Add support for HLE macros and accelerate MultiDrawElementsIndirectCount #2 (#2557)

* Add support for HLE macros and accelerate MultiDrawElementsIndirectCount

* Add missing barrier

* Fix index buffer count

* Add support check for each macro hle before use

* Add missing xml doc

Co-authored-by: gdkchan <gab.dark.100@gmail.com>

Ryujinx.Graphics.Gpu/Engine/MME/IMacroEE.cs

@@ -4,6 +4,33 @@ using System.Collections.Generic;
 
 namespace Ryujinx.Graphics.Gpu.Engine.MME
 {
+    /// <summary>
+    /// FIFO word.
+    /// </summary>
+    struct FifoWord
+    {
+        /// <summary>
+        /// GPU virtual address where the word is located in memory.
+        /// </summary>
+        public ulong GpuVa { get; }
+
+        /// <summary>
+        /// Word value.
+        /// </summary>
+        public int Word { get; }
+
+        /// <summary>
+        /// Creates a new FIFO word.
+        /// </summary>
+        /// <param name="gpuVa">GPU virtual address where the word is located in memory</param>
+        /// <param name="word">Word value</param>
+        public FifoWord(ulong gpuVa, int word)
+        {
+            GpuVa = gpuVa;
+            Word = word;
+        }
+    }
+
     /// <summary>
     /// Macro Execution Engine interface.
     /// </summary>
@@ -12,7 +39,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// <summary>
         /// Arguments FIFO.
         /// </summary>
-        Queue<int> Fifo { get; }
+        Queue<FifoWord> Fifo { get; }
 
         /// <summary>
         /// Should execute the GPU Macro code being passed.

Ryujinx.Graphics.Gpu/Engine/MME/Macro.cs

@@ -1,4 +1,6 @@
 using Ryujinx.Graphics.Device;
+using Ryujinx.Graphics.Gpu.Engine.GPFifo;
+using Ryujinx.Graphics.Gpu.Memory;
 using System;
 
 namespace Ryujinx.Graphics.Gpu.Engine.MME
@@ -13,10 +15,10 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// </summary>
         public int Position { get; }
 
+        private IMacroEE _executionEngine;
         private bool _executionPending;
         private int _argument;
-
-        private readonly IMacroEE _executionEngine;
+        private MacroHLEFunctionName _hleFunction;
 
         /// <summary>
         /// Creates a new instance of the GPU cached macro program.
@@ -26,28 +28,47 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         {
             Position = position;
 
+            _executionEngine = null;
             _executionPending = false;
             _argument = 0;
-
-            if (GraphicsConfig.EnableMacroJit)
-            {
-                _executionEngine = new MacroJit();
-            }
-            else
-            {
-                _executionEngine = new MacroInterpreter();
-            }
+            _hleFunction = MacroHLEFunctionName.None;
         }
 
         /// <summary>
         /// Sets the first argument for the macro call.
         /// </summary>
+        /// <param name="context">GPU context where the macro code is being executed</param>
+        /// <param name="processor">GPU GP FIFO command processor</param>
+        /// <param name="code">Code to be executed</param>
         /// <param name="argument">First argument</param>
-        public void StartExecution(int argument)
+        public void StartExecution(GpuContext context, GPFifoProcessor processor, ReadOnlySpan<int> code, int argument)
         {
             _argument = argument;
 
             _executionPending = true;
+
+            if (_executionEngine == null)
+            {
+                if (GraphicsConfig.EnableMacroHLE && MacroHLETable.TryGetMacroHLEFunction(code.Slice(Position), context.Capabilities, out _hleFunction))
+                {
+                    _executionEngine = new MacroHLE(processor, _hleFunction);
+                }
+                else if (GraphicsConfig.EnableMacroJit)
+                {
+                    _executionEngine = new MacroJit();
+                }
+                else
+                {
+                    _executionEngine = new MacroInterpreter();
+                }
+            }
+
+            if (_hleFunction == MacroHLEFunctionName.MultiDrawElementsIndirectCount)
+            {
+                // We don't consume the parameter buffer value, so we don't need to flush it.
+                // Doing so improves performance if the value was written by a GPU shader.
+                context.GPFifo.SetFlushSkips(2);
+            }
         }
 
