aboutsummaryrefslogtreecommitdiff
path: root/src
diff options
context:
space:
mode:
authorRSDuck <rsduck@users.noreply.github.com>2019-06-30 13:35:03 +0200
committerRSDuck <rsduck@users.noreply.github.com>2020-04-26 13:02:56 +0200
commit5f932cdf48681414465512fb47d619ad73414137 (patch)
tree233af2ee874ba03c51d1b66db600e02ffe019841 /src
parentff901141e77ad6c8d2910d77bef2b7c5674fcc7f (diff)
JIT: compilation of word load and store
Diffstat (limited to 'src')
-rw-r--r--src/ARMJIT.cpp4
-rw-r--r--src/ARMJIT.h3
-rw-r--r--src/ARMJIT_RegCache.h2
-rw-r--r--src/ARMJIT_x64/ARMJIT_ALU.cpp4
-rw-r--r--src/ARMJIT_x64/ARMJIT_Compiler.cpp111
-rw-r--r--src/ARMJIT_x64/ARMJIT_Compiler.h19
-rw-r--r--src/ARMJIT_x64/ARMJIT_LoadStore.cpp600
-rw-r--r--src/ARM_InstrInfo.h8
-rw-r--r--src/CMakeLists.txt1
-rw-r--r--src/dolphin/x64ABI.h3
10 files changed, 712 insertions, 43 deletions
diff --git a/src/ARMJIT.cpp b/src/ARMJIT.cpp
index 74e154b..4da781c 100644
--- a/src/ARMJIT.cpp
+++ b/src/ARMJIT.cpp
@@ -40,8 +40,7 @@ static ptrdiff_t JIT_MEM[2][32] = {
/* 2X*/ DUP2(offsetof(BlockCache, MainRAM)),
/* 3X*/ offsetof(BlockCache, SWRAM),
offsetof(BlockCache, ARM7_WRAM),
- /* 4X*/ -1,
- offsetof(BlockCache, ARM7_WIRAM),
+ /* 4X*/ DUP2(-1),
/* 5X*/ DUP2(-1),
/* 6X*/ DUP2(offsetof(BlockCache, ARM7_WVRAM)), /* contrary to Gbatek, melonDS and itself,
DeSmuME doesn't mirror the 64 MB region at 0x6800000 */
@@ -183,7 +182,6 @@ void ResetBlocks()
memset(cache.ARM9_ITCM, 0, sizeof(cache.ARM9_ITCM));
memset(cache.ARM9_LCDC, 0, sizeof(cache.ARM9_LCDC));
memset(cache.ARM7_BIOS, 0, sizeof(cache.ARM7_BIOS));
- memset(cache.ARM7_WIRAM, 0, sizeof(cache.ARM7_WIRAM));
memset(cache.ARM7_WRAM, 0, sizeof(cache.ARM7_WRAM));
memset(cache.ARM7_WVRAM, 0, sizeof(cache.ARM7_WVRAM));
}
diff --git a/src/ARMJIT.h b/src/ARMJIT.h
index 2ca29e8..45bb4ed 100644
--- a/src/ARMJIT.h
+++ b/src/ARMJIT.h
@@ -63,14 +63,13 @@ struct BlockCache
{
CompiledBlock* AddrMapping[2][0x4000] = {0};
- CompiledBlock MainRAM[16*1024*1024/2];
+ CompiledBlock MainRAM[4*1024*1024/2];
CompiledBlock SWRAM[0x8000/2]; // Shared working RAM
CompiledBlock ARM9_ITCM[0x8000/2];
CompiledBlock ARM9_LCDC[0xA4000/2];
CompiledBlock ARM9_BIOS[0x8000/2];
CompiledBlock ARM7_BIOS[0x4000/2];
CompiledBlock ARM7_WRAM[0x10000/2]; // dedicated ARM7 WRAM
- CompiledBlock ARM7_WIRAM[0x10000/2]; // Wifi
CompiledBlock ARM7_WVRAM[0x40000/2]; // VRAM allocated as Working RAM
};
diff --git a/src/ARMJIT_RegCache.h b/src/ARMJIT_RegCache.h
index e18d50f..ea9fb30 100644
--- a/src/ARMJIT_RegCache.h
+++ b/src/ARMJIT_RegCache.h
@@ -30,7 +30,7 @@ public:
assert(Mapping[reg] != -1);
if (DirtyRegs & (1 << reg))
- Compiler->UnloadReg(reg, Mapping[reg]);
+ Compiler->SaveReg(reg, Mapping[reg]);
DirtyRegs &= ~(1 << reg);
LoadedRegs &= ~(1 << reg);
diff --git a/src/ARMJIT_x64/ARMJIT_ALU.cpp b/src/ARMJIT_x64/ARMJIT_ALU.cpp
index dc82af7..6294e1d 100644
--- a/src/ARMJIT_x64/ARMJIT_ALU.cpp
+++ b/src/ARMJIT_x64/ARMJIT_ALU.cpp
@@ -255,8 +255,8 @@ OpArg Compiler::Comp_RegShiftReg(int op, Gen::OpArg rs, Gen::OpArg rm, bool S, b
if (S)
{
XOR(32, R(RSCRATCH2), R(RSCRATCH2));
- BT(32, R(RCPSR), Imm8(29));
- SETcc(CC_C, R(RSCRATCH2));
+ TEST(32, R(RCPSR), Imm32(1 << 29));
+ SETcc(CC_NZ, R(RSCRATCH2));
}
MOV(32, R(RSCRATCH), rm);
diff --git a/src/ARMJIT_x64/ARMJIT_Compiler.cpp b/src/ARMJIT_x64/ARMJIT_Compiler.cpp
index f51d4d9..9096397 100644
--- a/src/ARMJIT_x64/ARMJIT_Compiler.cpp
+++ b/src/ARMJIT_x64/ARMJIT_Compiler.cpp
@@ -9,13 +9,43 @@ using namespace Gen;
namespace ARMJIT
{
template <>
-const X64Reg RegCache<Compiler, X64Reg>::NativeRegAllocOrder[] = {RBX, RSI, RDI, R12, R13};
+const X64Reg RegCache<Compiler, X64Reg>::NativeRegAllocOrder[] =
+{
+#ifdef _WIN32
+ RBX, RSI, RDI, R12, R13
+#else
+ RBX, R12, R13
+#endif
+};
template <>
