From 79aa764d03744a3d2ad08487303f24902a2970af Mon Sep 17 00:00:00 2001
From: kuba <xnetcat.dev@gmail.com>
Date: Fri, 31 Jul 2026 11:08:56 +0200
Subject: [PATCH] Metal: skip waitUntilCompleted when the command buffer is
 already Completed (#709)

Tiny write-back batches often finish before the wait; checking status avoids redundant ordered-queue round-trips.
---
 .../Gpu/Metal/MetalVideoPresenter.Compute.cs  |  2 +-
 .../Gpu/Metal/MetalVideoPresenter.Draws.cs    | 28 +++++++++++++++++--
 2 files changed, 27 insertions(+), 3 deletions(-)

diff --git a/src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.Compute.cs b/src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.Compute.cs
index 97f1b3dd..04d44b74 100644
--- a/src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.Compute.cs
+++ b/src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.Compute.cs
@@ -253,7 +253,7 @@ internal static partial class MetalVideoPresenter
         if (writeBackBuffers.Count > 0)
         {
             var committed = FlushBatchedGuestCommands();
-            MetalNative.SendVoid(committed, MetalNative.Selector("waitUntilCompleted"));
+            WaitForCommittedCommandBuffer(committed);
             WriteBuffersBackToGuest(writeBackBuffers);
         }
 
diff --git a/src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.Draws.cs b/src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.Draws.cs
index 5214e501..502d4bef 100644
--- a/src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.Draws.cs
+++ b/src/SharpEmu.Libs/Gpu/Metal/MetalVideoPresenter.Draws.cs
@@ -580,7 +580,7 @@ internal static partial class MetalVideoPresenter
         if (writeBackBuffers.Count > 0)
         {
             var committed = FlushBatchedGuestCommands();
-            MetalNative.SendVoid(committed, MetalNative.Selector("waitUntilCompleted"));
+            WaitForCommittedCommandBuffer(committed);
             WriteBuffersBackToGuest(writeBackBuffers);
         }
 
@@ -668,7 +668,7 @@ internal static partial class MetalVideoPresenter
         TagSnapshotResources(commandBuffer);
         if (writeBackBuffers.Count > 0)
         {
-            MetalNative.SendVoid(commandBuffer, MetalNative.Selector("waitUntilCompleted"));
+            WaitForCommittedCommandBuffer(commandBuffer);
             WriteBuffersBackToGuest(writeBackBuffers);
         }
 
@@ -2082,6 +2082,30 @@ internal static partial class MetalVideoPresenter
         }
     }
 
+
+    /// <summary>
+    /// Blocks until a committed command buffer finishes. Skips the ObjC wait
+    /// when status is already Completed (common for tiny label/writeback
+    /// batches), avoiding redundant waitUntilCompleted round-trips on the
+    /// ordered queue.
+    /// </summary>
+    private static void WaitForCommittedCommandBuffer(nint commandBuffer)
+    {
+        if (commandBuffer == 0)
+        {
+            return;
+        }
+
+        // MTLCommandBufferStatusCompleted = 4.
+        const nint completed = 4;
+        if (MetalNative.Send(commandBuffer, MetalNative.Selector("status")) >= completed)
+        {
+            return;
+        }
+
+        MetalNative.SendVoid(commandBuffer, MetalNative.Selector("waitUntilCompleted"));
+    }
+
     private static void ReturnPooledGuestData(TranslatedGuestDraw draw)
     {
         foreach (var buffer in draw.GlobalMemoryBuffers)