Implement HLE macro for DrawElementsIndirect (#3748)

* Implement HLE macro for DrawElementsIndirect

* Shader cache version bump

* Use GL_ARB_shader_draw_parameters extension on OpenGL

* Fix DrawIndexedIndirectCount on Vulkan when extension is not supported

* Implement DrawIndex

* Alignment

* Fix some validation errors

* Rename BaseIds to DrawParameters

* Fix incorrect index buffer and vertex buffer size in some cases

* Add HLE macros for DrawArraysInstanced and DrawElementsInstanced

* Perform a regular draw when indirect data is not modified

* Use non-indirect draw methods if indirect buffer was not GPU modified

* Only check if draw parameters match if the shader actually uses them

* Expose Macro HLE setting on GUI

* Reset FirstVertex and FirstInstance after draw

* Update shader cache version again since some people already tested this

* PR feedback

Co-authored-by: riperiperi <rhy3756547@hotmail.com>

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

@@ -62,10 +62,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
                 }
             }
 
-            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.
+            if (_hleFunction == MacroHLEFunctionName.DrawElementsIndirect)
+            {
+                context.GPFifo.SetFlushSkips(1);
+            }
+            else 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);
             }
         }

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

@@ -16,6 +16,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         private const int ColorStructSize = 0x40;
         private const int ZetaLayerCountOffset = 0x1230;
 
+        private const int IndirectDataEntrySize = 0x10;
+        private const int IndirectIndexedDataEntrySize = 0x14;
+
         private readonly GPFifoProcessor _processor;
         private readonly MacroHLEFunctionName _functionName;
 
@@ -27,9 +30,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         /// <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="processor">GPU GP FIFO command processor</param>
         /// <param name="functionName">Name of the HLE macro function to be called</param>
         public MacroHLE(GPFifoProcessor processor, MacroHLEFunctionName functionName)
         {
@@ -55,12 +56,24 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
                 case MacroHLEFunctionName.ClearDepthStencil:
                     ClearDepthStencil(state, arg0);
                     break;
+                case MacroHLEFunctionName.DrawArraysInstanced:
+                    DrawArraysInstanced(state, arg0);
+                    break;
+                case MacroHLEFunctionName.DrawElementsInstanced:
+                    DrawElementsInstanced(state, arg0);
+                    break;
+                case MacroHLEFunctionName.DrawElementsIndirect:
+                    DrawElementsIndirect(state, arg0);
+                    break;
                 case MacroHLEFunctionName.MultiDrawElementsIndirectCount:
                     MultiDrawElementsIndirectCount(state, arg0);
                     break;
                 default:
                     throw new NotImplementedException(_functionName.ToString());
             }
+
+            // It should be empty at this point, but clear it just to be safe.
+            Fifo.Clear();
         }
 
         /// <summary>
@@ -89,7 +102,118 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         }
 
