jit: branch instructions

10 files changed, 363 insertions, 187 deletions

diff --git a/src/ARM.cpp b/src/ARM.cpp
index df58ce3..3c2253c 100644
--- a/src/ARM.cpp
+++ b/src/ARM.cpp
@@ -564,11 +564,8 @@ void ARMv5::Execute()
             printf("aaarg ungempappter raum %x\n", R[15]);*/
 
         ARMJIT::CompiledBlock block = ARMJIT::LookUpBlock(0, R[15] - ((CPSR&0x20)?2:4));
-        if (block == NULL)
-            ARMJIT::CompileBlock(this);
-        else
-            Cycles += block();
-
+        Cycles += (block ? block : ARMJIT::CompileBlock(this))();
+        
         // TODO optimize this shit!!!
         if (Halted)
         {
@@ -650,10 +647,7 @@ void ARMv4::Execute()
             printf("aaarg ungempappter raum %x\n", R[15]);*/
 
         ARMJIT::CompiledBlock block = ARMJIT::LookUpBlock(1, R[15] - ((CPSR&0x20)?2:4));
-        if (block == NULL)
-            ARMJIT::CompileBlock(this);
-        else
-            Cycles += block();
+        Cycles += (block ? block : ARMJIT::CompileBlock(this))();
 
         // TODO optimize this shit!!!
         if (Halted)
diff --git a/src/ARMJIT.cpp b/src/ARMJIT.cpp
index 6afa967..47b425f 100644
--- a/src/ARMJIT.cpp
+++ b/src/ARMJIT.cpp
@@ -121,7 +121,7 @@ void DeInit()
 	delete compiler;
 }
 
-void CompileBlock(ARM* cpu)
+CompiledBlock CompileBlock(ARM* cpu)
 {
     bool thumb = cpu->CPSR & 0x20;
 
@@ -171,6 +171,8 @@ void CompileBlock(ARM* cpu)
     CompiledBlock block = compiler->CompileBlock(cpu, instrs, i);
 
     InsertBlock(cpu->Num, r15Initial - (thumb ? 2 : 4), block);
+
+	return block;
 }
 
 void ResetBlocks()
diff --git a/src/ARMJIT.h b/src/ARMJIT.h
index 71188f9..45bb4ed 100644
--- a/src/ARMJIT.h
+++ b/src/ARMJIT.h
@@ -109,7 +109,7 @@ inline void InsertBlock(u32 num, u32 addr, CompiledBlock func)
 void Init();
 void DeInit();
 
-void CompileBlock(ARM* cpu);
+CompiledBlock CompileBlock(ARM* cpu);
 
 void ResetBlocks();
 