         /// <summary>
@@ -60,7 +81,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
             if (_executionPending)
             {
                 _executionPending = false;
-
                 _executionEngine?.Execute(code.Slice(Position), state, _argument);
             }
         }
@@ -68,10 +88,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// <summary>
         /// Pushes an argument to the macro call argument FIFO.
         /// </summary>
+        /// <param name="gpuVa">GPU virtual address where the command word is located</param>
         /// <param name="argument">Argument to be pushed</param>
-        public void PushArgument(int argument)
+        public void PushArgument(ulong gpuVa, int argument)
         {
-            _executionEngine?.Fifo.Enqueue(argument);
+            _executionEngine?.Fifo.Enqueue(new FifoWord(gpuVa, argument));
         }
     }
 }

Ryujinx.Graphics.Gpu/Engine/MME/MacroHLE.cs

@@ -0,0 +1,142 @@
+using Ryujinx.Common.Logging;
+using Ryujinx.Graphics.Device;
+using Ryujinx.Graphics.GAL;
+using Ryujinx.Graphics.Gpu.Engine.GPFifo;
+using Ryujinx.Graphics.Gpu.Engine.Threed;
+using Ryujinx.Graphics.Gpu.Memory;
+using System;
+using System.Collections.Generic;
+
+namespace Ryujinx.Graphics.Gpu.Engine.MME
+{
+    /// <summary>
+    /// Macro High-level emulation.
+    /// </summary>
+    class MacroHLE : IMacroEE
+    {
+        private readonly GPFifoProcessor _processor;
+        private readonly MacroHLEFunctionName _functionName;
+
+        /// <summary>
+        /// Arguments FIFO.
+        /// </summary>
+        public Queue<FifoWord> Fifo { get; }
+
+        /// <summary>
+        /// Creates a new instance of the HLE macro handler.
+        /// </summary>
+        /// <param name="context">GPU context the macro is being executed on</param>
+        /// <param name="memoryManager">GPU memory manager</param>
+        /// <param name="engine">3D engine where this macro is being called</param>
+        /// <param name="functionName">Name of the HLE macro function to be called</param>
+        public MacroHLE(GPFifoProcessor processor, MacroHLEFunctionName functionName)
+        {
+            _processor = processor;
+            _functionName = functionName;
+
+            Fifo = new Queue<FifoWord>();
+        }
+
+        /// <summary>
+        /// Executes a macro program until it exits.
+        /// </summary>
+        /// <param name="code">Code of the program to execute</param>
+        /// <param name="state">GPU state at the time of the call</param>
+        /// <param name="arg0">Optional argument passed to the program, 0 if not used</param>
+        public void Execute(ReadOnlySpan<int> code, IDeviceState state, int arg0)
+        {
+            switch (_functionName)
+            {
+                case MacroHLEFunctionName.MultiDrawElementsIndirectCount:
+                    MultiDrawElementsIndirectCount(state, arg0);
+                    break;
+                default:
+                    throw new NotImplementedException(_functionName.ToString());
+            }
+        }
+
+        /// <summary>
+        /// Performs a indirect multi-draw, with parameters from a GPU buffer.
