Merge pull request #2605 from dolphin-emu/revert-2358-pie

Revert "x64: build a Position-Independent Executable (PIE)"

CMakeLists.txt

@@ -11,7 +11,6 @@ option(USE_UPNP "Enables UPnP port mapping support" ON)
 option(DISABLE_WX "Disable wxWidgets (use Qt or CLI interface)" OFF)
 option(ENABLE_QT "Enable Qt (use the experimental Qt interface)" OFF)
 option(ENABLE_PCH "Use PCH to speed up compilation" ON)
-option(ENABLE_PIE "Build a Position-Independent Executable (PIE)" ON)
 option(ENABLE_LTO "Enables Link Time Optimization" OFF)
 option(ENABLE_GENERIC "Enables generic build that should run on any little-endian host" OFF)
 if(APPLE)
@@ -215,11 +214,6 @@ if(UNIX AND NOT APPLE)
 	check_and_add_flag(VISIBILITY_HIDDEN -fvisibility=hidden)
 endif()
 
-if(ENABLE_PIE)
-	add_definitions(-fPIE)
-	set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -pie")
-endif()
-
 if(ENABLE_LTO)
 	check_and_add_flag(LTO -flto)
 	if(CMAKE_CXX_COMPILER_ID STREQUAL GNU)

Source/Core/Common/CodeBlock.h

@@ -28,10 +28,10 @@ public:
 	virtual ~CodeBlock() { if (region) FreeCodeSpace(); }
 
 	// Call this before you generate any code.
-	void AllocCodeSpace(int size, void* hint = nullptr)
+	void AllocCodeSpace(int size)
 	{
 		region_size = size;
-		region = (u8*)AllocateExecutableMemory(region_size, hint);
+		region = (u8*)AllocateExecutableMemory(region_size);
 		T::SetCodePtr(region);
 	}
 

Source/Core/Common/MemoryUtil.cpp

@@ -27,29 +27,71 @@
 #endif
 #endif
 
-void* AllocateExecutableMemory(size_t size, void* map_hint)
+// Valgrind doesn't support MAP_32BIT.
+// Uncomment the following line to be able to run Dolphin in Valgrind.
+//#undef MAP_32BIT
+
+#if !defined(_WIN32) && defined(_M_X86_64) && !defined(MAP_32BIT)
+#include <unistd.h>
+#define PAGE_MASK     (getpagesize() - 1)
+#define round_page(x) ((((unsigned long)(x)) + PAGE_MASK) & ~(PAGE_MASK))
+#endif
+
+// This is purposely not a full wrapper for virtualalloc/mmap, but it
+// provides exactly the primitive operations that Dolphin needs.
+
+void* AllocateExecutableMemory(size_t size, bool low)
 {
 #if defined(_WIN32)
 	void* ptr = VirtualAlloc(0, size, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
 #else
+	static char *map_hint = nullptr;
+#if defined(_M_X86_64) && !defined(MAP_32BIT)
+	// This OS has no flag to enforce allocation below the 4 GB boundary,
+	// but if we hint that we want a low address it is very likely we will
+	// get one.
+	// An older version of this code used MAP_FIXED, but that has the side
+	// effect of discarding already mapped pages that happen to be in the
+	// requested virtual memory range (such as the emulated RAM, sometimes).
+	if (low && (!map_hint))
+		map_hint = (char*)round_page(512*1024*1024); /* 0.5 GB rounded up to the next page */
+#endif
 	void* ptr = mmap(map_hint, size, PROT_READ | PROT_WRITE | PROT_EXEC,
-	                 MAP_ANON | MAP_PRIVATE, -1, 0);
+		MAP_ANON | MAP_PRIVATE
+#if defined(_M_X86_64) && defined(MAP_32BIT)
+		| (low ? MAP_32BIT : 0)
+#endif
+		, -1, 0);
 #endif /* defined(_WIN32) */
 
+	// printf("Mapped executable memory at %p (size %ld)\n", ptr,
+	//	(unsigned long)size);
+
 #ifdef _WIN32
 	if (ptr == nullptr)
+	{
 #else
 	if (ptr == MAP_FAILED)
-#endif
 	{
 		ptr = nullptr;
-		PanicAlert("Failed to allocate executable memory.");
+#endif
+		PanicAlert("Failed to allocate executable memory. If you are running Dolphin in Valgrind, try '#undef MAP_32BIT'.");
 	}
+#if !defined(_WIN32) && defined(_M_X86_64) && !defined(MAP_32BIT)
+	else
+	{
+		if (low)
+		{
+			map_hint += size;
+			map_hint = (char*)round_page(map_hint); /* round up to the next page */
+			// printf("Next map will (hopefully) be at %p\n", map_hint);
+		}
+	}
+#endif
 
