Fix EFB reads in fastmem mode by removing the EFB mapping (forces it to go through the right handlers). Enable fastmem mode in x64. 1-2% speed boost overall maybe :p

Disable the Redundant MOV warning in release mode, it's only enabled in debug builds now.

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@4870 8ced0084-cf51-0410-be5f-012b33b47a6e

Source/Core/Common/Src/x64Emitter.cpp

@@ -927,8 +927,10 @@ void XEmitter::OR  (int bits, const OpArg &a1, const OpArg &a2) {WriteNormalOp(t
 void XEmitter::XOR (int bits, const OpArg &a1, const OpArg &a2) {WriteNormalOp(this, bits, nrmXOR, a1, a2);}
 void XEmitter::MOV (int bits, const OpArg &a1, const OpArg &a2) 
 {
-	_assert_msg_(DYNA_REC, !a1.IsSimpleReg() || !a2.IsSimpleReg() || a1.GetSimpleReg() != a2.GetSimpleReg(), "Redundant MOV @ %p", 
+#ifdef _DEBUG
+	_assert_msg_(DYNA_REC, !a1.IsSimpleReg() || !a2.IsSimpleReg() || a1.GetSimpleReg() != a2.GetSimpleReg(), "Redundant MOV @ %p - bug in JIT?", 
 				 code); 
+#endif
 	WriteNormalOp(this, bits, nrmMOV, a1, a2);
 }
 void XEmitter::TEST(int bits, const OpArg &a1, const OpArg &a2) {WriteNormalOp(this, bits, nrmTEST, a1, a2);}

Source/Core/Core/Src/HW/Memmap.cpp

@@ -78,7 +78,6 @@ bool m_IsInitialized = false; // Save the Init(), Shutdown() state
 // 64-bit: Pointers to low-mem (sub-0x10000000) mirror
 // 32-bit: Same as the corresponding physical/virtual pointers.
 u8 *m_pRAM;
-u8 *m_pEFB; 
 u8 *m_pL1Cache;
 u8 *m_pEXRAM;
 u8 *m_pFakeVMEM;
@@ -329,7 +328,9 @@ static const MemoryView views[] =
 	{NULL,         &m_pVirtualCachedRAM,     0x80000000, RAM_SIZE, MV_MIRROR_PREVIOUS},
 	{NULL,         &m_pVirtualUncachedRAM,   0xC0000000, RAM_SIZE, MV_MIRROR_PREVIOUS},
 
-	{&m_pEFB,      &m_pVirtualEFB,           0xC8000000, EFB_SIZE, 0},
+//  Don't map any memory for the EFB. We want all access to this area to go
+//  through the hardware access handlers.
+//	{&m_pEFB,      &m_pVirtualEFB,           0xC8000000, EFB_SIZE, 0},
 	{&m_pL1Cache,  &m_pVirtualL1Cache,       0xE0000000, L1_CACHE_SIZE, 0},
 
 	{&m_pFakeVMEM, &m_pVirtualFakeVMEM,      0x7E000000, FAKEVMEM_SIZE, MV_FAKE_VMEM},
@@ -390,8 +391,6 @@ void Clear()
 		memset(m_pRAM, 0, RAM_SIZE);
 	if (m_pL1Cache)
 		memset(m_pL1Cache, 0, L1_CACHE_SIZE);
-	if (m_pEFB)
-		memset(m_pEFB, 0, EFB_SIZE);
 	if (Core::GetStartupParameter().bWii && m_pEXRAM)
 		memset(m_pEXRAM, 0, EXRAM_SIZE);
 }
@@ -777,7 +776,8 @@ u8 *GetPointer(const u32 _Address)
 			return 0;
 
 	case 0xC8:
-		return m_pEFB + (_Address & 0xFFFFFF);
+		return 0;  // EFB. We don't want to return a pointer here since we have no memory mapped for it.
+
 	case 0xCC:
 	case 0xCD:
 		_dbg_assert_msg_(MEMMAP, 0, "Memory", "GetPointer from IO Bridge doesnt work");

Source/Core/Core/Src/HW/Memmap.h

@@ -135,9 +135,12 @@ u32 Read_U32(const u32 _Address);
 u64 Read_U64(const u32 _Address);
 
 // used by JIT. Return zero-extended 32bit values
-u32  Read_U8_ZX(const u32 _Address);
+u32 Read_U8_ZX(const u32 _Address);
 u32 Read_U16_ZX(const u32 _Address);
 