diff --git a/src/ARMJIT_x64/ARMJIT_Branch.cpp b/src/ARMJIT_x64/ARMJIT_Branch.cpp
new file mode 100644
index 0000000..fb2acba
--- /dev/null
+++ b/src/ARMJIT_x64/ARMJIT_Branch.cpp
@@ -0,0 +1,267 @@
+#include "ARMJIT_Compiler.h"
+
+using namespace Gen;
+
+namespace ARMJIT
+{
+    
+void Compiler::Comp_JumpTo(u32 addr, bool forceNonConstantCycles)
+{
+    // we can simplify constant branches by a lot
+    // it's not completely safe to assume stuff like, which instructions to preload
+    // we'll see how it works out
+
+    u32 newPC;
+    u32 nextInstr[2];
+    u32 cycles = 0;
+    bool setupRegion = false;
+
+    if (addr & 0x1 && !Thumb)
+    {
+        CPSRDirty = true;
+        OR(32, R(RCPSR), Imm8(0x20));
+    }
+    else if (!(addr & 0x1) && Thumb)
+    {
+        CPSRDirty = true;
+        AND(32, R(RCPSR), Imm32(~0x20));
+    }
+
+    if (Num == 0)
+    {
+        ARMv5* cpu9 = (ARMv5*)CurCPU;
+
+        u32 oldregion = R15 >> 24;
+        u32 newregion = addr >> 24;
+
+        u32 regionCodeCycles = cpu9->MemTimings[addr >> 12][0];
+        cpu9->RegionCodeCycles = regionCodeCycles;
+
+        MOV(32, MDisp(RCPU, offsetof(ARMv5, RegionCodeCycles)), Imm32(regionCodeCycles));
+
+        setupRegion = newregion != oldregion;
+        if (setupRegion)
+            cpu9->SetupCodeMem(addr);
+
+        if (addr & 0x1)
+        {
+            addr &= ~0x1;
+            newPC = addr+2;
+
+            // two-opcodes-at-once fetch
+            // doesn't matter if we put garbage in the MSbs there
+            if (addr & 0x2)
+            {
+                nextInstr[0] = cpu9->CodeRead32(addr-2, true) >> 16;
+                cycles += CurCPU->CodeCycles;
+                nextInstr[1] = cpu9->CodeRead32(addr+2, false);
+                cycles += CurCPU->CodeCycles;
+            }
+            else
+            {
+                nextInstr[0] = cpu9->CodeRead32(addr, true);
+                nextInstr[1] = nextInstr[0] >> 16;
+                cycles += CurCPU->CodeCycles;
+            }
+        }
+        else
+        {
+            addr &= ~0x3;
+            newPC = addr+4;
+
+            nextInstr[0] = cpu9->CodeRead32(addr, true);
+            cycles += cpu9->CodeCycles;
+            nextInstr[1] = cpu9->CodeRead32(addr+4, false);
+            cycles += cpu9->CodeCycles;
+        }
+    }
+    else
+    {
+        ARMv4* cpu7 = (ARMv4*)CurCPU;
+
+        u32 codeRegion = addr >> 24;
+        u32 codeCycles = addr >> 15; // cheato
+
+        cpu7->CodeRegion = codeRegion;
+        cpu7->CodeCycles = codeCycles;
+
+        MOV(32, MDisp(RCPU, offsetof(ARM, CodeRegion)), Imm32(codeRegion));
+        MOV(32, MDisp(RCPU, offsetof(ARM, CodeRegion)), Imm32(codeCycles));
+
+        if (addr & 0x1)
+        {
+            addr &= ~0x1;
+            newPC = addr+2;
+
+            nextInstr[0] = ((ARMv4*)CurCPU)->CodeRead16(addr);
+            nextInstr[1] = ((ARMv4*)CurCPU)->CodeRead16(addr+2);
+            cycles += NDS::ARM7MemTimings[codeCycles][0] + NDS::ARM7MemTimings[codeCycles][1];
+        }
+        else
+        {
+            addr &= ~0x3;
+            newPC = addr+4;
+
+            nextInstr[0] = cpu7->CodeRead32(addr);
+            nextInstr[1] = cpu7->CodeRead32(addr+4);
+            cycles += NDS::ARM7MemTimings[codeCycles][2] + NDS::ARM7MemTimings[codeCycles][3];
+        }
+    }
+
+    MOV(32, MDisp(RCPU, offsetof(ARM, R[15])), Imm32(newPC));
+    MOV(32, MDisp(RCPU, offsetof(ARM, NextInstr[0])), Imm32(nextInstr[0]));
+    MOV(32, MDisp(RCPU, offsetof(ARM, NextInstr[1])), Imm32(nextInstr[1]));
+    if ((Thumb || CurInstr.Cond() >= 0xE) && !forceNonConstantCycles)
+        ConstantCycles += cycles;
+    else
+        ADD(32, MDisp(RCPU, offsetof(ARM, Cycles)), Imm8(cycles));
+
+    if (setupRegion)
+    {
+        MOV(32, R(ABI_PARAM1), R(RCPU));
+        MOV(32, R(ABI_PARAM2), Imm32(newPC));
+        CALL((void*)&ARMv5::SetupCodeMem);
+    }
+}
+
+void Compiler::Comp_JumpTo(Gen::X64Reg addr, bool restoreCPSR)
+{
+    BitSet16 hiRegsLoaded(RegCache.DirtyRegs & 0xFFFF0000);
+    bool previouslyDirty = CPSRDirty;
+    SaveCPSR();
+
+    if (restoreCPSR)
+    {
+        if (Thumb || CurInstr.Cond() >= 0xE)
+        {
+            for (int reg : hiRegsLoaded)
+                RegCache.UnloadRegister(reg);
+        }
+        else
+        {
+            // the ugly way...
+            // we only save them, to load and save them again
+            for (int reg : hiRegsLoaded)
+                SaveReg(reg, RegCache.Mapping[reg]);
+        }
+    }
+
+    MOV(64, R(ABI_PARAM1), R(RCPU));
+    MOV(32, R(ABI_PARAM2), R(addr));
+    if (!restoreCPSR)
+        XOR(32, R(ABI_PARAM3), R(ABI_PARAM3));
+    else
+        MOV(32, R(ABI_PARAM3), Imm32(restoreCPSR));
+    if (Num == 0)
+        CALL((void*)&ARMv5::JumpTo);
+    else
+        CALL((void*)&ARMv4::JumpTo);
+    
+    if (!Thumb && restoreCPSR && CurInstr.Cond() < 0xE)
+    {
+        for (int reg : hiRegsLoaded)
+            LoadReg(reg, RegCache.Mapping[reg]);
+    }
+
+    if (previouslyDirty)
+        LoadCPSR();
+    CPSRDirty = previouslyDirty;
+}
+
+void Compiler::A_Comp_BranchImm()
+{
+    int op = (CurInstr.Instr >> 24) & 1;
+    s32 offset = (s32)(CurInstr.Instr << 8) >> 6;
+    u32 target = R15 + offset;
+    bool link = op;
+
+    if (CurInstr.Cond() == 0xF) // BLX_imm
+    {
+        target += (op << 1) + 1;
+        link = true;
+    }
+
+    if (link)
+        MOV(32, MapReg(14), Imm32(R15 - 4));
+
+    Comp_JumpTo(target);
+}
+
+void Compiler::A_Comp_BranchXchangeReg()
+{
+    OpArg rn = MapReg(CurInstr.A_Reg(0));
+    if ((CurInstr.Instr & 0xF0) == 0x30) // BLX_reg
+        MOV(32, MapReg(14), Imm32(R15 - 4));
+    Comp_JumpTo(rn.GetSimpleReg());
+}
+
+void Compiler::T_Comp_BCOND()
+{
+    u32 cond = (CurInstr.Instr >> 8) & 0xF;
+    FixupBranch skipExecute = CheckCondition(cond);
+
+    s32 offset = (s32)(CurInstr.Instr << 24) >> 23;
+    Comp_JumpTo(R15 + offset + 1, true);
+
+    FixupBranch skipFailed = J();
+    SetJumpTarget(skipExecute);
+    Comp_AddCycles_C(true);
+    SetJumpTarget(skipFailed);
+}
+
+void Compiler::T_Comp_B()
+{
+    s32 offset = (s32)((CurInstr.Instr & 0x7FF) << 21) >> 20;
+    Comp_JumpTo(R15 + offset + 1);
+}
+
+void Compiler::T_Comp_BranchXchangeReg()
+{
+    bool link = CurInstr.Instr & (1 << 7);
+    if (link && Num == 1)
+    {
+        printf("BLX unsupported on ARM7!!!\n");
+        return;
+    }
+
+    OpArg rn = MapReg(CurInstr.A_Reg(3));
+    if (link)
+        MOV(32, MapReg(14), Imm32(R15 - 1));
+    Comp_JumpTo(rn.GetSimpleReg());
+}
+
+void Compiler::T_Comp_BL_LONG_1()
+{
+    s32 offset = (s32)((CurInstr.Instr & 0x7FF) << 21) >> 9;
+    MOV(32, MapReg(14), Imm32(R15 + offset));
+    Comp_AddCycles_C();
+}
+
+void Compiler::T_Comp_BL_LONG_2()
+{
+    OpArg lr = MapReg(14);
+    s32 offset = (CurInstr.Instr & 0x7FF) << 1;
+    LEA(32, RSCRATCH, MDisp(lr.GetSimpleReg(), offset));
+    MOV(32, lr, Imm32((R15 - 2) | 1));
+    if (Num == 1 || CurInstr.Instr & (1 << 12))
+        OR(32, R(RSCRATCH), Imm8(1));
+    Comp_JumpTo(RSCRATCH);
+}
+
+void Compiler::T_Comp_BL_Merged(FetchedInstr part1)
+{
+    assert(part1.Info.Kind == ARMInstrInfo::tk_BL_LONG_1);
+    Comp_AddCycles_C();
+
+    u32 target = (R15 - 2) + ((s32)((part1.Instr & 0x7FF) << 21) >> 9);
+    target += (CurInstr.Instr & 0x7FF) << 1;
+
+    if (Num == 1 || CurInstr.Instr & (1 << 12))
+        target |= 1;
+
+    MOV(32, MapReg(14), Imm32((R15 - 2) | 1));
+    
+    Comp_JumpTo(target);
+}
+
+}
\ No newline at end of file
diff --git a/src/ARMJIT_x64/ARMJIT_Compiler.cpp b/src/ARMJIT_x64/ARMJIT_Compiler.cpp
index 4fe0c70..6799a90 100644
--- a/src/ARMJIT_x64/ARMJIT_Compiler.cpp
+++ b/src/ARMJIT_x64/ARMJIT_Compiler.cpp
@@ -50,50 +50,6 @@ Compiler::Compiler()
     ResetStart = GetWritableCodePtr();
 }
 
