From 1a550e810c71670e5a2f032ec136fb2f5f57ac9c Mon Sep 17 00:00:00 2001
From: gdkchan <gab.dark.100@gmail.com>
Date: Wed, 11 Dec 2019 16:43:28 -0300
Subject: [PATCH] Copy 16 bytes at a time for layout conversion, if possible

---
 Ryujinx.Graphics.Texture/BlockLinearLayout.cs | 41 ++++++++-----------
 Ryujinx.Graphics.Texture/LayoutConverter.cs   | 15 ++++++-
 2 files changed, 32 insertions(+), 24 deletions(-)

diff --git a/Ryujinx.Graphics.Texture/BlockLinearLayout.cs b/Ryujinx.Graphics.Texture/BlockLinearLayout.cs
index 3d3f13ba..b95db702 100644
--- a/Ryujinx.Graphics.Texture/BlockLinearLayout.cs
+++ b/Ryujinx.Graphics.Texture/BlockLinearLayout.cs
@@ -1,5 +1,5 @@
 using Ryujinx.Common;
-using System;
+using System.Runtime.CompilerServices;
 
 using static Ryujinx.Graphics.Texture.BlockLinearConstants;
 
@@ -19,11 +19,6 @@ namespace Ryujinx.Graphics.Texture
             }
         }
 
-        private int _texWidth;
-        private int _texHeight;
-        private int _texDepth;
-        private int _texGobBlocksInY;
-        private int _texGobBlocksInZ;
         private int _texBpp;
 
         private int _bhMask;
@@ -46,12 +41,7 @@ namespace Ryujinx.Graphics.Texture
             int gobBlocksInZ,
             int bpp)
         {
-            _texWidth        = width;
-            _texHeight       = height;
-            _texDepth        = depth;
-            _texGobBlocksInY = gobBlocksInY;
-            _texGobBlocksInZ = gobBlocksInZ;
-            _texBpp          = bpp;
+            _texBpp = bpp;
 
             _bppShift = BitUtils.CountTrailingZeros32(bpp);
 
@@ -80,27 +70,32 @@ namespace Ryujinx.Graphics.Texture
             return new RobAndSliceSizes(robSize, sliceSize);
         }
 
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
         public int GetOffset(int x, int y, int z)
         {
-            x <<= _bppShift;
+            return GetOffsetWithLineOffset(x << _bppShift, y, z);
+        }
 
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public int GetOffsetWithLineOffset(int x, int y, int z)
+        {
             int yh = y / GobHeight;
 
-            int position = (z >> _bdShift) * _sliceSize + (yh >> _bhShift) * _robSize;
+            int offset = (z >> _bdShift) * _sliceSize + (yh >> _bhShift) * _robSize;
 
-            position += (x / GobStride) << _xShift;
+            offset += (x / GobStride) << _xShift;
 
-            position += (yh & _bhMask) * GobSize;
+            offset += (yh & _bhMask) * GobSize;
 
-            position += ((z & _bdMask) * GobSize) << _bhShift;
+            offset += ((z & _bdMask) * GobSize) << _bhShift;
 
-            position += ((x & 0x3f) >> 5) << 8;
-            position += ((y & 0x07) >> 1) << 6;
-            position += ((x & 0x1f) >> 4) << 5;
-            position += ((y & 0x01) >> 0) << 4;
-            position += ((x & 0x0f) >> 0) << 0;
+            offset += ((x & 0x3f) >> 5) << 8;
+            offset += ((y & 0x07) >> 1) << 6;
+            offset += ((x & 0x1f) >> 4) << 5;
+            offset += ((y & 0x01) >> 0) << 4;
+            offset += ((x & 0x0f) >> 0) << 0;
 
-            return position;
+            return offset;
         }
     }
 }
\ No newline at end of file
diff --git a/Ryujinx.Graphics.Texture/LayoutConverter.cs b/Ryujinx.Graphics.Texture/LayoutConverter.cs
index c270b494..2c3d641b 100644
--- a/Ryujinx.Graphics.Texture/LayoutConverter.cs
+++ b/Ryujinx.Graphics.Texture/LayoutConverter.cs
@@ -62,6 +62,10 @@ namespace Ryujinx.Graphics.Texture
                     mipGobBlocksInZ >>= 1;
                 }
 
+                int strideTrunc = BitUtils.AlignDown(w * bytesPerPixel, 16);
+
+                int xStart = strideTrunc / bytesPerPixel;
+
                 int stride   = BitUtils.AlignUp(w * bytesPerPixel, HostStrideAlignment);
                 int wAligned = BitUtils.AlignUp(w, wAlignment);
 
@@ -80,7 +84,16 @@ namespace Ryujinx.Graphics.Texture
                     for (int z = 0; z < d; z++)
                     for (int y = 0; y < h; y++)
                     {
-                        for (int x = 0; x < w; x++)
+                        for (int x = 0; x < strideTrunc; x += 16)
+                        {
+                            int offset = inBaseOffset + layoutConverter.GetOffsetWithLineOffset(x, y, z);
+
+                            Span<byte> dest = output.Slice(outOffs + x, 16);
+
+                            data.Slice(offset, 16).CopyTo(dest);
+                        }
+
+                        for (int x = xStart; x < w; x++)
                         {
                             int offset = inBaseOffset + layoutConverter.GetOffset(x, y, z);