Merge branch 'JitCache' of https://skidau@code.google.com/p/dolphin-emu/

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

@@ -265,6 +265,7 @@ inline void WriteToHardware(u32 em_address, const T data, u32 effective_address,
 		((em_address & 0xF0000000) == 0xC0000000) ||
 		((em_address & 0xF0000000) == 0x00000000))
 	{
+		PowerPC::ppcState.iCache.InvalidateBlock(em_address);
 		*(T*)&m_pRAM[em_address & RAM_MASK] = bswap(data);
 		return;
 	}
@@ -272,6 +273,7 @@ inline void WriteToHardware(u32 em_address, const T data, u32 effective_address,
 		((em_address & 0xF0000000) == 0xD0000000) ||
 		((em_address & 0xF0000000) == 0x10000000))
 	{
+		PowerPC::ppcState.iCache.InvalidateBlock(em_address);
 		*(T*)&m_pEXRAM[em_address & EXRAM_MASK] = bswap(data);
 		return;
 	}
@@ -284,6 +286,7 @@ inline void WriteToHardware(u32 em_address, const T data, u32 effective_address,
 		(bFakeVMEM && ((em_address &0xF0000000) == 0x40000000)))
 	{
 		// fake VMEM
+		PowerPC::ppcState.iCache.InvalidateBlock(em_address);
 		*(T*)&m_pFakeVMEM[em_address & FAKEVMEM_MASK] = bswap(data);
 	}
 	else

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

@@ -363,12 +363,23 @@ void Interpreter::dcbf(UGeckoInstruction _inst)
 	{
 		NPC = PC + 12;
 	}*/
+	// Invalidate the icache on dcbf
+	if (jit)
+	{
+		u32 address = Helper_Get_EA_X(_inst);
+		jit->GetBlockCache()->InvalidateICache(address & ~0x1f);
+	}
 }
 
 void Interpreter::dcbi(UGeckoInstruction _inst)
 {
-	// Removes a block from data cache. Since we don't emulate the data cache, we don't need to do anything.
+	// Removes a block from data cache. Since we don't emulate the data cache, we don't need to do anything to the data cache
-	// Seen used during initialization.
+	// However, we invalidate the icache on dcbi
+	if (jit)
+	{
+		u32 address = Helper_Get_EA_X(_inst);
+		jit->GetBlockCache()->InvalidateICache(address & ~0x1f);
+	}
 }
 
 void Interpreter::dcbst(UGeckoInstruction _inst)

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

@@ -393,6 +393,7 @@ void STACKALIGN Jit64::Jit(u32 em_address)
 	{
 		ClearCache();
 	}
+
 	int block_num = blocks.AllocateBlock(em_address);
 	JitBlock *b = blocks.GetBlock(block_num);
 	blocks.FinalizeBlock(block_num, jo.enableBlocklink, DoJit(em_address, &code_buffer, b));
@@ -612,14 +613,14 @@ const u8* Jit64::DoJit(u32 em_address, PPCAnalyst::CodeBuffer *code_buf, JitBloc
 #ifdef _M_IX86
 		if (js.compilerPC & JIT_ICACHE_VMEM_BIT)
 			MOV(32, M((jit->GetBlockCache()->GetICacheVMEM() + (js.compilerPC & JIT_ICACHE_MASK))), Imm32(JIT_ICACHE_INVALID_WORD));
-		else if (js.blockStart & JIT_ICACHE_EXRAM_BIT)
+		else if (js.compilerPC & JIT_ICACHE_EXRAM_BIT)
 			MOV(32, M((jit->GetBlockCache()->GetICacheEx() + (js.compilerPC & JIT_ICACHEEX_MASK))), Imm32(JIT_ICACHE_INVALID_WORD));
 		else
 			MOV(32, M((jit->GetBlockCache()->GetICache() + (js.compilerPC & JIT_ICACHE_MASK))), Imm32(JIT_ICACHE_INVALID_WORD));
 #else
 		if (js.compilerPC & JIT_ICACHE_VMEM_BIT)
 			MOV(64, R(RAX), ImmPtr(jit->GetBlockCache()->GetICacheVMEM() + (js.compilerPC & JIT_ICACHE_MASK)));
-		else if (js.blockStart & JIT_ICACHE_EXRAM_BIT)
+		else if (js.compilerPC & JIT_ICACHE_EXRAM_BIT)
 			MOV(64, R(RAX), ImmPtr(jit->GetBlockCache()->GetICacheEx() + (js.compilerPC & JIT_ICACHEEX_MASK)));
 		else
 			MOV(64, R(RAX), ImmPtr(jit->GetBlockCache()->GetICache() + (js.compilerPC & JIT_ICACHE_MASK)));