         /// <summary>
-        /// Performs a indirect multi-draw, with parameters from a GPU buffer.
+        /// Performs a draw.
+        /// </summary>
+        /// <param name="state">GPU state at the time of the call</param>
+        /// <param name="arg0">First argument of the call</param>
+        private void DrawArraysInstanced(IDeviceState state, int arg0)
+        {
+            var topology = (PrimitiveTopology)arg0;
+
+            var count = FetchParam();
+            var instanceCount = FetchParam();
+            var firstVertex = FetchParam();
+            var firstInstance = FetchParam();
+
+            if (ShouldSkipDraw(state, instanceCount.Word))
+            {
+                return;
+            }
+
+            _processor.ThreedClass.Draw(
+                topology,
+                count.Word,
+                instanceCount.Word,
+                0,
+                firstVertex.Word,
+                firstInstance.Word,
+                indexed: false);
+        }
+
+        /// <summary>
+        /// Performs a indexed draw.
+        /// </summary>
+        /// <param name="state">GPU state at the time of the call</param>
+        /// <param name="arg0">First argument of the call</param>
+        private void DrawElementsInstanced(IDeviceState state, int arg0)
+        {
+            var topology = (PrimitiveTopology)arg0;
+
+            var count = FetchParam();
+            var instanceCount = FetchParam();
+            var firstIndex = FetchParam();
+            var firstVertex = FetchParam();
+            var firstInstance = FetchParam();
+
+            if (ShouldSkipDraw(state, instanceCount.Word))
+            {
+                return;
+            }
+
+            _processor.ThreedClass.Draw(
+                topology,
+                count.Word,
+                instanceCount.Word,
+                firstIndex.Word,
+                firstVertex.Word,
+                firstInstance.Word,
+                indexed: true);
+        }
+
+        /// <summary>
+        /// Performs a indirect indexed 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 DrawElementsIndirect(IDeviceState state, int arg0)
+        {
+            var topology = (PrimitiveTopology)arg0;
+
+            var count = FetchParam();
+            var instanceCount = FetchParam();
+            var firstIndex = FetchParam();
+            var firstVertex = FetchParam();
+            var firstInstance = FetchParam();
+
+            ulong indirectBufferGpuVa = count.GpuVa;
+
+            var bufferCache = _processor.MemoryManager.Physical.BufferCache;
+
+            bool useBuffer = bufferCache.CheckModified(_processor.MemoryManager, indirectBufferGpuVa, IndirectIndexedDataEntrySize, out ulong indirectBufferAddress);
+
+            if (useBuffer)
+            {
+                int indexCount = firstIndex.Word + count.Word;
+
+                _processor.ThreedClass.DrawIndirect(
+                    topology,
+                    indirectBufferAddress,
+                    0,
+                    1,
+                    IndirectIndexedDataEntrySize,
+                    indexCount,
+                    Threed.IndirectDrawType.DrawIndexedIndirect);
+            }
+            else
+            {
+                if (ShouldSkipDraw(state, instanceCount.Word))
+                {
+                    return;
+                }
+
+                _processor.ThreedClass.Draw(
+                    topology,
+                    count.Word,
+                    instanceCount.Word,
+                    firstIndex.Word,
+                    firstVertex.Word,
+                    firstInstance.Word,
+                    indexed: true);
+            }
+        }
+
+        /// <summary>
+        /// Performs a indirect indexed 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>
@@ -132,8 +256,6 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
                 return;
             }
 
-            int indirectBufferSize = maxDrawCount * stride;
-
             ulong indirectBufferGpuVa = 0;
             int indexCount = 0;
 
@@ -142,8 +264,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
                 var count = FetchParam();
                 var instanceCount = FetchParam();
                 var firstIndex = FetchParam();
-                var baseVertex = FetchParam();
-                var baseInstance = FetchParam();
+                var firstVertex = FetchParam();
+                var firstInstance = FetchParam();
 
                 if (i == 0)
                 {
@@ -161,15 +283,32 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
                 }
             }
 
-            // It should be empty at this point, but clear it just to be safe.
-            Fifo.Clear();
-
             var bufferCache = _processor.MemoryManager.Physical.BufferCache;
 
-            var parameterBuffer = bufferCache.GetGpuBufferRange(_processor.MemoryManager, parameterBufferGpuVa, 4);
-            var indirectBuffer = bufferCache.GetGpuBufferRange(_processor.MemoryManager, indirectBufferGpuVa, (ulong)indirectBufferSize);
+            ulong indirectBufferSize = (ulong)maxDrawCount * (ulong)stride;
 