-void* Compiler::Gen_ChangeCPSRRoutine()
-{
-    void* res = (void*)GetWritableCodePtr();
-
-    MOV(32, R(RSCRATCH), R(RCPSR));
-    AND(32, R(RSCRATCH), Imm8(0x1F));
-    CMP(32, R(RSCRATCH), Imm8(0x11));
-    FixupBranch fiq = J_CC(CC_E);
-    CMP(32, R(RSCRATCH), Imm8(0x12));
-    FixupBranch irq = J_CC(CC_E);
-    CMP(32, R(RSCRATCH), Imm8(0x13));
-    FixupBranch svc = J_CC(CC_E);
-    CMP(32, R(RSCRATCH), Imm8(0x17));
-    FixupBranch abt = J_CC(CC_E);
-    CMP(32, R(RSCRATCH), Imm8(0x1B));
-    FixupBranch und = J_CC(CC_E);
-
-    SetJumpTarget(fiq);
-
-    SetJumpTarget(irq);
-
-    SetJumpTarget(svc);
-
-    SetJumpTarget(abt);
-
-    SetJumpTarget(und);
-
-    return res;
-}
-
-DataRegion Compiler::ClassifyAddress(u32 addr)
-{
-    if (Num == 0 && addr >= ((ARMv5*)CurCPU)->DTCMBase && addr < ((ARMv5*)CurCPU)->DTCMBase)
-        return dataRegionDTCM;
-    switch (addr & 0xFF000000)
-    {
-        case 0x02000000: return dataRegionMainRAM;
-        case 0x03000000: return Num == 1 && (addr & 0xF00000) == 0x800000 ? dataRegionWRAM7 : dataRegionSWRAM;
-        case 0x04000000: return dataRegionIO;
-        case 0x06000000: return dataRegionVRAM;
-    }
-    return dataRegionGeneric;
-}
-
 void Compiler::LoadCPSR()
 {
     assert(!CPSRDirty);
@@ -123,6 +79,29 @@ void Compiler::SaveReg(int reg, X64Reg nativeReg)
     MOV(32, MDisp(RCPU, offsetof(ARM, R[reg])), R(nativeReg));
 }
 
