OpenTK (#6143 )

Implement SQSHL (immediate) CPU instruction (#6155 )
* Implement SQSHL (immediate) CPU instruction * Fix test
2024-01-25 19:25:47 +01:00 · 2024-01-24 23:50:43 +01:00 · 2024-01-24 19:33:52 -03:00 · 2024-01-24 19:26:59 -03:00 · 2024-01-22 23:10:25 +01:00
10 changed files with 292 additions and 10 deletions
--- a/Directory.Packages.props
+++ b/Directory.Packages.props
@ -28,10 +28,10 @@
    <PackageVersion Include="NetCoreServer" Version="7.0.0" />
    <PackageVersion Include="NUnit" Version="3.13.3" />
    <PackageVersion Include="NUnit3TestAdapter" Version="4.1.0" />
-    <PackageVersion Include="OpenTK.Core" Version="4.8.1" />
-    <PackageVersion Include="OpenTK.Graphics" Version="4.8.1" />
-    <PackageVersion Include="OpenTK.Audio.OpenAL" Version="4.8.1" />
-    <PackageVersion Include="OpenTK.Windowing.GraphicsLibraryFramework" Version="4.8.1" />
+    <PackageVersion Include="OpenTK.Core" Version="4.8.2" />
+    <PackageVersion Include="OpenTK.Graphics" Version="4.8.2" />
+    <PackageVersion Include="OpenTK.Audio.OpenAL" Version="4.8.2" />
+    <PackageVersion Include="OpenTK.Windowing.GraphicsLibraryFramework" Version="4.8.2" />
    <PackageVersion Include="Ryujinx.Audio.OpenAL.Dependencies" Version="1.21.0.1" />
    <PackageVersion Include="Ryujinx.Graphics.Nvdec.Dependencies" Version="5.0.1-build13" />
    <PackageVersion Include="Ryujinx.Graphics.Vulkan.Dependencies.MoltenVK" Version="1.2.0" />
--- a/distribution/macos/shortcut-launch-script.sh
+++ b/distribution/macos/shortcut-launch-script.sh
@ -0,0 +1,8 @@
+#!/bin/sh
+launch_arch="$(uname -m)"
+if [ "$(sysctl -in sysctl.proc_translated)" = "1" ]
+then
+    launch_arch="arm64"
+fi
+
+arch -$launch_arch {0} {1}
--- a/src/ARMeilleure/Decoders/OpCodeTable.cs
+++ b/src/ARMeilleure/Decoders/OpCodeTable.cs
@ -517,7 +517,10 @@ namespace ARMeilleure.Decoders
            SetA64("0x00111100>>>xxx100111xxxxxxxxxx", InstName.Sqrshrn_V,       InstEmit.Sqrshrn_V,       OpCodeSimdShImm.Create);
            SetA64("0111111100>>>xxx100011xxxxxxxxxx", InstName.Sqrshrun_S,      InstEmit.Sqrshrun_S,      OpCodeSimdShImm.Create);
            SetA64("0x10111100>>>xxx100011xxxxxxxxxx", InstName.Sqrshrun_V,      InstEmit.Sqrshrun_V,      OpCodeSimdShImm.Create);
+            SetA64("010111110>>>>xxx011101xxxxxxxxxx", InstName.Sqshl_Si,        InstEmit.Sqshl_Si,        OpCodeSimdShImm.Create);
            SetA64("0>001110<<1xxxxx010011xxxxxxxxxx", InstName.Sqshl_V,         InstEmit.Sqshl_V,         OpCodeSimdReg.Create);
+            SetA64("0000111100>>>xxx011101xxxxxxxxxx", InstName.Sqshl_Vi,        InstEmit.Sqshl_Vi,        OpCodeSimdShImm.Create);
+            SetA64("010011110>>>>xxx011101xxxxxxxxxx", InstName.Sqshl_Vi,        InstEmit.Sqshl_Vi,        OpCodeSimdShImm.Create);
            SetA64("0101111100>>>xxx100101xxxxxxxxxx", InstName.Sqshrn_S,        InstEmit.Sqshrn_S,        OpCodeSimdShImm.Create);
            SetA64("0x00111100>>>xxx100101xxxxxxxxxx", InstName.Sqshrn_V,        InstEmit.Sqshrn_V,        OpCodeSimdShImm.Create);
            SetA64("0111111100>>>xxx100001xxxxxxxxxx", InstName.Sqshrun_S,       InstEmit.Sqshrun_S,       OpCodeSimdShImm.Create);
--- a/src/ARMeilleure/Instructions/InstEmitSimdShift.cs
+++ b/src/ARMeilleure/Instructions/InstEmitSimdShift.cs
@ -116,7 +116,7 @@ namespace ARMeilleure.Instructions
            }
            else if (shift >= eSize)
            {
-                if ((op.RegisterSize == RegisterSize.Simd64))
+                if (op.RegisterSize == RegisterSize.Simd64)
                {
                    Operand res = context.VectorZeroUpper64(GetVec(op.Rd));

