                     // because the shader code is not expecting scaled values.
                     res = context.FPDivide(res, context.Load(StorageKind.ConstantBuffer, SupportBuffer.Binding, Const((int)SupportBufferField.RenderScale), Const(0)));
+
+                    if (op.Imm10 == AttributeConsts.PositionY && context.Config.Options.TargetApi != TargetApi.OpenGL)
+                    {
+                        // If YNegate is enabled, we need to flip the fragment coordinates vertically, unless
+                        // the API supports changing the origin (only OpenGL does).
+                        if (context.Config.GpuAccessor.QueryYNegateEnabled())
+                        {
+                            Operand viewportHeight = context.Load(StorageKind.ConstantBuffer, 0, Const((int)SupportBufferField.ViewportSize), Const(1));
+
+                            res = context.FPSubtract(viewportHeight, res);
+                        }
+                    }
                 }
                 else if (op.Imm10 == AttributeConsts.FrontFacing && context.Config.GpuAccessor.QueryHostHasFrontFacingBug())
                 {

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

@@ -218,7 +218,7 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 return context.Copy(Register(srcB++, RegisterType.Gpr));
             }
 
-            Operand d = dest != RegisterConsts.RegisterZeroIndex ? Register(dest, RegisterType.Gpr) : null;
+            Operand d = Register(dest, RegisterType.Gpr);
 
             List<Operand> sourcesList = new();
 
@@ -304,6 +304,11 @@ namespace Ryujinx.Graphics.Shader.Instructions
             bool byteAddress,
             bool isBindless)
         {
+            if (srcB == RegisterConsts.RegisterZeroIndex)
+            {
+                return;
+            }
+
             context.Config.SetUsedFeature(FeatureFlags.IntegerSampling);
 
             SamplerType type = ConvertSamplerType(dimensions);

src/Ryujinx.Graphics.Shader/ShaderProgramInfo.cs

@@ -13,6 +13,7 @@ namespace Ryujinx.Graphics.Shader
         public ShaderIdentification Identification { get; }
         public int GpLayerInputAttribute { get; }
         public ShaderStage Stage { get; }
+        public bool UsesFragCoord { get; }
         public bool UsesInstanceId { get; }
         public bool UsesDrawParameters { get; }
         public bool UsesRtLayer { get; }
@@ -27,6 +28,7 @@ namespace Ryujinx.Graphics.Shader
             ShaderIdentification identification,
             int gpLayerInputAttribute,
             ShaderStage stage,
+            bool usesFragCoord,
             bool usesInstanceId,
             bool usesDrawParameters,
             bool usesRtLayer,
@@ -41,6 +43,7 @@ namespace Ryujinx.Graphics.Shader
             Identification = identification;
             GpLayerInputAttribute = gpLayerInputAttribute;
             Stage = stage;
+            UsesFragCoord = usesFragCoord;
             UsesInstanceId = usesInstanceId;
             UsesDrawParameters = usesDrawParameters;
             UsesRtLayer = usesRtLayer;

src/Ryujinx.Graphics.Shader/StructuredIr/ShaderProperties.cs

@@ -18,6 +18,8 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
         public IReadOnlyDictionary<int, MemoryDefinition> LocalMemories => _localMemories;
         public IReadOnlyDictionary<int, MemoryDefinition> SharedMemories => _sharedMemories;
 
+        public readonly bool OriginUpperLeft;
+
         public ShaderProperties()
         {
             _constantBuffers = new Dictionary<int, BufferDefinition>();
@@ -28,6 +30,11 @@ namespace Ryujinx.Graphics.Shader.StructuredIr
             _sharedMemories = new Dictionary<int, MemoryDefinition>();
         }
 
+        public ShaderProperties(bool originUpperLeft) : this()
+        {
+            OriginUpperLeft = originUpperLeft;
+        }
+
         public void AddOrUpdateConstantBuffer(int binding, BufferDefinition definition)
         {
             _constantBuffers[binding] = definition;

src/Ryujinx.Graphics.Shader/SupportBuffer.cs

@@ -19,6 +19,7 @@ namespace Ryujinx.Graphics.Shader
         FragmentAlphaTest,
         FragmentIsBgra,
         ViewportInverse,
+        ViewportSize,
         FragmentRenderScaleCount,
         RenderScale,
     }
@@ -33,6 +34,7 @@ namespace Ryujinx.Graphics.Shader
         public static readonly int FragmentAlphaTestOffset;
         public static readonly int FragmentIsBgraOffset;
         public static readonly int ViewportInverseOffset;
+        public static readonly int ViewportSizeOffset;
         public static readonly int FragmentRenderScaleCountOffset;
         public static readonly int GraphicsRenderScaleOffset;
         public static readonly int ComputeRenderScaleOffset;
@@ -56,6 +58,7 @@ namespace Ryujinx.Graphics.Shader
             FragmentAlphaTestOffset = OffsetOf(ref instance, ref instance.FragmentAlphaTest);
             FragmentIsBgraOffset = OffsetOf(ref instance, ref instance.FragmentIsBgra);
             ViewportInverseOffset = OffsetOf(ref instance, ref instance.ViewportInverse);
+            ViewportSizeOffset = OffsetOf(ref instance, ref instance.ViewportSize);
             FragmentRenderScaleCountOffset = OffsetOf(ref instance, ref instance.FragmentRenderScaleCount);
             GraphicsRenderScaleOffset = OffsetOf(ref instance, ref instance.RenderScale);
             ComputeRenderScaleOffset = GraphicsRenderScaleOffset + FieldSize;
@@ -68,6 +71,7 @@ namespace Ryujinx.Graphics.Shader
                 new StructureField(AggregateType.U32, "s_alpha_test"),
                 new StructureField(AggregateType.Array | AggregateType.U32, "s_is_bgra", FragmentIsBgraCount),
                 new StructureField(AggregateType.Vector4 | AggregateType.FP32, "s_viewport_inverse"),
+                new StructureField(AggregateType.Vector4 | AggregateType.FP32, "s_viewport_size"),
                 new StructureField(AggregateType.S32, "s_frag_scale_count"),
                 new StructureField(AggregateType.Array | AggregateType.FP32, "s_render_scale", RenderScaleMaxCount),
             });
@@ -76,6 +80,7 @@ namespace Ryujinx.Graphics.Shader
         public Vector4<int> FragmentAlphaTest;
         public Array8<Vector4<int>> FragmentIsBgra;
         public Vector4<float> ViewportInverse;
+        public Vector4<float> ViewportSize;
         public Vector4<int> FragmentRenderScaleCount;
 