+// invalidates RSCRATCH and RSCRATCH3
+Gen::FixupBranch Compiler::CheckCondition(u32 cond)
+{
+    if (cond >= 0x8)
+    {
+        static_assert(RSCRATCH3 == ECX);
+        MOV(32, R(RSCRATCH3), R(RCPSR));
+        SHR(32, R(RSCRATCH3), Imm8(28));
+        MOV(32, R(RSCRATCH), Imm32(1));
+        SHL(32, R(RSCRATCH), R(RSCRATCH3));
+        TEST(32, R(RSCRATCH), Imm32(ARM::ConditionTable[cond]));
+
+        return J_CC(CC_Z);
+    }
+    else
+    {
+        // could have used a LUT, but then where would be the fun?
+        TEST(32, R(RCPSR), Imm32(1 << (28 + ((~(cond >> 1) & 1) << 1 | (cond >> 2 & 1) ^ (cond >> 1 & 1)))));
+
+        return J_CC(cond & 1 ? CC_NZ : CC_Z);
+    }
+}
+
 CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrsCount)
 {
     if (IsAlmostFull())
@@ -140,6 +119,8 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
     CodeRegion = cpu->CodeRegion;
     CurCPU = cpu;
 
+    bool mergedThumbBL = false;
+
     ABI_PushRegistersAndAdjustStack({ABI_ALL_CALLEE_SAVED & ABI_ALL_GPRS}, 8, 16);
 
     MOV(64, R(RCPU), ImmPtr(cpu));
@@ -167,17 +148,10 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
                 MOV(32, MDisp(RCPU, offsetof(ARM, NextInstr[1])), Imm32(CurInstr.NextInstr[1]));
             }
 
-            if (comp == NULL || CurInstr.Info.Branches())
+            if (comp == NULL)
                 SaveCPSR();
         }
-
-        // run interpreter
-        cpu->CodeCycles = CurInstr.CodeCycles;
-        cpu->R[15] = R15;
-        cpu->CurInstr = CurInstr.Instr;
-        cpu->NextInstr[0] = CurInstr.NextInstr[0];
-        cpu->NextInstr[1] = CurInstr.NextInstr[1];
-
+        
         if (comp != NULL)
             RegCache.Prepare(i);
         else
@@ -185,58 +159,44 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
 
         if (Thumb)
         {
-            u32 icode = (CurInstr.Instr >> 6) & 0x3FF;
-            if (comp == NULL)
+            if (i < instrsCount - 1 && CurInstr.Info.Kind == ARMInstrInfo::tk_BL_LONG_1
+                && instrs[i + 1].Info.Kind == ARMInstrInfo::tk_BL_LONG_2)
+                mergedThumbBL = true;
+            else
             {
-                MOV(64, R(ABI_PARAM1), R(RCPU));
+                u32 icode = (CurInstr.Instr >> 6) & 0x3FF;
+                if (comp == NULL)
+                {
+                    MOV(64, R(ABI_PARAM1), R(RCPU));
 
