/* Copyright 2016-2023 melonDS team This file is part of melonDS. melonDS is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. melonDS is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with melonDS. If not, see http://www.gnu.org/licenses/. */ #include "ARMJIT.h" #include #include #include #define XXH_STATIC_LINKING_ONLY #include "xxhash/xxhash.h" #include "Platform.h" #include "ARMJIT_Internal.h" #include "ARMJIT_Memory.h" #include "ARMJIT_Compiler.h" #include "ARMInterpreter_ALU.h" #include "ARMInterpreter_LoadStore.h" #include "ARMInterpreter_Branch.h" #include "ARMInterpreter.h" #include "DSi.h" #include "GPU.h" #include "GPU3D.h" #include "SPU.h" #include "Wifi.h" #include "NDSCart.h" #include "Platform.h" using Platform::Log; using Platform::LogLevel; #include "ARMJIT_x64/ARMJIT_Offsets.h" static_assert(offsetof(ARM, CPSR) == ARM_CPSR_offset, ""); static_assert(offsetof(ARM, Cycles) == ARM_Cycles_offset, ""); static_assert(offsetof(ARM, StopExecution) == ARM_StopExecution_offset, ""); namespace ARMJIT { #define JIT_DEBUGPRINT(msg, ...) //#define JIT_DEBUGPRINT(msg, ...) Platform::Log(Platform::LogLevel::Debug, msg, ## __VA_ARGS__) Compiler* JITCompiler; int MaxBlockSize; bool LiteralOptimizations; bool BranchOptimizations; bool FastMemory; std::unordered_map JitBlocks9; std::unordered_map JitBlocks7; std::unordered_map RestoreCandidates; TinyVector InvalidLiterals; AddressRange CodeIndexITCM[ITCMPhysicalSize / 512]; AddressRange CodeIndexMainRAM[NDS::MainRAMMaxSize / 512]; AddressRange CodeIndexSWRAM[NDS::SharedWRAMSize / 512]; AddressRange CodeIndexVRAM[0x100000 / 512]; AddressRange CodeIndexARM9BIOS[sizeof(NDS::ARM9BIOS) / 512]; AddressRange CodeIndexARM7BIOS[sizeof(NDS::ARM7BIOS) / 512]; AddressRange CodeIndexARM7WRAM[NDS::ARM7WRAMSize / 512]; AddressRange CodeIndexARM7WVRAM[0x40000 / 512]; AddressRange CodeIndexBIOS9DSi[0x10000 / 512]; AddressRange CodeIndexBIOS7DSi[0x10000 / 512]; AddressRange CodeIndexNWRAM_A[DSi::NWRAMSize / 512]; AddressRange CodeIndexNWRAM_B[DSi::NWRAMSize / 512]; AddressRange CodeIndexNWRAM_C[DSi::NWRAMSize / 512]; u64 FastBlockLookupITCM[ITCMPhysicalSize / 2]; u64 FastBlockLookupMainRAM[NDS::MainRAMMaxSize / 2]; u64 FastBlockLookupSWRAM[NDS::SharedWRAMSize / 2]; u64 FastBlockLookupVRAM[0x100000 / 2]; u64 FastBlockLookupARM9BIOS[sizeof(NDS::ARM9BIOS) / 2]; u64 FastBlockLookupARM7BIOS[sizeof(NDS::ARM7BIOS) / 2]; u64 FastBlockLookupARM7WRAM[NDS::ARM7WRAMSize / 2]; u64 FastBlockLookupARM7WVRAM[0x40000 / 2]; u64 FastBlockLookupBIOS9DSi[0x10000 / 2]; u64 FastBlockLookupBIOS7DSi[0x10000 / 2]; u64 FastBlockLookupNWRAM_A[DSi::NWRAMSize / 2]; u64 FastBlockLookupNWRAM_B[DSi::NWRAMSize / 2]; u64 FastBlockLookupNWRAM_C[DSi::NWRAMSize / 2]; const u32 CodeRegionSizes[ARMJIT_Memory::memregions_Count] = { 0, ITCMPhysicalSize, 0, sizeof(NDS::ARM9BIOS), NDS::MainRAMMaxSize, NDS::SharedWRAMSize, 0, 0x100000, sizeof(NDS::ARM7BIOS), NDS::ARM7WRAMSize, 0, 0, 0x40000, 0x10000, 0x10000, DSi::NWRAMSize, DSi::NWRAMSize, DSi::NWRAMSize, }; AddressRange* const CodeMemRegions[ARMJIT_Memory::memregions_Count] = { NULL, CodeIndexITCM, NULL, CodeIndexARM9BIOS, CodeIndexMainRAM, CodeIndexSWRAM, NULL, CodeIndexVRAM, CodeIndexARM7BIOS, CodeIndexARM7WRAM, NULL, NULL, CodeIndexARM7WVRAM, CodeIndexBIOS9DSi, CodeIndexBIOS7DSi, CodeIndexNWRAM_A, CodeIndexNWRAM_B, CodeIndexNWRAM_C }; u64* const FastBlockLookupRegions[ARMJIT_Memory::memregions_Count] = { NULL, FastBlockLookupITCM, NULL, FastBlockLookupARM9BIOS, FastBlockLookupMainRAM, FastBlockLookupSWRAM, NULL, FastBlockLookupVRAM, FastBlockLookupARM7BIOS, FastBlockLookupARM7WRAM, NULL, NULL, FastBlockLookupARM7WVRAM, FastBlockLookupBIOS9DSi, FastBlockLookupBIOS7DSi, FastBlockLookupNWRAM_A, FastBlockLookupNWRAM_B, FastBlockLookupNWRAM_C }; u32 LocaliseCodeAddress(u32 num, u32 addr) { int region = num == 0 ? ARMJIT_Memory::ClassifyAddress9(addr) : ARMJIT_Memory::ClassifyAddress7(addr); if (CodeMemRegions[region]) return ARMJIT_Memory::LocaliseAddress(region, num, addr); return 