Fix CPU FCVTN instruction implementation (slow path) (#4159)

* Fix CPU FCVTN instruction implementation (slow path)

* PPTC version bump

ARMeilleure/Instructions/InstEmitSimdCvt.cs

@@ -381,7 +381,7 @@ namespace ARMeilleure.Instructions
 
                 for (int index = 0; index < elems; index++)
                 {
-                    Operand ne = context.VectorExtract(type, GetVec(op.Rn), 0);
+                    Operand ne = context.VectorExtract(type, GetVec(op.Rn), index);
 
                     if (sizeF == 0)
                     {
@@ -389,8 +389,6 @@ namespace ARMeilleure.Instructions
                         Operand e = context.Call(typeof(SoftFloat32_16).GetMethod(nameof(SoftFloat32_16.FPConvert)), ne);
                         context.LoadFromContext();
 
-                        e = context.ZeroExtend16(OperandType.I64, e);
-
                         res = EmitVectorInsert(context, res, e, part + index, 1);
                     }
                     else /* if (sizeF == 1) */

ARMeilleure/IntermediateRepresentation/Operand.cs

@@ -378,14 +378,7 @@ namespace ARMeilleure.IntermediateRepresentation
 
         public override int GetHashCode()
         {
-            if (Kind == OperandKind.LocalVariable)
-            {
-                return base.GetHashCode();
-            }
-            else
-            {
-                return (int)Value ^ ((int)Kind << 16) ^ ((int)Type << 20);
-            }
+            return ((ulong)_data).GetHashCode();
         }
 
         public bool Equals(Operand operand)

ARMeilleure/Translation/PTC/Ptc.cs

@@ -27,7 +27,7 @@ namespace ARMeilleure.Translation.PTC
         private const string OuterHeaderMagicString = "PTCohd\0\0";
         private const string InnerHeaderMagicString = "PTCihd\0\0";
 
-        private const uint InternalVersion = 4140; //! To be incremented manually for each change to the ARMeilleure project.
+        private const uint InternalVersion = 4159; //! To be incremented manually for each change to the ARMeilleure project.
 
         private const string ActualDir = "0";
         private const string BackupDir = "1";

Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs

@@ -142,6 +142,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
             _drawState.FirstIndex = firstIndex;
             _drawState.IndexCount = indexCount;
+            _drawState.DrawFirstVertex = drawFirstVertex;
+            _drawState.DrawVertexCount = drawVertexCount;
             _currentSpecState.SetHasConstantBufferDrawParameters(false);
 
             engine.UpdateState();
@@ -163,10 +165,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
                 _instancedIndexCount = ibCount != 0 ? ibCount : indexCount;
 
-                var drawState = _state.State.VertexBufferDrawState;
-
-                _instancedDrawStateFirst = drawState.First;
-                _instancedDrawStateCount = drawState.Count;
+                _instancedDrawStateFirst = drawFirstVertex;
+                _instancedDrawStateCount = drawVertexCount;
 
                 _drawState.DrawIndexed = false;
 
@@ -415,6 +415,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             bool oldDrawIndexed = _drawState.DrawIndexed;
 
             _drawState.DrawIndexed = false;
+            engine.ForceStateDirty(VertexBufferFirstMethodOffset * 4);
 
             DrawEnd(engine, 0, 0, firstVertex, vertexCount);
 
@@ -526,8 +527,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             }
             else
             {
-                _state.State.VertexBufferDrawState.First = firstVertex;
-                _state.State.VertexBufferDrawState.Count = count;
+                _drawState.DrawFirstVertex = firstVertex;
+                _drawState.DrawVertexCount = count;
                 engine.ForceStateDirty(VertexBufferFirstMethodOffset * 4);
             }
 

Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs

@@ -17,6 +17,16 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// </summary>
         public int IndexCount;
 
+        /// <summary>
+        /// First vertex used on non-indexed draws. This value is stored somewhere else on indexed draws.
+        /// </summary>
+        public int DrawFirstVertex;
+
+        /// <summary>
+        /// Vertex count used on non-indexed draws. Indexed draws have a index count instead.
+        /// </summary>
+        public int DrawVertexCount;
+
         /// <summary>
         /// Indicates if the next draw will be a indexed draw.
         /// </summary>

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

@@ -989,6 +989,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
             bool drawIndexed = _drawState.DrawIndexed;
             bool drawIndirect = _drawState.DrawIndirect;
+            int drawFirstVertex = _drawState.DrawFirstVertex;
+            int drawVertexCount = _drawState.DrawVertexCount;
             uint vbEnableMask = 0;
 
             for (int index = 0; index < Constants.TotalVertexBuffers; index++)
@@ -1050,9 +1052,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
                     int firstInstance = (int)_state.State.FirstInstance;
 
-                    var drawState = _state.State.VertexBufferDrawState;
-
-                    size = Math.Min(vbSize, (ulong)((firstInstance + drawState.First + drawState.Count) * stride));
+                    size = Math.Min(vbSize, (ulong)((firstInstance + drawFirstVertex + drawVertexCount) * stride));
                 }
 