+// used by JIT (Jit64::lXz)
+u32 EFB_Read(const u32 addr);
+
 void Write_U8(const u8 _Data, const u32 _Address);
 void Write_U16(const u16 _Data, const u32 _Address);
 void Write_U32(const u32 _Data, const u32 _Address);

Source/Core/Core/Src/HW/MemmapFunctions.cpp

@@ -125,6 +125,26 @@ inline void hwWriteIOBridge(u16 var, u32 addr) {WII_IOBridge::Write16(var, addr)
 inline void hwWriteIOBridge(u32 var, u32 addr) {WII_IOBridge::Write32(var, addr);}
 inline void hwWriteIOBridge(u64 var, u32 addr) {PanicAlert("hwWriteIOBridge: There's no 64-bit HW write. %08x", addr);}
 
+// Nasty but necessary. Super Mario Galaxy pointer relies on this stuff.
+u32 EFB_Read(const u32 addr) 
+{
+	u32 var = 0;
+	// Convert address to coordinates. It's possible that this should be done
+	// differently depending on color depth, especially regarding PEEK_COLOR.
+	int x = (addr & 0xfff) >> 2;
+	int y = (addr >> 12) & 0x3ff;
+
+	if (addr & 0x00400000) {
+		var = CPluginManager::GetInstance().GetVideo()->Video_AccessEFB(PEEK_Z, x, y);
+		DEBUG_LOG(MEMMAP, "EFB Z Read @ %i, %i\t= 0x%08x", x, y, var);
+	} else {
+		var = CPluginManager::GetInstance().GetVideo()->Video_AccessEFB(PEEK_COLOR, x, y);
+		DEBUG_LOG(MEMMAP, "EFB Color Read @ %i, %i\t= 0x%08x", x, y, var);
+	}
+
+	return var;
+}
+
 template <class T>
 inline void ReadFromHardware(T &_var, u32 em_address, u32 effective_address, Memory::XCheckTLBFlag flag)
 {
@@ -132,17 +152,7 @@ inline void ReadFromHardware(T &_var, u32 em_address, u32 effective_address, Mem
 	if ((em_address & 0xC8000000) == 0xC8000000)
 	{
 		if (em_address < 0xcc000000)
-		{
+			_var = EFB_Read(em_address);
-			int x = (em_address & 0xfff) >> 2;
-			int y = (em_address >> 12) & 0x3ff;
-			if (em_address & 0x00400000) {
-				_var = CPluginManager::GetInstance().GetVideo()->Video_AccessEFB(PEEK_Z, x, y);
-				DEBUG_LOG(MEMMAP, "EFB Z Read @ %i, %i\t= 0x%08x", x, y, _var);
-			} else {
-				_var = CPluginManager::GetInstance().GetVideo()->Video_AccessEFB(PEEK_COLOR, x, y);
-				DEBUG_LOG(MEMMAP, "EFB Color Read @ %i, %i\t= 0x%08x", x, y, _var);
-			}
-		}
 		else if (em_address <= 0xcc009000)
 			hwRead(_var, em_address);
 		/* WIIMODE */
@@ -356,7 +366,6 @@ u32 Read_U32(const u32 _Address)
 	return _var;
 }
 
-
 u64 Read_U64(const u32 _Address)
 {
 	u64 _var = 0;

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

@@ -405,8 +405,6 @@ void fresx(UGeckoInstruction _inst)
 	if (_inst.Rc) Helper_UpdateCR1(rPS0(_inst.FD));
 }
 
-// #define USE_ACCURATE_FRSQRTEX
-
 void frsqrtex(UGeckoInstruction _inst)
 {
 	double b = rPS0(_inst.FB);
@@ -417,7 +415,7 @@ void frsqrtex(UGeckoInstruction _inst)
 	} 
 	else 
 	{
-#ifdef USE_ACCURATE_FRSQRTEX
+#ifdef VERY_ACCURATE_FP
 		if (b == 0.0) {
 			SetFPException(FPSCR_ZX);
 			riPS0(_inst.FD) = 0x7ff0000000000000;

Source/Core/Core/Src/PowerPC/Jit64/Jit_LoadStore.cpp

@@ -193,7 +193,7 @@ void Jit64::lXz(UGeckoInstruction inst)
 	}
 
 	//Still here? Do regular path.
-#if defined(_M_X64) && 0 // This is disabled, it's unaccurate
+#if defined(_M_X64)
 	if (accessSize == 8 || accessSize == 16 || !jo.enableFastMem) {
 #else
 	if (true) {