0; } template T SlowRead9(u32 addr, ARMv5* cpu) { u32 offset = addr & 0x3; addr &= ~(sizeof(T) - 1); T val; if (addr < cpu->ITCMSize) val = *(T*)&cpu->ITCM[addr & 0x7FFF]; else if ((addr & cpu->DTCMMask) == cpu->DTCMBase) val = *(T*)&cpu->DTCM[addr & 0x3FFF]; else if (std::is_same::value) val = (ConsoleType == 0 ? NDS::ARM9Read32 : DSi::ARM9Read32)(addr); else if (std::is_same::value) val = (ConsoleType == 0 ? NDS::ARM9Read16 : DSi::ARM9Read16)(addr); else val = (ConsoleType == 0 ? NDS::ARM9Read8 : DSi::ARM9Read8)(addr); if (std::is_same::value) return ROR(val, offset << 3); else return val; } template void SlowWrite9(u32 addr, ARMv5* cpu, u32 val) { addr &= ~(sizeof(T) - 1); if (addr < cpu->ITCMSize) { CheckAndInvalidate<0, ARMJIT_Memory::memregion_ITCM>(addr); *(T*)&cpu->ITCM[addr & 0x7FFF] = val; } else if ((addr & cpu->DTCMMask) == cpu->DTCMBase) { *(T*)&cpu->DTCM[addr & 0x3FFF] = val; } else if (std::is_same::value) { (ConsoleType == 0 ? NDS::ARM9Write32 : DSi::ARM9Write32)(addr, val); } else if (std::is_same::value) { (ConsoleType == 0 ? NDS::ARM9Write16 : DSi::ARM9Write16)(addr, val); } else { (ConsoleType == 0 ? NDS::ARM9Write8 : DSi::ARM9Write8)(addr, val); } } template T SlowRead7(u32 addr) { u32 offset = addr & 0x3; addr &= ~(sizeof(T) - 1); T val; if (std::is_same::value) val = (ConsoleType == 0 ? NDS::ARM7Read32 : DSi::ARM7Read32)(addr); else if (std::is_same::value) val = (ConsoleType == 0 ? NDS::ARM7Read16 : DSi::ARM7Read16)(addr); else val = (ConsoleType == 0 ? NDS::ARM7Read8 : DSi::ARM7Read8)(addr); if (std::is_same::value) return ROR(val, offset << 3); else return val; } template void SlowWrite7(u32 addr, u32 val) { addr &= ~(sizeof(T) - 1); if (std::is_same::value) (ConsoleType == 0 ? NDS::ARM7Write32 : DSi::ARM7Write32)(addr, val); else if (std::is_same::value) (ConsoleType == 0 ? NDS::ARM7Write16 : DSi::ARM7Write16)(addr, val); else (ConsoleType == 0 ? NDS::ARM7Write8 : DSi::ARM7Write8)(addr, val); } template void SlowBlockTransfer9(u32 addr, u64* data, u32 num, ARMv5* cpu) { addr &= ~0x3; for (u32 i = 0; i < num; i++) { if (Write) SlowWrite9(addr, cpu, data[i]); else data[i] = SlowRead9(addr, cpu); addr += 4; } } template void SlowBlockTransfer7(u32 addr, u64* data, u32 num) { addr &= ~0x3; for (u32 i = 0; i < num; i++) { if (Write) SlowWrite7(addr, data[i]); else data[i] = SlowRead7(addr); addr += 4; } } #define INSTANTIATE_SLOWMEM(consoleType) \ template void SlowWrite9(u32, ARMv5*, u32); \ template void SlowWrite9(u32, ARMv5*, u32); \ template void SlowWrite9(u32, ARMv5*, u32); \ \ template u32 SlowRead9(u32, ARMv5*); \ template u16 SlowRead9(u32, ARMv5*); \ template u8 SlowRead9(u32, ARMv5*); \ \ template void SlowWrite7(u32, u32); \ template void SlowWrite7(u32, u32); \ template void SlowWrite7(u32, u32); \ \ template u32 SlowRead7(u32); \ template u16 SlowRead7(u32); \ template u8 SlowRead7(u32); \ \ template void SlowBlockTransfer9(u32, u64*, u32, ARMv5*); \ template void SlowBlockTransfer9(u32, u64*, u32, ARMv5*); \ template void SlowBlockTransfer7(u32 addr, u64* data, u32 num); \ template void SlowBlockTransfer7(u32 addr, u64* data, u32 num); \ INSTANTIATE_SLOWMEM(0) INSTANTIATE_SLOWMEM(1) void Init() { JITCompiler = new Compiler(); ARMJIT_Memory::Init(); } void DeInit() { JitEnableWrite(); ResetBlockCache(); ARMJIT_Memory::DeInit(); delete JITCompiler; JITCompiler = nullptr; } void Reset() { MaxBlockSize = Platform::GetConfigInt(Platform::JIT_MaxBlockSize); LiteralOptimizations = Platform::GetConfigBool(Platform::JIT_LiteralOptimizations); BranchOptimizations = Platform::GetConfigBool(Platform::JIT_BranchOptimizations); FastMemory = Platform::GetConfigBool(Platform::JIT_FastMemory); if (MaxBlockSize < 1) MaxBlockSize = 1; if (MaxBlockSize > 32) MaxBlockSize = 32; JitEnableWrite(); ResetBlockCache(); ARMJIT_Memory::Reset(); } void FloodFillSetFlags(FetchedInstr instrs[], int start, u8 flags) { for (int j = start; j >= 0; j--) { u8 match = instrs[j].Info.WriteFlags & flags; u8 matchMaybe = (instrs[j].Info.WriteFlags >> 4) & flags; if (matchMaybe) // writes flags maybe instrs[j].SetFlags |= matchMaybe; if (match) { instrs[j].SetFlags |= match; flags &= ~match; if (!flags) return; } } } bool DecodeLiteral(bool thumb, const FetchedInstr& instr, u32& addr) { if (!thumb) { switch (instr.Info.Kind) { case ARMInstrInfo::ak_LDR_IMM: case ARMInstrInfo::ak_LDRB_IMM: addr = (instr.Addr + 8) + ((instr.Instr & 0xFFF) * (instr.Instr & (1 << 23) ? 