Do not add more code after alpha test discard on fragment shader (#5529)

* Do not add more code after alpha test discard on fragment shader

* Shader cache version bump

.github/workflows/build.yml

@@ -3,10 +3,6 @@ name: Build job
 on:
   workflow_call:
 
-concurrency:
-  group: pr-builds-${{ github.event.number }}
-  cancel-in-progress: true
-
 env:
   POWERSHELL_TELEMETRY_OPTOUT: 1
   DOTNET_CLI_TELEMETRY_OPTOUT: 1
@@ -54,7 +50,11 @@ jobs:
         run: dotnet build -c "${{ 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 +141,4 @@ 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()

.github/workflows/pr_triage.yml

@@ -28,6 +28,7 @@ jobs:
           dot: true
 
       - name: Assign reviewers
+        if: ! github.event.pull_request.draft
         run: |
           pip3 install PyGithub
           python3 .github/update_reviewers.py ${{ secrets.GITHUB_TOKEN }} ${{ github.repository }} ${{ github.event.pull_request.number }} .github/reviewers.yml

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/StateUpdater.cs

@@ -331,7 +331,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 UpdateShaderState();
             }
 
-            _channel.BufferManager.CommitGraphicsBindings();
+            _channel.BufferManager.CommitGraphicsBindings(_drawState.DrawIndexed);
         }
 
         /// <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/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 = 4675;
+        private const uint CodeGenVersion = 5529;
 
         private const string SharedTocFileName = "shared.toc";
         private const string SharedDataFileName = "shared.data";

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

@@ -162,8 +162,10 @@ namespace Ryujinx.Graphics.Shader.Instructions
 
             if (op.Ccc == Ccc.T)
             {
-                context.PrepareForReturn();
-                context.Return();
+                if (context.PrepareForReturn())
+                {
+                    context.Return();
+                }
             }
             else
             {
@@ -174,8 +176,12 @@ namespace Ryujinx.Graphics.Shader.Instructions
                 {
                     Operand lblSkip = Label();
                     context.BranchIfFalse(lblSkip, cond);
-                    context.PrepareForReturn();
-                    context.Return();
+
+                    if (context.PrepareForReturn())
+                    {
+                        context.Return();
+                    }
+
                     context.MarkLabel(lblSkip);
                 }
             }

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/Translation/EmitterContext.cs

@@ -304,11 +304,11 @@ namespace Ryujinx.Graphics.Shader.Translation
             PrepareForVertexReturn();
         }
 
-        public void PrepareForReturn()
+        public bool PrepareForReturn()
         {
             if (IsNonMain)
             {
-                return;
+                return true;
             }
 
             if (Config.LastInVertexPipeline &&
@@ -383,13 +383,13 @@ namespace Ryujinx.Graphics.Shader.Translation
 
                 AlphaTestOp alphaTestOp = Config.GpuAccessor.QueryAlphaTestCompare();
 
-                if (alphaTestOp != AlphaTestOp.Always && (Config.OmapTargets & 8) != 0)
+                if (alphaTestOp != AlphaTestOp.Always)
                 {
                     if (alphaTestOp == AlphaTestOp.Never)
                     {
                         this.Discard();
                     }
-                    else
+                    else if ((Config.OmapTargets & 8) != 0)
                     {
                         Instruction comparator = alphaTestOp switch
                         {
@@ -415,6 +415,12 @@ namespace Ryujinx.Graphics.Shader.Translation
                     }
                 }
 
+                // We don't need to output anything if alpha test always fails.
+                if (alphaTestOp == AlphaTestOp.Never)
+                {
+                    return false;
+                }
+
                 int regIndexBase = 0;
 
                 for (int rtIndex = 0; rtIndex < 8; rtIndex++)
@@ -462,6 +468,8 @@ namespace Ryujinx.Graphics.Shader.Translation
                     }
                 }
             }
+
+            return true;
         }
 
         private void GenerateAlphaToCoverageDitherDiscard()

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
             };
         }