-#ifdef _X86_64
+#if _M_X86_64
-	ptrdiff_t ofs = (u8*)ptr - (u8*)map_hint;
+	if ((u64)ptr >= 0x80000000 && low == true)
-	if (ofs < -0x80000000ll || ofs + size > 0x80000000ll)
+		PanicAlert("Executable memory ended up above 2GB!");
-		PanicAlert("Executable range can't be used for RIP-relative addressing.");
 #endif
 
 	return ptr;
@@ -75,12 +117,18 @@ void* AllocateMemoryPages(size_t size)
 
 void* AllocateAlignedMemory(size_t size, size_t alignment)
 {
-	void* ptr = nullptr;
 #ifdef _WIN32
-	if (!(ptr = _aligned_malloc(size, alignment)))
+	void* ptr = _aligned_malloc(size, alignment);
 #else
-	if (posix_memalign(&ptr, alignment, size))
+	void* ptr = nullptr;
+	if (posix_memalign(&ptr, alignment, size) != 0)
+		ERROR_LOG(MEMMAP, "Failed to allocate aligned memory");
 #endif
+
+	// printf("Mapped memory at %p (size %ld)\n", ptr,
+	//	(unsigned long)size);
+
+	if (ptr == nullptr)
 		PanicAlert("Failed to allocate aligned memory");
 
 	return ptr;
@@ -88,13 +136,24 @@ void* AllocateAlignedMemory(size_t size, size_t alignment)
 
 void FreeMemoryPages(void* ptr, size_t size)
 {
+	if (ptr)
+	{
+		bool error_occurred = false;
+
 #ifdef _WIN32
-	if (ptr && !VirtualFree(ptr, 0, MEM_RELEASE))
+		if (!VirtualFree(ptr, 0, MEM_RELEASE))
+			error_occurred = true;
 #else
-	if (ptr && munmap(ptr, size))
+		int retval = munmap(ptr, size);
+
+		if (retval != 0)
+			error_occurred = true;
 #endif
+
+		if (error_occurred)
 			PanicAlert("FreeMemoryPages failed!\n%s", GetLastErrorMsg().c_str());
 	}
+}
 
 void FreeAlignedMemory(void* ptr)
 {
@@ -110,34 +169,58 @@ void FreeAlignedMemory(void* ptr)
 
 void ReadProtectMemory(void* ptr, size_t size)
 {
+	bool error_occurred = false;
+
 #ifdef _WIN32
 	DWORD oldValue;
 	if (!VirtualProtect(ptr, size, PAGE_NOACCESS, &oldValue))
+		error_occurred = true;
 #else
-	if (mprotect(ptr, size, PROT_NONE))
+	int retval = mprotect(ptr, size, PROT_NONE);
+
+	if (retval != 0)
+		error_occurred = true;
 #endif
+
+	if (error_occurred)
 		PanicAlert("ReadProtectMemory failed!\n%s", GetLastErrorMsg().c_str());
 }
 
 void WriteProtectMemory(void* ptr, size_t size, bool allowExecute)
 {
+	bool error_occurred = false;
+
 #ifdef _WIN32
 	DWORD oldValue;
 	if (!VirtualProtect(ptr, size, allowExecute ? PAGE_EXECUTE_READ : PAGE_READONLY, &oldValue))
+		error_occurred = true;
 #else
-	if (mprotect(ptr, size, PROT_READ | (allowExecute ? PROT_EXEC : 0)))
+	int retval = mprotect(ptr, size, allowExecute ? (PROT_READ | PROT_EXEC) : PROT_READ);
+
+	if (retval != 0)
+		error_occurred = true;
 #endif
+
+	if (error_occurred)
 		PanicAlert("WriteProtectMemory failed!\n%s", GetLastErrorMsg().c_str());
 }
 
 void UnWriteProtectMemory(void* ptr, size_t size, bool allowExecute)
 {
+	bool error_occurred = false;
+
 #ifdef _WIN32
 	DWORD oldValue;
 	if (!VirtualProtect(ptr, size, allowExecute ? PAGE_EXECUTE_READWRITE : PAGE_READWRITE, &oldValue))
+		error_occurred = true;
 #else
-	if (mprotect(ptr, size, PROT_READ | PROT_WRITE | (allowExecute ? PROT_EXEC : 0)))
+	int retval = mprotect(ptr, size, allowExecute ? (PROT_READ | PROT_WRITE | PROT_EXEC) : PROT_WRITE | PROT_READ);
+
+	if (retval != 0)
+		error_occurred = true;
 #endif
+
+	if (error_occurred)
 		PanicAlert("UnWriteProtectMemory failed!\n%s", GetLastErrorMsg().c_str());
 }
 
@@ -153,8 +236,7 @@ std::string MemUsage()
 	// Print information about the memory usage of the process.
 
 	hProcess = OpenProcess(PROCESS_QUERY_INFORMATION | PROCESS_VM_READ, FALSE, processID);
-	if (nullptr == hProcess)
+	if (nullptr == hProcess) return "MemUsage Error";
-		return "MemUsage Error";
 
 	if (GetProcessMemoryInfo(hProcess, &pmc, sizeof(pmc)))
 		Ret = StringFromFormat("%s K", ThousandSeparate(pmc.WorkingSetSize / 1024, 7).c_str());

Source/Core/Common/MemoryUtil.h

@@ -7,7 +7,7 @@
 #include <cstddef>
 #include <string>
 
-void* AllocateExecutableMemory(size_t size, void* map_hint);
+void* AllocateExecutableMemory(size_t size, bool low = true);
 void* AllocateMemoryPages(size_t size);
 void FreeMemoryPages(void* ptr, size_t size);
 void* AllocateAlignedMemory(size_t size,size_t alignment);

