From 6fa020470984970f22f810ceb09f1eeb8fb051f2 Mon Sep 17 00:00:00 2001
From: Sergey Lipskiy <gonetz@ngs.ru>
Date: Thu, 9 Apr 2015 20:33:40 +0600
Subject: [PATCH] Replace Adler32 by more faster textureCRC from Glide64.
 textureCRC is good enough to calculate checksum of frame buffer.

---
 CRC.cpp         | 34 +++++++++++++++-------------------
 CRC.h           |  4 +++-
 FrameBuffer.cpp |  5 +++--
 gDP.cpp         |  2 +-
 4 files changed, 22 insertions(+), 23 deletions(-)

diff --git a/CRC.cpp b/CRC.cpp
index 5223b002..d7df9dbd 100644
--- a/CRC.cpp
+++ b/CRC.cpp
@@ -59,28 +59,24 @@ u32 CRC_CalculatePalette(u32 crc, const void * buffer, u32 count )
 	return crc ^ orig;
 }
 
-u32 Adler32(u32 crc, const void *buffer, u32 count)
+u32 textureCRC(u8 * addr, u32 height, u32 stride)
 {
-	register u32 s1 = crc & 0xFFFF;
-	register u32 s2 = (crc >> 16) & 0xFFFF;
-	int k;
-	const u8 *Buffer = (const u8*)buffer;
+	const u32 width = stride / 8;
+	const u32 line = stride % 8;
+	u64 crc = 0;
+	u64 twopixel_crc;
 
-	if (Buffer == NULL)
-		return 0;
-
-	while (count > 0) {
-		/* 5552 is the largest n such that 255n(n+1)/2 + (n+1)(BASE-1) <= 2^32-1 */
-		k = (count < 5552 ? count : 5552);
-		count -= k;
-		while (k--) {
-			s1 += *Buffer++;
-			s2 += s1;
+	u32 *  pixelpos = (u32*)addr;
+	for (; height; height--) {
+		int col = 0;
+		for (u32 i = width; i; --i) {
+			twopixel_crc = i * ((u64)(pixelpos[1] & 0xFFFEFFFE) + (u64)(pixelpos[0] & 0xFFFEFFFE) + crc);
+			crc = (twopixel_crc >> 32) + twopixel_crc;
+			pixelpos += 2;
 		}
-		/* 65521 is the largest prime smaller than 65536 */
-		s1 %= 65521;
-		s2 %= 65521;
+		crc = (height * crc >> 32) + height * crc;
+		pixelpos = (u32*)((u8*)pixelpos + line);
 	}
 
-	return (s2 << 16) | s1;
+	return crc&0xFFFFFFFF;
 }
diff --git a/CRC.h b/CRC.h
index 2094f781..a9a818fc 100644
--- a/CRC.h
+++ b/CRC.h
@@ -2,6 +2,8 @@
 
 void CRC_BuildTable();
 
+// CRC32
 u32 CRC_Calculate( u32 crc, const void *buffer, u32 count );
 u32 CRC_CalculatePalette( u32 crc, const void *buffer, u32 count );
-u32 Adler32(u32 crc, const void *buffer, u32 count);
+// Fast checksum calculation from Glide64
+u32 textureCRC(u8 * addr, u32 height, u32 stride);
diff --git a/FrameBuffer.cpp b/FrameBuffer.cpp
index 4bd6a5a4..2de1f6f4 100644
--- a/FrameBuffer.cpp
+++ b/FrameBuffer.cpp
@@ -811,7 +811,8 @@ void FrameBufferToRDRAM::_copyWhite(FrameBuffer * _pBuffer)
 			}
 		}
 	}
-	_pBuffer->m_RdramCrc = Adler32(0, RDRAM + _pBuffer->m_startAddress, (VI.width*VI.height) << _pBuffer->m_size >> 1);
+	_pBuffer->m_RdramCrc = textureCRC(RDRAM + _pBuffer->m_startAddress, _pBuffer->m_height, _pBuffer->m_width << _pBuffer->m_size >> 1);
+
 	_pBuffer->m_cleared = false;
 }
 
@@ -880,7 +881,7 @@ void FrameBufferToRDRAM::CopyToRDRAM(u32 _address)
 			}
 		}
 	}
-	pBuffer->m_RdramCrc = Adler32(0, RDRAM + _address, (VI.width*VI.height) << pBuffer->m_size >> 1);
+	pBuffer->m_RdramCrc = textureCRC(RDRAM + pBuffer->m_startAddress, pBuffer->m_height, pBuffer->m_width << pBuffer->m_size >> 1);
 	pBuffer->m_cleared = false;
 #ifndef GLES2
 	glUnmapBuffer(GL_PIXEL_PACK_BUFFER);
diff --git a/gDP.cpp b/gDP.cpp
index 59b985b2..0212a150 100644
--- a/gDP.cpp
+++ b/gDP.cpp
@@ -427,7 +427,7 @@ bool CheckForFrameBufferTexture(u32 _address, u32 _bytes)
 				else
 					frameBufferList().removeBuffer(pBuffer->m_startAddress);
 			} else if (pBuffer->m_RdramCrc != 0) {
-				const u32 crc = Adler32(0, RDRAM + pBuffer->m_startAddress, (VI.width*VI.height) << pBuffer->m_size >> 1);
+				const u32 crc = textureCRC(RDRAM + pBuffer->m_startAddress, pBuffer->m_height, pBuffer->m_width << pBuffer->m_size >> 1);
 				bRes = (pBuffer->m_RdramCrc == crc);
 				if (bRes)
 					pBuffer->m_validityChecked = RSP.DList;