From f3cc2e5703e5df5c359ce1789a4fb0d73fb9a637 Mon Sep 17 00:00:00 2001
From: riperiperi <rhy3756547@hotmail.com>
Date: Thu, 24 Nov 2022 01:56:55 +0000
Subject: [PATCH] GPU: Access non-prefetch command buffers directly (#3882)

* GPU: Access non-prefetch command buffers directly

Saves allocating new arrays for them constantly - they can be quite small so it can be very wasteful. About 0.4% of GPU thread in SMO, but was a bit higher in S/V when I checked.

Assumes that non-prefetch command buffers won't be randomly clobbered before they finish executing, though that's probably a safe bet.

* Small change while I'm here

* Address feedback
---
 .../Engine/GPFifo/GPFifoDevice.cs             | 35 ++++++++++++++-----
 1 file changed, 27 insertions(+), 8 deletions(-)

diff --git a/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoDevice.cs b/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoDevice.cs
index b3de738d69..cd29a9da0e 100644
--- a/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoDevice.cs
+++ b/Ryujinx.Graphics.Gpu/Engine/GPFifo/GPFifoDevice.cs
@@ -51,16 +51,35 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
             /// </summary>
             public uint EntryCount;
 
+            /// <summary>
+            /// Get the entries for the command buffer from memory.
+            /// </summary>
+            /// <param name="memoryManager">The memory manager used to fetch the data</param>
+            /// <param name="flush">If true, flushes potential GPU written data before reading the command buffer</param>
+            /// <returns>The fetched data</returns>
+            private ReadOnlySpan<int> GetWords(MemoryManager memoryManager, bool flush)
+            {
+                return MemoryMarshal.Cast<byte, int>(memoryManager.GetSpan(EntryAddress, (int)EntryCount * 4, flush));
+            }
+
+            /// <summary>
+            /// Prefetch the command buffer.
+            /// </summary>
+            /// <param name="memoryManager">The memory manager used to fetch the data</param>
+            public void Prefetch(MemoryManager memoryManager)
+            {
+                Words = GetWords(memoryManager, true).ToArray();
+            }
+
             /// <summary>
             /// Fetch the command buffer.
             /// </summary>
+            /// <param name="memoryManager">The memory manager used to fetch the data</param>
             /// <param name="flush">If true, flushes potential GPU written data before reading the command buffer</param>
-            public void Fetch(MemoryManager memoryManager, bool flush = true)
+            /// <returns>The command buffer words</returns>
+            public ReadOnlySpan<int> Fetch(MemoryManager memoryManager, bool flush)
             {
-                if (Words == null)
-                {
-                    Words = MemoryMarshal.Cast<byte, int>(memoryManager.GetSpan(EntryAddress, (int)EntryCount * 4, flush)).ToArray();
-                }
+                return Words ?? GetWords(memoryManager, flush);
             }
         }
 
@@ -158,7 +177,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
 
                 if (beforeBarrier && commandBuffer.Type == CommandBufferType.Prefetch)
                 {
-                    commandBuffer.Fetch(processor.MemoryManager);
+                    commandBuffer.Prefetch(processor.MemoryManager);
                 }
 
                 if (commandBuffer.Type == CommandBufferType.NoPrefetch)
@@ -199,7 +218,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
                 }
 
                 _currentCommandBuffer = entry;
-                _currentCommandBuffer.Fetch(entry.Processor.MemoryManager, flushCommandBuffer);
+                ReadOnlySpan<int> words = entry.Fetch(entry.Processor.MemoryManager, flushCommandBuffer);
 
                 // If we are changing the current channel,
                 // we need to force all the host state to be updated.
@@ -209,7 +228,7 @@ namespace Ryujinx.Graphics.Gpu.Engine.GPFifo
                     entry.Processor.ForceAllDirty();
                 }
 
-                entry.Processor.Process(entry.EntryAddress, _currentCommandBuffer.Words);
+                entry.Processor.Process(entry.EntryAddress, words);
             }
 
             _interrupt = false;