Source/Core/Core/Src/PowerPC/Jit64/Jit64_Tables.cpp

@@ -210,12 +210,12 @@ static GekkoOPTemplate table31[] =
 	{824, &Jit64::srawix}, //"srawix", OPTYPE_INTEGER, FL_OUT_A | FL_IN_B | FL_IN_S | FL_RC_BIT}},
 	{24,  &Jit64::slwx}, //"slwx",   OPTYPE_INTEGER, FL_OUT_A | FL_IN_B | FL_IN_S | FL_RC_BIT}},
 
-	{54,   &Jit64::Default}, //"dcbst",  OPTYPE_DCACHE, 0, 4}},
+	{54,   &Jit64::DoNothing}, //"dcbst",  OPTYPE_DCACHE, 0, 4}},
-	{86,   &Jit64::DoNothing}, //"dcbf",   OPTYPE_DCACHE, 0, 4}},
+	{86,   &Jit64::Default}, //"dcbf",   OPTYPE_DCACHE, 0, 4}},
-	{246,  &Jit64::Default}, //"dcbtst", OPTYPE_DCACHE, 0, 1}},
+	{246,  &Jit64::DoNothing}, //"dcbtst", OPTYPE_DCACHE, 0, 1}},
-	{278,  &Jit64::Default}, //"dcbt",   OPTYPE_DCACHE, 0, 1}},
+	{278,  &Jit64::DoNothing}, //"dcbt",   OPTYPE_DCACHE, 0, 1}},
 	{470,  &Jit64::Default}, //"dcbi",   OPTYPE_DCACHE, 0, 4}},
-	{758,  &Jit64::Default}, //"dcba",   OPTYPE_DCACHE, 0, 4}},
+	{758,  &Jit64::DoNothing}, //"dcba",   OPTYPE_DCACHE, 0, 4}},
 	{1014, &Jit64::dcbz}, //"dcbz",   OPTYPE_DCACHE, 0, 4}},
 
 	//load word

Source/Core/Core/Src/PowerPC/Jit64IL/JitIL.cpp

@@ -494,7 +494,7 @@ void JitIL::Trace()
 	}
 #endif	
 
-	NOTICE_LOG(DYNA_REC, "JITIL PC: %08x SRR0: %08x SRR1: %08x CRfast: %02x%02x%02x%02x%02x%02x%02x%02x FPSCR: %08x MSR: %08x LR: %08x %s %s", 
+	DEBUG_LOG(DYNA_REC, "JITIL PC: %08x SRR0: %08x SRR1: %08x CRfast: %02x%02x%02x%02x%02x%02x%02x%02x FPSCR: %08x MSR: %08x LR: %08x %s %s", 
 		PC, SRR0, SRR1, PowerPC::ppcState.cr_fast[0], PowerPC::ppcState.cr_fast[1], PowerPC::ppcState.cr_fast[2], PowerPC::ppcState.cr_fast[3], 
 		PowerPC::ppcState.cr_fast[4], PowerPC::ppcState.cr_fast[5], PowerPC::ppcState.cr_fast[6], PowerPC::ppcState.cr_fast[7], PowerPC::ppcState.fpscr, 
 		PowerPC::ppcState.msr, PowerPC::ppcState.spr[8], regs, fregs);

Source/Core/Core/Src/PowerPC/Jit64IL/JitIL_Tables.cpp

@@ -212,12 +212,12 @@ static GekkoOPTemplate table31[] =
 	{824, &JitIL::srawix}, //"srawix", OPTYPE_INTEGER, FL_OUT_A | FL_IN_B | FL_IN_S | FL_RC_BIT}},
 	{24,  &JitIL::slwx}, //"slwx",   OPTYPE_INTEGER, FL_OUT_A | FL_IN_B | FL_IN_S | FL_RC_BIT}},
 
