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>

6 files changed, 263 insertions, 43 deletions

diff --git a/Ryujinx.Graphics.Gpu/Engine/Threed/ConditionalRendering.cs b/Ryujinx.Graphics.Gpu/Engine/Threed/ConditionalRendering.cs
index 85f66985..a6b62a4a 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Threed/ConditionalRendering.cs
+++ b/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;
             }
diff --git a/Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs b/Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs
index e02c8cdc..a7acb469 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Threed/DrawManager.cs
+++ b/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 +399,104 @@ 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="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,
+            PrimitiveTopology topology,
+            int count,
+            int instanceCount,
+            int firstIndex,
+            int firstVertex,
+            int firstInstance,
+            bool indexed)
+        {
+            UpdateTopology(topology);
+
+            ConditionalRenderEnabled renderEnable = ConditionalRendering.GetRenderEnable(
+                _context,
+                _channel.MemoryManager,
+                _state.State.RenderEnableAddress,
+                _state.State.RenderEnableCondition);
+
+            if (renderEnable == ConditionalRenderEnabled.False)
+            {
+                _drawState.DrawIndexed = false;
+                return;
+            }
+
+            if (indexed)
+            {
+                _drawState.FirstIndex = firstIndex;
+                _drawState.IndexCount = count;
+                _state.State.FirstVertex = (uint)firstVertex;
+                engine.ForceStateDirty(IndexBufferCountMethodOffset * 4);
+            }
+            else
+            {
+                _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="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="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 element on the <paramref name="indirectBuffer"/> array</param>
-        public void MultiDrawIndirectCount(
+        /// <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,
-            int indexCount,
             PrimitiveTopology topology,
-            BufferRange indirectBuffer,
-            BufferRange parameterBuffer,
+            ulong indirectBufferAddress,
+            ulong parameterBufferAddress,
             int maxDrawCount,
-            int stride)
+            int stride,
+            int indexCount,
+            IndirectDrawType drawType)
         {
-            engine.Write(IndexBufferCountMethodOffset * 4, indexCount);
-
-            _context.Renderer.Pipeline.SetPrimitiveTopology(topology);
-            _drawState.Topology = topology;
-            _topologySet = true;
+            UpdateTopology(topology);
 
             ConditionalRenderEnabled renderEnable = ConditionalRendering.GetRenderEnable(
                 _context,
@@ -418,21 +510,56 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
                 return;
             }
 
-            _drawState.FirstIndex = _state.State.IndexBufferState.First;
-            _drawState.IndexCount = indexCount;
+            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 (_drawState.DrawIndexed)
+            if (hasCount)
             {
-                _context.Renderer.Pipeline.MultiDrawIndexedIndirectCount(indirectBuffer, parameterBuffer, maxDrawCount, stride);
+                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
             {
-                _context.Renderer.Pipeline.MultiDrawIndirectCount(indirectBuffer, parameterBuffer, maxDrawCount, stride);
+                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)
             {
diff --git a/Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs b/Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs
index ff186acc..fd1cb0ea 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/Threed/DrawState.cs
@@ -23,6 +23,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         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>
         public bool VsUsesInstanceId;
@@ -33,6 +38,11 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         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>
         public PrimitiveTopology Topology;
diff --git a/Ryujinx.Graphics.Gpu/Engine/Threed/IndirectDrawType.cs b/Ryujinx.Graphics.Gpu/Engine/Threed/IndirectDrawType.cs
new file mode 100644
index 00000000..d78aa498
--- /dev/null
+++ b/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
+    }
+}
\ No newline at end of file
diff --git a/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs b/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs
index fdf8f822..3f71172c 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Threed/StateUpdater.cs
+++ b/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>
diff --git a/Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs b/Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs
index 8e222e71..106a6f3f 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/Threed/ThreedClass.cs
@@ -498,34 +498,58 @@ namespace Ryujinx.Graphics.Gpu.Engine.Threed
         }
 
         /// <summary>
-        /// Clears the current color and depth-stencil buffers.
-        /// Which buffers should be cleared can also specified with the arguments.
+        /// Performs a indexed or non-indexed draw.
         /// </summary>
-        /// <param name="argument">Method call argument</param>
-        /// <param name="layerCount">For array and 3D textures, indicates how many layers should be cleared</param>
-        public void Clear(int argument, int layerCount)
+        /// <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.Clear(this, argument, layerCount);
+            _drawManager.Draw(this, topology, count, instanceCount, firstIndex, firstVertex, firstInstance, indexed);
         }
 
         /// <summary>
-        /// Performs a indirect multi-draw, with parameters from a GPU buffer.
+        /// Performs a indirect 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="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 element on the <paramref name="indirectBuffer"/> array</param>
-        public void MultiDrawIndirectCount(
-            int indexCount,
+        /// <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,
-            BufferRange indirectBuffer,
-            BufferRange parameterBuffer,
+            ulong indirectBufferAddress,
+            ulong parameterBufferAddress,
             int maxDrawCount,
-            int stride)
+            int stride,
+            int indexCount,
+            IndirectDrawType drawType)
         {
-            _drawManager.MultiDrawIndirectCount(this, indexCount, topology, indirectBuffer, parameterBuffer, maxDrawCount, stride);
+            _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.
+        /// </summary>
+        /// <param name="argument">Method call argument</param>
+        /// <param name="layerCount">For array and 3D textures, indicates how many layers should be cleared</param>
+        public void Clear(int argument, int layerCount)
+        {
+            _drawManager.Clear(this, argument, layerCount);
         }
     }
 }