-            _processor.ThreedClass.MultiDrawIndirectCount(indexCount, topology, indirectBuffer, parameterBuffer, maxDrawCount, stride);
+            ulong indirectBufferAddress = bufferCache.TranslateAndCreateBuffer(_processor.MemoryManager, indirectBufferGpuVa, indirectBufferSize);
+            ulong parameterBufferAddress = bufferCache.TranslateAndCreateBuffer(_processor.MemoryManager, parameterBufferGpuVa, 4);
+
+            _processor.ThreedClass.DrawIndirect(
+                topology,
+                indirectBufferAddress,
+                parameterBufferAddress,
+                maxDrawCount,
+                stride,
+                indexCount,
+                Threed.IndirectDrawType.DrawIndexedIndirectCount);
+        }
+
+        /// <summary>
+        /// Checks if the draw should be skipped, because the masked instance count is zero.
+        /// </summary>
+        /// <param name="state">Current GPU state</param>
+        /// <param name="instanceCount">Draw instance count</param>
+        /// <returns>True if the draw should be skipped, false otherwise</returns>
+        private static bool ShouldSkipDraw(IDeviceState state, int instanceCount)
+        {
+            return (Read(state, 0xd1b) & instanceCount) == 0;
         }
 
         /// <summary>
@@ -189,14 +328,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         }
 
         /// <summary>
-        /// Performs a GPU method call.
+        /// Reads data from a GPU register.
         /// </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)
+        /// <param name="reg">Register offset to read</param>
+        /// <returns>GPU register value</returns>
+        private static int Read(IDeviceState state, int reg)
         {
-            state.Write(methAddr * 4, value);
+            return state.Read(reg * 4);
         }
     }
 }

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

@@ -8,6 +8,9 @@
         None,
         ClearColor,
         ClearDepthStencil,
+        DrawArraysInstanced,
+        DrawElementsInstanced,
+        DrawElementsIndirect,
         MultiDrawElementsIndirectCount
     }
 }

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

@@ -44,17 +44,29 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
             }
         }
 
-        private static readonly TableEntry[] Table = new TableEntry[]
+        private static readonly TableEntry[] _table = new TableEntry[]
         {
             new TableEntry(MacroHLEFunctionName.ClearColor, new Hash128(0xA9FB28D1DC43645A, 0xB177E5D2EAE67FB0), 0x28),
             new TableEntry(MacroHLEFunctionName.ClearDepthStencil, new Hash128(0x1B96CB77D4879F4F, 0x8557032FE0C965FB), 0x24),
+            new TableEntry(MacroHLEFunctionName.DrawArraysInstanced, new Hash128(0x197FB416269DBC26, 0x34288C01DDA82202), 0x48),
+            new TableEntry(MacroHLEFunctionName.DrawElementsInstanced, new Hash128(0x1A501FD3D54EC8E0, 0x6CF570CF79DA74D6), 0x5c),
+            new TableEntry(MacroHLEFunctionName.DrawElementsIndirect, new Hash128(0x86A3E8E903AF8F45, 0xD35BBA07C23860A4), 0x7c),
             new TableEntry(MacroHLEFunctionName.MultiDrawElementsIndirectCount, new Hash128(0x890AF57ED3FB1C37, 0x35D0C95C61F5386F), 0x19C)
         };
 
+        /// <summary>
+        /// Checks if the host supports all features required by the HLE macro.
+        /// </summary>
+        /// <param name="caps">Host capabilities</param>
+        /// <param name="name">Name of the HLE macro to be checked</param>
+        /// <returns>True if the host supports the HLE macro, false otherwise</returns>
         private static bool IsMacroHLESupported(Capabilities caps, MacroHLEFunctionName name)
         {
             if (name == MacroHLEFunctionName.ClearColor ||
-                name == MacroHLEFunctionName.ClearDepthStencil)
+                name == MacroHLEFunctionName.ClearDepthStencil ||
+                name == MacroHLEFunctionName.DrawArraysInstanced ||
+                name == MacroHLEFunctionName.DrawElementsInstanced ||
+                name == MacroHLEFunctionName.DrawElementsIndirect)
             {
                 return true;
             }
@@ -77,15 +89,20 @@ namespace Ryujinx.Graphics.Gpu.Engine.MME
         {
             var mc = MemoryMarshal.Cast<int, byte>(code);
 