-                ABI_CallFunction(ARMInterpreter::THUMBInstrTable[icode]);
+                    ABI_CallFunction(ARMInterpreter::THUMBInstrTable[icode]);
+                }
+                else if (mergedThumbBL)
+                    T_Comp_BL_Merged(instrs[i - 1]);
+                else
+                    (this->*comp)();
             }
-            else
-                (this->*comp)();
-
-            ARMInterpreter::THUMBInstrTable[icode](cpu);
         }
         else
         {
             u32 cond = CurInstr.Cond();
             if (CurInstr.Info.Kind == ARMInstrInfo::ak_BLX_IMM)
             {
-                MOV(64, R(ABI_PARAM1), R(RCPU));
-                ABI_CallFunction(ARMInterpreter::A_BLX_IMM);
-
-                ARMInterpreter::A_BLX_IMM(cpu);
+                if (comp)
+                    (this->*comp)();
+                else
+                {
+                    MOV(64, R(ABI_PARAM1), R(RCPU));
+                    ABI_CallFunction(ARMInterpreter::A_BLX_IMM);
+                }
             }
             else if (cond == 0xF)
-            {
                 Comp_AddCycles_C();
-                cpu->AddCycles_C();
-            }
             else
             {
                 FixupBranch skipExecute;
                 if (cond < 0xE)
-                {
-                    if (cond >= 0x8)
-                    {
-                        static_assert(RSCRATCH3 == ECX);
-                        MOV(32, R(RSCRATCH3), R(RCPSR));
-                        SHR(32, R(RSCRATCH3), Imm8(28));
-                        MOV(32, R(RSCRATCH), Imm32(1));
-                        SHL(32, R(RSCRATCH), R(RSCRATCH3));
-                        TEST(32, R(RSCRATCH), Imm32(ARM::ConditionTable[cond]));
-
-                        skipExecute = J_CC(CC_Z);
-                    }
-                    else
-                    {
-                        // could have used a LUT, but then where would be the fun?
-                        TEST(32, R(RCPSR), Imm32(1 << (28 + ((~(cond >> 1) & 1) << 1 | (cond >> 2 & 1) ^ (cond >> 1 & 1)))));
-
-                        skipExecute = J_CC(cond & 1 ? CC_NZ : CC_Z);
-                    }
-
-                }
+                    skipExecute = CheckCondition(cond);
 
                 u32 icode = ((CurInstr.Instr >> 4) & 0xF) | ((CurInstr.Instr >> 16) & 0xFF0);
                 if (comp == NULL)
@@ -258,19 +218,9 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
 
                     SetJumpTarget(skipFailed);
                 }
-
-                if (cpu->CheckCondition(cond))
-                    ARMInterpreter::ARMInstrTable[icode](cpu);
-                else
-                    cpu->AddCycles_C();
             }
         }
 
-        /*
-            we don't need to collect the interpreted cycles,
-            since cpu->Cycles is taken into account by the dispatcher.
-        */
-
         if (comp == NULL && i != instrsCount - 1)
             LoadCPSR();
     }
@@ -367,7 +317,7 @@ CompileFunc Compiler::GetCompFunc(int kind)
         // LDM/STM
         NULL, NULL,
         // Branch
-        NULL, NULL, NULL, NULL, NULL,
+        A_Comp_BranchImm, A_Comp_BranchImm, A_Comp_BranchImm, A_Comp_BranchXchangeReg, A_Comp_BranchXchangeReg,
         // system stuff
         NULL, NULL, NULL, NULL, NULL, NULL, NULL,
     };
@@ -389,7 +339,7 @@ CompileFunc Compiler::GetCompFunc(int kind)
         // pc/sp relative
         T_Comp_RelAddr, T_Comp_RelAddr, T_Comp_AddSP,
         // LDR pcrel
-        NULL,
+        T_Comp_LoadPCRel,
         // LDR/STR reg offset
         T_Comp_MemReg, T_Comp_MemReg, T_Comp_MemReg, T_Comp_MemReg,
         // LDR/STR sign extended, half
@@ -399,25 +349,27 @@ CompileFunc Compiler::GetCompFunc(int kind)
         // LDR/STR half imm offset
         T_Comp_MemImmHalf, T_Comp_MemImmHalf,
         // LDR/STR sp rel
-        NULL, NULL,
+        T_Comp_MemSPRel, T_Comp_MemSPRel,
         // PUSH/POP
-        NULL, NULL, 
+        T_Comp_PUSH_POP, T_Comp_PUSH_POP, 
         // LDMIA, STMIA
-        NULL, NULL, 
-        NULL, NULL,
-        NULL, NULL, NULL, NULL, NULL, NULL
+        T_Comp_LDMIA_STMIA, T_Comp_LDMIA_STMIA, 
+        // Branch
+        T_Comp_BCOND, T_Comp_BranchXchangeReg, T_Comp_BranchXchangeReg, T_Comp_B, T_Comp_BL_LONG_1, T_Comp_BL_LONG_2, 
+        // Unk, SVC
+        NULL, NULL
     };
 