Source/Core/Common/x64Emitter.h

@@ -218,17 +218,17 @@ inline OpArg M(const T* ptr)    {return OpArg((u64)(const void*)ptr, (int)SCALE_
 inline OpArg R(X64Reg value)    {return OpArg(0, SCALE_NONE, value);}
 inline OpArg MatR(X64Reg value) {return OpArg(0, SCALE_ATREG, value);}
 
-inline OpArg MDisp(X64Reg value, ptrdiff_t offset)
+inline OpArg MDisp(X64Reg value, int offset)
 {
-	return OpArg(offset, SCALE_ATREG, value);
+	return OpArg((u32)offset, SCALE_ATREG, value);
 }
 
-inline OpArg MComplex(X64Reg base, X64Reg scaled, int scale, ptrdiff_t offset)
+inline OpArg MComplex(X64Reg base, X64Reg scaled, int scale, int offset)
 {
 	return OpArg(offset, scale, base, scaled);
 }
 
-inline OpArg MScaled(X64Reg scaled, int scale, ptrdiff_t offset)
+inline OpArg MScaled(X64Reg scaled, int scale, int offset)
 {
 	if (scale == SCALE_1)
 		return OpArg(offset, SCALE_ATREG, scaled);
@@ -247,10 +247,17 @@ inline OpArg Imm32(u32 imm) {return OpArg(imm, SCALE_IMM32);}
 inline OpArg Imm64(u64 imm) {return OpArg(imm, SCALE_IMM64);}
 inline OpArg ImmPtr(const void* imm) {return Imm64((u64)imm);}
 
-inline bool FitsInS32(const ptrdiff_t distance)
+inline u32 PtrOffset(const void* ptr, const void* base)
 {
-	return distance <   0x80000000LL &&
+	s64 distance = (s64)ptr-(s64)base;
-	       distance >= -0x80000000LL;
+	if (distance >= 0x80000000LL ||
+	    distance < -0x80000000LL)
+	{
+		_assert_msg_(DYNA_REC, 0, "pointer offset out of range");
+		return 0;
+	}
+
+	return (u32)distance;
 }
 
 //usage: int a[]; ARRAY_OFFSET(a,10)

Source/Core/Core/DSP/Jit/DSPJitUtil.cpp

@@ -20,16 +20,14 @@ void DSPEmitter::dsp_reg_stack_push(int stack_reg)
 	AND(8, R(AL), Imm8(DSP_STACK_MASK));
 	MOV(8, M(&g_dsp.reg_stack_ptr[stack_reg]), R(AL));
 
-	X64Reg tmp1, tmp2;
+	X64Reg tmp1;
 	gpr.getFreeXReg(tmp1);
-	gpr.getFreeXReg(tmp2);
 	//g_dsp.reg_stack[stack_reg][g_dsp.reg_stack_ptr[stack_reg]] = g_dsp.r[DSP_REG_ST0 + stack_reg];
 	MOV(16, R(tmp1), M(&g_dsp.r.st[stack_reg]));
 	MOVZX(64, 8, RAX, R(AL));
-	MOV(64, R(tmp2), ImmPtr(g_dsp.reg_stack[stack_reg]));
+	MOV(16, MComplex(EAX, EAX, 1,
-	MOV(16, MComplex(tmp2, EAX, SCALE_2, 0), R(tmp1));
+			 PtrOffset(&g_dsp.reg_stack[stack_reg][0],nullptr)), R(tmp1));
 	gpr.putXReg(tmp1);
-	gpr.putXReg(tmp2);
 }
 
 //clobbers:
@@ -39,15 +37,13 @@ void DSPEmitter::dsp_reg_stack_pop(int stack_reg)
 {
 	//g_dsp.r[DSP_REG_ST0 + stack_reg] = g_dsp.reg_stack[stack_reg][g_dsp.reg_stack_ptr[stack_reg]];
 	MOV(8, R(AL), M(&g_dsp.reg_stack_ptr[stack_reg]));
-	X64Reg tmp1, tmp2;
+	X64Reg tmp1;
 	gpr.getFreeXReg(tmp1);
-	gpr.getFreeXReg(tmp2);
 	MOVZX(64, 8, RAX, R(AL));
-	MOV(64, R(tmp2), ImmPtr(g_dsp.reg_stack[stack_reg]));
+	MOV(16, R(tmp1), MComplex(EAX, EAX, 1,
-	MOV(16, R(tmp1), MComplex(tmp2, EAX, SCALE_2, 0));
+				  PtrOffset(&g_dsp.reg_stack[stack_reg][0],nullptr)));
 	MOV(16, M(&g_dsp.r.st[stack_reg]), R(tmp1));
 	gpr.putXReg(tmp1);
-	gpr.putXReg(tmp2);
 
 	//g_dsp.reg_stack_ptr[stack_reg]--;
 	//g_dsp.reg_stack_ptr[stack_reg] &= DSP_STACK_MASK;

Source/Core/Core/PowerPC/Jit64/JitAsm.cpp

@@ -40,7 +40,7 @@ void Jit64AsmRoutineManager::Generate()
 