                 _pipeline.VertexBuffers[index] = new BufferPipelineDescriptor(_channel.MemoryManager.IsMapped(address), stride, divisor);

Ryujinx.Graphics.Gpu/Image/TextureBindingsManager.cs

@@ -320,10 +320,15 @@ namespace Ryujinx.Graphics.Gpu.Image
 
             // Check if the texture pool has been modified since bindings were last committed.
             // If it wasn't, then it's possible to avoid looking up textures again when the handle remains the same.
-            bool poolModified = _cachedTexturePool != texturePool || _cachedSamplerPool != samplerPool;
+            if (_cachedTexturePool != texturePool || _cachedSamplerPool != samplerPool)
+            {
+                Rebind();
 
-            _cachedTexturePool = texturePool;
-            _cachedSamplerPool = samplerPool;
+                _cachedTexturePool = texturePool;
+                _cachedSamplerPool = samplerPool;
+            }
+
+            bool poolModified = false;
 
             if (texturePool != null)
             {

Ryujinx.HLE/HOS/Services/Nv/INvDrvServices.cs

@@ -8,6 +8,8 @@ using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostAsGpu;
 using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostChannel;
 using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostCtrl;
 using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostCtrlGpu;
+using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostDbgGpu;
+using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostProfGpu;
 using Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvMap;
 using Ryujinx.HLE.HOS.Services.Nv.Types;
 using Ryujinx.Memory;
@@ -23,7 +25,13 @@ namespace Ryujinx.HLE.HOS.Services.Nv
     [Service("nvdrv:t")]
     class INvDrvServices : IpcService
     {
-        private static Dictionary<string, Type> _deviceFileRegistry = new Dictionary<string, Type>()
+        private static readonly List<string> _deviceFileDebugRegistry = new List<string>()
+        {
+            "/dev/nvhost-dbg-gpu",
+            "/dev/nvhost-prof-gpu"
+        };
+
+        private static readonly Dictionary<string, Type> _deviceFileRegistry = new Dictionary<string, Type>()
         {
             { "/dev/nvmap",           typeof(NvMapDeviceFile)         },
             { "/dev/nvhost-ctrl",     typeof(NvHostCtrlDeviceFile)    },
@@ -35,6 +43,8 @@ namespace Ryujinx.HLE.HOS.Services.Nv
             //{ "/dev/nvhost-nvjpg",    typeof(NvHostChannelDeviceFile) },
             { "/dev/nvhost-vic",      typeof(NvHostChannelDeviceFile) },
             //{ "/dev/nvhost-display",  typeof(NvHostChannelDeviceFile) },
+            { "/dev/nvhost-dbg-gpu",  typeof(NvHostDbgGpuDeviceFile)  },
+            { "/dev/nvhost-prof-gpu", typeof(NvHostProfGpuDeviceFile) },
         };
 
         public static IdDictionary DeviceFileIdRegistry = new IdDictionary();
@@ -44,13 +54,23 @@ namespace Ryujinx.HLE.HOS.Services.Nv
 
         private bool _transferMemInitialized = false;
 
+        // TODO: This should call set:sys::GetDebugModeFlag
+        private bool _debugModeEnabled = false;
+
         public INvDrvServices(ServiceCtx context) : base(context.Device.System.NvDrvServer)
         {
             _owner = 0;
         }
 
-        private int Open(ServiceCtx context, string path)
+        private NvResult Open(ServiceCtx context, string path, out int fd)
         {
+            fd = -1;
+
+            if (!_debugModeEnabled && _deviceFileDebugRegistry.Contains(path))
+            {
+                return NvResult.NotSupported;
+            }
+
             if (_deviceFileRegistry.TryGetValue(path, out Type deviceFileClass))
             {
                 ConstructorInfo constructor = deviceFileClass.GetConstructor(new Type[] { typeof(ServiceCtx), typeof(IVirtualMemoryManager), typeof(ulong) });
@@ -59,14 +79,14 @@ namespace Ryujinx.HLE.HOS.Services.Nv
 
                 deviceFile.Path = path;
 
-                return DeviceFileIdRegistry.Add(deviceFile);
-            }
-            else
-            {
-                Logger.Warning?.Print(LogClass.ServiceNv, $"Cannot find file device \"{path}\"!");
+                fd = DeviceFileIdRegistry.Add(deviceFile);
+
+                return NvResult.Success;
             }
 
-            return -1;
+            Logger.Warning?.Print(LogClass.ServiceNv, $"Cannot find file device \"{path}\"!");
+
+            return NvResult.FileOperationFailed;
         }
 
         private NvResult GetIoctlArgument(ServiceCtx context, NvIoctl ioctlCommand, out Span<byte> arguments)
@@ -229,12 +249,7 @@ namespace Ryujinx.HLE.HOS.Services.Nv
 
                 string path = MemoryHelper.ReadAsciiString(context.Memory, pathPtr, (long)pathSize);
 