1 : -1)); return true; case ARMInstrInfo::ak_LDRH_IMM: addr = (instr.Addr + 8) + (((instr.Instr & 0xF00) >> 4 | (instr.Instr & 0xF)) * (instr.Instr & (1 << 23) ? 1 : -1)); return true; default: break; } } else if (instr.Info.Kind == ARMInstrInfo::tk_LDR_PCREL) { addr = ((instr.Addr + 4) & ~0x2) + ((instr.Instr & 0xFF) << 2); return true; } JIT_DEBUGPRINT("Literal %08x %x not recognised %d\n", instr.Instr, instr.Addr, instr.Info.Kind); return false; } bool DecodeBranch(bool thumb, const FetchedInstr& instr, u32& cond, bool hasLink, u32 lr, bool& link, u32& linkAddr, u32& targetAddr) { if (thumb) { u32 r15 = instr.Addr + 4; cond = 0xE; link = instr.Info.Kind == ARMInstrInfo::tk_BL_LONG; linkAddr = instr.Addr + 4; if (instr.Info.Kind == ARMInstrInfo::tk_BL_LONG && !(instr.Instr & (1 << 12))) { targetAddr = r15 + ((s32)((instr.Instr & 0x7FF) << 21) >> 9); targetAddr += ((instr.Instr >> 16) & 0x7FF) << 1; return true; } else if (instr.Info.Kind == ARMInstrInfo::tk_B) { s32 offset = (s32)((instr.Instr & 0x7FF) << 21) >> 20; targetAddr = r15 + offset; return true; } else if (instr.Info.Kind == ARMInstrInfo::tk_BCOND) { cond = (instr.Instr >> 8) & 0xF; s32 offset = (s32)(instr.Instr << 24) >> 23; targetAddr = r15 + offset; return true; } else if (hasLink && instr.Info.Kind == ARMInstrInfo::tk_BX && instr.A_Reg(3) == 14) { JIT_DEBUGPRINT("returning!\n"); targetAddr = lr; return true; } } else { link = instr.Info.Kind == ARMInstrInfo::ak_BL; linkAddr = instr.Addr + 4; cond = instr.Cond(); if (instr.Info.Kind == ARMInstrInfo::ak_BL || instr.Info.Kind == ARMInstrInfo::ak_B) { s32 offset = (s32)(instr.Instr << 8) >> 6; u32 r15 = instr.Addr + 8; targetAddr = r15 + offset; return true; } else if (hasLink && instr.Info.Kind == ARMInstrInfo::ak_BX && instr.A_Reg(0) == 14) { JIT_DEBUGPRINT("returning!\n"); targetAddr = lr; return true; } } return false; } bool IsIdleLoop(bool thumb, FetchedInstr* instrs, int instrsCount) { // see https://github.com/dolphin-emu/dolphin/blob/master/Source/Core/Core/PowerPC/PPCAnalyst.cpp#L678 // it basically checks if one iteration of a loop depends on another // the rules are quite simple JIT_DEBUGPRINT("checking potential idle loop\n"); u16 regsWrittenTo = 0; u16 regsDisallowedToWrite = 0; for (int i = 0; i < instrsCount; i++) { JIT_DEBUGPRINT("instr %d %08x regs(%x %x) %x %x\n", i, instrs[i].Instr, instrs[i].Info.DstRegs, instrs[i].Info.SrcRegs, regsWrittenTo, regsDisallowedToWrite); if (instrs[i].Info.SpecialKind == ARMInstrInfo::special_WriteMem) return false; if (!thumb && instrs[i].Info.Kind >= ARMInstrInfo::ak_MSR_IMM && instrs[i].Info.Kind <= ARMInstrInfo::ak_MRC) return false; if (i < instrsCount - 1 && instrs[i].Info.Branches()) return false; u16 srcRegs = instrs[i].Info.SrcRegs & ~(1 << 15); u16 dstRegs = instrs[i].Info.DstRegs & ~(1 << 15); regsDisallowedToWrite |= srcRegs & ~regsWrittenTo; if (dstRegs & regsDisallowedToWrite) return false; regsWrittenTo |= dstRegs; } return true; } typedef void (*InterpreterFunc)(ARM* cpu); void NOP(ARM* cpu) {} #define F(x) &ARMInterpreter::A_##x #define F_ALU(name, s) \ F(name##_REG_LSL_IMM##s), F(name##_REG_LSR_IMM##s), F(name##_REG_ASR_IMM##s), F(name##_REG_ROR_IMM##s), \ F(name##_REG_LSL_REG##s), F(name##_REG_LSR_REG##s), F(name##_REG_ASR_REG##s), F(name##_REG_ROR_REG##s), F(name##_IMM##s) #define F_MEM_WB(name) \ F(name##_REG_LSL), F(name##_REG_LSR), F(name##_REG_ASR), F(name##_REG_ROR), F(name##_IMM), \ F(name##_POST_REG_LSL), F(name##_POST_REG_LSR), F(name##_POST_REG_ASR), F(name##_POST_REG_ROR), F(name##_POST_IMM) #define