-            for (int i = 0; i < Table.Length; i++)
+            for (int i = 0; i < _table.Length; i++)
             {
-                ref var entry = ref Table[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);
+                    if (IsMacroHLESupported(caps, entry.Name))
+                    {
+                        name = entry.Name;
+                        return true;
+                    }
+
+                    break;
                 }
             }
 

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

@@ -63,7 +63,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             else
             {
                 evt.Flush();
-                return (memoryManager.Read<ulong>(gpuVa) != 0) ? ConditionalRenderEnabled.True : ConditionalRenderEnabled.False;
+                return (memoryManager.Read<ulong>(gpuVa, true) != 0) ? ConditionalRenderEnabled.True : ConditionalRenderEnabled.False;
             }
         }
 
@@ -108,8 +108,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 evt?.Flush();
                 evt2?.Flush();
 
-                ulong x = memoryManager.Read<ulong>(gpuVa);
-                ulong y = memoryManager.Read<ulong>(gpuVa + 16);
+                ulong x = memoryManager.Read<ulong>(gpuVa, true);
+                ulong y = memoryManager.Read<ulong>(gpuVa + 16, true);
 
                 return (isEqual ? x == y : x != y) ? ConditionalRenderEnabled.True : ConditionalRenderEnabled.False;
             }

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

@@ -1,5 +1,6 @@
 using Ryujinx.Graphics.GAL;
 using Ryujinx.Graphics.Gpu.Engine.Types;
+using Ryujinx.Graphics.Gpu.Memory;
 using System;
 
 namespace Ryujinx.Graphics.Gpu.Engine.Threed
@@ -9,6 +10,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
     /// </summary>
     class DrawManager
     {
+        // Since we don't know the index buffer size for indirect draws,
+        // we must assume a minimum and maximum size and use that for buffer data update purposes.
+        private const int MinIndirectIndexCount = 0x10000;
+        private const int MaxIndirectIndexCount = 0x4000000;
+
         private readonly GpuContext _context;
         private readonly GpuChannel _channel;
         private readonly DeviceStateWithShadow<ThreedClassState> _state;
@@ -28,6 +34,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
         private int _instanceIndex;
 
+        private const int VertexBufferFirstMethodOffset = 0x35d;
         private const int IndexBufferCountMethodOffset = 0x5f8;
 
         /// <summary>
@@ -237,6 +244,15 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 _instanceIndex = 0;
             }
 
+            UpdateTopology(topology);
+        }
+
+        /// <summary>
+        /// Updates the current primitive topology if needed.
+        /// </summary>
+        /// <param name="topology">New primitive topology</param>
+        private void UpdateTopology(PrimitiveTopology topology)
+        {
             if (_drawState.Topology != topology || !_topologySet)
             {
                 _context.Renderer.Pipeline.SetPrimitiveTopology(topology);
@@ -383,28 +399,27 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         }
 
         /// <summary>
-        /// Performs a indirect multi-draw, with parameters from a GPU buffer.
+        /// Performs a indexed or non-indexed draw.
         /// </summary>
         /// <param name="engine">3D engine where this method is being called</param>
         /// <param name="topology">Primitive topology</param>
-        /// <param name="indirectBuffer">GPU buffer with the draw parameters, such as count, first index, etc</param>
-        /// <param name="parameterBuffer">GPU buffer with the draw count</param>
-        /// <param name="maxDrawCount">Maximum number of draws that can be made</param>
-        /// <param name="stride">Distance in bytes between each element on the <paramref name="indirectBuffer"/> array</param>
-        public void MultiDrawIndirectCount(
+        /// <param name="count">Index count for indexed draws, vertex count for non-indexed draws</param>
+        /// <param name="instanceCount">Instance count</param>
+        /// <param name="firstIndex">First index on the index buffer for indexed draws, ignored for non-indexed draws</param>
+        /// <param name="firstVertex">First vertex on the vertex buffer</param>
+        /// <param name="firstInstance">First instance</param>
+        /// <param name="indexed">True if the draw is indexed, false otherwise</param>
+        public void Draw(
             ThreedClass engine,
-            int indexCount,
             PrimitiveTopology topology,
-            BufferRange indirectBuffer,
-            BufferRange parameterBuffer,
-            int maxDrawCount,
-            int stride)
+            int count,
+            int instanceCount,
+            int firstIndex,
+            int firstVertex,
+            int firstInstance,
+            bool indexed)
         {
-            engine.Write(IndexBufferCountMethodOffset * 4, indexCount);
-
-            _context.Renderer.Pipeline.SetPrimitiveTopology(topology);
-            _drawState.Topology = topology;
-            _topologySet = true;
+            UpdateTopology(topology);
 