-const int RegCache<Compiler, X64Reg>::NativeRegsAvailable = 5;
+const int RegCache<Compiler, X64Reg>::NativeRegsAvailable =
+#ifdef _WIN32
+ 5
+#else
+ 3
+#endif
+;
Compiler::Compiler()
{
- AllocCodeSpace(1024 * 1024 * 4);
+ AllocCodeSpace(1024 * 1024 * 16);
+
+ for (int i = 0; i < 15; i++)
+ {
+ ReadMemFuncs9[i] = Gen_MemoryRoutine9(false, 32, 0x1000000 * i);
+ WriteMemFuncs9[i] = Gen_MemoryRoutine9(true, 32, 0x1000000 * i);
+ for (int j = 0; j < 2; j++)
+ {
+ ReadMemFuncs7[j][i] = Gen_MemoryRoutine7(false, 32, j, 0x1000000 * i);
+ WriteMemFuncs7[j][i] = Gen_MemoryRoutine7(true, 32, j, 0x1000000 * i);
+ }
+ }
+ ReadMemFuncs9[15] = Gen_MemoryRoutine9(false, 32, 0xFF000000);
+ WriteMemFuncs9[15] = Gen_MemoryRoutine9(true, 32, 0xFF000000);
+ ReadMemFuncs7[15][0] = ReadMemFuncs7[15][1] = Gen_MemoryRoutine7(false, 32, false, 0xFF000000);
+ WriteMemFuncs7[15][0] = WriteMemFuncs7[15][1] = Gen_MemoryRoutine7(true, 32, false, 0xFF000000);
+
+ ResetStart = GetWritableCodePtr();
}
void Compiler::LoadCPSR()
@@ -42,7 +72,7 @@ void Compiler::LoadReg(int reg, X64Reg nativeReg)
MOV(32, R(nativeReg), Imm32(R15));
}
-void Compiler::UnloadReg(int reg, X64Reg nativeReg)
+void Compiler::SaveReg(int reg, X64Reg nativeReg)
{
MOV(32, MDisp(RCPU, offsetof(ARM, R[reg])), R(nativeReg));
}
@@ -52,7 +82,7 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
if (IsAlmostFull())
{
ResetBlocks();
- ResetCodePtr();
+ SetCodePtr((u8*)ResetStart);
}
CompiledBlock res = (CompiledBlock)GetWritableCodePtr();
@@ -61,8 +91,9 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
Thumb = cpu->CPSR & 0x20;
Num = cpu->Num;
R15 = cpu->R[15];
+ CodeRegion = cpu->CodeRegion;
- ABI_PushRegistersAndAdjustStack({ABI_ALL_CALLEE_SAVED}, 8, 0);
+ ABI_PushRegistersAndAdjustStack({ABI_ALL_CALLEE_SAVED & ABI_ALL_GPRS}, 8, 16);
MOV(64, R(RCPU), ImmPtr(cpu));
XOR(32, R(RCycles), R(RCycles));
@@ -142,9 +173,9 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
else
{
// could have used a LUT, but then where would be the fun?
- BT(32, R(RCPSR), Imm8(28 + ((~(cond >> 1) & 1) << 1 | (cond >> 2 & 1) ^ (cond >> 1 & 1))));
+ TEST(32, R(RCPSR), Imm32(1 << (28 + ((~(cond >> 1) & 1) << 1 | (cond >> 2 & 1) ^ (cond >> 1 & 1)))));
- skipExecute = J_CC(cond & 1 ? CC_C : CC_NC);
+ skipExecute = J_CC(cond & 1 ? CC_NZ : CC_Z);
}
}
@@ -187,7 +218,7 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
LEA(32, RAX, MDisp(RCycles, ConstantCycles));
- ABI_PopRegistersAndAdjustStack({ABI_ALL_CALLEE_SAVED}, 8, 0);
+ ABI_PopRegistersAndAdjustStack({ABI_ALL_CALLEE_SAVED & ABI_ALL_GPRS}, 8, 16);
RET();
return res;
@@ -243,23 +274,38 @@ CompileFunc Compiler::GetCompFunc(int kind)
A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp,
// CMN
A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp, A_Comp_CmpOp,
+ // Mul
+ NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
+ // ARMv5 stuff
+ NULL, NULL, NULL, NULL, NULL,
+ // STR
+ A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB,
+ // STRB
+ NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
+ // LDR
+ A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB, A_Comp_MemWB,
+ // LDRB
+ NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
+ // STRH
+ NULL, NULL, NULL, NULL,
+ // LDRD
+ NULL, NULL, NULL, NULL,
+ // STRD
+ NULL, NULL, NULL, NULL,
+ // LDRH
+ NULL, NULL, NULL, NULL,
+ // LDRSB
NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
+ // LDRSH
+ NULL, NULL, NULL, NULL,
+ // swap
+ NULL, NULL,
+ // LDM/STM
+ NULL, NULL,
+ // Branch
+ NULL, NULL, NULL, NULL, NULL,
+ // system stuff
+ NULL, NULL, NULL, NULL, NULL, NULL, NULL,
};
const CompileFunc T_Comp[ARMInstrInfo::tk_Count] = {
@@ -278,10 +324,17 @@ CompileFunc Compiler::GetCompFunc(int kind)
T_Comp_ALU_HiReg, T_Comp_ALU_HiReg, T_Comp_ALU_HiReg,
// pc/sp relative
NULL, NULL, NULL,
- // mem...
- NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL,
- NULL, NULL, NULL, NULL, NULL, NULL,
+ // LDR pcrel
+ NULL,
+ // LDR/STR reg offset
+ T_Comp_MemReg, NULL, T_Comp_MemReg, NULL,
+ // LDR/STR sign extended, half
+ NULL, NULL, NULL, NULL,
+ // LDR/STR imm offset
+ T_Comp_MemImm, T_Comp_MemImm, NULL, NULL,
+ // LDR/STR half imm offset
+ NULL, NULL,
+ // branch, etc.
NULL, NULL, NULL, NULL, NULL, NULL,
NULL, NULL, NULL, NULL, NULL, NULL,
NULL, NULL
diff --git a/src/ARMJIT_x64/ARMJIT_Compiler.h b/src/ARMJIT_x64/ARMJIT_Compiler.h
index 9b454f4..7ab9b25 100644
--- a/src/ARMJIT_x64/ARMJIT_Compiler.h
+++ b/src/ARMJIT_x64/ARMJIT_Compiler.h
@@ -29,7 +29,7 @@ public:
CompiledBlock CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrsCount);
void LoadReg(int reg, Gen::X64Reg nativeReg);
- void UnloadReg(int reg, Gen::X64Reg nativeReg);
+ void SaveReg(int reg, Gen::X64Reg nativeReg);
private:
CompileFunc GetCompFunc(int kind);
@@ -51,12 +51,17 @@ private:
void A_Comp_MovOp();
void A_Comp_CmpOp();
+ void A_Comp_MemWB();
+
void T_Comp_ShiftImm();
void T_Comp_AddSub_();
void T_Comp_ALU_Imm8();
void T_Comp_ALU();
void T_Comp_ALU_HiReg();
+ void T_Comp_MemReg();
+ void T_Comp_MemImm();
+
void Comp_ArithTriOp(void (Compiler::*op)(int, const Gen::OpArg&, const Gen::OpArg&),
Gen::OpArg rd, Gen::OpArg rn, Gen::OpArg op2, bool carryUsed, int opFlags);
void Comp_ArithTriOpReverse(void (Compiler::*op)(int, const Gen::OpArg&, const Gen::OpArg&),
@@ -65,10 +70,14 @@ private:
void Comp_RetriveFlags(bool sign, bool retriveCV, bool carryUsed);
+ void* Gen_MemoryRoutine9(bool store, int size, u32 region);
+ void* Gen_MemoryRoutine7(bool store, int size, bool mainRAMCode, u32 region);
+
Gen::OpArg Comp_RegShiftImm(int op, int amount, Gen::OpArg rm, bool S, bool& carryUsed);
Gen::OpArg Comp_RegShiftReg(int op, Gen::OpArg rs, Gen::OpArg rm, bool S, bool& carryUsed);
Gen::OpArg A_Comp_GetALUOp2(bool S, bool& carryUsed);
+ Gen::OpArg A_Comp_GetMemWBOffset();
void LoadCPSR();
void SaveCPSR();
@@ -82,6 +91,8 @@ private:
return Gen::R(RegCache.Mapping[reg]);
}
+ void* ResetStart;
+
bool CPSRDirty = false;
FetchedInstr CurrentInstr;
@@ -91,10 +102,16 @@ private:
bool Thumb;
u32 Num;
u32 R15;
+ u32 CodeRegion;
u32 ConstantCycles;
};
+extern void* ReadMemFuncs9[16];
+extern void* ReadMemFuncs7[2][16];
+extern void* WriteMemFuncs9[16];
+extern void* WriteMemFuncs7[2][16];
+
}
#endif \ No newline at end of file
diff --git a/src/ARMJIT_x64/ARMJIT_LoadStore.cpp b/src/ARMJIT_x64/ARMJIT_LoadStore.cpp
index e69de29..d534269 100644
--- a/src/ARMJIT_x64/ARMJIT_LoadStore.cpp
+++ b/src/ARMJIT_x64/ARMJIT_LoadStore.cpp
@@ -0,0 +1,600 @@
+#include "ARMJIT_Compiler.h"
+
+#include "../GPU.h"
+#include "../Wifi.h"
+
+namespace NDS
+{
+#define MAIN_RAM_SIZE 0x400000
+extern u8* SWRAM_ARM9;
+extern u32 SWRAM_ARM9Mask;
+extern u8* SWRAM_ARM7;
+extern u32 SWRAM_ARM7Mask;
+extern u8 ARM7WRAM[];
+extern u16 ARM7BIOSProt;
+}
+
+using namespace Gen;
+
+namespace ARMJIT
+{
+
+void* ReadMemFuncs9[16];
+void* ReadMemFuncs7[2][16];
+void* WriteMemFuncs9[16];
+void* WriteMemFuncs7[2][16];
+
+template <typename T>
+int squeezePointer(T* ptr)
+{
+ int truncated = (int)((u64)ptr);
+ assert((T*)((u64)truncated) == ptr);
+ return truncated;
+}