 	// Two statically allocated registers.
 	//MOV(64, R(RMEM), Imm64((u64)Memory::physical_base));
-	MOV(64, R(RPPCSTATE), ImmPtr(PPCSTATE_BASE));
+	MOV(64, R(RPPCSTATE), Imm64((u64)&PowerPC::ppcState + 0x80));
 
 	const u8* outerLoop = GetCodePtr();
 		ABI_PushRegistersAndAdjustStack({}, 0);
@@ -103,9 +103,9 @@ void Jit64AsmRoutineManager::Generate()
 			// optimizations safe, because IR and DR are usually set/cleared together.
 			// TODO: Branching based on the 20 most significant bits of instruction
 			// addresses without translating them is wrong.
-			u8* icache = jit->GetBlockCache()->iCache.data();
+			u64 icache = (u64)jit->GetBlockCache()->iCache.data();
-			u8* icacheVmem = jit->GetBlockCache()->iCacheVMEM.data();
+			u64 icacheVmem = (u64)jit->GetBlockCache()->iCacheVMEM.data();
-			u8* icacheEx = jit->GetBlockCache()->iCacheEx.data();
+			u64 icacheEx = (u64)jit->GetBlockCache()->iCacheEx.data();
 			u32 mask = 0;
 			FixupBranch no_mem;
 			FixupBranch exit_mem;
@@ -117,13 +117,13 @@ void Jit64AsmRoutineManager::Generate()
 			no_mem = J_CC(CC_NZ);
 			AND(32, R(RSCRATCH), Imm32(JIT_ICACHE_MASK));
 
-			if (FitsInS32(PPCSTATE_OFS(icache)))
+			if (icache <= INT_MAX)
 			{
-				MOV(32, R(RSCRATCH), MPIC(icache, RSCRATCH));
+				MOV(32, R(RSCRATCH), MDisp(RSCRATCH, (s32)icache));
 			}
 			else
 			{
-				MOV(64, R(RSCRATCH2), ImmPtr(icache));
+				MOV(64, R(RSCRATCH2), Imm64(icache));
 				MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH));
 			}
 
@@ -132,14 +132,13 @@ void Jit64AsmRoutineManager::Generate()
 			TEST(32, R(RSCRATCH), Imm32(JIT_ICACHE_VMEM_BIT));
 			FixupBranch no_vmem = J_CC(CC_Z);
 			AND(32, R(RSCRATCH), Imm32(JIT_ICACHE_MASK));
-
+			if (icacheVmem <= INT_MAX)
-			if (FitsInS32(PPCSTATE_OFS(icacheVmem)))
 			{
-				MOV(32, R(RSCRATCH), MPIC(icacheVmem, RSCRATCH));
+				MOV(32, R(RSCRATCH), MDisp(RSCRATCH, (s32)icacheVmem));
 			}
 			else
 			{
-				MOV(64, R(RSCRATCH2), ImmPtr(icacheVmem));
+				MOV(64, R(RSCRATCH2), Imm64(icacheVmem));
 				MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH));
 			}
 
@@ -150,13 +149,14 @@ void Jit64AsmRoutineManager::Generate()
 				TEST(32, R(RSCRATCH), Imm32(JIT_ICACHE_EXRAM_BIT));
 				FixupBranch no_exram = J_CC(CC_Z);
 				AND(32, R(RSCRATCH), Imm32(JIT_ICACHEEX_MASK));
-				if (FitsInS32(PPCSTATE_OFS(icacheEx)))
+
+				if (icacheEx <= INT_MAX)
 				{
-					MOV(32, R(RSCRATCH), MPIC(icacheEx, RSCRATCH));
+					MOV(32, R(RSCRATCH), MDisp(RSCRATCH, (s32)icacheEx));
 				}
 				else
 				{
-					MOV(64, R(RSCRATCH2), ImmPtr(icacheEx));
+					MOV(64, R(RSCRATCH2), Imm64(icacheEx));
 					MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH));
 				}
 
@@ -169,17 +169,16 @@ void Jit64AsmRoutineManager::Generate()
 			TEST(32, R(RSCRATCH), R(RSCRATCH));
 			FixupBranch notfound = J_CC(CC_L);
 			//grab from list and jump to it
-			const u8** codePointers = jit->GetBlockCache()->GetCodePointers();
+			u64 codePointers = (u64)jit->GetBlockCache()->GetCodePointers();
-			if (FitsInS32(PPCSTATE_OFS(codePointers)))
+			if (codePointers <= INT_MAX)
 			{
-				JMPptr(MPIC(codePointers, RSCRATCH, SCALE_8));
+				JMPptr(MScaled(RSCRATCH, SCALE_8, (s32)codePointers));
 			}
 			else
 			{
-				MOV(64, R(RSCRATCH2), ImmPtr(codePointers));
+				MOV(64, R(RSCRATCH2), Imm64(codePointers));
 				JMPptr(MComplex(RSCRATCH2, RSCRATCH, SCALE_8, 0));
 			}
-
 			SetJumpTarget(notfound);
 
 			//Ok, no block, let's jit
@@ -272,7 +271,7 @@ void Jit64AsmRoutineManager::GenerateCommon()
 	CMP(32, R(ABI_PARAM2), Imm32(0xCC008000));
 	FixupBranch skip_fast_write = J_CC(CC_NE, false);
 	MOV(32, RSCRATCH, M(&m_gatherPipeCount));
-	MOV(8, MPIC(&m_gatherPipe, RSCRATCH), ABI_PARAM1);
+	MOV(8, MDisp(RSCRATCH, (u32)&m_gatherPipe), ABI_PARAM1);
 	ADD(32, 1, M(&m_gatherPipeCount));
 	RET();
 	SetJumpTarget(skip_fast_write);