F_MEM_HD(name) \ F(name##_REG), F(name##_IMM), F(name##_POST_REG), F(name##_POST_IMM) InterpreterFunc InterpretARM[ARMInstrInfo::ak_Count] = { F_ALU(AND,), F_ALU(AND,_S), F_ALU(EOR,), F_ALU(EOR,_S), F_ALU(SUB,), F_ALU(SUB,_S), F_ALU(RSB,), F_ALU(RSB,_S), F_ALU(ADD,), F_ALU(ADD,_S), F_ALU(ADC,), F_ALU(ADC,_S), F_ALU(SBC,), F_ALU(SBC,_S), F_ALU(RSC,), F_ALU(RSC,_S), F_ALU(ORR,), F_ALU(ORR,_S), F_ALU(MOV,), F_ALU(MOV,_S), F_ALU(BIC,), F_ALU(BIC,_S), F_ALU(MVN,), F_ALU(MVN,_S), F_ALU(TST,), F_ALU(TEQ,), F_ALU(CMP,), F_ALU(CMN,), F(MUL), F(MLA), F(UMULL), F(UMLAL), F(SMULL), F(SMLAL), F(SMLAxy), F(SMLAWy), F(SMULWy), F(SMLALxy), F(SMULxy), F(CLZ), F(QADD), F(QSUB), F(QDADD), F(QDSUB), F_MEM_WB(STR), F_MEM_WB(STRB), F_MEM_WB(LDR), F_MEM_WB(LDRB), F_MEM_HD(STRH), F_MEM_HD(LDRD), F_MEM_HD(STRD), F_MEM_HD(LDRH), F_MEM_HD(LDRSB), F_MEM_HD(LDRSH), F(SWP), F(SWPB), F(LDM), F(STM), F(B), F(BL), F(BLX_IMM), F(BX), F(BLX_REG), F(UNK), F(MSR_IMM), F(MSR_REG), F(MRS), F(MCR), F(MRC), F(SVC), NOP }; #undef F_ALU #undef F_MEM_WB #undef F_MEM_HD #undef F void T_BL_LONG(ARM* cpu) { ARMInterpreter::T_BL_LONG_1(cpu); cpu->R[15] += 2; ARMInterpreter::T_BL_LONG_2(cpu); } #define F(x) ARMInterpreter::T_##x InterpreterFunc InterpretTHUMB[ARMInstrInfo::tk_Count] = { F(LSL_IMM), F(LSR_IMM), F(ASR_IMM), F(ADD_REG_), F(SUB_REG_), F(ADD_IMM_), F(SUB_IMM_), F(MOV_IMM), F(CMP_IMM), F(ADD_IMM), F(SUB_IMM), F(AND_REG), F(EOR_REG), F(LSL_REG), F(LSR_REG), F(ASR_REG), F(ADC_REG), F(SBC_REG), F(ROR_REG), F(TST_REG), F(NEG_REG), F(CMP_REG), F(CMN_REG), F(ORR_REG), F(MUL_REG), F(BIC_REG), F(MVN_REG), F(ADD_HIREG), F(CMP_HIREG), F(MOV_HIREG), F(ADD_PCREL), F(ADD_SPREL), F(ADD_SP), F(LDR_PCREL), F(STR_REG), F(STRB_REG), F(LDR_REG), F(LDRB_REG), F(STRH_REG), F(LDRSB_REG), F(LDRH_REG), F(LDRSH_REG), F(STR_IMM), F(LDR_IMM), F(STRB_IMM), F(LDRB_IMM), F(STRH_IMM), F(LDRH_IMM), F(STR_SPREL), F(LDR_SPREL), F(PUSH), F(POP), F(LDMIA), F(STMIA), F(BCOND), F(BX), F(BLX_REG), F(B), F(BL_LONG_1), F(BL_LONG_2), F(UNK), F(SVC), T_BL_LONG // BL_LONG psudo opcode }; #undef F void RetireJitBlock(JitBlock* block) { auto it = RestoreCandidates.find(block->InstrHash); if (it != RestoreCandidates.end()) { delete it->second; it->second = block; } else { RestoreCandidates[block->InstrHash] = block; } } void CompileBlock(ARM* cpu) { bool thumb = cpu->CPSR & 0x20; u32 blockAddr = cpu->R[15] - (thumb ? 2 : 4); u32 localAddr = LocaliseCodeAddress(cpu->Num, blockAddr); if (!localAddr) { Log(LogLevel::Warn, "trying to compile non executable code? %x\n", blockAddr); } auto& map = cpu->Num == 0 ? JitBlocks9 : JitBlocks7; auto existingBlockIt = map.find(blockAddr); if (existingBlockIt != map.end()) { // there's already a block, though it's not inside the fast map // could be that there are two blocks at the same physical addr // but different mirrors u32 otherLocalAddr = existingBlockIt->second->StartAddrLocal; if (localAddr == otherLocalAddr) { JIT_DEBUGPRINT("switching out block %x %x %x\n", localAddr, blockAddr, existingBlockIt->second->StartAddr); u64* entry = &FastBlockLookupRegions[localAddr >> 27][(localAddr & 0x7FFFFFF) / 2]; *entry = ((u64)blockAddr | cpu->Num) << 32; *entry |= JITCompiler->SubEntryOffset(existingBlockIt->second->EntryPoint); return; } // some memory has been remapped RetireJitBlock(existingBlockIt->second); map.erase(existingBlockIt); } FetchedInstr instrs[MaxBlockSize]; int i = 0; u32 r15 = cpu->R[15]; u32 addressRanges[MaxBlockSize]; u32 addressMasks[MaxBlockSize]; memset(addressMasks, 0, MaxBlockSize * sizeof(u32)); u32 numAddressRanges = 0; u32 numLiterals = 0; u32 literalLoadAddrs[MaxBlockSize]; // they are going to be hashed u32 literalValues[MaxBlockSize]; u32 instrValues[MaxBlockSize]; // due to instruction merging i might not reflect the amount of actual instructions u32 numInstrs = 0; u32 writeAddrs[MaxBlockSize]; u32 numWriteAddrs = 0, writeAddrsTranslated = 0; cpu->FillPipeline(); u32 nextInstr[2] = {cpu->NextInstr[0], cpu->NextInstr[1]}; u32 nextInstrAddr[2] = {blockAddr, r15}; JIT_DEBUGPRINT("start block %x %08x (%x)\n", blockAddr, cpu->CPSR, localAddr); u32 lastSegmentStart = blockAddr; u32 lr; bool hasLink = false; bool hasMemoryInstr = false; do { r15 += thumb ? 