-	{54,   &JitIL::Default}, //"dcbst",  OPTYPE_DCACHE, 0, 4}},
+	{54,   &JitIL::DoNothing}, //"dcbst",  OPTYPE_DCACHE, 0, 4}},
-	{86,   &JitIL::DoNothing}, //"dcbf",   OPTYPE_DCACHE, 0, 4}},
+	{86,   &JitIL::Default}, //"dcbf",   OPTYPE_DCACHE, 0, 4}},
-	{246,  &JitIL::Default}, //"dcbtst", OPTYPE_DCACHE, 0, 1}},
+	{246,  &JitIL::DoNothing}, //"dcbtst", OPTYPE_DCACHE, 0, 1}},
-	{278,  &JitIL::Default}, //"dcbt",   OPTYPE_DCACHE, 0, 1}},
+	{278,  &JitIL::DoNothing}, //"dcbt",   OPTYPE_DCACHE, 0, 1}},
 	{470,  &JitIL::Default}, //"dcbi",   OPTYPE_DCACHE, 0, 4}},
-	{758,  &JitIL::Default}, //"dcba",   OPTYPE_DCACHE, 0, 4}},
+	{758,  &JitIL::DoNothing}, //"dcba",   OPTYPE_DCACHE, 0, 4}},
 	{1014, &JitIL::dcbz}, //"dcbz",   OPTYPE_DCACHE, 0, 4}},
 	//load word
 	{23,  &JitIL::lXzx}, //"lwzx",  OPTYPE_LOAD, FL_OUT_D | FL_IN_A0 | FL_IN_B}},

Source/Core/Core/Src/PowerPC/JitCommon/JitCache.cpp

@@ -126,16 +126,16 @@ bool JitBlock::ContainsAddress(u32 em_address)
 	// This clears the JIT cache. It's called from JitCache.cpp when the JIT cache
 	// is full and when saving and loading states.
 	void JitBlockCache::Clear()
-	{		
+	{
 		Core::DisplayMessage("Clearing code cache.", 3000);
 		for (int i = 0; i < num_blocks; i++)
 		{
 			DestroyBlock(i, false);
 		}
 		links_to.clear();
-		block_map.clear();
 		num_blocks = 0;
 		memset(blockCodePointers, 0, sizeof(u8*)*MAX_NUM_BLOCKS);
+		ClearSafe();
 	}
 
 	void JitBlockCache::ClearSafe()
@@ -207,8 +207,9 @@ bool JitBlock::ContainsAddress(u32 em_address)
 		blockCodePointers[block_num] = code_ptr;
 		JitBlock &b = blocks[block_num];
 		b.originalFirstOpcode = Memory::Read_Opcode_JIT(b.originalAddress);
+		if ((b.originalAddress + b.originalSize) > code_high)
+			code_high = b.originalAddress + b.originalSize;
 		Memory::Write_Opcode_JIT(b.originalAddress, (JIT_OPCODE << 26) | block_num);
-		block_map[std::make_pair(b.originalAddress + 4 * b.originalSize - 1, b.originalAddress)] = block_num;
 		if (block_link)
 		{
 			for (int i = 0; i < 2; i++)
@@ -355,79 +356,44 @@ bool JitBlock::ContainsAddress(u32 em_address)
 
 	void JitBlockCache::DestroyBlock(int block_num, bool invalidate)
 	{
-		if (block_num < 0 || block_num >= num_blocks)
-		{
-			PanicAlert("DestroyBlock: Invalid block number %d", block_num);
-			return;
-		}
 		JitBlock &b = blocks[block_num];
 		if (b.invalid)
 		{
-			if (invalidate)
-				PanicAlert("Invalidating invalid block %d", block_num);
 			return;
 		}
 		b.invalid = true;
 #ifdef JIT_UNLIMITED_ICACHE
-		Memory::Write_Opcode_JIT(b.originalAddress, b.originalFirstOpcode?b.originalFirstOpcode:JIT_ICACHE_INVALID_WORD);
+		Memory::Write_Opcode_JIT(b.originalAddress, JIT_ICACHE_INVALID_WORD);
 #else
 		if (Memory::ReadFast32(b.originalAddress) == block_num)
 			Memory::WriteUnchecked_U32(b.originalFirstOpcode, b.originalAddress);
 #endif
-
-		// We don't unlink blocks, we just send anyone who tries to run them back to the dispatcher.
-		// Not entirely ideal, but .. pretty good.		
-		// Spurious entrances from previously linked blocks can only come through checkedEntry
-		XEmitter emit((u8 *)b.checkedEntry);
-		emit.MOV(32, M(&PC), Imm32(b.originalAddress));
-		emit.JMP(jit->GetAsmRoutines()->dispatcher, true);
-		// this is not needed really
-		/*
-		emit.SetCodePtr((u8 *)blockCodePointers[blocknum]);
-		emit.MOV(32, M(&PC), Imm32(b.originalAddress));
-		emit.JMP(asm_routines.dispatcher, true);
-		*/
 	}
 
