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>

4 files changed, 191 insertions, 28 deletions

diff --git a/Ryujinx.Graphics.Gpu/Engine/MME/Macro.cs b/Ryujinx.Graphics.Gpu/Engine/MME/Macro.cs
index 1d054969..12a3ac02 100644
--- a/Ryujinx.Graphics.Gpu/Engine/MME/Macro.cs
+++ b/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);
             }
         }
diff --git a/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLE.cs b/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLE.cs
index 5f238a71..8630bbc4 100644
--- a/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLE.cs
+++ b/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;
+
+            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);
+        }
 
-            _processor.ThreedClass.MultiDrawIndirectCount(indexCount, topology, indirectBuffer, parameterBuffer, maxDrawCount, stride);
+        /// <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);
         }
     }
 }
diff --git a/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLEFunctionName.cs b/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLEFunctionName.cs
index 4cce07fa..751867fc 100644
--- a/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLEFunctionName.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLEFunctionName.cs
@@ -8,6 +8,9 @@
         None,
         ClearColor,
         ClearDepthStencil,
+        DrawArraysInstanced,
+        DrawElementsInstanced,
+        DrawElementsIndirect,
         MultiDrawElementsIndirectCount
     }
 }
diff --git a/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLETable.cs b/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLETable.cs
index c5d98848..ab6f54ef 100644
--- a/Ryujinx.Graphics.Gpu/Engine/MME/MacroHLETable.cs
+++ b/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;
                 }
             }