+
+u32 ReadVRAM9(u32 addr)
+{
+ switch (addr & 0x00E00000)
+ {
+ case 0x00000000: return GPU::ReadVRAM_ABG<u32>(addr);
+ case 0x00200000: return GPU::ReadVRAM_BBG<u32>(addr);
+ case 0x00400000: return GPU::ReadVRAM_AOBJ<u32>(addr);
+ case 0x00600000: return GPU::ReadVRAM_BOBJ<u32>(addr);
+ default: return GPU::ReadVRAM_LCDC<u32>(addr);
+ }
+}
+
+void WriteVRAM9(u32 addr, u32 val)
+{
+ switch (addr & 0x00E00000)
+ {
+ case 0x00000000: GPU::WriteVRAM_ABG<u32>(addr, val); return;
+ case 0x00200000: GPU::WriteVRAM_BBG<u32>(addr, val); return;
+ case 0x00400000: GPU::WriteVRAM_AOBJ<u32>(addr, val); return;
+ case 0x00600000: GPU::WriteVRAM_BOBJ<u32>(addr, val); return;
+ default: GPU::WriteVRAM_LCDC<u32>(addr, val); return;
+ }
+}
+
+/*
+ R11 - data to write (store only)
+ RSCRATCH2 - address
+ RSCRATCH3 - code cycles
+*/
+void* Compiler::Gen_MemoryRoutine9(bool store, int size, u32 region)
+{
+ AlignCode4();
+ void* res = (void*)GetWritableCodePtr();
+
+ if (!store)
+ {
+ MOV(32, R(RSCRATCH), R(RSCRATCH2));
+ AND(32, R(RSCRATCH), Imm8(0x3));
+ SHL(32, R(RSCRATCH), Imm8(3));
+ // enter the shadow realm!
+ MOV(32, MDisp(RSP, 8), R(RSCRATCH));
+ }
+
+ // cycle counting!
+ // this is AddCycles_CDI
+ MOV(32, R(R10), R(RSCRATCH2));
+ SHR(32, R(R10), Imm8(12));
+ MOVZX(32, 8, R10, MComplex(RCPU, R10, SCALE_1, offsetof(ARMv5, MemTimings) + 2));
+ LEA(32, RSCRATCH, MComplex(RSCRATCH3, R10, SCALE_1, -6));
+ CMP(32, R(R10), R(RSCRATCH3));
+ CMOVcc(32, RSCRATCH3, R(R10), CC_G);
+ CMP(32, R(RSCRATCH), R(RSCRATCH3));
+ CMOVcc(32, RSCRATCH3, R(RSCRATCH), CC_G);
+ ADD(32, R(RCycles), R(RSCRATCH3));
+
+ if (!store)
+ XOR(32, R(RSCRATCH), R(RSCRATCH));
+ AND(32, R(RSCRATCH2), Imm32(~3));
+
+ {
+ MOV(32, R(RSCRATCH3), R(RSCRATCH2));
+ SUB(32, R(RSCRATCH2), MDisp(RCPU, offsetof(ARMv5, DTCMBase)));
+ CMP(32, R(RSCRATCH2), MDisp(RCPU, offsetof(ARMv5, DTCMSize)));
+ FixupBranch outsideDTCM = J_CC(CC_AE);
+ AND(32, R(RSCRATCH2), Imm32(0x3FFF));
+ if (!store)
+ {
+ MOV(32, R(RSCRATCH), MComplex(RCPU, RSCRATCH2, SCALE_1, offsetof(ARMv5, DTCM)));
+ MOV(32, R(ECX), MDisp(RSP, 8));
+ ROR_(32, R(RSCRATCH), R(ECX));
+ }
+ else
+ MOV(32, MComplex(RCPU, RSCRATCH2, SCALE_1, offsetof(ARMv5, DTCM)), R(R11));
+ RET();
+ SetJumpTarget(outsideDTCM);
+ MOV(32, R(RSCRATCH2), R(RSCRATCH3));
+ }
+
+ switch (region)
+ {
+ case 0x00000000:
+ case 0x01000000:
+ {
+ CMP(32, R(RSCRATCH2), MDisp(RCPU, offsetof(ARMv5, ITCMSize)));
+ FixupBranch insideITCM = J_CC(CC_B);
+ RET();
+ SetJumpTarget(insideITCM);
+ AND(32, R(RSCRATCH2), Imm32(0x7FFF));
+ if (!store)
+ MOV(32, R(RSCRATCH), MComplex(RCPU, RSCRATCH2, SCALE_1, offsetof(ARMv5, ITCM)));
+ else
+ {
+ MOV(32, MComplex(RCPU, RSCRATCH2, SCALE_1, offsetof(ARMv5, ITCM)), R(R11));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.ARM9_ITCM)), Imm32(0));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.ARM9_ITCM) + 8), Imm32(0));
+ }
+ }
+ break;
+ case 0x02000000:
+ AND(32, R(RSCRATCH2), Imm32(MAIN_RAM_SIZE - 1));
+ if (!store)
+ MOV(32, R(RSCRATCH), MDisp(RSCRATCH2, squeezePointer(NDS::MainRAM)));
+ else
+ {
+ MOV(32, MDisp(RSCRATCH2, squeezePointer(NDS::MainRAM)), R(R11));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.MainRAM)), Imm32(0));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.MainRAM) + 8), Imm32(0));