         // Render scale max count: 1 + 64 + 8. First scale is fragment output scale, others are textures/image inputs.

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

@@ -429,6 +429,11 @@ namespace Ryujinx.Graphics.Shader.Translation
             return context.Add(Instruction.FP32 | Instruction.SquareRoot, Local(), a);
         }
 
+        public static Operand FPSubtract(this EmitterContext context, Operand a, Operand b, Instruction fpType = Instruction.FP32)
+        {
+            return context.Add(fpType | Instruction.Subtract, Local(), a, b);
+        }
+
         public static Operand FPTruncate(this EmitterContext context, Operand a, Instruction fpType = Instruction.FP32)
         {
             return context.Add(fpType | Instruction.Truncate, Local(), a);

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

@@ -123,7 +123,20 @@ namespace Ryujinx.Graphics.Shader.Translation
             UsedInputAttributesPerPatch = new HashSet<int>();
             UsedOutputAttributesPerPatch = new HashSet<int>();
 
-            ResourceManager = new ResourceManager(stage, gpuAccessor, new ShaderProperties());
+            ShaderProperties properties;
+
+            switch (stage)
+            {
+                case ShaderStage.Fragment:
+                    bool originUpperLeft = options.TargetApi == TargetApi.Vulkan || gpuAccessor.QueryYNegateEnabled();
+                    properties = new ShaderProperties(originUpperLeft);
+                    break;
+                default:
+                    properties = new ShaderProperties();
+                    break;
+            }
+
+            ResourceManager = new ResourceManager(stage, gpuAccessor, properties);
 
             if (!gpuAccessor.QueryHostSupportsTransformFeedback() && gpuAccessor.QueryTransformFeedbackEnabled())
             {
@@ -135,7 +148,7 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                 BufferDefinition tfeInfoBuffer = new(BufferLayout.Std430, 1, Constants.TfeInfoBinding, "tfe_info", tfeInfoStruct);
 
-                Properties.AddOrUpdateStorageBuffer(Constants.TfeInfoBinding, tfeInfoBuffer);
+                properties.AddOrUpdateStorageBuffer(Constants.TfeInfoBinding, tfeInfoBuffer);
 
                 StructureType tfeDataStruct = new(new StructureField[]
                 {
@@ -146,7 +159,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 {
                     int binding = Constants.TfeBufferBaseBinding + i;
                     BufferDefinition tfeDataBuffer = new(BufferLayout.Std430, 1, binding, $"tfe_data{i}", tfeDataStruct);
-                    Properties.AddOrUpdateStorageBuffer(binding, tfeDataBuffer);
+                    properties.AddOrUpdateStorageBuffer(binding, tfeDataBuffer);
                 }
             }
         }
@@ -615,6 +628,7 @@ namespace Ryujinx.Graphics.Shader.Translation
                 identification,
                 GpLayerInputAttribute,
                 Stage,
+                UsedFeatures.HasFlag(FeatureFlags.FragCoordXY),
                 UsedFeatures.HasFlag(FeatureFlags.InstanceId),
                 UsedFeatures.HasFlag(FeatureFlags.DrawParameters),
                 UsedFeatures.HasFlag(FeatureFlags.RtLayer),

src/Ryujinx.Graphics.Vulkan/VulkanInitialization.cs

@@ -42,6 +42,7 @@ namespace Ryujinx.Graphics.Vulkan
             "VK_NV_viewport_array2",
             "VK_EXT_depth_clip_control",
             "VK_KHR_portability_subset", // As per spec, we should enable this if present.
+            "VK_EXT_4444_formats",
         };
 
         private static readonly string[] _requiredExtensions = {

src/Ryujinx.Tests/Cpu/CpuTestSimd.cs

@@ -764,7 +764,9 @@ namespace Ryujinx.Tests.Cpu
             {
                 0x7E30D820u, // FADDP   S0, V1.2S
                 0x7E30C820u, // FMAXNMP S0, V1.2S
+                0x7E30F820u, // FMAXP   S0, V1.2S
                 0x7EB0C820u, // FMINNMP S0, V1.2S
+                0x7EB0F820u, // FMINP   S0, V1.2S
             };
         }
 
@@ -774,7 +776,9 @@ namespace Ryujinx.Tests.Cpu
             {
                 0x7E70D820u, // FADDP   D0, V1.2D
                 0x7E70C820u, // FMAXNMP D0, V1.2D
+                0x7E70F820u, // FMAXP   D0, V1.2D
                 0x7EF0C820u, // FMINNMP D0, V1.2D
+                0x7EF0F820u, // FMINP   D0, V1.2D
             };
         }