             ConditionalRenderEnabled renderEnable = ConditionalRendering.GetRenderEnable(
                 _context,
@@ -418,21 +433,133 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 return;
             }
 
-            _drawState.FirstIndex = _state.State.IndexBufferState.First;
-            _drawState.IndexCount = indexCount;
-
-            engine.UpdateState();
-
-            if (_drawState.DrawIndexed)
+            if (indexed)
             {
-                _context.Renderer.Pipeline.MultiDrawIndexedIndirectCount(indirectBuffer, parameterBuffer, maxDrawCount, stride);
+                _drawState.FirstIndex = firstIndex;
+                _drawState.IndexCount = count;
+                _state.State.FirstVertex = (uint)firstVertex;
+                engine.ForceStateDirty(IndexBufferCountMethodOffset * 4);
             }
             else
             {
-                _context.Renderer.Pipeline.MultiDrawIndirectCount(indirectBuffer, parameterBuffer, maxDrawCount, stride);
+                _state.State.VertexBufferDrawState.First = firstVertex;
+                _state.State.VertexBufferDrawState.Count = count;
+                engine.ForceStateDirty(VertexBufferFirstMethodOffset * 4);
+            }
+
+            _state.State.FirstInstance = (uint)firstInstance;
+
+            _drawState.DrawIndexed = indexed;
+            _drawState.HasConstantBufferDrawParameters = true;
+
+            engine.UpdateState();
+
+            if (indexed)
+            {
+                _context.Renderer.Pipeline.DrawIndexed(count, instanceCount, firstIndex, firstVertex, firstInstance);
+                _state.State.FirstVertex = 0;
+            }
+            else
+            {
+                _context.Renderer.Pipeline.Draw(count, instanceCount, firstVertex, firstInstance);
+            }
+
+            _state.State.FirstInstance = 0;
+
+            _drawState.DrawIndexed = false;
+            _drawState.HasConstantBufferDrawParameters = false;
+
+            if (renderEnable == ConditionalRenderEnabled.Host)
+            {
+                _context.Renderer.Pipeline.EndHostConditionalRendering();
+            }
+        }
+
+        /// <summary>
+        /// Performs a indirect draw, with parameters from a GPU buffer.
+        /// </summary>
+        /// <param name="engine">3D engine where this method is being called</param>
+        /// <param name="topology">Primitive topology</param>
+        /// <param name="indirectBufferAddress">Address of the buffer with the draw parameters, such as count, first index, etc</param>
+        /// <param name="parameterBufferAddress">Address of the buffer with the draw count</param>
+        /// <param name="maxDrawCount">Maximum number of draws that can be made</param>
+        /// <param name="stride">Distance in bytes between each entry on the data pointed to by <paramref name="indirectBufferAddress"/></param>
+        /// <param name="indexCount">Maximum number of indices that the draw can consume</param>
+        /// <param name="drawType">Type of the indirect draw, which can be indexed or non-indexed, with or without a draw count</param>
+        public void DrawIndirect(
+            ThreedClass engine,
+            PrimitiveTopology topology,
+            ulong indirectBufferAddress,
+            ulong parameterBufferAddress,
+            int maxDrawCount,
+            int stride,
+            int indexCount,
+            IndirectDrawType drawType)
+        {
+            UpdateTopology(topology);
+
+            ConditionalRenderEnabled renderEnable = ConditionalRendering.GetRenderEnable(
+                _context,
+                _channel.MemoryManager,
+                _state.State.RenderEnableAddress,
+                _state.State.RenderEnableCondition);
+
+            if (renderEnable == ConditionalRenderEnabled.False)
+            {
+                _drawState.DrawIndexed = false;
+                return;
+            }
+
+            PhysicalMemory memory = _channel.MemoryManager.Physical;
+
+            bool hasCount = (drawType & IndirectDrawType.Count) != 0;
+            bool indexed = (drawType & IndirectDrawType.Indexed) != 0;
+
+            if (indexed)
+            {
+                indexCount = Math.Clamp(indexCount, MinIndirectIndexCount, MaxIndirectIndexCount);
+                _drawState.FirstIndex = 0;
+                _drawState.IndexCount = indexCount;
+                engine.ForceStateDirty(IndexBufferCountMethodOffset * 4);
+            }
+
+            _drawState.DrawIndexed = indexed;
+            _drawState.DrawIndirect = true;
+            _drawState.HasConstantBufferDrawParameters = true;
+
+            engine.UpdateState();
+
+            if (hasCount)
+            {
+                var indirectBuffer = memory.BufferCache.GetBufferRange(indirectBufferAddress, (ulong)maxDrawCount * (ulong)stride);
+                var parameterBuffer = memory.BufferCache.GetBufferRange(parameterBufferAddress, 4);
+
+                if (indexed)
+                {
+                    _context.Renderer.Pipeline.DrawIndexedIndirectCount(indirectBuffer, parameterBuffer, maxDrawCount, stride);
+                }
+                else
+                {
+                    _context.Renderer.Pipeline.DrawIndirectCount(indirectBuffer, parameterBuffer, maxDrawCount, stride);
+                }
+            }
+            else
+            {
+                var indirectBuffer = memory.BufferCache.GetBufferRange(indirectBufferAddress, (ulong)stride);
+
+                if (indexed)
+                {
+                    _context.Renderer.Pipeline.DrawIndexedIndirect(indirectBuffer);
+                }
+                else
+                {
+                    _context.Renderer.Pipeline.DrawIndirect(indirectBuffer);
+                }
             }
 