@ -359,6 +359,16 @@ namespace ARMeilleure.Instructions
            }
        }

+        public static void Sqshl_Si(ArmEmitterContext context)
+        {
+            EmitShlImmOp(context, signedDst: true, ShlRegFlags.Signed | ShlRegFlags.Scalar | ShlRegFlags.Saturating);
+        }
+
+        public static void Sqshl_Vi(ArmEmitterContext context)
+        {
+            EmitShlImmOp(context, signedDst: true, ShlRegFlags.Signed | ShlRegFlags.Saturating);
+        }
+
        public static void Sqshrn_S(ArmEmitterContext context)
        {
            if (Optimizations.UseAdvSimd)
@ -1593,6 +1603,99 @@ namespace ARMeilleure.Instructions
            Saturating = 1 << 3,
        }

+        private static void EmitShlImmOp(ArmEmitterContext context, bool signedDst, ShlRegFlags flags = ShlRegFlags.None)
+        {
+            bool scalar = flags.HasFlag(ShlRegFlags.Scalar);
+            bool signed = flags.HasFlag(ShlRegFlags.Signed);
+            bool saturating = flags.HasFlag(ShlRegFlags.Saturating);
+
+            OpCodeSimdShImm op = (OpCodeSimdShImm)context.CurrOp;
+
+            Operand res = context.VectorZero();
+
+            int elems = !scalar ? op.GetBytesCount() >> op.Size : 1;
+
+            for (int index = 0; index < elems; index++)
+            {
+                Operand ne = EmitVectorExtract(context, op.Rn, index, op.Size, signed);
+
+                Operand e = !saturating
+                    ? EmitShlImm(context, ne, GetImmShl(op), op.Size)
+                    : EmitShlImmSatQ(context, ne, GetImmShl(op), op.Size, signed, signedDst);
+
+                res = EmitVectorInsert(context, res, e, index, op.Size);
+            }
+
+            context.Copy(GetVec(op.Rd), res);
+        }
+
+        private static Operand EmitShlImm(ArmEmitterContext context, Operand op, int shiftLsB, int size)
+        {
+            int eSize = 8 << size;
+
+            Debug.Assert(op.Type == OperandType.I64);
+            Debug.Assert(eSize == 8 || eSize == 16 || eSize == 32 || eSize == 64);
+
+            Operand res = context.AllocateLocal(OperandType.I64);
+
+            if (shiftLsB >= eSize)
+            {
+                Operand shl = context.ShiftLeft(op, Const(shiftLsB));
+                context.Copy(res, shl);
+            }
+            else
+            {
+                Operand zeroL = Const(0L);
+                context.Copy(res, zeroL);
+            }
+
+            return res;
+        }
+
+        private static Operand EmitShlImmSatQ(ArmEmitterContext context, Operand op, int shiftLsB, int size, bool signedSrc, bool signedDst)
+        {
+            int eSize = 8 << size;
+
+            Debug.Assert(op.Type == OperandType.I64);
+            Debug.Assert(eSize == 8 || eSize == 16 || eSize == 32 || eSize == 64);
+
+            Operand lblEnd = Label();
+
+            Operand res = context.Copy(context.AllocateLocal(OperandType.I64), op);
+
+            if (shiftLsB >= eSize)
+            {
+                context.Copy(res, signedSrc
+                    ? EmitSignedSignSatQ(context, op, size)
+                    : EmitUnsignedSignSatQ(context, op, size));
+            }
+            else
+            {
+                Operand shl = context.ShiftLeft(op, Const(shiftLsB));
+                if (eSize == 64)
+                {
+                    Operand sarOrShr = signedSrc
+                        ? context.ShiftRightSI(shl, Const(shiftLsB))
+                        : context.ShiftRightUI(shl, Const(shiftLsB));
+                    context.Copy(res, shl);
+                    context.BranchIf(lblEnd, sarOrShr, op, Comparison.Equal);
+                    context.Copy(res, signedSrc
+                        ? EmitSignedSignSatQ(context, op, size)
+                        : EmitUnsignedSignSatQ(context, op, size));
+                }
+                else
+                {
+                    context.Copy(res, signedSrc
+                        ? EmitSignedSrcSatQ(context, shl, size, signedDst)
+                        : EmitUnsignedSrcSatQ(context, shl, size, signedDst));
+                }
+            }
+
+            context.MarkLabel(lblEnd);
+
+            return res;
+        }
+
        private static void EmitShlRegOp(ArmEmitterContext context, ShlRegFlags flags = ShlRegFlags.None)
        {
            bool scalar = flags.HasFlag(ShlRegFlags.Scalar);
--- a/src/ARMeilleure/Instructions/InstName.cs
+++ b/src/ARMeilleure/Instructions/InstName.cs
@ -384,7 +384,9 @@ namespace ARMeilleure.Instructions
        Sqrshrn_V,
        Sqrshrun_S,
        Sqrshrun_V,
+        Sqshl_Si,
        Sqshl_V,
+        Sqshl_Vi,
        Sqshrn_S,
        Sqshrn_V,
        Sqshrun_S,
--- a/src/Ryujinx.Graphics.Vulkan/Queries/CounterQueue.cs
+++ b/src/Ryujinx.Graphics.Vulkan/Queries/CounterQueue.cs
@ -67,9 +67,18 @@ namespace Ryujinx.Graphics.Vulkan.Queries
            lock (_queryPool)
            {
                count = Math.Min(count, _queryPool.Count);
-                for (int i = 0; i < count; i++)
+
+                if (count > 0)
                {
-                    _queryPool.ElementAt(i).PoolReset(cmd, ResetSequence);
+                    foreach (BufferedQuery query in _queryPool)
+                    {
+                        query.PoolReset(cmd, ResetSequence);
+
+                        if (--count == 0)
+                        {
+                            break;
+                        }
+                    }
                }
            }
        }
--- a/src/Ryujinx.HLE/FileSystem/VirtualFileSystem.cs
+++ b/src/Ryujinx.HLE/FileSystem/VirtualFileSystem.cs
@ -186,7 +186,12 @@ namespace Ryujinx.HLE.FileSystem