 
 	void JitBlockCache::InvalidateICache(u32 address)
-	{			
+	{
 		address &= ~0x1f;
-		// destroy JIT blocks
-		// !! this works correctly under assumption that any two overlapping blocks end at the same address
-		std::map<pair<u32,u32>, u32>::iterator it1 = block_map.lower_bound(std::make_pair(address, 0)), it2 = it1, it;		
-		while (it2 != block_map.end() && it2->first.second < address + 0x20)
-		{
-			DestroyBlock(it2->second, true);
-			it2++;
-		}
-		if (it1 != it2)
-		{
-			block_map.erase(it1, it2);
-		}
-
 #ifdef JIT_UNLIMITED_ICACHE
-		// invalidate iCache.
-		// icbi can be called with any address, so we should check
-		if ((address & ~JIT_ICACHE_MASK) != 0x80000000 && (address & ~JIT_ICACHE_MASK) != 0x00000000 &&
-			(address & ~JIT_ICACHE_MASK) != 0x7e000000 && // TLB area
-			(address & ~JIT_ICACHEEX_MASK) != 0x90000000 && (address & ~JIT_ICACHEEX_MASK) != 0x10000000)
-		{
-			return;
-		}
 		if (address & JIT_ICACHE_VMEM_BIT)
 		{
 			u32 cacheaddr = address & JIT_ICACHE_MASK;
+			if (cacheaddr > (code_high & JIT_ICACHE_MASK))
+				return;
 			memset(iCacheVMEM + cacheaddr, JIT_ICACHE_INVALID_BYTE, 32);
 		}
 		else if (address & JIT_ICACHE_EXRAM_BIT)
 		{
 			u32 cacheaddr = address & JIT_ICACHEEX_MASK;
+			if (cacheaddr > (code_high & JIT_ICACHEEX_MASK))
+				return;
 			memset(iCacheEx + cacheaddr, JIT_ICACHE_INVALID_BYTE, 32);
 		}
 		else
 		{
 			u32 cacheaddr = address & JIT_ICACHE_MASK;
+			if (cacheaddr > (code_high & JIT_ICACHE_MASK))
+				return;
 			memset(iCache + cacheaddr, JIT_ICACHE_INVALID_BYTE, 32);
 		}
 #endif

Source/Core/Core/Src/PowerPC/JitCommon/JitCache.h

@@ -76,7 +76,6 @@ class JitBlockCache
 	JitBlock *blocks;
 	int num_blocks;
 	std::multimap<u32, int> links_to;
-	std::map<std::pair<u32,u32>, u32> block_map; // (end_addr, start_addr) -> number
 #ifdef JIT_UNLIMITED_ICACHE
 	u8 *iCache;
 	u8 *iCacheEx;
@@ -105,6 +104,7 @@ public:
 	void Reset();
 
 	bool IsFull() const;
+	u32 code_high;
 
 	// Code Cache
 	JitBlock *GetBlock(int block_num);

Source/Core/Core/Src/PowerPC/PPCCache.cpp

@@ -82,8 +82,6 @@ namespace PowerPC
 
 	void InstructionCache::Invalidate(u32 addr)
 	{
-		if (jit)
-			jit->GetBlockCache()->InvalidateICache(addr);
 		if (!HID0.ICE)
 			return;
 		// invalidates the whole set
@@ -101,6 +99,14 @@ namespace PowerPC
 			}
 #endif
 		valid[set] = 0;
+		if (jit)
+			jit->GetBlockCache()->InvalidateICache(addr);
+	}
+
+	void InstructionCache::InvalidateBlock(u32 addr)
+	{
+		if (jit)
+			jit->GetBlockCache()->InvalidateICache(addr);
 	}
 
 	u32 InstructionCache::ReadInstruction(u32 addr)

Source/Core/Core/Src/PowerPC/PPCCache.h

@@ -53,6 +53,7 @@ namespace PowerPC
 		void Reset();
 		u32 ReadInstruction(u32 addr);
 		void Invalidate(u32 addr);
+		void InvalidateBlock(u32 addr);
 	};
 
 }