     return Thumb ? T_Comp[kind] : A_Comp[kind];
 }
 
-void Compiler::Comp_AddCycles_C()
+void Compiler::Comp_AddCycles_C(bool forceNonConstant)
 {
     s32 cycles = Num ?
         NDS::ARM7MemTimings[CurInstr.CodeCycles][Thumb ? 1 : 3]
         : ((R15 & 0x2) ? 0 : CurInstr.CodeCycles);
 
-    if (CurInstr.Cond() < 0xE)
+    if ((!Thumb && CurInstr.Cond() < 0xE) || forceNonConstant)
         ADD(32, MDisp(RCPU, offsetof(ARM, Cycles)), Imm8(cycles));
     else
         ConstantCycles += cycles;
@@ -429,25 +381,10 @@ void Compiler::Comp_AddCycles_CI(u32 i)
         NDS::ARM7MemTimings[CurInstr.CodeCycles][Thumb ? 0 : 2]
         : ((R15 & 0x2) ? 0 : CurInstr.CodeCycles)) + i;
 
-    if (CurInstr.Cond() < 0xE)
+    if (!Thumb && CurInstr.Cond() < 0xE)
         ADD(32, MDisp(RCPU, offsetof(ARM, Cycles)), Imm8(cycles));
     else
         ConstantCycles += cycles;
 }
 
-void Compiler::Comp_JumpTo(Gen::X64Reg addr, bool restoreCPSR)
-{
-    // potentieller Bug: falls ein Register das noch gecacht ist, beim Modeswitch gespeichert
-    // wird der alte Wert gespeichert
-    SaveCPSR();
-
-    MOV(64, R(ABI_PARAM1), R(RCPU));
-    MOV(32, R(ABI_PARAM2), R(addr));
-    MOV(32, R(ABI_PARAM3), Imm32(restoreCPSR));
-    if (Num == 0)
-        CALL((void*)&ARMv5::JumpTo);
-    else
-        CALL((void*)&ARMv4::JumpTo);
-}
-
 }
\ No newline at end of file
diff --git a/src/ARMJIT_x64/ARMJIT_Compiler.h b/src/ARMJIT_x64/ARMJIT_Compiler.h
index a751737..45b488a 100644
--- a/src/ARMJIT_x64/ARMJIT_Compiler.h
+++ b/src/ARMJIT_x64/ARMJIT_Compiler.h
@@ -22,19 +22,6 @@ class Compiler;
 
 typedef void (Compiler::*CompileFunc)();
 
-enum DataRegion
-{
-    dataRegionGeneric, // hey, that's me!
-    dataRegionMainRAM,
-    dataRegionSWRAM,
-    dataRegionVRAM,
-    dataRegionIO,
-    dataRegionExclusive,
-    dataRegionsCount,
-    dataRegionDTCM = dataRegionExclusive,
-    dataRegionWRAM7 = dataRegionExclusive,
-};
-
 class Compiler : public Gen::X64CodeBlock
 {
 public:
@@ -49,8 +36,9 @@ private:
     CompileFunc GetCompFunc(int kind);
 
     void Comp_JumpTo(Gen::X64Reg addr, bool restoreCPSR = false);
+    void Comp_JumpTo(u32 addr, bool forceNonConstantCycles = false);
 
-    void Comp_AddCycles_C();
+    void Comp_AddCycles_C(bool forceNonConstant = false);
     void Comp_AddCycles_CI(u32 i);
 
     enum
@@ -63,8 +51,6 @@ private:
         opInvertOp2 = 1 << 5,
     };
 
-    DataRegion ClassifyAddress(u32 addr);
-
     void A_Comp_Arith();
     void A_Comp_MovOp();
     void A_Comp_CmpOp();
@@ -73,6 +59,9 @@ private:
     void A_Comp_MemHalf();
     void A_Comp_LDM_STM();
 
+    void A_Comp_BranchImm();
+    void A_Comp_BranchXchangeReg();
+
     void T_Comp_ShiftImm();
     void T_Comp_AddSub_();
     void T_Comp_ALU_Imm8();
@@ -91,6 +80,13 @@ private:
     void T_Comp_PUSH_POP();
     void T_Comp_LDMIA_STMIA();
 
