Implement a new JIT for Arm devices (#6057)

* Implement a new JIT for Arm devices

* Auto-format

* Make a lot of Assembler members read-only

* More read-only

* Fix more warnings

* ObjectDisposedException.ThrowIf

* New JIT cache for platforms that enforce W^X, currently unused

* Remove unused using

* Fix assert

* Pass memory manager type around

* Safe memory manager mode support + other improvements

* Actual safe memory manager mode masking support

* PR feedback

5 files changed, 281 insertions, 0 deletions

diff --git a/src/Ryujinx.Cpu/LightningJit/Graph/DataFlow.cs b/src/Ryujinx.Cpu/LightningJit/Graph/DataFlow.cs
new file mode 100644
index 00000000..e5b369d6
--- /dev/null
+++ b/src/Ryujinx.Cpu/LightningJit/Graph/DataFlow.cs
@@ -0,0 +1,171 @@
+namespace Ryujinx.Cpu.LightningJit.Graph
+{
+    static class DataFlow
+    {
+        public static (RegisterMask[], RegisterMask[]) GetGlobalUses(IBlockList blocks)
+        {
+            // Compute local register inputs and outputs used inside blocks.
+            RegisterMask[] localInputs = new RegisterMask[blocks.Count];
+            RegisterMask[] localOutputs = new RegisterMask[blocks.Count];
+
+            for (int index = 0; index < blocks.Count; index++)
+            {
+                IBlock block = blocks[index];
+
+                RegisterUse use = block.ComputeUseMasks();
+
+                localInputs[block.Index] = use.Read;
+                localOutputs[block.Index] = use.Write;
+            }
+
+            // Compute global register inputs and outputs used across blocks.
+            RegisterMask[] globalInputs = new RegisterMask[blocks.Count];
+            RegisterMask[] globalOutputs = new RegisterMask[blocks.Count];
+
+            bool modified;
+
+            // Compute register outputs.
+            do
+            {
+                modified = false;
+
+                for (int index = 0; index < blocks.Count; index++)
+                {
+                    IBlock block = blocks[index];
+
+                    int firstPIndex = GetFirstPredecessorIndex(block);
+                    if (firstPIndex >= 0)
+                    {
+                        IBlock predecessor = block.GetPredecessor(firstPIndex);
+
+                        RegisterMask outputs = globalOutputs[predecessor.Index];
+
+                        for (int pIndex = firstPIndex + 1; pIndex < block.PredecessorsCount; pIndex++)
+                        {
+                            predecessor = block.GetPredecessor(pIndex);
+
+                            if (predecessor.EndsWithContextStore())
+                            {
+                                // If a block ended with a context store, then we don't need to care
+                                // about any of it's outputs, as they have already been saved to the context.
+                                // Common outputs must be reset as doing a context stores indicates we will
+                                // do a function call and wipe all register values.
+
+                                continue;
+                            }
+
+                            outputs |= globalOutputs[predecessor.Index];
+                        }
+
+                        outputs |= localOutputs[block.Index];
+                        modified |= Exchange(globalOutputs, block.Index, globalOutputs[block.Index] | outputs);
+                    }
+                    else
+                    {
+                        modified |= Exchange(globalOutputs, block.Index, localOutputs[block.Index]);
+                    }
+                }
+            }
+            while (modified);
+
+            // Compute register inputs.
+            do
+            {
+                modified = false;
+
+                for (int index = blocks.Count - 1; index >= 0; index--)
+                {
+                    IBlock block = blocks[index];
+
+                    RegisterMask cmnOutputs = RegisterMask.Zero;
+                    RegisterMask allOutputs = RegisterMask.Zero;
+
+                    int firstPIndex = GetFirstPredecessorIndex(block);
+                    if (firstPIndex == 0)
+                    {
+                        IBlock predecessor = block.GetPredecessor(0);
+
+                        // Assumes that block index 0 is the entry block.
+                        cmnOutputs = block.Index != 0 ? globalOutputs[predecessor.Index] : RegisterMask.Zero;
+                        allOutputs = globalOutputs[predecessor.Index];
+
+                        for (int pIndex = 1; pIndex < block.PredecessorsCount; pIndex++)
+                        {
+                            predecessor = block.GetPredecessor(pIndex);
+
+                            if (!predecessor.EndsWithContextStore())
+                            {
+                                RegisterMask outputs = globalOutputs[predecessor.Index];
+
+                                cmnOutputs &= outputs;
+                                allOutputs |= outputs;
+                            }
+                            else
+                            {
+                                cmnOutputs = RegisterMask.Zero;
+                            }
+                        }
+                    }
+                    else if (firstPIndex > 0)
+                    {
+                        IBlock predecessor = block.GetPredecessor(firstPIndex);
+
+                        allOutputs = globalOutputs[predecessor.Index];
+
+                        for (int pIndex = firstPIndex + 1; pIndex < block.PredecessorsCount; pIndex++)
+                        {
+                            predecessor = block.GetPredecessor(pIndex);
+
+                            if (!predecessor.EndsWithContextStore())
+                            {