Source/Core/Core/PowerPC/Jit64/JitAsm.h

@@ -34,7 +34,7 @@ public:
 		m_stack_top = stack_top;
 		// NOTE: When making large additions to the AsmCommon code, you might
 		// want to ensure this number is big enough.
-		AllocCodeSpace(16384, PPCSTATE_BASE);
+		AllocCodeSpace(16384);
 		Generate();
 		WriteProtect();
 	}

Source/Core/Core/PowerPC/Jit64/Jit_Integer.cpp

@@ -45,7 +45,7 @@ void Jit64::GenerateOverflow()
 	//rare).
 	static const u8 ovtable[4] = {0, 0, XER_SO_MASK, XER_SO_MASK};
 	MOVZX(32, 8, RSCRATCH, PPCSTATE(xer_so_ov));
-	MOV(8, R(RSCRATCH), MPIC(ovtable, RSCRATCH));
+	MOV(8, R(RSCRATCH), MDisp(RSCRATCH, (u32)(u64)ovtable));
 	MOV(8, PPCSTATE(xer_so_ov), R(RSCRATCH));
 	SetJumpTarget(exit);
 }

Source/Core/Core/PowerPC/Jit64/Jit_LoadStorePaired.cpp

@@ -132,13 +132,13 @@ void Jit64::psq_stXX(UGeckoInstruction inst)
 	{
 		// One value
 		CVTSD2SS(XMM0, fpr.R(s));
-		CALLptr(MPIC(asm_routines.singleStoreQuantized, RSCRATCH, SCALE_8));
+		CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)asm_routines.singleStoreQuantized));
 	}
 	else
 	{
 		// Pair of values
 		CVTPD2PS(XMM0, fpr.R(s));
-		CALLptr(MPIC(asm_routines.pairedStoreQuantized, RSCRATCH, SCALE_8));
+		CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)asm_routines.pairedStoreQuantized));
 	}
 
 	if (update && jo.memcheck)
@@ -306,7 +306,7 @@ void Jit64::psq_lXX(UGeckoInstruction inst)
 	AND(32, R(RSCRATCH2), gqr);
 	MOVZX(32, 8, RSCRATCH, R(RSCRATCH2));
 
-	CALLptr(MPIC(&asm_routines.pairedLoadQuantized[w * 8], RSCRATCH, SCALE_8));
+	CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)(&asm_routines.pairedLoadQuantized[w * 8])));
 
 	MemoryExceptionCheck();
 	CVTPS2PD(fpr.RX(s), R(XMM0));

Source/Core/Core/PowerPC/Jit64/Jit_SystemRegisters.cpp

@@ -458,7 +458,7 @@ void Jit64::mtcrf(UGeckoInstruction inst)
 						SHR(32, R(RSCRATCH), Imm8(28 - (i * 4)));
 					if (i != 0)
 						AND(32, R(RSCRATCH), Imm8(0xF));
-					MOV(64, R(RSCRATCH), MPIC(m_crTable, RSCRATCH, SCALE_8));
+					MOV(64, R(RSCRATCH), MScaled(RSCRATCH, SCALE_8, (u32)(u64)m_crTable));
 					MOV(64, PPCSTATE(cr_val[i]), R(RSCRATCH));
 				}
 			}
@@ -493,7 +493,7 @@ void Jit64::mcrxr(UGeckoInstruction inst)
 	// [SO OV CA 0] << 3
 	SHL(32, R(RSCRATCH), Imm8(4));
 
-	MOV(64, R(RSCRATCH), MPIC(m_crTable, RSCRATCH));
+	MOV(64, R(RSCRATCH), MDisp(RSCRATCH, (u32)(u64)m_crTable));
 	MOV(64, PPCSTATE(cr_val[inst.CRFD]), R(RSCRATCH));
 
 	// Clear XER[0-3]

Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.cpp

@@ -24,8 +24,10 @@ void CommonAsmRoutines::GenFifoWrite(int size)
 	const void* start = GetCodePtr();
 
 	// Assume value in RSCRATCH
+	u32 gather_pipe = (u32)(u64)GPFifo::m_gatherPipe;
+	_assert_msg_(DYNA_REC, gather_pipe <= 0x7FFFFFFF, "Gather pipe not in low 2GB of memory!");
 	MOV(32, R(RSCRATCH2), M(&GPFifo::m_gatherPipeCount));
-	SwapAndStore(size, MPIC(GPFifo::m_gatherPipe, RSCRATCH2), RSCRATCH);
+	SwapAndStore(size, MDisp(RSCRATCH2, gather_pipe), RSCRATCH);
 	ADD(32, R(RSCRATCH2), Imm8(size >> 3));
 	MOV(32, M(&GPFifo::m_gatherPipeCount), R(RSCRATCH2));
 	RET();
@@ -66,8 +68,8 @@ void CommonAsmRoutines::GenFrsqrte()
 