+ }
+ break;
+ case 0x03000000:
+ {
+ MOV(64, R(RSCRATCH3), M(&NDS::SWRAM_ARM9));
+ TEST(64, R(RSCRATCH3), R(RSCRATCH3));
+ FixupBranch notMapped = J_CC(CC_Z);
+ AND(32, R(RSCRATCH2), M(&NDS::SWRAM_ARM9Mask));
+ if (!store)
+ MOV(32, R(RSCRATCH), MRegSum(RSCRATCH2, RSCRATCH3));
+ else
+ {
+ MOV(32, MRegSum(RSCRATCH2, RSCRATCH3), R(R11));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.SWRAM)), Imm32(0));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.SWRAM) + 8), Imm32(0));
+ }
+ SetJumpTarget(notMapped);
+ }
+ break;
+ case 0x04000000:
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ if (!store)
+ {
+ ABI_PushRegistersAndAdjustStack({}, 8, 0);
+ ABI_CallFunction(NDS::ARM9IORead32);
+ ABI_PopRegistersAndAdjustStack({}, 8, 0);
+ }
+ else
+ {
+ MOV(32, R(ABI_PARAM2), R(R11));
+ JMP((u8*)NDS::ARM9IOWrite32, true);
+ }
+ break;
+ case 0x05000000:
+ {
+ MOV(32, R(RSCRATCH), Imm32(1<<1));
+ MOV(32, R(RSCRATCH3), Imm32(1<<9));
+ TEST(32, R(RSCRATCH2), Imm32(0x400));
+ CMOVcc(32, RSCRATCH, R(RSCRATCH3), CC_NZ);
+ TEST(16, R(RSCRATCH), M(&NDS::PowerControl9));
+ FixupBranch available = J_CC(CC_NZ);
+ RET();
+ SetJumpTarget(available);
+ AND(32, R(RSCRATCH2), Imm32(0x7FF));
+ if (!store)
+ MOV(32, R(RSCRATCH), MDisp(RSCRATCH2, squeezePointer(GPU::Palette)));
+ else
+ MOV(32, MDisp(RSCRATCH2, squeezePointer(GPU::Palette)), R(R11));
+ }
+ break;
+ case 0x06000000:
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ if (!store)
+ {
+ ABI_PushRegistersAndAdjustStack({}, 8);
+ ABI_CallFunction(ReadVRAM9);
+ ABI_PopRegistersAndAdjustStack({}, 8);
+ }
+ else
+ {
+ MOV(32, R(ABI_PARAM2), R(R11));
+ JMP((u8*)WriteVRAM9, true);
+ }
+ break;
+ case 0x07000000:
+ {
+ MOV(32, R(RSCRATCH), Imm32(1<<1));
+ MOV(32, R(RSCRATCH3), Imm32(1<<9));
+ TEST(32, R(RSCRATCH2), Imm32(0x400));
+ CMOVcc(32, RSCRATCH, R(RSCRATCH3), CC_NZ);
+ TEST(16, R(RSCRATCH), M(&NDS::PowerControl9));
+ FixupBranch available = J_CC(CC_NZ);
+ RET();
+ SetJumpTarget(available);
+ AND(32, R(RSCRATCH2), Imm32(0x7FF));
+ if (!store)
+ MOV(32, R(RSCRATCH), MDisp(RSCRATCH2, squeezePointer(GPU::OAM)));
+ else
+ MOV(32, MDisp(RSCRATCH2, squeezePointer(GPU::OAM)), R(R11));
+ }
+ break;
+ case 0x08000000:
+ case 0x09000000:
+ case 0x0A000000:
+ if (!store)
+ MOV(32, R(RSCRATCH), Imm32(0xFFFFFFFF));
+ break;
+ case 0xFF000000:
+ if (!store)
+ {
+ AND(32, R(RSCRATCH2), Imm32(0xFFF));
+ MOV(32, R(RSCRATCH), MDisp(RSCRATCH2, squeezePointer(NDS::ARM9BIOS)));
+ }
+ break;
+ default:
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ if (!store)
+ {
+ ABI_PushRegistersAndAdjustStack({}, 8, 0);
+ ABI_CallFunction(NDS::ARM9Read32);
+ ABI_PopRegistersAndAdjustStack({}, 8, 0);
+ }
+ else
+ {
+ MOV(32, R(ABI_PARAM2), R(R11));
+ JMP((u8*)NDS::ARM9Write32, true);
+ }
+ break;
+ }
+
+ if (!store)
+ {
+ MOV(32, R(ECX), MDisp(RSP, 8));
+ ROR_(32, R(RSCRATCH), R(ECX));
+ }
+
+ RET();
+
+ return res;
+}
+
+void* Compiler::Gen_MemoryRoutine7(bool store, int size, bool mainRAMCode, u32 region)
+{
+ AlignCode4();
+ void* res = GetWritableCodePtr();
+
+ if (!store)
+ {
+ MOV(32, R(RSCRATCH), R(RSCRATCH2));
+ AND(32, R(RSCRATCH), Imm8(0x3));
+ SHL(32, R(RSCRATCH), Imm8(3));
+ // enter the shadow realm!