             _drawState.DrawIndexed = false;
+            _drawState.DrawIndirect = false;
+            _drawState.HasConstantBufferDrawParameters = false;
 
             if (renderEnable == ConditionalRenderEnabled.Host)
             {

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

@@ -22,6 +22,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// </summary>
         public bool DrawIndexed;
 
+        /// <summary>
+        /// Indicates if the next draw will be a indirect draw.
+        /// </summary>
+        public bool DrawIndirect;
+
         /// <summary>
         /// Indicates if any of the currently used vertex shaders reads the instance ID.
         /// </summary>
@@ -32,6 +37,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         /// </summary>
         public bool IsAnyVbInstanced;
 
+        /// <summary>
+        /// Indicates that the draw is writing the base vertex, base instance and draw index to Constant Buffer 0.
+        /// </summary>
+        public bool HasConstantBufferDrawParameters;
+
         /// <summary>
         /// Primitive topology for the next draw.
         /// </summary>

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

@@ -0,0 +1,38 @@
+namespace Ryujinx.Graphics.Gpu.Engine.Threed
+{
+    /// <summary>
+    /// Indirect draw type, which can be indexed or non-indexed, with or without a draw count.
+    /// </summary>
+    enum IndirectDrawType
+    {
+        /// <summary>
+        /// Non-indexed draw without draw count.
+        /// </summary>
+        DrawIndirect = 0,
+
+        /// <summary>
+        /// Indexed draw without draw count.
+        /// </summary>
+        DrawIndexedIndirect = Indexed,
+
+        /// <summary>
+        /// Non-indexed draw with draw count.
+        /// </summary>
+        DrawIndirectCount = Count,
+
+        /// <summary>
+        /// Indexed draw with draw count.
+        /// </summary>
+        DrawIndexedIndirectCount = Indexed | Count,
+
+        /// <summary>
+        /// Indexed flag.
+        /// </summary>
+        Indexed = 1 << 0,
+
+        /// <summary>
+        /// Draw count flag.
+        /// </summary>
+        Count = 1 << 1
+    }
+}