+                                allOutputs |= globalOutputs[predecessor.Index];
+                            }
+                        }
+                    }
+
+                    RegisterMask inputs = localInputs[block.Index];
+
+                    // If this block will load from context at the end,
+                    // we don't need to care about what comes next.
+                    if (!block.EndsWithContextLoad())
+                    {
+                        for (int sIndex = 0; sIndex < block.SuccessorsCount; sIndex++)
+                        {
+                            inputs |= globalInputs[block.GetSuccessor(sIndex).Index] & ~localOutputs[block.Index];
+                        }
+                    }
+
+                    inputs |= allOutputs & ~localOutputs[block.Index];
+                    inputs &= ~cmnOutputs;
+
+                    modified |= Exchange(globalInputs, block.Index, globalInputs[block.Index] | inputs);
+                }
+            }
+            while (modified);
+
+            return (globalInputs, globalOutputs);
+        }
+
+        private static bool Exchange(RegisterMask[] masks, int blkIndex, RegisterMask value)
+        {
+            ref RegisterMask curValue = ref masks[blkIndex];
+            bool changed = curValue != value;
+            curValue = value;
+
+            return changed;
+        }
+
+        private static int GetFirstPredecessorIndex(IBlock block)
+        {
+            for (int pIndex = 0; pIndex < block.PredecessorsCount; pIndex++)
+            {
+                if (!block.GetPredecessor(pIndex).EndsWithContextStore())
+                {
+                    return pIndex;
+                }
+            }
+
+            return -1;
+        }
+    }
+}
diff --git a/src/Ryujinx.Cpu/LightningJit/Graph/IBlock.cs b/src/Ryujinx.Cpu/LightningJit/Graph/IBlock.cs
new file mode 100644
index 00000000..2e5d7132
--- /dev/null
+++ b/src/Ryujinx.Cpu/LightningJit/Graph/IBlock.cs
@@ -0,0 +1,17 @@
+namespace Ryujinx.Cpu.LightningJit.Graph
+{
+    interface IBlock
+    {
+        int Index { get; }
+
+        int PredecessorsCount { get; }
+        int SuccessorsCount { get; }
+
+        IBlock GetSuccessor(int index);
+        IBlock GetPredecessor(int index);
+
+        RegisterUse ComputeUseMasks();
+        bool EndsWithContextLoad();
+        bool EndsWithContextStore();
+    }
+}
diff --git a/src/Ryujinx.Cpu/LightningJit/Graph/IBlockList.cs b/src/Ryujinx.Cpu/LightningJit/Graph/IBlockList.cs
new file mode 100644
index 00000000..aeec799c
--- /dev/null
+++ b/src/Ryujinx.Cpu/LightningJit/Graph/IBlockList.cs
@@ -0,0 +1,9 @@
+namespace Ryujinx.Cpu.LightningJit.Graph
+{
+    interface IBlockList
+    {
+        int Count { get; }
+
+        IBlock this[int index] { get; }
+    }
+}
diff --git a/src/Ryujinx.Cpu/LightningJit/Graph/RegisterMask.cs b/src/Ryujinx.Cpu/LightningJit/Graph/RegisterMask.cs
new file mode 100644
index 00000000..cde324d5
--- /dev/null
+++ b/src/Ryujinx.Cpu/LightningJit/Graph/RegisterMask.cs
@@ -0,0 +1,60 @@
+using System;
+
+namespace Ryujinx.Cpu.LightningJit.Graph
+{
+    readonly struct RegisterMask : IEquatable<RegisterMask>
+    {
+        public readonly uint GprMask;
+        public readonly uint FpSimdMask;
+        public readonly uint PStateMask;
+
+        public static RegisterMask Zero => new(0u, 0u, 0u);
+
+        public RegisterMask(uint gprMask, uint fpSimdMask, uint pStateMask)
+        {
+            GprMask = gprMask;
+            FpSimdMask = fpSimdMask;
+            PStateMask = pStateMask;
+        }
+
+        public static RegisterMask operator &(RegisterMask x, RegisterMask y)
+        {
+            return new(x.GprMask & y.GprMask, x.FpSimdMask & y.FpSimdMask, x.PStateMask & y.PStateMask);
+        }
+
+        public static RegisterMask operator |(RegisterMask x, RegisterMask y)
+        {
+            return new(x.GprMask | y.GprMask, x.FpSimdMask | y.FpSimdMask, x.PStateMask | y.PStateMask);
+        }
+
+        public static RegisterMask operator ~(RegisterMask x)
+        {
+            return new(~x.GprMask, ~x.FpSimdMask, ~x.PStateMask);
+        }
+
+        public static bool operator ==(RegisterMask x, RegisterMask y)
+        {
+            return x.Equals(y);
+        }
+
+        public static bool operator !=(RegisterMask x, RegisterMask y)
+        {
+            return !x.Equals(y);
+        }
+
+        public override bool Equals(object obj)
+        {
+            return obj is RegisterMask regMask && Equals(regMask);
+        }
+
+        public bool Equals(RegisterMask other)
+        {
+            return GprMask == other.GprMask && FpSimdMask == other.FpSimdMask && PStateMask == other.PStateMask;
+        }
+
+        public override int GetHashCode()
+        {
+            return HashCode.Combine(GprMask, FpSimdMask, PStateMask);
+        }
+    }
+}
diff --git a/src/Ryujinx.Cpu/LightningJit/Graph/RegisterUse.cs b/src/Ryujinx.Cpu/LightningJit/Graph/RegisterUse.cs
new file mode 100644
index 00000000..83ef4192
--- /dev/null
+++ b/src/Ryujinx.Cpu/LightningJit/Graph/RegisterUse.cs
@@ -0,0 +1,24 @@
+namespace Ryujinx.Cpu.LightningJit.Graph
+{
+    readonly struct RegisterUse
+    {
+        public readonly RegisterMask Read;
+        public readonly RegisterMask Write;
+
+        public RegisterUse(RegisterMask read, RegisterMask write)
+        {
+            Read = read;
+            Write = write;
+        }
+
+        public RegisterUse(
+            uint gprReadMask,
+            uint gprWriteMask,
+            uint fpSimdReadMask,
+            uint fpSimdWriteMask,
+            uint pStateReadMask,
+            uint pStateWriteMask) : this(new(gprReadMask, fpSimdReadMask, pStateReadMask), new(gprWriteMask, fpSimdWriteMask, pStateWriteMask))
+        {
+        }
+    }
+}