 	SHR(64, R(RSCRATCH), Imm8(37));
 	AND(32, R(RSCRATCH), Imm32(0x7FF));
-	IMUL(32, RSCRATCH, MPIC(MathUtil::frsqrte_expected_dec, RSCRATCH_EXTRA, SCALE_4));
+	IMUL(32, RSCRATCH, MScaled(RSCRATCH_EXTRA, SCALE_4, (u32)(u64)MathUtil::frsqrte_expected_dec));
-	MOV(32, R(RSCRATCH_EXTRA), MPIC(MathUtil::frsqrte_expected_base, RSCRATCH_EXTRA, SCALE_4));
+	MOV(32, R(RSCRATCH_EXTRA), MScaled(RSCRATCH_EXTRA, SCALE_4, (u32)(u64)MathUtil::frsqrte_expected_base));
 	SUB(32, R(RSCRATCH_EXTRA), R(RSCRATCH));
 	SHL(64, R(RSCRATCH_EXTRA), Imm8(26));
 	OR(64, R(RSCRATCH2), R(RSCRATCH_EXTRA));     // vali |= (s64)(frsqrte_expected_base[index] - frsqrte_expected_dec[index] * (i % 2048)) << 26;
@@ -134,11 +136,11 @@ void CommonAsmRoutines::GenFres()
 	AND(32, R(RSCRATCH), Imm32(0x3FF)); // i % 1024
 	AND(32, R(RSCRATCH2), Imm8(0x1F));   // i / 1024
 
-	IMUL(32, RSCRATCH, MPIC(MathUtil::fres_expected_dec, RSCRATCH2, SCALE_4));
+	IMUL(32, RSCRATCH, MScaled(RSCRATCH2, SCALE_4, (u32)(u64)MathUtil::fres_expected_dec));
 	ADD(32, R(RSCRATCH), Imm8(1));
 	SHR(32, R(RSCRATCH), Imm8(1));
 
-	MOV(32, R(RSCRATCH2), MPIC(MathUtil::fres_expected_base, RSCRATCH2, SCALE_4));
+	MOV(32, R(RSCRATCH2), MScaled(RSCRATCH2, SCALE_4, (u32)(u64)MathUtil::fres_expected_base));
 	SUB(32, R(RSCRATCH2), R(RSCRATCH));
 	SHL(64, R(RSCRATCH2), Imm8(29));
 	OR(64, R(RSCRATCH2), R(RSCRATCH_EXTRA));     // vali |= (s64)(fres_expected_base[i / 1024] - (fres_expected_dec[i / 1024] * (i % 1024) + 1) / 2) << 29
@@ -197,7 +199,7 @@ void CommonAsmRoutines::GenMfcr()
 		// SO: Bit 61 set; set flag bit 0
 		// LT: Bit 62 set; set flag bit 3
 		SHR(64, R(cr_val), Imm8(61));
-		OR(32, R(dst), MPIC(m_flagTable, cr_val, SCALE_4));
+		OR(32, R(dst), MScaled(cr_val, SCALE_4, (u32)(u64)m_flagTable));
 	}
 	RET();
 
@@ -245,7 +247,7 @@ void CommonAsmRoutines::GenQuantizedStores()
 
 	const u8* storePairedU8 = AlignCode4();
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MOVQ_xmm(XMM1, MPIC(m_quantizeTableS, RSCRATCH2));
+	MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS));
 	MULPS(XMM0, R(XMM1));
 #ifdef QUANTIZE_OVERFLOW_SAFE
 	MINPS(XMM0, M(m_65535));
@@ -260,7 +262,7 @@ void CommonAsmRoutines::GenQuantizedStores()
 
 	const u8* storePairedS8 = AlignCode4();
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MOVQ_xmm(XMM1, MPIC(m_quantizeTableS, RSCRATCH2));
+	MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS));
 	MULPS(XMM0, R(XMM1));
 #ifdef QUANTIZE_OVERFLOW_SAFE
 	MINPS(XMM0, M(m_65535));
@@ -276,7 +278,7 @@ void CommonAsmRoutines::GenQuantizedStores()
 
 	const u8* storePairedU16 = AlignCode4();
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MOVQ_xmm(XMM1, MPIC(m_quantizeTableS, RSCRATCH2));
+	MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS));
 	MULPS(XMM0, R(XMM1));
 
 	if (cpu_info.bSSE4_1)
@@ -308,7 +310,7 @@ void CommonAsmRoutines::GenQuantizedStores()
 
 	const u8* storePairedS16 = AlignCode4();
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MOVQ_xmm(XMM1, MPIC(m_quantizeTableS, RSCRATCH2));
+	MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS));
 	MULPS(XMM0, R(XMM1));
 #ifdef QUANTIZE_OVERFLOW_SAFE
 	MINPS(XMM0, M(m_65535));
@@ -353,7 +355,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores()
 
 	const u8* storeSingleU8 = AlignCode4();  // Used by MKWii
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MULSS(XMM0, MPIC(m_quantizeTableS, RSCRATCH2));
+	MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS));
 	XORPS(XMM1, R(XMM1));
 	MAXSS(XMM0, R(XMM1));
 	MINSS(XMM0, M(&m_255));
@@ -363,7 +365,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores()
 
 	const u8* storeSingleS8 = AlignCode4();
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MULSS(XMM0, MPIC(m_quantizeTableS, RSCRATCH2));
+	MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS));
 	MAXSS(XMM0, M(&m_m128));
 	MINSS(XMM0, M(&m_127));
 	CVTTSS2SI(RSCRATCH, R(XMM0));
@@ -372,7 +374,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores()
 