+    void T_Comp_BCOND();
+    void T_Comp_B();
+    void T_Comp_BranchXchangeReg();
+    void T_Comp_BL_LONG_1();
+    void T_Comp_BL_LONG_2();
+    void T_Comp_BL_Merged(FetchedInstr prefix);
+
     void Comp_MemAccess(Gen::OpArg rd, bool signExtend, bool store, int size);
     s32 Comp_MemAccessBlock(Gen::OpArg rb, BitSet16 regs, bool store, bool preinc, bool decrement, bool usermode);
 
@@ -119,6 +115,8 @@ private:
     void LoadCPSR();
     void SaveCPSR();
 
+    Gen::FixupBranch CheckCondition(u32 cond);
+
     Gen::OpArg MapReg(int reg)
     {
         if (reg == 15 && RegCache.Mapping[reg] == Gen::INVALID_REG)
diff --git a/src/ARMJIT_x64/ARMJIT_LoadStore.cpp b/src/ARMJIT_x64/ARMJIT_LoadStore.cpp
index 20e1893..69b324c 100644
--- a/src/ARMJIT_x64/ARMJIT_LoadStore.cpp
+++ b/src/ARMJIT_x64/ARMJIT_LoadStore.cpp
@@ -462,38 +462,10 @@ s32 Compiler::Comp_MemAccessBlock(OpArg rb, BitSet16 regs, bool store, bool prei
 {
     int regsCount = regs.Count();
 
-    const u8 userModeOffsets[] =
-    {
-        offsetof(ARM, R[8]), offsetof(ARM, R[9]), offsetof(ARM, R[10]), offsetof(ARM, R[11]),
-        offsetof(ARM, R[12]), offsetof(ARM, R[13]), offsetof(ARM, R[14]), 0,
-
-        offsetof(ARM, R_FIQ[0]), offsetof(ARM, R_FIQ[1]), offsetof(ARM, R_FIQ[2]), offsetof(ARM, R_FIQ[3]),
-        offsetof(ARM, R_FIQ[4]), offsetof(ARM, R_FIQ[5]), offsetof(ARM, R_FIQ[6]), 0,
-
-        offsetof(ARM, R[8]), offsetof(ARM, R[9]), offsetof(ARM, R[10]), offsetof(ARM, R[11]),
-        offsetof(ARM, R[12]), offsetof(ARM, R_IRQ[13]), offsetof(ARM, R_IRQ[14]), 0,
-
-        offsetof(ARM, R[8]), offsetof(ARM, R[9]), offsetof(ARM, R[10]), offsetof(ARM, R[11]),
-        offsetof(ARM, R[12]), offsetof(ARM, R_SVC[13]), offsetof(ARM, R_SVC[14]), 0,
-
-        0, 0, 0, 0, 0, 0, 0, 0,
-        0, 0, 0, 0, 0, 0, 0, 0,
-
-        offsetof(ARM, R[8]), offsetof(ARM, R[9]), offsetof(ARM, R[10]), offsetof(ARM, R[11]),
-        offsetof(ARM, R[12]), offsetof(ARM, R_ABT[13]), offsetof(ARM, R_ABT[14]), 0,
-
-        0, 0, 0, 0, 0, 0, 0, 0,
-        0, 0, 0, 0, 0, 0, 0, 0,
-        0, 0, 0, 0, 0, 0, 0, 0,
-
-        offsetof(ARM, R[8]), offsetof(ARM, R[9]), offsetof(ARM, R[10]), offsetof(ARM, R[11]),
-        offsetof(ARM, R[12]), offsetof(ARM, R_UND[13]), offsetof(ARM, R_UND[14]), 0,
-    };
-
     if (decrement)
     {
         MOV_sum(32, ABI_PARAM1, rb, Imm32(-regsCount * 4));
-        preinc = !preinc;
+        preinc ^= true;
     }
     else
         MOV(32, R(ABI_PARAM1), rb);
@@ -516,16 +488,16 @@ s32 Compiler::Comp_MemAccessBlock(OpArg rb, BitSet16 regs, bool store, bool prei
         {
             if (regs[reg])
             {
-                if (usermode && reg >= 8 && reg < 15)
+                /*if (usermode && reg >= 8 && reg < 15)
                 {
                     MOV(32, R(RSCRATCH2), R(RCPSR));
                     AND(32, R(RSCRATCH2), Imm8(0x1F));
                     // (RSCRATCH2 - 0x11) * 8 + squeezePointer(userModeOffsets) + (reg - 8), algebra is great!
-                    MOVZX(32, 8, RSCRATCH2, MScaled(RSCRATCH2, SCALE_8, squeezePointer(userModeOffsets) - 0x11 * 8 + (reg - 8)));
+                    MOVZX(32, 8, RSCRATCH2, MScaled(RSCRATCH2, SCALE_8, squeezePointer(userModeOffsets) - 0x10 * 8 + (reg - 8)));
                     POP(RSCRATCH);
                     MOV(32, MRegSum(RCPU, RSCRATCH2), R(RSCRATCH));
                 }
-                else if (RegCache.Mapping[reg] == INVALID_REG)
+                else */if (RegCache.Mapping[reg] == INVALID_REG)
                 {
                     assert(reg != 15);
 