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

@@ -34,10 +34,12 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
         private ProgramPipelineState _pipeline;
 
+        private bool _vsUsesDrawParameters;
         private bool _vtgWritesRtLayer;
         private byte _vsClipDistancesWritten;
 
         private bool _prevDrawIndexed;
+        private bool _prevDrawIndirect;
         private IndexType _prevIndexType;
         private uint _prevFirstVertex;
         private bool _prevTfEnable;
@@ -210,7 +212,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             // of the shader for the new state.
             if (_shaderSpecState != null)
             {
-                if (!_shaderSpecState.MatchesGraphics(_channel, GetPoolState(), GetGraphicsState(), false))
+                if (!_shaderSpecState.MatchesGraphics(_channel, GetPoolState(), GetGraphicsState(), _vsUsesDrawParameters, false))
                 {
                     ForceShaderUpdate();
                 }
@@ -237,6 +239,15 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 _prevDrawIndexed = _drawState.DrawIndexed;
             }
 
+            // Some draw parameters are used to restrict the vertex buffer size,
+            // but they can't be used on indirect draws because their values are unknown in this case.
+            // When switching between indirect and non-indirect draw, we need to
+            // make sure the vertex buffer sizes are still correct.
+            if (_drawState.DrawIndirect != _prevDrawIndirect)
+            {
+                _updateTracker.ForceDirty(VertexBufferStateIndex);
+            }
+
             // In some cases, the index type is also used to guess the
             // vertex buffer size, so we must update it if the type changed too.
             if (_drawState.DrawIndexed &&
@@ -938,6 +949,9 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
 
             _drawState.IsAnyVbInstanced = false;
 
+            bool drawIndexed = _drawState.DrawIndexed;
+            bool drawIndirect = _drawState.DrawIndirect;
+
             for (int index = 0; index < Constants.TotalVertexBuffers; index++)
             {
                 var vertexBuffer = _state.State.VertexBufferState[index];
@@ -965,14 +979,14 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 ulong vbSize = endAddress.Pack() - address + 1;
                 ulong size;
 
-                if (_drawState.IbStreamer.HasInlineIndexData || _drawState.DrawIndexed || stride == 0 || instanced)
+                if (_drawState.IbStreamer.HasInlineIndexData || drawIndexed || stride == 0 || instanced)
                 {
                     // This size may be (much) larger than the real vertex buffer size.
                     // Avoid calculating it this way, unless we don't have any other option.
 
                     size = vbSize;
 
-                    if (stride > 0 && indexTypeSmall && _drawState.DrawIndexed && !instanced)
+                    if (stride > 0 && indexTypeSmall && drawIndexed && !drawIndirect && !instanced)
                     {
                         // If the index type is a small integer type, then we might be still able
                         // to reduce the vertex buffer size based on the maximum possible index value.
@@ -1207,6 +1221,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             byte oldVsClipDistancesWritten = _vsClipDistancesWritten;
 
             _drawState.VsUsesInstanceId = gs.Shaders[1]?.Info.UsesInstanceId ?? false;
+            _vsUsesDrawParameters = gs.Shaders[1]?.Info.UsesDrawParameters ?? false;
             _vsClipDistancesWritten = gs.Shaders[1]?.Info.ClipDistancesWritten ?? 0;
 
             if (oldVsClipDistancesWritten != _vsClipDistancesWritten)
@@ -1222,6 +1237,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             _context.Renderer.Pipeline.SetProgram(gs.HostProgram);
         }
 
