drkIIRaziel reversed engineered the frsqrte instruction providing hardware accurate emulation.

The algorithm and lookup table are 99% accurate which is far better than the approximation that Dolphin had previously.

The frsqrte instruction is used for a lot of 3D graphics functions so this change may help fix graphics glitches.

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@4864 8ced0084-cf51-0410-be5f-012b33b47a6e

Source/Core/Core/Core.vcproj

@@ -939,6 +939,14 @@
 				RelativePath=".\Src\PowerPC\Gekko.h"
 				>
 			</File>
+			<File
+				RelativePath=".\Src\PowerPC\LUT_frsqrtex.cpp"
+				>
+			</File>
+			<File
+				RelativePath=".\Src\PowerPC\LUT_frsqrtex.h"
+				>
+			</File>
 			<File
 				RelativePath=".\Src\PowerPC\PowerPC.cpp"
 				>
@@ -1029,6 +1037,14 @@
 							AssemblerOutput="4"
 						/>
 					</FileConfiguration>
+					<FileConfiguration
+						Name="Debug_JITIL|Win32"
+						>
+						<Tool
+							Name="VCCLCompilerTool"
+							AssemblerOutput="4"
+						/>
+					</FileConfiguration>
 				</File>
 				<File
 					RelativePath=".\Src\PowerPC\Interpreter\Interpreter_FPUtils.h"

Source/Core/Core/Src/PowerPC/Interpreter/Interpreter_FloatingPoint.cpp

@@ -36,6 +36,7 @@
 #include "Interpreter.h"
 #include "MathUtil.h"
 #include "Interpreter_FPUtils.h"
+#include "../LUT_frsqrtex.h"
 
 using namespace MathUtil;
 
@@ -414,8 +415,27 @@ void frsqrtex(UGeckoInstruction _inst)
 	} 
 	else 
 	{
-		if (b == 0.0) SetFPException(FPSCR_ZX);
+		if (b == 0.0) {
-		rPS0(_inst.FD) = ForceDouble(1.0 / sqrt(b));
+			SetFPException(FPSCR_ZX);
+			riPS0(_inst.FD) = 0x7ff0000000000000;
+		}
+		else
+		{
+			u32 fsa = Common::swap32(Common::swap64(riPS0(_inst.FB)));
+			u32 fsb = Common::swap32(Common::swap64(riPS0(_inst.FB)) >> 32);
+			u32 idx=(fsa >> 5) % (sizeof(frsqrtex_lut) / sizeof(frsqrtex_lut[0]));
+
+			s32 e = fsa >> (32-12);
+			e &= 2047;
+			e -= 1023;
+			s32 oe =- ((e + 1) / 2);
+			oe -= ((e + 1) & 1);
+
+			u32 outb = frsqrtex_lut[idx] << 20;
+			u32 outa = ((oe + 1023) & 2047) << 20;
+			outa |= frsqrtex_lut[idx] >> 12;
+			riPS0(_inst.FD) = ((u64)outa << 32) + (u64)outb;
+		}
 	}
  	UpdateFPRF(rPS0(_inst.FD));
 	if (_inst.Rc) Helper_UpdateCR1(rPS0(_inst.FD));

Source/Core/Core/Src/PowerPC/LUT_frsqrtex.h

@@ -0,0 +1,25 @@
+// Copyright (C) 2003 Dolphin Project.
+
+// This program is free software: you can redistribute it and/or modify
+// it under the terms of the GNU General Public License as published by
+// the Free Software Foundation, version 2.0.
+
+// This program is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+// GNU General Public License 2.0 for more details.
+
+// A copy of the GPL 2.0 should have been included with the program.
+// If not, see http://www.gnu.org/licenses/
+
+// Official SVN repository and contact information can be found at
+// http://code.google.com/p/dolphin-emu/
+// Gekko related unions, structs, ...
+//
+
+#ifndef _LUT_frsqrtex_h_
+#define _LUT_frsqrtex_h_
+
+extern const unsigned int frsqrtex_lut[65536];
+
+#endif //_LUT_frsqrtex_h_

Source/Core/Core/Src/SConscript

@@ -73,6 +73,7 @@ files = ["ActionReplay.cpp",
          "PowerPC/PowerPC.cpp",
          "PowerPC/PPCAnalyst.cpp",
          "PowerPC/PPCTables.cpp",
+         "PowerPC/LUT_frsqrtex.cpp",
          "PowerPC/PPCCache.cpp",
          "PowerPC/Profiler.cpp",
          "PowerPC/SignatureDB.cpp",