@@ -552,16 +524,16 @@ s32 Compiler::Comp_MemAccessBlock(OpArg rb, BitSet16 regs, bool store, bool prei
     {
         for (int reg : regs)
         {
-            if (usermode && reg >= 8 && reg < 15)
+            /*if (usermode && reg >= 8 && reg < 15)
             {
                 MOV(32, R(RSCRATCH), R(RCPSR));
                 AND(32, R(RSCRATCH), Imm8(0x1F));
                 // (RSCRATCH2 - 0x11) * 8 + squeezePointer(userModeOffsets) + (reg - 8), algebra is great!
-                MOVZX(32, 8, RSCRATCH, MScaled(RSCRATCH, SCALE_8, squeezePointer(userModeOffsets) - 0x11 * 8 + (reg - 8)));
+                MOVZX(32, 8, RSCRATCH, MScaled(RSCRATCH, SCALE_8, squeezePointer(userModeOffsets) - 0x10 * 8 + (reg - 8)));
                 MOV(32, R(RSCRATCH), MRegSum(RCPU, RSCRATCH));
                 PUSH(RSCRATCH);
             }
-            else if (RegCache.Mapping[reg] == INVALID_REG)
+            else */if (RegCache.Mapping[reg] == INVALID_REG)
             {
                 LoadReg(reg, RSCRATCH);
                 PUSH(RSCRATCH);
diff --git a/src/ARM_InstrInfo.cpp b/src/ARM_InstrInfo.cpp
index c519229..b8dff00 100644
--- a/src/ARM_InstrInfo.cpp
+++ b/src/ARM_InstrInfo.cpp
@@ -255,7 +255,7 @@ const u32 T_STMIA = T_Read8 | T_Write8 | tk(tk_STMIA);
 
 const u32 T_BCOND = T_BranchAlways | tk(tk_BCOND);
 const u32 T_BX = T_BranchAlways | T_ReadHi3 | tk(tk_BX);
-const u32 T_BLX_REG = T_BranchAlways | T_ReadR15 | T_WriteR14 | T_ReadHi3 | tk(tk_BLX_REG);
+const u32 T_BLX_REG = T_BranchAlways | T_WriteR14 | T_ReadHi3 | tk(tk_BLX_REG);
 const u32 T_B = T_BranchAlways | tk(tk_B);
 const u32 T_BL_LONG_1 = T_WriteR14 | T_ReadR15 | tk(tk_BL_LONG_1);
 const u32 T_BL_LONG_2 = T_BranchAlways | T_ReadR14 | T_WriteR14 | T_ReadR15 | tk(tk_BL_LONG_2);
@@ -301,6 +301,10 @@ Info Decode(bool thumb, u32 num, u32 instr)
             res.DstRegs |= (1 << 13);
         if (data & T_ReadR15)
             res.SrcRegs |= (1 << 15);
+        if (data & T_WriteR14)
+            res.DstRegs |= (1 << 14);
+        if (data & T_ReadR14)
+            res.SrcRegs |= (1 << 14);
 
         if (data & T_BranchAlways)
             res.DstRegs |= (1 << 15);
diff --git a/src/ARM_InstrInfo.h b/src/ARM_InstrInfo.h
index dcd938b..51dcfa2 100644
--- a/src/ARM_InstrInfo.h
+++ b/src/ARM_InstrInfo.h
@@ -202,6 +202,7 @@ enum
     tk_POP,
     tk_LDMIA,
     tk_STMIA,
+    
     tk_BCOND,
     tk_BX,
     tk_BLX_REG,
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index ae04ffb..75fa42c 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -54,6 +54,7 @@ add_library(core STATIC
 	ARMJIT_x64/ARMJIT_Compiler.cpp
 	ARMJIT_x64/ARMJIT_ALU.cpp
 	ARMJIT_x64/ARMJIT_LoadStore.cpp
+	ARMJIT_x64/ARMJIT_Branch.cpp
 
 	dolphin/CommonFuncs.cpp
 	dolphin/x64ABI.cpp