 	const u8* storeSingleU16 = AlignCode4();  // Used by MKWii
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MULSS(XMM0, MPIC(m_quantizeTableS, RSCRATCH2));
+	MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS));
 	XORPS(XMM1, R(XMM1));
 	MAXSS(XMM0, R(XMM1));
 	MINSS(XMM0, M(m_65535));
@@ -382,7 +384,7 @@ void CommonAsmRoutines::GenQuantizedSingleStores()
 
 	const u8* storeSingleS16 = AlignCode4();
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MULSS(XMM0, MPIC(m_quantizeTableS, RSCRATCH2));
+	MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_quantizeTableS));
 	MAXSS(XMM0, M(&m_m32768));
 	MINSS(XMM0, M(&m_32767));
 	CVTTSS2SI(RSCRATCH, R(XMM0));
@@ -482,7 +484,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
 	}
 	CVTDQ2PS(XMM0, R(XMM0));
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MOVQ_xmm(XMM1, MPIC(m_dequantizeTableS, RSCRATCH2));
+	MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS));
 	MULPS(XMM0, R(XMM1));
 	RET();
 
@@ -493,7 +495,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
 		UnsafeLoadRegToRegNoSwap(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 8, 0); // RSCRATCH_EXTRA = 0x000000xx
 	CVTSI2SS(XMM0, R(RSCRATCH_EXTRA));
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MULSS(XMM0, MPIC(m_dequantizeTableS, RSCRATCH2));
+	MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS));
 	UNPCKLPS(XMM0, M(m_one));
 	RET();
 
@@ -521,7 +523,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
 	}
 	CVTDQ2PS(XMM0, R(XMM0));
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MOVQ_xmm(XMM1, MPIC(m_dequantizeTableS, RSCRATCH2));
+	MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS));
 	MULPS(XMM0, R(XMM1));
 	RET();
 
@@ -532,7 +534,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
 		UnsafeLoadRegToRegNoSwap(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 8, 0, true);
 	CVTSI2SS(XMM0, R(RSCRATCH_EXTRA));
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MULSS(XMM0, MPIC(m_dequantizeTableS, RSCRATCH2));
+	MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS));
 	UNPCKLPS(XMM0, M(m_one));
 	RET();
 
@@ -555,7 +557,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
 	}
 	CVTDQ2PS(XMM0, R(XMM0));
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MOVQ_xmm(XMM1, MPIC(m_dequantizeTableS, RSCRATCH2));
+	MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS));
 	MULPS(XMM0, R(XMM1));
 	RET();
 
@@ -566,7 +568,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
 		UnsafeLoadRegToReg(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 16, 0, false);
 	CVTSI2SS(XMM0, R(RSCRATCH_EXTRA));
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MULSS(XMM0, MPIC(m_dequantizeTableS, RSCRATCH2));
+	MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS));
 	UNPCKLPS(XMM0, M(m_one));
 	RET();
 
@@ -588,7 +590,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
 	}
 	CVTDQ2PS(XMM0, R(XMM0));
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MOVQ_xmm(XMM1, MPIC(m_dequantizeTableS, RSCRATCH2));
+	MOVQ_xmm(XMM1, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS));
 	MULPS(XMM0, R(XMM1));
 	RET();
 
@@ -599,7 +601,7 @@ void CommonAsmRoutines::GenQuantizedLoads()
 		UnsafeLoadRegToReg(RSCRATCH_EXTRA, RSCRATCH_EXTRA, 16, 0, true);
 	CVTSI2SS(XMM0, R(RSCRATCH_EXTRA));
 	SHR(32, R(RSCRATCH2), Imm8(5));
-	MULSS(XMM0, MPIC(m_dequantizeTableS, RSCRATCH2));
+	MULSS(XMM0, MDisp(RSCRATCH2, (u32)(u64)m_dequantizeTableS));
 	UNPCKLPS(XMM0, M(m_one));
 	RET();
 

Source/Core/Core/PowerPC/Jit64IL/IR_X86.cpp

@@ -1611,7 +1611,7 @@ static void DoWriteCode(IRBuilder* ibuild, JitIL* Jit, u32 exitAddress)
 			Jit->OR(32, R(RSCRATCH), Imm8(w << 3));
 
 			Jit->MOV(32, R(RSCRATCH_EXTRA), regLocForInst(RI, getOp1(I)));
-			Jit->CALLptr(MPIC(Jit->asm_routines.pairedLoadQuantized, RSCRATCH, SCALE_8));
+			Jit->CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)(Jit->asm_routines.pairedLoadQuantized)));
 			Jit->MOVAPD(reg, R(XMM0));
 			RI.fregs[reg] = I;
 			regNormalRegClear(RI, I);
@@ -1669,7 +1669,7 @@ static void DoWriteCode(IRBuilder* ibuild, JitIL* Jit, u32 exitAddress)
 
 			Jit->MOV(32, R(RSCRATCH_EXTRA), regLocForInst(RI, getOp2(I)));
 			Jit->MOVAPD(XMM0, fregLocForInst(RI, getOp1(I)));
-			Jit->CALLptr(MPIC(Jit->asm_routines.pairedStoreQuantized, RSCRATCH, SCALE_8));
+			Jit->CALLptr(MScaled(RSCRATCH, SCALE_8, (u32)(u64)(Jit->asm_routines.pairedStoreQuantized)));
 			if (RI.IInfo[I - RI.FirstI] & 4)
 				fregClearInst(RI, getOp1(I));
 			if (RI.IInfo[I - RI.FirstI] & 8)