+ MOV(32, MDisp(RSP, 8), R(RSCRATCH));
+ }
+
+ // AddCycles_CDI
+ MOV(32, R(RSCRATCH), R(RSCRATCH2));
+ SHR(32, R(RSCRATCH), Imm8(15));
+ MOVZX(32, 8, RSCRATCH, MDisp(RSCRATCH, squeezePointer(NDS::ARM7MemTimings + 2)));
+ if ((region == 0x02000000 && mainRAMCode) || (region != 0x02000000 && !mainRAMCode))
+ {
+ if (!store && region != 0x02000000)
+ LEA(32, RSCRATCH3, MComplex(RSCRATCH, RSCRATCH3, SCALE_1, 1));
+ ADD(32, R(RCycles), R(RSCRATCH3));
+ }
+ else
+ {
+ if (!store)
+ ADD(32, R(region == 0x02000000 ? RSCRATCH2 : RSCRATCH), Imm8(1));
+ LEA(32, R10, MComplex(RSCRATCH, RSCRATCH3, SCALE_1, -3));
+ CMP(32, R(RSCRATCH3), R(RSCRATCH));
+ CMOVcc(32, RSCRATCH, R(RSCRATCH3), CC_G);
+ CMP(32, R(R10), R(RSCRATCH));
+ CMOVcc(32, RSCRATCH, R(R10), CC_G);
+ ADD(32, R(RCycles), R(RSCRATCH));
+ }
+
+ if (!store)
+ XOR(32, R(RSCRATCH), R(RSCRATCH));
+ AND(32, R(RSCRATCH2), Imm32(~3));
+
+ switch (region)
+ {
+ case 0x00000000:
+ if (!store) {
+ CMP(32, R(RSCRATCH2), Imm32(0x4000));
+ FixupBranch outsideBIOS1 = J_CC(CC_AE);
+
+ MOV(32, R(RSCRATCH), MDisp(RCPU, offsetof(ARM, R[15])));
+ CMP(32, R(RSCRATCH), Imm32(0x4000));
+ FixupBranch outsideBIOS2 = J_CC(CC_AE);
+ MOV(32, R(RSCRATCH3), M(&NDS::ARM7BIOSProt));
+ CMP(32, R(RSCRATCH2), R(RSCRATCH3));
+ FixupBranch notDenied1 = J_CC(CC_AE);
+ CMP(32, R(RSCRATCH), R(RSCRATCH3));
+ FixupBranch notDenied2 = J_CC(CC_B);
+ SetJumpTarget(outsideBIOS2);
+ MOV(32, R(RSCRATCH), Imm32(0xFFFFFFFF));
+ RET();
+
+ SetJumpTarget(notDenied1);
+ SetJumpTarget(notDenied2);
+ MOV(32, R(RSCRATCH), MDisp(RSCRATCH2, squeezePointer(NDS::ARM7BIOS)));
+ MOV(32, R(ECX), MDisp(RSP, 8));
+ ROR_(32, R(RSCRATCH), R(ECX));
+ RET();
+
+ SetJumpTarget(outsideBIOS1);
+ }
+ break;
+ case 0x02000000:
+ AND(32, R(RSCRATCH2), Imm32(MAIN_RAM_SIZE - 1));
+ if (!store)
+ MOV(32, R(RSCRATCH), MDisp(RSCRATCH2, squeezePointer(NDS::MainRAM)));
+ else
+ {
+ MOV(32, MDisp(RSCRATCH2, squeezePointer(NDS::MainRAM)), R(R11));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.MainRAM)), Imm32(0));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.MainRAM) + 8), Imm32(0));
+ }
+ break;
+ case 0x03000000:
+ {
+ TEST(32, R(RSCRATCH2), Imm32(0x800000));
+ FixupBranch region = J_CC(CC_NZ);
+ MOV(64, R(RSCRATCH), M(&NDS::SWRAM_ARM7));
+ TEST(64, R(RSCRATCH), R(RSCRATCH));
+ FixupBranch notMapped = J_CC(CC_Z);
+ AND(32, R(RSCRATCH2), M(&NDS::SWRAM_ARM7Mask));
+ if (!store)
+ {
+ MOV(32, R(RSCRATCH), MRegSum(RSCRATCH, RSCRATCH2));
+ MOV(32, R(ECX), MDisp(RSP, 8));
+ ROR_(32, R(RSCRATCH), R(ECX));
+ }
+ else
+ {
+ MOV(32, MRegSum(RSCRATCH, RSCRATCH2), R(R11));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.SWRAM)), Imm32(0));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.SWRAM) + 8), Imm32(0));
+ }
+ RET();
+ SetJumpTarget(region);
+ SetJumpTarget(notMapped);
+ AND(32, R(RSCRATCH2), Imm32(0xFFFF));
+ if (!store)
+ MOV(32, R(RSCRATCH), MDisp(RSCRATCH2, squeezePointer(NDS::ARM7WRAM)));
+ else
+ {
+ MOV(32, MDisp(RSCRATCH2, squeezePointer(NDS::ARM7WRAM)), R(R11));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.ARM7_WRAM)), Imm32(0));
+ MOV(64, MScaled(RSCRATCH2, SCALE_4, squeezePointer(cache.ARM7_WRAM) + 8), Imm32(0));
+ }
+ }
+ break;
+ case 0x04000000:
+ {
+ TEST(32, R(RSCRATCH2), Imm32(0x800000));
+ FixupBranch region = J_CC(CC_NZ);
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ if (!store)
+ {
+ ABI_PushRegistersAndAdjustStack({}, 8);
+ ABI_CallFunction(NDS::ARM7IORead32);