        public void InitializeFsServer(LibHac.Horizon horizon, out HorizonClient fsServerClient)
        {
-            LocalFileSystem serverBaseFs = new(AppDataManager.BaseDirPath);
+            LocalFileSystem serverBaseFs = new(useUnixTimeStamps: true);
+            Result result = serverBaseFs.Initialize(AppDataManager.BaseDirPath, LocalFileSystem.PathMode.DefaultCaseSensitivity, ensurePathExists: true);
+            if (result.IsFailure())
+            {
+                throw new HorizonResultException(result, "Error creating LocalFileSystem.");
+            }

            fsServerClient = horizon.CreatePrivilegedHorizonClient();
            var fsServer = new FileSystemServer(fsServerClient);
--- a/src/Ryujinx.Tests/Cpu/CpuTestSimdShImm.cs
+++ b/src/Ryujinx.Tests/Cpu/CpuTestSimdShImm.cs
@ -311,6 +311,46 @@ namespace Ryujinx.Tests.Cpu
            };
        }

+        private static uint[] _ShlImm_S_D_()
+        {
+            return new[]
+            {
+                0x5F407400u, // SQSHL D0, D0, #0
+            };
+        }
+
+        private static uint[] _ShlImm_V_8B_16B_()
+        {
+            return new[]
+            {
+                0x0F087400u, // SQSHL V0.8B, V0.8B, #0
+            };
+        }
+
+        private static uint[] _ShlImm_V_4H_8H_()
+        {
+            return new[]
+            {
+                0x0F107400u, // SQSHL V0.4H, V0.4H, #0
+            };
+        }
+
+        private static uint[] _ShlImm_V_2S_4S_()
+        {
+            return new[]
+            {
+                0x0F207400u, // SQSHL V0.2S, V0.2S, #0
+            };
+        }
+
+        private static uint[] _ShlImm_V_2D_()
+        {
+            return new[]
+            {
+                0x4F407400u, // SQSHL V0.2D, V0.2D, #0
+            };
+        }
+
        private static uint[] _ShrImm_Sri_S_D_()
        {
            return new[]
@ -813,6 +853,117 @@ namespace Ryujinx.Tests.Cpu
            CompareAgainstUnicorn();
        }