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

@@ -250,7 +250,7 @@ void JitIL::Init()
 	UpdateMemoryOptions();
 
 	trampolines.Init(jo.memcheck ? TRAMPOLINE_CODE_SIZE_MMU : TRAMPOLINE_CODE_SIZE);
-	AllocCodeSpace(CODE_SIZE, PPCSTATE_BASE);
+	AllocCodeSpace(CODE_SIZE);
 	blocks.Init();
 	asm_routines.Init(nullptr);
 

Source/Core/Core/PowerPC/JitCommon/JitBase.h

@@ -43,18 +43,6 @@
 // to address as much as possible in a one-byte offset form.
 #define RPPCSTATE RBP
 
-namespace Gen
-{
-
-inline OpArg MPIC(const void* address, X64Reg scale_reg, int scale = SCALE_1)
-{
-	ptrdiff_t offset = PPCSTATE_OFS(address);
-	_dbg_assert_(DYNA_REC, FitsInS32(offset));
-	return MComplex(RPPCSTATE, scale_reg, scale, offset);
-}
-
-}
-
 // Use these to control the instruction selection
 // #define INSTRUCTION_START FallBackToInterpreter(inst); return;
 // #define INSTRUCTION_START PPCTables::CountInstruction(inst);

Source/Core/Core/PowerPC/JitCommon/Jit_Util.cpp

@@ -173,11 +173,11 @@ private:
 		u32 all_ones = (1ULL << sbits) - 1;
 		if ((all_ones & mask) == all_ones)
 		{
-			MoveOpArgToReg(sbits, MatR(RSCRATCH));
+			MoveOpArgToReg(sbits, MDisp(RSCRATCH, 0));
 		}
 		else
 		{
-			m_code->MOVZX(32, sbits, m_dst_reg, MatR(RSCRATCH));
+			m_code->MOVZX(32, sbits, m_dst_reg, MDisp(RSCRATCH, 0));
 			m_code->AND(32, R(m_dst_reg), Imm32(mask));
 			if (m_sign_extend)
 				m_code->MOVSX(32, sbits, m_dst_reg, R(m_dst_reg));

Source/Core/Core/PowerPC/JitCommon/Jit_Util.h

@@ -9,15 +9,15 @@
 #include "Common/BitSet.h"
 #include "Common/CPUDetect.h"
 #include "Common/x64Emitter.h"
-#include "Core/PowerPC/PowerPC.h"
 
 namespace MMIO { class Mapping; }
 
 // We offset by 0x80 because the range of one byte memory offsets is
 // -0x80..0x7f.
-#define PPCSTATE_BASE ((u8*)&PowerPC::ppcState + 0x80)
+#define PPCSTATE(x) MDisp(RPPCSTATE, \
-#define PPCSTATE_OFS(x) ((u8*)(x) - PPCSTATE_BASE)
+	(int) ((char *) &PowerPC::ppcState.x - (char *) &PowerPC::ppcState) - 0x80)
-#define PPCSTATE(x) MDisp(RPPCSTATE, PPCSTATE_OFS(&PowerPC::ppcState.x))
+// In case you want to disable the ppcstate register:
+// #define PPCSTATE(x) M(&PowerPC::ppcState.x)
 #define PPCSTATE_LR PPCSTATE(spr[SPR_LR])
 #define PPCSTATE_CTR PPCSTATE(spr[SPR_CTR])
 #define PPCSTATE_SRR0 PPCSTATE(spr[SPR_SRR0])
@@ -31,7 +31,7 @@ private:
 	bool m_enabled = false;
 public:
 	bool Enabled() { return m_enabled; }
-	void Init(int size) { AllocCodeSpace(size, PPCSTATE_BASE); m_enabled = true; }
+	void Init(int size) { AllocCodeSpace(size); m_enabled = true; }
 	void Shutdown() { FreeCodeSpace(); m_enabled = false; }
 };
 

Source/Core/Core/PowerPC/JitCommon/TrampolineCache.cpp

@@ -22,7 +22,7 @@ using namespace Gen;
 
 void TrampolineCache::Init(int size)
 {
-	AllocCodeSpace(size, PPCSTATE_BASE);
+	AllocCodeSpace(size);
 }
 
 void TrampolineCache::ClearCodeSpace()

Source/UnitTests/Common/x64EmitterTest.cpp

@@ -19,7 +19,6 @@
 
 #include "Common/CPUDetect.h"
 #include "Common/x64Emitter.h"
-#include "Core/PowerPC/JitCommon/Jit_Util.h"
 
 namespace Gen
 {
@@ -95,7 +94,7 @@ protected:
 		memset(&cpu_info, 0xFF, sizeof (cpu_info));
 
 		emitter.reset(new X64CodeBlock());
-		emitter->AllocCodeSpace(4096, PPCSTATE_BASE);
+		emitter->AllocCodeSpace(4096);
 		code_buffer = emitter->GetWritableCodePtr();
 
 		disasm.reset(new disassembler);