From bea1fc2e8d40ec792964852f57e7b884dfbd8306 Mon Sep 17 00:00:00 2001
From: riperiperi <rhy3756547@hotmail.com>
Date: Sat, 13 Jun 2020 23:31:06 +0100
Subject: Optimize texture format conversion, and MethodCopyBuffer (#1274)

* Improve performance when converting texture formats.

Still more work to do.

* Speed up buffer -> texture copies.

No longer copies byte by byte. Fast path when formats are identical.

* Fix a few things, 64 byte block fast copy.

* Spacing cleanup, unrelated change.

* Fix base offset calculation for region copies.

* Fix Linear -> BlockLinear

* Fix some nits. (part 1 of review feedback)

* Use a generic version of the Convert* functions rather than lambdas.

This is some real monkey's paw shit.

* Remove unnecessary span constructor.

* Revert "Use a generic version of the Convert* functions rather than lambdas."

This reverts commit aa43dcfbe8bba291eea4e10c68569af7a56a5851.

* Fix bug with rectangle destination writing, better rectangle calculation for linear textures.
---
 Ryujinx.Graphics.Texture/BlockLinearLayout.cs | 94 +++++++++++++++++++++++++++
 1 file changed, 94 insertions(+)

(limited to 'Ryujinx.Graphics.Texture/BlockLinearLayout.cs')

diff --git a/Ryujinx.Graphics.Texture/BlockLinearLayout.cs b/Ryujinx.Graphics.Texture/BlockLinearLayout.cs
index b95db702..0b112242 100644
--- a/Ryujinx.Graphics.Texture/BlockLinearLayout.cs
+++ b/Ryujinx.Graphics.Texture/BlockLinearLayout.cs
@@ -33,6 +33,11 @@ namespace Ryujinx.Graphics.Texture
         private int _robSize;
         private int _sliceSize;
 
+        // Variables for built in iteration.
+        private int _yPart;
+        private int _yzPart;
+        private int _zPart;
+
         public BlockLinearLayout(
             int width,
             int height,
@@ -97,5 +102,94 @@ namespace Ryujinx.Graphics.Texture
 
             return offset;
         }
+
+        public (int offset, int size) GetRectangleRange(int x, int y, int width, int height)
+        {
+            // Justification:
+            // The 2D offset is a combination of separate x and y parts.
+            // Both components increase with input and never overlap bits.
+            // Therefore for each component, the minimum input value is the lowest that component can go.
+            // Minimum total value is minimum X component + minimum Y component. Similar goes for maximum.
+
+            int start = GetOffset(x, y, 0);
+            int end = GetOffset(x + width - 1, y + height - 1, 0) + _texBpp; // Cover the last pixel.
+            return (start, end - start);
+        }
+
+        public bool LayoutMatches(BlockLinearLayout other)
+        {
+            return _robSize == other._robSize &&
+                   _sliceSize == other._sliceSize &&
+                   _texBpp == other._texBpp &&
+                   _bhMask == other._bhMask &&
+                   _bdMask == other._bdMask;
+        }
+
+        // Functions for built in iteration.
+        // Components of the offset can be updated separately, and combined to save some time.
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public void SetY(int y)
+        {
+            int yh = y / GobHeight;
+            int offset = (yh >> _bhShift) * _robSize;
+
+            offset += (yh & _bhMask) * GobSize;
+
+            offset += ((y & 0x07) >> 1) << 6;
+            offset += ((y & 0x01) >> 0) << 4;
+
+            _yPart = offset;
+            _yzPart = offset + _zPart;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public void SetZ(int z)
+        {
+            int offset = (z >> _bdShift) * _sliceSize;
+
+            offset += ((z & _bdMask) * GobSize) << _bhShift;
+
+            _zPart = offset;
+            _yzPart = offset + _yPart;
+        }
+
+        /// <summary>
+        /// Optimized conversion for line offset in bytes to an absolute offset. Input x must be divisible by 16.
+        /// </summary>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public int GetOffsetWithLineOffset16(int x)
+        {
+            int offset = (x / GobStride) << _xShift;
+
+            offset += ((x & 0x3f) >> 5) << 8;
+            offset += ((x & 0x1f) >> 4) << 5;
+
+            return offset + _yzPart;
+        }
+
+        /// <summary>
+        /// Optimized conversion for line offset in bytes to an absolute offset. Input x must be divisible by 64.
+        /// </summary>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public int GetOffsetWithLineOffset64(int x)
+        {
+            int offset = (x / GobStride) << _xShift;
+
+            return offset + _yzPart;
+        }
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public int GetOffset(int x)
+        {
+            x <<= _bppShift;
+            int offset = (x / GobStride) << _xShift;
+
+            offset += ((x & 0x3f) >> 5) << 8;
+            offset += ((x & 0x1f) >> 4) << 5;
+            offset += (x & 0x0f);
+
+            return offset + _yzPart;
+        }
     }
 }
\ No newline at end of file
-- 
cgit v1.2.3