-                fd = Open(context, path);
-
-                if (fd == -1)
-                {
-                    errorCode = NvResult.FileOperationFailed;
-                }
+                errorCode = Open(context, path, out fd);
             }
 
             context.ResponseData.Write(fd);

Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostDbgGpu/NvHostDbgGpuDeviceFile.cs

@@ -0,0 +1,11 @@
+using Ryujinx.Memory;
+using System;
+namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostDbgGpu
+{
+    class NvHostDbgGpuDeviceFile : NvDeviceFile
+    {
+        public NvHostDbgGpuDeviceFile(ServiceCtx context, IVirtualMemoryManager memory, ulong owner) : base(context, owner) { }
+
+        public override void Close() { }
+    }
+}

Ryujinx.HLE/HOS/Services/Nv/NvDrvServices/NvHostProfGpu/NvHostProfGpuDeviceFile.cs

@@ -0,0 +1,11 @@
+using Ryujinx.Memory;
+
+namespace Ryujinx.HLE.HOS.Services.Nv.NvDrvServices.NvHostProfGpu
+{
+    class NvHostProfGpuDeviceFile : NvDeviceFile
+    {
+        public NvHostProfGpuDeviceFile(ServiceCtx context, IVirtualMemoryManager memory, ulong owner) : base(context, owner) { }
+
+        public override void Close() { }
+    }
+}

Ryujinx.HLE/HOS/SystemState/SystemStateMgr.cs

@@ -57,9 +57,9 @@ namespace Ryujinx.HLE.HOS.SystemState
             DesiredTitleLanguage = language switch
             {
                 SystemLanguage.Taiwanese or
-                SystemLanguage.TraditionalChinese => TitleLanguage.Taiwanese,
+                SystemLanguage.TraditionalChinese => TitleLanguage.TraditionalChinese,
                 SystemLanguage.Chinese or
-                SystemLanguage.SimplifiedChinese  => TitleLanguage.Chinese,
+                SystemLanguage.SimplifiedChinese  => TitleLanguage.SimplifiedChinese,
                 _                                 => Enum.Parse<TitleLanguage>(Enum.GetName<SystemLanguage>(language)),
             };
         }

Ryujinx.HLE/HOS/SystemState/TitleLanguage.cs

@@ -15,8 +15,8 @@
         Portuguese,
         Russian,
         Korean,
-        Taiwanese,
-        Chinese,
+        TraditionalChinese,
+        SimplifiedChinese,
         BrazilianPortuguese
     }
 }

Ryujinx.HLE/Ryujinx.HLE.csproj

@@ -11,9 +11,7 @@
     <ProjectReference Include="..\Ryujinx.Graphics.Host1x\Ryujinx.Graphics.Host1x.csproj" />
     <ProjectReference Include="..\Ryujinx.Graphics.Nvdec\Ryujinx.Graphics.Nvdec.csproj" />
     <ProjectReference Include="..\Ryujinx.Graphics.Vic\Ryujinx.Graphics.Vic.csproj" />
-    <ProjectReference Include="..\Ryujinx.Horizon.Generators\Ryujinx.Horizon.Generators.csproj"
-                      OutputItemType="Analyzer"
-                      ReferenceOutputAssembly="false" />
+    <ProjectReference Include="..\Ryujinx.Horizon.Generators\Ryujinx.Horizon.Generators.csproj" OutputItemType="Analyzer" ReferenceOutputAssembly="false" />
     <ProjectReference Include="..\Ryujinx.Memory\Ryujinx.Memory.csproj" />
     <ProjectReference Include="..\ARMeilleure\ARMeilleure.csproj" />
     <ProjectReference Include="..\Ryujinx.Graphics.Gpu\Ryujinx.Graphics.Gpu.csproj" />

Ryujinx.Tests/Cpu/CpuTestSimd.cs

@@ -2176,8 +2176,8 @@ namespace Ryujinx.Tests.Cpu
             opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
             opcodes |= ((q & 1) << 30);
 
-            V128 v0 = MakeVectorE0E1(z, z);
-            V128 v1 = MakeVectorE0E1(a, a);
+            V128 v0 = MakeVectorE0E1(z, a);
+            V128 v1 = MakeVectorE0E1(a, z);
 
             int rnd = (int)TestContext.CurrentContext.Random.NextUInt();
 
@@ -2202,8 +2202,8 @@ namespace Ryujinx.Tests.Cpu
             opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
             opcodes |= ((q & 1) << 30);
 
-            V128 v0 = MakeVectorE0E1(z, z);
-            V128 v1 = MakeVectorE0E1(a, a);
+            V128 v0 = MakeVectorE0E1(z, a);
+            V128 v1 = MakeVectorE0E1(a, z);
 
             SingleOpcode(opcodes, v0: v0, v1: v1);