+        /// </summary>
+        /// <param name="state">GPU state at the time of the call</param>
+        /// <param name="arg0">First argument of the call</param>
+        private void MultiDrawElementsIndirectCount(IDeviceState state, int arg0)
+        {
+            int arg1 = FetchParam().Word;
+            int arg2 = FetchParam().Word;
+            int arg3 = FetchParam().Word;
+
+            int startOffset = arg0;
+            int endOffset = arg1;
+            var topology = (PrimitiveTopology)arg2;
+            int paddingWords = arg3;
+            int maxDrawCount = endOffset - startOffset;
+            int stride = paddingWords * 4 + 0x14;
+            int indirectBufferSize = maxDrawCount * stride;
+
+            ulong parameterBufferGpuVa = FetchParam().GpuVa;
+            ulong indirectBufferGpuVa = 0;
+
+            int indexCount = 0;
+
+            for (int i = 0; i < maxDrawCount; i++)
+            {
+                var count = FetchParam();
+                var instanceCount = FetchParam();
+                var firstIndex = FetchParam();
+                var baseVertex = FetchParam();
+                var baseInstance = FetchParam();
+
+                if (i == 0)
+                {
+                    indirectBufferGpuVa = count.GpuVa;
+                }
+
+                indexCount = Math.Max(indexCount, count.Word + firstIndex.Word);
+
+                if (i != maxDrawCount - 1)
+                {
+                    for (int j = 0; j < paddingWords; j++)
+                    {
+                        FetchParam();
+                    }
+                }
+            }
+
+            // It should be empty at this point, but clear it just to be safe.
+            Fifo.Clear();
+
+            var parameterBuffer = _processor.MemoryManager.Physical.BufferCache.GetGpuBufferRange(_processor.MemoryManager, parameterBufferGpuVa, 4);
+            var indirectBuffer = _processor.MemoryManager.Physical.BufferCache.GetGpuBufferRange(_processor.MemoryManager, indirectBufferGpuVa, (ulong)indirectBufferSize);
+
+            _processor.ThreedClass.MultiDrawIndirectCount(indexCount, topology, indirectBuffer, parameterBuffer, maxDrawCount, stride);
+        }
+
+        /// <summary>
+        /// Fetches a arguments from the arguments FIFO.
+        /// </summary>
+        /// <returns>The call argument, or a 0 value with null address if the FIFO is empty</returns>
+        private FifoWord FetchParam()
+        {
+            if (!Fifo.TryDequeue(out var value))
+            {
+                Logger.Warning?.Print(LogClass.Gpu, "Macro attempted to fetch an inexistent argument.");
+
+                return new FifoWord(0UL, 0);
+            }
+
+            return value;
+        }
+
+        /// <summary>
+        /// Performs a GPU method call.
+        /// </summary>
+        /// <param name="state">Current GPU state</param>
+        /// <param name="methAddr">Address, in words, of the method</param>
+        /// <param name="value">Call argument</param>
+        private static void Send(IDeviceState state, int methAddr, int value)
+        {
+            state.Write(methAddr * 4, value);
+        }
+    }
+}