+ ABI_PopRegistersAndAdjustStack({}, 8);
+
+ MOV(32, R(ECX), MDisp(RSP, 8));
+ ROR_(32, R(RSCRATCH), R(ECX));
+ RET();
+ }
+ else
+ {
+ MOV(32, R(ABI_PARAM2), R(R11));
+ JMP((u8*)NDS::ARM7IOWrite32, true);
+ }
+ SetJumpTarget(region);
+
+ if (!store)
+ {
+ ABI_PushRegistersAndAdjustStack({RSCRATCH2}, 8);
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ ABI_CallFunction(Wifi::Read);
+ ABI_PopRegistersAndAdjustStack({RSCRATCH2}, 8);
+
+ ADD(32, R(RSCRATCH2), Imm8(2));
+ ABI_PushRegistersAndAdjustStack({EAX}, 8);
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ ABI_CallFunction(Wifi::Read);
+ MOV(32, R(RSCRATCH2), R(EAX));
+ SHL(32, R(RSCRATCH2), Imm8(16));
+ ABI_PopRegistersAndAdjustStack({EAX}, 8);
+ OR(32, R(EAX), R(RSCRATCH2));
+ }
+ else
+ {
+ ABI_PushRegistersAndAdjustStack({RSCRATCH2, R11}, 8);
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ MOVZX(32, 16, ABI_PARAM2, R(R11));
+ ABI_CallFunction(Wifi::Write);
+ ABI_PopRegistersAndAdjustStack({RSCRATCH2, R11}, 8);
+ SHR(32, R(R11), Imm8(16));
+ ADD(32, R(RSCRATCH2), Imm8(2));
+ ABI_PushRegistersAndAdjustStack({RSCRATCH2, R11}, 8);
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ MOVZX(32, 16, ABI_PARAM2, R(R11));
+ ABI_CallFunction(Wifi::Write);
+ ABI_PopRegistersAndAdjustStack({RSCRATCH2, R11}, 8);
+ }
+ }
+ break;
+ case 0x06000000:
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ if (!store)
+ {
+ ABI_PushRegistersAndAdjustStack({}, 8);
+ ABI_CallFunction(GPU::ReadVRAM_ARM7<u32>);
+ ABI_PopRegistersAndAdjustStack({}, 8);
+ }
+ else
+ {
+ AND(32, R(ABI_PARAM1), Imm32(0x40000 - 1));
+ MOV(64, MScaled(ABI_PARAM1, SCALE_4, squeezePointer(cache.ARM7_WVRAM)), Imm32(0));
+ MOV(64, MScaled(ABI_PARAM1, SCALE_4, squeezePointer(cache.ARM7_WVRAM) + 8), Imm32(0));
+ MOV(32, R(ABI_PARAM2), R(R11));
+ JMP((u8*)GPU::WriteVRAM_ARM7<u32>, true);
+ }
+ break;
+ case 0x08000000:
+ case 0x09000000:
+ case 0x0A000000:
+ if (!store)
+ MOV(32, R(RSCRATCH), Imm32(0xFFFFFFFF));
+ break;
+ /*default:
+ ABI_PushRegistersAndAdjustStack({}, 8, 0);
+ MOV(32, R(ABI_PARAM1), R(RSCRATCH2));
+ ABI_CallFunction(NDS::ARM7Read32);
+ ABI_PopRegistersAndAdjustStack({}, 8, 0);
+ break;*/
+ }
+
+ if (!store)
+ {
+ MOV(32, R(ECX), MDisp(RSP, 8));
+ ROR_(32, R(RSCRATCH), R(ECX));
+ }
+
+ RET();
+
+ return res;
+}
+
+OpArg Compiler::A_Comp_GetMemWBOffset()
+{
+ if (!(CurrentInstr.Instr & (1 << 25)))
+ return Imm32(CurrentInstr.Instr & 0xFFF);
+ else
+ {
+ int op = (CurrentInstr.Instr >> 5) & 0x3;
+ int amount = (CurrentInstr.Instr >> 7) & 0x1F;
+ OpArg rm = MapReg(CurrentInstr.A_Reg(0));
+ bool carryUsed;
+ return Comp_RegShiftImm(op, amount, rm, false, carryUsed);
+ }
+}
+
+void Compiler::A_Comp_MemWB()
+{
+ OpArg rn = MapReg(CurrentInstr.A_Reg(16));
+ OpArg rd = MapReg(CurrentInstr.A_Reg(12));
+ bool load = CurrentInstr.Instr & (1 << 20);
+
+ MOV(32, R(RSCRATCH2), rn);
+ if (CurrentInstr.Instr & (1 << 24))
+ {
+ OpArg offset = A_Comp_GetMemWBOffset();
+ if (CurrentInstr.Instr & (1 << 23))
+ ADD(32, R(RSCRATCH2), offset);
+ else
+ SUB(32, R(RSCRATCH2), offset);
+
+ if (CurrentInstr.Instr & (1 << 21))
+ MOV(32, rn, R(RSCRATCH2));
+ }
+
+ u32 cycles = Num ? NDS::ARM7MemTimings[CurrentInstr.CodeCycles][2] : CurrentInstr.CodeCycles;
+ MOV(32, R(RSCRATCH3), Imm32(cycles));
+ MOV(32, R(RSCRATCH), R(RSCRATCH2));
+ SHR(32, R(RSCRATCH), Imm8(24));
+ AND(32, R(RSCRATCH), Imm8(0xF));
+ void** funcArray;
+ if (load)
+ funcArray = Num ? ReadMemFuncs7[CodeRegion == 0x02] : ReadMemFuncs9;
