5 files changed, 58 insertions, 6 deletions

diff --git a/Ryujinx.Graphics.Gpu/Engine/Compute.cs b/Ryujinx.Graphics.Gpu/Engine/Compute.cs
index bcff5953..be317a7f 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Compute.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/Compute.cs
@@ -16,6 +16,8 @@ namespace Ryujinx.Graphics.Gpu.Engine
         /// <param name="argument">Method call argument</param>
         public void Dispatch(GpuState state, int argument)
         {
+            FlushUboDirty();
+
             uint qmdAddress = (uint)state.Get<int>(MethodOffset.DispatchParamsAddress);
 
             var qmd = _context.MemoryManager.Read<ComputeQmd>((ulong)qmdAddress << 8);
diff --git a/Ryujinx.Graphics.Gpu/Engine/MethodCopyBuffer.cs b/Ryujinx.Graphics.Gpu/Engine/MethodCopyBuffer.cs
index dd16cb2d..a1cf86ec 100644
--- a/Ryujinx.Graphics.Gpu/Engine/MethodCopyBuffer.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/MethodCopyBuffer.cs
@@ -69,6 +69,8 @@ namespace Ryujinx.Graphics.Gpu.Engine
                 return;
             }
 
+            FlushUboDirty();
+
             if (copy2D)
             {
                 // Buffer to texture copy.
diff --git a/Ryujinx.Graphics.Gpu/Engine/MethodUniformBufferBind.cs b/Ryujinx.Graphics.Gpu/Engine/MethodUniformBufferBind.cs
index 3fee1fcf..16fb31d6 100644
--- a/Ryujinx.Graphics.Gpu/Engine/MethodUniformBufferBind.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/MethodUniformBufferBind.cs
@@ -66,6 +66,8 @@ namespace Ryujinx.Graphics.Gpu.Engine
 
             int index = (argument >> 4) & 0x1f;
 
+            FlushUboDirty();
+
             if (enable)
             {
                 var uniformBuffer = state.Get<UniformBufferState>(MethodOffset.UniformBufferState);
diff --git a/Ryujinx.Graphics.Gpu/Engine/MethodUniformBufferUpdate.cs b/Ryujinx.Graphics.Gpu/Engine/MethodUniformBufferUpdate.cs
index 61772327..3e1dd151 100644
--- a/Ryujinx.Graphics.Gpu/Engine/MethodUniformBufferUpdate.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/MethodUniformBufferUpdate.cs
@@ -1,3 +1,4 @@
+using Ryujinx.Graphics.Gpu.Memory;
 using Ryujinx.Graphics.Gpu.State;
 using System;
 using System.Runtime.InteropServices;
@@ -6,6 +7,25 @@ namespace Ryujinx.Graphics.Gpu.Engine
 {
     partial class Methods
     {
+        // State associated with direct uniform buffer updates.
+        // This state is used to attempt to batch together consecutive updates.
+        private ulong _ubBeginCpuAddress = 0;
+        private ulong _ubFollowUpAddress = 0;
+        private ulong _ubByteCount = 0;
+
+        /// <summary>
+        /// Flushes any queued ubo updates.
+        /// </summary>
+        private void FlushUboDirty()
+        {
+            if (_ubFollowUpAddress != 0)
+            {
+                BufferManager.ForceDirty(_ubFollowUpAddress - _ubByteCount, _ubByteCount);
+
+                _ubFollowUpAddress = 0;
+            }
+        }
+
         /// <summary>
         /// Updates the uniform buffer data with inline data.
         /// </summary>
@@ -15,11 +35,22 @@ namespace Ryujinx.Graphics.Gpu.Engine
         {
             var uniformBuffer = state.Get<UniformBufferState>(MethodOffset.UniformBufferState);
 
-            _context.MemoryManager.Write(uniformBuffer.Address.Pack() + (uint)uniformBuffer.Offset, argument);
+            ulong address = uniformBuffer.Address.Pack() + (uint)uniformBuffer.Offset;
 
-            state.SetUniformBufferOffset(uniformBuffer.Offset + 4);
+            if (_ubFollowUpAddress != address)
+            {
+                FlushUboDirty();
+
+                _ubByteCount = 0;
+                _ubBeginCpuAddress = _context.MemoryManager.Translate(address);
+            }
 
-            _context.AdvanceSequence();
+            _context.PhysicalMemory.WriteUntracked(_ubBeginCpuAddress + _ubByteCount, MemoryMarshal.Cast<int, byte>(MemoryMarshal.CreateSpan(ref argument, 1)));
+
+            _ubFollowUpAddress = address + 4;
+            _ubByteCount += 4;
+
+            state.SetUniformBufferOffset(uniformBuffer.Offset + 4);
         }
 
         /// <summary>
@@ -31,11 +62,24 @@ namespace Ryujinx.Graphics.Gpu.Engine
         {
             var uniformBuffer = state.Get<UniformBufferState>(MethodOffset.UniformBufferState);
 
-            _context.MemoryManager.Write(uniformBuffer.Address.Pack() + (uint)uniformBuffer.Offset, MemoryMarshal.Cast<int, byte>(data));
+            ulong address = uniformBuffer.Address.Pack() + (uint)uniformBuffer.Offset;
 
-            state.SetUniformBufferOffset(uniformBuffer.Offset + data.Length * 4);
+            ulong size = (ulong)data.Length * 4;
+
+            if (_ubFollowUpAddress != address)
+            {
+                FlushUboDirty();
 
-            _context.AdvanceSequence();
+                _ubByteCount = 0;
+                _ubBeginCpuAddress = _context.MemoryManager.Translate(address);
+            }
+
+            _context.PhysicalMemory.WriteUntracked(_ubBeginCpuAddress + _ubByteCount, MemoryMarshal.Cast<int, byte>(data));
+
+            _ubFollowUpAddress = address + size;
+            _ubByteCount += size;
+
+            state.SetUniformBufferOffset(uniformBuffer.Offset + data.Length * 4);
         }
     }
 }
\ No newline at end of file
diff --git a/Ryujinx.Graphics.Gpu/Engine/Methods.cs b/Ryujinx.Graphics.Gpu/Engine/Methods.cs
index ae9bdb0d..431ea449 100644
--- a/Ryujinx.Graphics.Gpu/Engine/Methods.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/Methods.cs
@@ -130,6 +130,8 @@ namespace Ryujinx.Graphics.Gpu.Engine
                 _prevTfEnable = false;
             }
 
+            FlushUboDirty();
+
             // Shaders must be the first one to be updated if modified, because
             // some of the other state depends on information from the currently
             // bound shaders.