+        /// <summary>
+        /// Updates bindings consumed by the shader stage on the texture and buffer managers.
+        /// </summary>
+        /// <param name="stage">Shader stage to have the bindings updated</param>
+        /// <param name="info">Shader stage bindings info</param>
         private void UpdateStageBindings(int stage, ShaderProgramInfo info)
         {
             _currentProgramInfo[stage] = info;
@@ -1340,7 +1360,8 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 _state.State.AlphaTestEnable,
                 _state.State.AlphaTestFunc,
                 _state.State.AlphaTestRef,
-                ref attributeTypes);
+                ref attributeTypes,
+                _drawState.HasConstantBufferDrawParameters);
         }
 
         /// <summary>

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

@@ -497,6 +497,50 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
             return 0;
         }
 
+        /// <summary>
+        /// Performs a indexed or non-indexed draw.
+        /// </summary>
+        /// <param name="topology">Primitive topology</param>
+        /// <param name="count">Index count for indexed draws, vertex count for non-indexed draws</param>
+        /// <param name="instanceCount">Instance count</param>
+        /// <param name="firstIndex">First index on the index buffer for indexed draws, ignored for non-indexed draws</param>
+        /// <param name="firstVertex">First vertex on the vertex buffer</param>
+        /// <param name="firstInstance">First instance</param>
+        /// <param name="indexed">True if the draw is indexed, false otherwise</param>
+        public void Draw(
+            PrimitiveTopology topology,
+            int count,
+            int instanceCount,
+            int firstIndex,
+            int firstVertex,
+            int firstInstance,
+            bool indexed)
+        {
+            _drawManager.Draw(this, topology, count, instanceCount, firstIndex, firstVertex, firstInstance, indexed);
+        }
+
+        /// <summary>
+        /// Performs a indirect draw, with parameters from a GPU buffer.
+        /// </summary>
+        /// <param name="topology">Primitive topology</param>
+        /// <param name="indirectBufferAddress">Address of the buffer with the draw parameters, such as count, first index, etc</param>
+        /// <param name="parameterBufferAddress">Address of the buffer with the draw count</param>
+        /// <param name="maxDrawCount">Maximum number of draws that can be made</param>
+        /// <param name="stride">Distance in bytes between each entry on the data pointed to by <paramref name="indirectBufferAddress"/></param>
+        /// <param name="indexCount">Maximum number of indices that the draw can consume</param>
+        /// <param name="drawType">Type of the indirect draw, which can be indexed or non-indexed, with or without a draw count</param>
+        public void DrawIndirect(
+            PrimitiveTopology topology,
+            ulong indirectBufferAddress,
+            ulong parameterBufferAddress,
+            int maxDrawCount,
+            int stride,
+            int indexCount,
+            IndirectDrawType drawType)
+        {
+            _drawManager.DrawIndirect(this, topology, indirectBufferAddress, parameterBufferAddress, maxDrawCount, stride, indexCount, drawType);
+        }
+
         /// <summary>
         /// Clears the current color and depth-stencil buffers.
         /// Which buffers should be cleared can also specified with the arguments.
@@ -507,25 +551,5 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         {
             _drawManager.Clear(this, argument, layerCount);
         }
-
-        /// <summary>
-        /// Performs a indirect multi-draw, with parameters from a GPU buffer.
-        /// </summary>
-        /// <param name="indexCount">Index Buffer Count</param>
-        /// <param name="topology">Primitive topology</param>
-        /// <param name="indirectBuffer">GPU buffer with the draw parameters, such as count, first index, etc</param>
-        /// <param name="parameterBuffer">GPU buffer with the draw count</param>
-        /// <param name="maxDrawCount">Maximum number of draws that can be made</param>
-        /// <param name="stride">Distance in bytes between each element on the <paramref name="indirectBuffer"/> array</param>
-        public void MultiDrawIndirectCount(
-            int indexCount,
-            PrimitiveTopology topology,
-            BufferRange indirectBuffer,
-            BufferRange parameterBuffer,
-            int maxDrawCount,
-            int stride)
-        {
-            _drawManager.MultiDrawIndirectCount(this, indexCount, topology, indirectBuffer, parameterBuffer, maxDrawCount, stride);
-        }
     }
 }