+        [Test, Pairwise]
+        public void ShlImm_S_D([ValueSource(nameof(_ShlImm_S_D_))] uint opcodes,
+                               [Values(0u)] uint rd,
+                               [Values(1u, 0u)] uint rn,
+                               [ValueSource(nameof(_1D_))] ulong z,
+                               [ValueSource(nameof(_1D_))] ulong a,
+                               [Values(1u, 64u)] uint shift)
+        {
+            uint immHb = (64 + shift) & 0x7F;
+
+            opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
+            opcodes |= (immHb << 16);
+
+            V128 v0 = MakeVectorE0E1(z, z);
+            V128 v1 = MakeVectorE0(a);
+
+            SingleOpcode(opcodes, v0: v0, v1: v1);
+
+            CompareAgainstUnicorn();
+        }
+
+        [Test, Pairwise]
+        public void ShlImm_V_8B_16B([ValueSource(nameof(_ShlImm_V_8B_16B_))] uint opcodes,
+                                    [Values(0u)] uint rd,
+                                    [Values(1u, 0u)] uint rn,
+                                    [ValueSource(nameof(_8B_))] ulong z,
+                                    [ValueSource(nameof(_8B_))] ulong a,
+                                    [Values(1u, 8u)] uint shift,
+                                    [Values(0b0u, 0b1u)] uint q) // <8B, 16B>
+        {
+            uint immHb = (8 + shift) & 0x7F;
+
+            opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
+            opcodes |= (immHb << 16);
+            opcodes |= ((q & 1) << 30);
+
+            V128 v0 = MakeVectorE0E1(z, z);
+            V128 v1 = MakeVectorE0E1(a, a * q);
+
+            SingleOpcode(opcodes, v0: v0, v1: v1);
+
+            CompareAgainstUnicorn();
+        }
+
+        [Test, Pairwise]
+        public void ShlImm_V_4H_8H([ValueSource(nameof(_ShlImm_V_4H_8H_))] uint opcodes,
+                                   [Values(0u)] uint rd,
+                                   [Values(1u, 0u)] uint rn,
+                                   [ValueSource(nameof(_4H_))] ulong z,
+                                   [ValueSource(nameof(_4H_))] ulong a,
+                                   [Values(1u, 16u)] uint shift,
+                                   [Values(0b0u, 0b1u)] uint q) // <4H, 8H>
+        {
+            uint immHb = (16 + shift) & 0x7F;
+
+            opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
+            opcodes |= (immHb << 16);
+            opcodes |= ((q & 1) << 30);
+
+            V128 v0 = MakeVectorE0E1(z, z);
+            V128 v1 = MakeVectorE0E1(a, a * q);
+
+            SingleOpcode(opcodes, v0: v0, v1: v1);
+
+            CompareAgainstUnicorn();
+        }
+
+        [Test, Pairwise]
+        public void ShlImm_V_2S_4S([ValueSource(nameof(_ShlImm_V_2S_4S_))] uint opcodes,
+                                   [Values(0u)] uint rd,
+                                   [Values(1u, 0u)] uint rn,
+                                   [ValueSource(nameof(_2S_))] ulong z,
+                                   [ValueSource(nameof(_2S_))] ulong a,
+                                   [Values(1u, 32u)] uint shift,
+                                   [Values(0b0u, 0b1u)] uint q) // <2S, 4S>
+        {
+            uint immHb = (32 + shift) & 0x7F;
+
+            opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
+            opcodes |= (immHb << 16);
+            opcodes |= (((q | (immHb >> 6)) & 1) << 30);
+
+            V128 v0 = MakeVectorE0E1(z, z);
+            V128 v1 = MakeVectorE0E1(a, a * q);
+
+            SingleOpcode(opcodes, v0: v0, v1: v1);
+
+            CompareAgainstUnicorn();
+        }
+
+        [Test, Pairwise]
+        public void ShlImm_V_2D([ValueSource(nameof(_ShlImm_V_2D_))] uint opcodes,
+                                [Values(0u)] uint rd,
+                                [Values(1u, 0u)] uint rn,
+                                [ValueSource(nameof(_1D_))] ulong z,
+                                [ValueSource(nameof(_1D_))] ulong a,
+                                [Values(1u, 64u)] uint shift)
+        {
+            uint immHb = (64 + shift) & 0x7F;
+
+            opcodes |= ((rn & 31) << 5) | ((rd & 31) << 0);
+            opcodes |= (immHb << 16);
+
+            V128 v0 = MakeVectorE0E1(z, z);
+            V128 v1 = MakeVectorE0E1(a, a);
+
+            SingleOpcode(opcodes, v0: v0, v1: v1);
+
+            CompareAgainstUnicorn();
+        }
+
        [Test, Pairwise]
        public void ShrImm_Sri_S_D([ValueSource(nameof(_ShrImm_Sri_S_D_))] uint opcodes,
                                   [Values(0u)] uint rd,
--- a/src/Ryujinx.Ui.Common/Helper/ShortcutHelper.cs
+++ b/src/Ryujinx.Ui.Common/Helper/ShortcutHelper.cs
@ -54,6 +54,7 @@ namespace Ryujinx.Ui.Common.Helper
        {
            string basePath = Path.Combine(AppDomain.CurrentDomain.BaseDirectory, "Ryujinx");
            var plistFile = EmbeddedResources.ReadAllText("Ryujinx.Ui.Common/shortcut-template.plist");
+            var shortcutScript = EmbeddedResources.ReadAllText("Ryujinx.Ui.Common/shortcut-launch-script.sh");
            // Macos .App folder
            string contentFolderPath = Path.Combine("/Applications", cleanedAppName + ".app", "Contents");
            string scriptFolderPath = Path.Combine(contentFolderPath, "MacOS");
@ -68,8 +69,7 @@ namespace Ryujinx.Ui.Common.Helper
            string scriptPath = Path.Combine(scriptFolderPath, ScriptName);
            using StreamWriter scriptFile = new(scriptPath);