2 : 4; instrs[i].BranchFlags = 0; instrs[i].SetFlags = 0; instrs[i].Instr = nextInstr[0]; nextInstr[0] = nextInstr[1]; instrs[i].Addr = nextInstrAddr[0]; nextInstrAddr[0] = nextInstrAddr[1]; nextInstrAddr[1] = r15; JIT_DEBUGPRINT("instr %08x %x\n", instrs[i].Instr & (thumb ? 0xFFFF : ~0), instrs[i].Addr); instrValues[numInstrs++] = instrs[i].Instr; u32 translatedAddr = LocaliseCodeAddress(cpu->Num, instrs[i].Addr); assert(translatedAddr >> 27); u32 translatedAddrRounded = translatedAddr & ~0x1FF; if (i == 0 || translatedAddrRounded != addressRanges[numAddressRanges - 1]) { bool returning = false; for (u32 j = 0; j < numAddressRanges; j++) { if (addressRanges[j] == translatedAddrRounded) { std::swap(addressRanges[j], addressRanges[numAddressRanges - 1]); std::swap(addressMasks[j], addressMasks[numAddressRanges - 1]); returning = true; break; } } if (!returning) addressRanges[numAddressRanges++] = translatedAddrRounded; } addressMasks[numAddressRanges - 1] |= 1 << ((translatedAddr & 0x1FF) / 16); if (cpu->Num == 0) { ARMv5* cpuv5 = (ARMv5*)cpu; if (thumb && r15 & 0x2) { nextInstr[1] >>= 16; instrs[i].CodeCycles = 0; } else { nextInstr[1] = cpuv5->CodeRead32(r15, false); instrs[i].CodeCycles = cpu->CodeCycles; } } else { ARMv4* cpuv4 = (ARMv4*)cpu; if (thumb) nextInstr[1] = cpuv4->CodeRead16(r15); else nextInstr[1] = cpuv4->CodeRead32(r15); instrs[i].CodeCycles = cpu->CodeCycles; } instrs[i].Info = ARMInstrInfo::Decode(thumb, cpu->Num, instrs[i].Instr); hasMemoryInstr |= thumb ? (instrs[i].Info.Kind >= ARMInstrInfo::tk_LDR_PCREL && instrs[i].Info.Kind <= ARMInstrInfo::tk_STMIA) : (instrs[i].Info.Kind >= ARMInstrInfo::ak_STR_REG_LSL && instrs[i].Info.Kind <= ARMInstrInfo::ak_STM); cpu->R[15] = r15; cpu->CurInstr = instrs[i].Instr; cpu->CodeCycles = instrs[i].CodeCycles; if (instrs[i].Info.DstRegs & (1 << 14) || (!thumb && (instrs[i].Info.Kind == ARMInstrInfo::ak_MSR_IMM || instrs[i].Info.Kind == ARMInstrInfo::ak_MSR_REG) && instrs[i].Instr & (1 << 16))) hasLink = false; if (thumb) { InterpretTHUMB[instrs[i].Info.Kind](cpu); } else { if (cpu->Num == 0 && instrs[i].Info.Kind == ARMInstrInfo::ak_BLX_IMM) { ARMInterpreter::A_BLX_IMM(cpu); } else { u32 icode = ((instrs[i].Instr >> 4) & 0xF) | ((instrs[i].Instr >> 16) & 0xFF0); assert(InterpretARM[instrs[i].Info.Kind] == ARMInterpreter::ARMInstrTable[icode] || instrs[i].Info.Kind == ARMInstrInfo::ak_MOV_REG_LSL_IMM || instrs[i].Info.Kind == ARMInstrInfo::ak_Nop || instrs[i].Info.Kind == ARMInstrInfo::ak_UNK); if (cpu->CheckCondition(instrs[i].Cond())) InterpretARM[instrs[i].Info.Kind](cpu); else cpu->AddCycles_C(); } } instrs[i].DataCycles = cpu->DataCycles; instrs[i].DataRegion = cpu->DataRegion; u32 literalAddr; if (LiteralOptimizations && instrs[i].Info.SpecialKind == ARMInstrInfo::special_LoadLiteral && DecodeLiteral(thumb, instrs[i], literalAddr)) { u32 translatedAddr = LocaliseCodeAddress(cpu->Num, literalAddr); if (!translatedAddr) { Log(LogLevel::Warn,"literal in non executable memory?\n"); } if (InvalidLiterals.Find(translatedAddr) == -1) { u32 translatedAddrRounded = translatedAddr & ~0x1FF; u32 j = 0; for (; j < numAddressRanges; j++) if (addressRanges[j] == translatedAddrRounded) break; if (j == numAddressRanges) addressRanges[numAddressRanges++] = translatedAddrRounded; addressMasks[j] |= 1 << ((translatedAddr & 0x1FF) / 16); JIT_DEBUGPRINT("literal loading %08x %08x %08x %08x\n", literalAddr, translatedAddr, addressMasks[j], addressRanges[j]); cpu->DataRead32(literalAddr, &literalValues[numLiterals]); literalLoadAddrs[numLiterals++] = translatedAddr; } } else if (instrs[i].Info.SpecialKind == ARMInstrInfo::special_WriteMem) writeAddrs[numWriteAddrs++] = instrs[i].DataRegion; else if (thumb && instrs[i].Info.Kind == ARMInstrInfo::tk_BL_LONG_2 && i > 0 && instrs[i - 1].Info.Kind == ARMInstrInfo::tk_BL_LONG_1) { i--; instrs[i].Info.Kind = ARMInstrInfo::tk_BL_LONG; instrs[i].Instr = (instrs[i].Instr & 0xFFFF) | (instrs[i + 1].Instr << 16); instrs[i].Info.DstRegs = 0xC000; instrs[i].Info.SrcRegs = 0; instrs[i].Info.EndBlock = true; JIT_DEBUGPRINT("merged BL\n"); } if (instrs[i].Info.Branches() && BranchOptimizations && instrs[i].Info.Kind != (thumb ? ARMInstrInfo::tk_SVC : ARMInstrInfo::ak_SVC)) { bool hasBranched = cpu->R[15] != r15; bool link; u32 cond, target, linkAddr; bool staticBranch = DecodeBranch(thumb, instrs[i], cond, hasLink, lr, link, linkAddr, target); JIT_DEBUGPRINT("branch cond %x target %x (%d)\n", cond, target, hasBranched); if (staticBranch) { instrs[i].BranchFlags |= branch_StaticTarget; bool isBackJump = false; if (hasBranched) { for (int j = 0; j < i; j++) { if (instrs[j].Addr == target) { isBackJump = true; break; } } } if (cond < 0xE && target < instrs[i].Addr && target >= lastSegmentStart) { // we might have an idle loop u32 backwardsOffset = (instrs[i].Addr - target) / (thumb ? 2 : 4); if (IsIdleLoop(thumb, &instrs[i - backwardsOffset], backwardsOffset + 1)) { instrs[i].BranchFlags |= branch_IdleBranch; JIT_DEBUGPRINT("found %s idle loop %d in block %08x\n", thumb ? "thumb" : "arm", cpu->Num, blockAddr); } } else if (hasBranched && !isBackJump && i + 1 < MaxBlockSize) { if (link) { lr = linkAddr; hasLink = true; } r15 = target + (thumb ? 2 : 4); assert(r15 == cpu->R[15]); JIT_DEBUGPRINT("block lengthened by static branch (target %x)\n", target); nextInstr[0] = cpu->NextInstr[0]; nextInstr[1] = cpu->NextInstr[1]; nextInstrAddr[0] = target; nextInstrAddr[1] = r15; lastSegmentStart = target; instrs[i].Info.EndBlock = false; if (cond < 0xE) instrs[i].BranchFlags |= branch_FollowCondTaken; } } if (!hasBranched && cond < 0xE && i + 1 < MaxBlockSize) { JIT_DEBUGPRINT("block lengthened by untaken branch\n"); instrs[i].Info.EndBlock = false; instrs[i].BranchFlags |= branch_FollowCondNotTaken; } } i++; bool canCompile = JITCompiler->CanCompile(thumb, instrs[i - 1].Info.Kind); bool secondaryFlagReadCond = !canCompile || (instrs[i - 1].BranchFlags & (branch_FollowCondTaken | branch_FollowCondNotTaken)); if (instrs[i - 1].Info.ReadFlags != 0 || secondaryFlagReadCond) FloodFillSetFlags(instrs, i - 2, !secondaryFlagReadCond ? instrs[i - 1].Info.ReadFlags : 0xF); } while(!instrs[i - 1].Info.EndBlock && i < MaxBlockSize && !cpu->Halted && (!cpu->IRQ || (cpu->CPSR & 0x80))); if (numLiterals) { for (u32 j = 0; j < numWriteAddrs; j++) { u32 translatedAddr = LocaliseCodeAddress(cpu->Num, writeAddrs[j]); if (translatedAddr) { for (u32 k = 0; k < numLiterals; k++) { if (literalLoadAddrs[k] == translatedAddr) { if (InvalidLiterals.Find(translatedAddr) == -1) InvalidLiterals.Add(translatedAddr); break; } } } } } u32 literalHash = (u32)XXH3_64bits(literalValues, numLiterals * 4); u32 instrHash = (u32)XXH3_64bits(instrValues, numInstrs * 4); auto prevBlockIt = RestoreCandidates.find(instrHash); JitBlock* prevBlock = NULL; bool mayRestore = true; if (prevBlockIt != RestoreCandidates.end()) { prevBlock = prevBlockIt->second; RestoreCandidates.erase(prevBlockIt); mayRestore = prevBlock->StartAddr == blockAddr && prevBlock->LiteralHash == literalHash; if (mayRestore && prevBlock->NumAddresses == numAddressRanges) { for (u32 j = 0; j < numAddressRanges; j++) { if (prevBlock->AddressRanges()[j] != addressRanges[j] || prevBlock->AddressMasks()[j] != addressMasks[j]) { mayRestore = false; break; } } } else mayRestore = false; } else { mayRestore = false; } JitBlock* block; if (!mayRestore) { if (prevBlock) delete prevBlock; block = new JitBlock(cpu->Num, i, numAddressRanges, numLiterals); block->LiteralHash = literalHash; block->InstrHash = instrHash; for (u32 j = 0; j < numAddressRanges; j++) block->AddressRanges()[j] = addressRanges[j]; for (u32 j = 0; j < numAddressRanges; j++) block->AddressMasks()[j] = addressMasks[j]; for (int j = 0; j < numLiterals; j++) block->Literals()[j] = literalLoadAddrs[j]; block->StartAddr = blockAddr; block->StartAddrLocal = localAddr; FloodFillSetFlags(instrs, i - 1, 0xF); JitEnableWrite(); block->EntryPoint = JITCompiler->CompileBlock(cpu, thumb, instrs, i, hasMemoryInstr); JitEnableExecute(); JIT_DEBUGPRINT("block start %p\n", block->EntryPoint); } else { JIT_DEBUGPRINT("restored! %p\n", prevBlock); block = prevBlock; } assert((localAddr & 1) == 0); for (u32 j = 0; j < numAddressRanges; j++) { assert(addressRanges[j] == block->AddressRanges()[j]); assert(addressMasks[j] == block->AddressMasks()[j]); assert(addressMasks[j] != 0); AddressRange* region = CodeMemRegions[addressRanges[j] >> 27]; if (!PageContainsCode(®ion[(addressRanges[j] & 0x7FFF000) / 512])) ARMJIT_Memory::SetCodeProtection(addressRanges[j] >> 27, addressRanges[j] & 0x7FFFFFF, true); AddressRange* range = ®ion[(addressRanges[j] & 0x7FFFFFF) / 512]; range->Code |= addressMasks[j]; range->Blocks.Add(block); } if (cpu->Num == 0) JitBlocks9[blockAddr] = block; else JitBlocks7[blockAddr] = block; u64* entry = &FastBlockLookupRegions[(localAddr >> 27)][(localAddr & 0x7FFFFFF) / 2]; *entry = ((u64)blockAddr | cpu->Num) << 32; *entry |= JITCompiler->SubEntryOffset(block->EntryPoint); } void InvalidateByAddr(u32 localAddr) { JIT_DEBUGPRINT("invalidating by addr %x\n", localAddr); AddressRange* region = CodeMemRegions[localAddr >> 27]; AddressRange* range = ®ion[(localAddr & 0x7FFFFFF) / 512]; u32 mask = 1 << ((localAddr & 0x1FF) / 16); range->Code = 0; for (int i = 0; i < range->Blocks.Length;) { JitBlock* block = range->Blocks[i]; bool invalidated = false; u32 mask = 0; for (int j = 0; j < block->NumAddresses; j++) { if (block->AddressRanges()[j] == (localAddr & ~0x1FF)) { mask = block->AddressMasks()[j]; invalidated = block->AddressMasks()[j] & mask; assert(mask); break; } } assert(mask); if (!invalidated) { range->Code |= mask; i++; continue; } range->Blocks.Remove(i); if (range->Blocks.Length == 0 && !PageContainsCode(®ion[(localAddr & 0x7FFF000) / 512])) { ARMJIT_Memory::SetCodeProtection(localAddr >> 27, localAddr & 0x7FFFFFF, false); } bool literalInvalidation = false; for (int j = 0; j < block->NumLiterals; j++) { u32 addr = block->Literals()[j]; if (addr == localAddr) { if (InvalidLiterals.Find(localAddr) == -1) { InvalidLiterals.Add(localAddr); JIT_DEBUGPRINT("found invalid literal %d\n", InvalidLiterals.Length); } literalInvalidation = true; break; } } for (int j = 0; j < block->NumAddresses; j++) { u32 addr = block->AddressRanges()[j]; if ((addr / 512) != (localAddr / 512)) { AddressRange* otherRegion = CodeMemRegions[addr >> 27]; AddressRange* otherRange = &otherRegion[(addr & 0x7FFFFFF) / 512]; assert(otherRange != range); bool removed = otherRange->Blocks.RemoveByValue(block); assert(removed); if (otherRange->Blocks.Length == 0) { if (!PageContainsCode(&otherRegion[(addr & 0x7FFF000) / 512])) ARMJIT_Memory::SetCodeProtection(addr >> 27, addr & 0x7FFFFFF, false); otherRange->Code = 0; } } } FastBlockLookupRegions[block->StartAddrLocal >> 27][(block->StartAddrLocal & 0x7FFFFFF) / 2] = (u64)UINT32_MAX << 32; if (block->Num == 0) JitBlocks9.erase(block->StartAddr); else JitBlocks7.erase(block->StartAddr); if (!literalInvalidation) { RetireJitBlock(block); } else { delete block; } } } void CheckAndInvalidateITCM() { for (u32 i = 0; i < ITCMPhysicalSize; i+=512) { if (CodeIndexITCM[i / 512].Code) { // maybe using bitscan would be better here? // The thing is that in densely populated sets // The old fashioned way can actually be faster for (u32 j = 0; j < 512; j += 16) { if (CodeIndexITCM[i / 512].Code & (1 << ((j & 0x1FF) / 16))) InvalidateByAddr((i+j) | (ARMJIT_Memory::memregion_ITCM << 27)); } } } } void CheckAndInvalidateWVRAM(int bank) { u32 start = bank == 1 ? 