+ else
+ {
+ funcArray = Num ? WriteMemFuncs7[CodeRegion == 0x02] : WriteMemFuncs9;
+ MOV(32, R(R11), rd);
+ }
+ CALLptr(MScaled(RSCRATCH, SCALE_8, squeezePointer(funcArray)));
+
+ if (load)
+ MOV(32, R(RSCRATCH2), R(RSCRATCH));
+
+ if (!(CurrentInstr.Instr & (1 << 24)))
+ {
+ OpArg offset = A_Comp_GetMemWBOffset();
+
+ if (CurrentInstr.Instr & (1 << 23))
+ ADD(32, rn, offset);
+ else
+ SUB(32, rn, offset);
+ }
+
+ if (load)
+ MOV(32, rd, R(RSCRATCH2));
+}
+
+void Compiler::T_Comp_MemReg()
+{
+ OpArg rd = MapReg(CurrentInstr.T_Reg(0));
+ OpArg rb = MapReg(CurrentInstr.T_Reg(3));
+ OpArg ro = MapReg(CurrentInstr.T_Reg(6));
+
+ int op = (CurrentInstr.Instr >> 10) & 0x3;
+ bool load = op & 0x2;
+
+ MOV(32, R(RSCRATCH2), rb);
+ ADD(32, R(RSCRATCH2), ro);
+
+ u32 cycles = Num ? NDS::ARM7MemTimings[CurrentInstr.CodeCycles][0] : (R15 & 0x2 ? 0 : CurrentInstr.CodeCycles);
+ MOV(32, R(RSCRATCH3), Imm32(cycles));
+ MOV(32, R(RSCRATCH), R(RSCRATCH2));
+ SHR(32, R(RSCRATCH), Imm8(24));
+ AND(32, R(RSCRATCH), Imm8(0xF));
+ void** funcArray;
+ if (load)
+ funcArray = Num ? ReadMemFuncs7[CodeRegion == 0x02] : ReadMemFuncs9;
+ else
+ {
+ funcArray = Num ? WriteMemFuncs7[CodeRegion == 0x02] : WriteMemFuncs9;
+ MOV(32, R(R11), rd);
+ }
+ CALLptr(MScaled(RSCRATCH, SCALE_8, squeezePointer(funcArray)));
+
+ if (load)
+ MOV(32, rd, R(RSCRATCH));
+}
+
+void Compiler::T_Comp_MemImm()
+{
+ // TODO: aufräumen!!!
+ OpArg rd = MapReg(CurrentInstr.T_Reg(0));
+ OpArg rb = MapReg(CurrentInstr.T_Reg(3));
+
+ int op = (CurrentInstr.Instr >> 11) & 0x3;
+ u32 offset = ((CurrentInstr.Instr >> 6) & 0x1F) * 4;
+ bool load = op & 0x1;
+
+ LEA(32, RSCRATCH2, MDisp(rb.GetSimpleReg(), offset));
+ u32 cycles = Num ? NDS::ARM7MemTimings[CurrentInstr.CodeCycles][0] : (R15 & 0x2 ? 0 : CurrentInstr.CodeCycles);
+ MOV(32, R(RSCRATCH3), Imm32(cycles));
+ MOV(32, R(RSCRATCH), R(RSCRATCH2));
+ SHR(32, R(RSCRATCH), Imm8(24));
+ AND(32, R(RSCRATCH), Imm8(0xF));
+ void** funcArray;
+ if (load)
+ funcArray = Num ? ReadMemFuncs7[CodeRegion == 0x02] : ReadMemFuncs9;
+ else
+ {
+ funcArray = Num ? WriteMemFuncs7[CodeRegion == 0x02] : WriteMemFuncs9;
+ MOV(32, R(R11), rd);
+ }
+ CALLptr(MScaled(RSCRATCH, SCALE_8, squeezePointer(funcArray)));
+
+ if (load)
+ MOV(32, rd, R(RSCRATCH));
+}
+
+} \ No newline at end of file
diff --git a/src/ARM_InstrInfo.h b/src/ARM_InstrInfo.h
index e717664..dcd938b 100644
--- a/src/ARM_InstrInfo.h
+++ b/src/ARM_InstrInfo.h
@@ -83,10 +83,10 @@ enum
ak_ALU(BIC),
ak_ALU(MVN),
- ak_ALU(TST),
- ak_ALU(TEQ),
- ak_ALU(CMP),
- ak_ALU(CMN),
+ ak_Test(TST),
+ ak_Test(TEQ),
+ ak_Test(CMP),
+ ak_Test(CMN),
ak_MUL,
ak_MLA,
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index d88638a..662ed5c 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -34,6 +34,7 @@ add_library(core STATIC
ARMJIT.cpp
ARMJIT_x64/ARMJIT_Compiler.cpp
ARMJIT_x64/ARMJIT_ALU.cpp
+ ARMJIT_x64/ARMJIT_LoadStore.cpp
dolphin/CommonFuncs.cpp
dolphin/x64ABI.cpp
diff --git a/src/dolphin/x64ABI.h b/src/dolphin/x64ABI.h
index 997782e..94336d0 100644
--- a/src/dolphin/x64ABI.h
+++ b/src/dolphin/x64ABI.h
@@ -37,7 +37,8 @@
// xmm0-xmm15 use the upper 16 bits in the functions that push/pop registers.
#define ABI_ALL_CALLER_SAVED \
- (BitSet32{RAX, RCX, RDX, R8, R9, R10, R11})
+ (BitSet32{RAX, RCX, RDX, R8, R9, R10, R11, XMM0 + 16, XMM1 + 16, XMM2 + 16, XMM3 + 16, \
+ XMM4 + 16, XMM5 + 16})
#else // 64-bit Unix / OS X
#define ABI_PARAM1 RDI