-            scriptFile.WriteLine("#!/bin/sh");
-            scriptFile.WriteLine($"{basePath} {GetArgsString(appFilePath)}");
+            scriptFile.Write(shortcutScript, basePath, GetArgsString(appFilePath));

            // Set execute permission
            FileInfo fileInfo = new(scriptPath);
--- a/src/Ryujinx.Ui.Common/Ryujinx.Ui.Common.csproj
+++ b/src/Ryujinx.Ui.Common/Ryujinx.Ui.Common.csproj
@ -51,6 +51,7 @@

  <ItemGroup Condition="'$(RuntimeIdentifier)' == 'osx-x64' OR '$(RuntimeIdentifier)' == 'osx-arm64' OR '$(RuntimeIdentifier)' == ''">
    <EmbeddedResource Include="..\..\distribution\macos\shortcut-template.plist" />
+    <EmbeddedResource Include="..\..\distribution\macos\shortcut-launch-script.sh" />
  </ItemGroup>

  <ItemGroup>
Author	SHA1	Message	Date
Isaac Marovitz	dd2e851e95	OpenTK (#6143 )	2024-01-25 19:25:47 +01:00
gdkchan	2ca70eb9a0	Implement SQSHL (immediate) CPU instruction (#6155 ) * Implement SQSHL (immediate) CPU instruction * Fix test	2024-01-24 23:50:43 +01:00
riperiperi	6575952432	Vulkan: Enumerate Query Pool properly (#6167 ) Turns out that ElementAt for Queue<T> runs the default implementation as it doesn't implement IList, which enumerates elements of the queue up to the given index. This code was creating `count` enumerators and iterating way more queue items than it needed to at higher counts. The solution is just to use one enumerator and break out of the loop when we get the count that we need. 3.5% of backend time was being spent _just_ enumerating at the usual spot in SMO.	2024-01-24 19:33:52 -03:00
gdkchan	9a28ba72b1	Use unix timestamps on GetFileTimeStampRaw (#6169 )	2024-01-24 19:26:59 -03:00
Alex0007	34a9922b57	Fix architecture preference for MacOS game shortcuts (#6145 ) * Fix architecture preference for MacOS game shortcuts * Detect arch and then pass it to script Co-authored-by: jcm <john.moody@coloradocollege.edu> * Remove old script write call in `CreateShortcutMacos` * Turn launch script into EmbeddedResource * Added final newline --------- Co-authored-by: jcm <john.moody@coloradocollege.edu>	2024-01-22 23:10:25 +01:00