0x20000 : 0; for (u32 i = start; i < start+0x20000; i+=512) { if (CodeIndexARM7WVRAM[i / 512].Code) { for (u32 j = 0; j < 512; j += 16) { if (CodeIndexARM7WVRAM[i / 512].Code & (1 << ((j & 0x1FF) / 16))) InvalidateByAddr((i+j) | (ARMJIT_Memory::memregion_VWRAM << 27)); } } } } template void CheckAndInvalidate(u32 addr) { u32 localAddr = ARMJIT_Memory::LocaliseAddress(region, num, addr); if (CodeMemRegions[region][(localAddr & 0x7FFFFFF) / 512].Code & (1 << ((localAddr & 0x1FF) / 16))) InvalidateByAddr(localAddr); } JitBlockEntry LookUpBlock(u32 num, u64* entries, u32 offset, u32 addr) { u64* entry = &entries[offset / 2]; if (*entry >> 32 == (addr | num)) return JITCompiler->AddEntryOffset((u32)*entry); return NULL; } void blockSanityCheck(u32 num, u32 blockAddr, JitBlockEntry entry) { u32 localAddr = LocaliseCodeAddress(num, blockAddr); assert(JITCompiler->AddEntryOffset((u32)FastBlockLookupRegions[localAddr >> 27][(localAddr & 0x7FFFFFF) / 2]) == entry); } bool SetupExecutableRegion(u32 num, u32 blockAddr, u64*& entry, u32& start, u32& size) { // amazingly ignoring the DTCM is the proper behaviour for code fetches int region = num == 0 ? ARMJIT_Memory::ClassifyAddress9(blockAddr) : ARMJIT_Memory::ClassifyAddress7(blockAddr); u32 memoryOffset; if (FastBlockLookupRegions[region] && ARMJIT_Memory::GetMirrorLocation(region, num, blockAddr, memoryOffset, start, size)) { //printf("setup exec region %d %d %08x %08x %x %x\n", num, region, blockAddr, start, size, memoryOffset); entry = FastBlockLookupRegions[region] + memoryOffset / 2; return true; } return false; } template void CheckAndInvalidate<0, ARMJIT_Memory::memregion_MainRAM>(u32); template void CheckAndInvalidate<1, ARMJIT_Memory::memregion_MainRAM>(u32); template void CheckAndInvalidate<0, ARMJIT_Memory::memregion_SharedWRAM>(u32); template void CheckAndInvalidate<1, ARMJIT_Memory::memregion_SharedWRAM>(u32); template void CheckAndInvalidate<1, ARMJIT_Memory::memregion_WRAM7>(u32); template void CheckAndInvalidate<1, ARMJIT_Memory::memregion_VWRAM>(u32); template void CheckAndInvalidate<0, ARMJIT_Memory::memregion_VRAM>(u32); template void CheckAndInvalidate<0, ARMJIT_Memory::memregion_ITCM>(u32); template void CheckAndInvalidate<0, ARMJIT_Memory::memregion_NewSharedWRAM_A>(u32); template void CheckAndInvalidate<1, ARMJIT_Memory::memregion_NewSharedWRAM_A>(u32); template void CheckAndInvalidate<0, ARMJIT_Memory::memregion_NewSharedWRAM_B>(u32); template void CheckAndInvalidate<1, ARMJIT_Memory::memregion_NewSharedWRAM_B>(u32); template void CheckAndInvalidate<0, ARMJIT_Memory::memregion_NewSharedWRAM_C>(u32); template void CheckAndInvalidate<1, ARMJIT_Memory::memregion_NewSharedWRAM_C>(u32); void ResetBlockCache() { Log(LogLevel::Debug, "Resetting JIT block cache...\n"); // could be replace through a function which only resets // the permissions but we're too lazy ARMJIT_Memory::Reset(); InvalidLiterals.Clear(); for (int i = 0; i < ARMJIT_Memory::memregions_Count; i++) { if (FastBlockLookupRegions[i]) memset(FastBlockLookupRegions[i], 0xFF, CodeRegionSizes[i] * sizeof(u64) / 2); } for (auto it = RestoreCandidates.begin(); it != RestoreCandidates.end(); it++) delete it->second; RestoreCandidates.clear(); for (auto it : JitBlocks9) { JitBlock* block = it.second; for (int j = 0; j < block->NumAddresses; j++) { u32 addr = block->AddressRanges()[j]; AddressRange* range = &CodeMemRegions[addr >> 27][(addr & 0x7FFFFFF) / 512]; range->Blocks.Clear(); range->Code = 0; } delete block; } for (auto it : JitBlocks7) { JitBlock* block = it.second; for (int j = 0; j < block->NumAddresses; j++) { u32 addr = block->AddressRanges()[j]; AddressRange* range = &CodeMemRegions[addr >> 27][(addr & 0x7FFFFFF) / 512]; range->Blocks.Clear(); range->Code = 0; } delete block; } JitBlocks9.clear(); JitBlocks7.clear(); JITCompiler->Reset(); } void JitEnableWrite() { #if defined(__APPLE__) && defined(__aarch64__) if (__builtin_available(macOS 11.0, *)) pthread_jit_write_protect_np(false); #endif } void JitEnableExecute() { #if defined(__APPLE__) && defined(__aarch64__) if (__builtin_available(macOS 11.0, *)) pthread_jit_write_protect_np(true); #endif } }