Ryujinx.Graphics.Gpu/Engine/MME/MacroHLEFunctionName.cs

@@ -0,0 +1,11 @@
+namespace Ryujinx.Graphics.Gpu.Engine.MME
+{
+    /// <summary>
+    /// Name of the High-level implementation of a Macro function.
+    /// </summary>
+    enum MacroHLEFunctionName
+    {
+        None,
+        MultiDrawElementsIndirectCount
+    }
+}

Ryujinx.Graphics.Gpu/Engine/MME/MacroHLETable.cs

@@ -0,0 +1,89 @@
+using Ryujinx.Common;
+using Ryujinx.Graphics.GAL;
+using System;
+using System.Runtime.InteropServices;
+
+namespace Ryujinx.Graphics.Gpu.Engine.MME
+{
+    /// <summary>
+    /// Table with information about High-level implementations of GPU Macro code.
+    /// </summary>
+    static class MacroHLETable
+    {
+        /// <summary>
+        /// Macroo High-level implementation table entry.
+        /// </summary>
+        struct TableEntry
+        {
+            /// <summary>
+            /// Name of the Macro function.
+            /// </summary>
+            public MacroHLEFunctionName Name { get; }
+
+            /// <summary>
+            /// Hash of the original binary Macro function code.
+            /// </summary>
+            public Hash128 Hash { get; }
+
+            /// <summary>
+            /// Size (in bytes) of the original binary Macro function code.
+            /// </summary>
+            public int Length { get; }
+
+            /// <summary>
+            /// Creates a new table entry.
+            /// </summary>
+            /// <param name="name">Name of the Macro function</param>
+            /// <param name="hash">Hash of the original binary Macro function code</param>
+            /// <param name="length">Size (in bytes) of the original binary Macro function code</param>
+            public TableEntry(MacroHLEFunctionName name, Hash128 hash, int length)
+            {
+                Name = name;
+                Hash = hash;
+                Length = length;
+            }
+        }
+
+        private static readonly TableEntry[] Table = new TableEntry[]
+        {
+            new TableEntry(MacroHLEFunctionName.MultiDrawElementsIndirectCount, new Hash128(0x890AF57ED3FB1C37, 0x35D0C95C61F5386F), 0x19C)
+        };
+
+        private static bool IsMacroHLESupported(Capabilities caps, MacroHLEFunctionName name)
+        {
+            if (name == MacroHLEFunctionName.MultiDrawElementsIndirectCount)
+            {
+                return caps.SupportsIndirectParameters;
+            }
+
+            return false;
+        }
+
+        /// <summary>
+        /// Checks if there's a fast, High-level implementation of the specified Macro code available.
+        /// </summary>
+        /// <param name="code">Macro code to be checked</param>
+        /// <param name="caps">Renderer capabilities to check for this macro HLE support</param>
+        /// <param name="name">Name of the function if a implementation is available and supported, otherwise <see cref="MacroHLEFunctionName.None"/></param>
+        /// <returns>True if there is a implementation available and supported, false otherwise</returns>
+        public static bool TryGetMacroHLEFunction(ReadOnlySpan<int> code, Capabilities caps, out MacroHLEFunctionName name)
+        {
+            var mc = MemoryMarshal.Cast<int, byte>(code);
+
+            for (int i = 0; i < Table.Length; i++)
+            {
+                ref var entry = ref Table[i];
+
+                var hash = XXHash128.ComputeHash(mc.Slice(0, entry.Length));
+                if (hash == entry.Hash)
+                {
+                    name = entry.Name;
+                    return IsMacroHLESupported(caps, name);
+                }
+            }
+
+            name = MacroHLEFunctionName.None;
+            return false;
+        }
+    }
+}

Ryujinx.Graphics.Gpu/Engine/MME/MacroInterpreter.cs

@@ -13,7 +13,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// <summary>
         /// Arguments FIFO.
         /// </summary>
-        public Queue<int> Fifo { get; }
+        public Queue<FifoWord> Fifo { get; }
 
         private int[] _gprs;
 
@@ -34,7 +34,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// </summary>
         public MacroInterpreter()
         {
-            Fifo = new Queue<int>();
+            Fifo = new Queue<FifoWord>();
 
             _gprs = new int[8];
         }
@@ -364,14 +364,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// <returns>The call argument, or 0 if the FIFO is empty</returns>
         private int FetchParam()
         {
-            if (!Fifo.TryDequeue(out int value))
+            if (!Fifo.TryDequeue(out var value))
             {
                 Logger.Warning?.Print(LogClass.Gpu, "Macro attempted to fetch an inexistent argument.");
 
                 return 0;
             }
 
-            return value;
+            return value.Word;
         }
 
         /// <summary>

Ryujinx.Graphics.Gpu/Engine/MME/MacroJit.cs

@@ -14,7 +14,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// <summary>
         /// Arguments FIFO.
         /// </summary>
-        public Queue<int> Fifo => _context.Fifo;
+        public Queue<FifoWord> Fifo => _context.Fifo;
 
         private MacroJitCompiler.MacroExecute _execute;
 

Ryujinx.Graphics.Gpu/Engine/MME/MacroJitContext.cs

@@ -12,22 +12,22 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// <summary>
         /// Arguments FIFO.
         /// </summary>
-        public Queue<int> Fifo { get; } = new Queue<int>();
+        public Queue<FifoWord> Fifo { get; } = new Queue<FifoWord>();
 
         /// <summary>
         /// Fetches a arguments from the arguments FIFO.
         /// </summary>
-        /// <returns></returns>
+        /// <returns>The call argument, or 0 if the FIFO is empty</returns>
         public int FetchParam()
         {
-            if (!Fifo.TryDequeue(out int value))
+            if (!Fifo.TryDequeue(out var value))
             {
                 Logger.Warning?.Print(LogClass.Gpu, "Macro attempted to fetch an inexistent argument.");
 
                 return 0;
             }
 
-            return value;
+            return value.Word;
         }
 
         /// <summary>