new block cache and much more...

- more reliable code invalidation detection
- blocks aren't stopped at any branch, but are being followed
if possible to get larger blocks
- idle loop recognition
- optimised literal loads, load/store cycle counting
 and loads/stores from constant addresses
This commit is contained in:
RSDuck
2019-10-03 01:10:59 +02:00
parent 5338c28f40
commit a687be9879
19 changed files with 1557 additions and 696 deletions

View File

@ -72,12 +72,15 @@ Compiler::Compiler()
for (int i = 0; i < 3; i++)
{
for (int j = 0; j < 2; j++)
{
MemoryFuncs9[i][j] = Gen_MemoryRoutine9(j, 8 << i);
MemoryFuncs7[i][j][0] = Gen_MemoryRoutine7(j, false, 8 << i);
MemoryFuncs7[i][j][1] = Gen_MemoryRoutine7(j, true, 8 << i);
}
}
MemoryFuncs7[0][0] = (void*)NDS::ARM7Read8;
MemoryFuncs7[0][1] = (void*)NDS::ARM7Write8;
MemoryFuncs7[1][0] = (void*)NDS::ARM7Read16;
MemoryFuncs7[1][1] = (void*)NDS::ARM7Write16;
MemoryFuncs7[2][0] = (void*)NDS::ARM7Read32;
MemoryFuncs7[2][1] = (void*)NDS::ARM7Write32;
for (int i = 0; i < 2; i++)
for (int j = 0; j < 2; j++)
{
@ -179,12 +182,13 @@ void Compiler::LoadCPSR()
MOV(32, R(RCPSR), MDisp(RCPU, offsetof(ARM, CPSR)));
}
void Compiler::SaveCPSR()
void Compiler::SaveCPSR(bool flagClean)
{
if (CPSRDirty)
{
MOV(32, MDisp(RCPU, offsetof(ARM, CPSR)), R(RCPSR));
CPSRDirty = false;
if (flagClean)
CPSRDirty = false;
}
}
@ -204,6 +208,9 @@ void Compiler::SaveReg(int reg, X64Reg nativeReg)
// invalidates RSCRATCH and RSCRATCH3
Gen::FixupBranch Compiler::CheckCondition(u32 cond)
{
// hack, ldm/stm can get really big TODO: make this better
bool ldmStm = !Thumb &&
(CurInstr.Info.Kind == ARMInstrInfo::ak_LDM || CurInstr.Info.Kind == ARMInstrInfo::ak_STM);
if (cond >= 0x8)
{
static_assert(RSCRATCH3 == ECX, "RSCRATCH has to be equal to ECX!");
@ -213,14 +220,14 @@ Gen::FixupBranch Compiler::CheckCondition(u32 cond)
SHL(32, R(RSCRATCH), R(RSCRATCH3));
TEST(32, R(RSCRATCH), Imm32(ARM::ConditionTable[cond]));
return J_CC(CC_Z);
return J_CC(CC_Z, ldmStm);
}
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);
return J_CC(cond & 1 ? CC_NZ : CC_Z, ldmStm);
}
}
@ -354,25 +361,34 @@ void Compiler::Reset()
SetCodePtr(ResetStart);
}
CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrsCount)
void Compiler::Comp_SpecialBranchBehaviour()
{
if (CurInstr.BranchFlags & branch_IdleBranch)
OR(32, MDisp(RCPU, offsetof(ARM, Halted)), Imm8(0x20));
if (CurInstr.BranchFlags & branch_FollowCondNotTaken)
{
RegCache.PrepareExit();
SaveCPSR(false);
MOV(32, R(RAX), Imm32(ConstantCycles));
ABI_PopRegistersAndAdjustStack(BitSet32(ABI_ALL_CALLEE_SAVED & ABI_ALL_GPRS & ~BitSet32({RSP})), 8);
RET();
}
}
JitBlockEntry Compiler::CompileBlock(ARM* cpu, bool thumb, FetchedInstr instrs[], int instrsCount)
{
if (CodeMemSize - (GetWritableCodePtr() - ResetStart) < 1024 * 32) // guess...
InvalidateBlockCache();
ResetBlockCache();
ConstantCycles = 0;
Thumb = cpu->CPSR & 0x20;
Thumb = thumb;
Num = cpu->Num;
CodeRegion = cpu->CodeRegion;
CodeRegion = instrs[0].Addr >> 24;
CurCPU = cpu;
CompiledBlock res = (CompiledBlock)GetWritableCodePtr();
if (!(Num == 0
? IsMapped<0>(instrs[0].Addr - (Thumb ? 2 : 4))
: IsMapped<1>(instrs[0].Addr - (Thumb ? 2 : 4))))
{
printf("Trying to compile a block in unmapped memory\n");
}
JitBlockEntry res = (JitBlockEntry)GetWritableCodePtr();
ABI_PushRegistersAndAdjustStack(BitSet32(ABI_ALL_CALLEE_SAVED & ABI_ALL_GPRS & ~BitSet32({RSP})), 8);
@ -380,7 +396,6 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
LoadCPSR();
// TODO: this is ugly as a whole, do better
RegCache = RegisterCache<Compiler, X64Reg>(this, instrs, instrsCount);
for (int i = 0; i < instrsCount; i++)
@ -388,21 +403,25 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
CurInstr = instrs[i];
R15 = CurInstr.Addr + (Thumb ? 4 : 8);
Exit = i == instrsCount - 1 || (CurInstr.BranchFlags & branch_FollowCondNotTaken);
CompileFunc comp = Thumb
? T_Comp[CurInstr.Info.Kind]
: A_Comp[CurInstr.Info.Kind];
bool isConditional = Thumb ? CurInstr.Info.Kind == ARMInstrInfo::tk_BCOND : CurInstr.Cond() < 0xE;
if (comp == NULL || (i == instrsCount - 1 && (!CurInstr.Info.Branches() || isConditional)))
if (comp == NULL || (CurInstr.BranchFlags & branch_FollowCondTaken) || (i == instrsCount - 1 && (!CurInstr.Info.Branches() || isConditional)))
{
MOV(32, MDisp(RCPU, offsetof(ARM, R[15])), Imm32(R15));
MOV(32, MDisp(RCPU, offsetof(ARM, CodeCycles)), Imm32(CurInstr.CodeCycles));
MOV(32, MDisp(RCPU, offsetof(ARM, CurInstr)), Imm32(CurInstr.Instr));
if (comp == NULL)
{
MOV(32, MDisp(RCPU, offsetof(ARM, CodeCycles)), Imm32(CurInstr.CodeCycles));
MOV(32, MDisp(RCPU, offsetof(ARM, CurInstr)), Imm32(CurInstr.Instr));
SaveCPSR();
}
}
if (comp != NULL)
RegCache.Prepare(Thumb, i);
else
@ -410,12 +429,11 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
if (Thumb)
{
u32 icode = (CurInstr.Instr >> 6) & 0x3FF;
if (comp == NULL)
{
MOV(64, R(ABI_PARAM1), R(RCPU));
ABI_CallFunction(ARMInterpreter::THUMBInstrTable[icode]);
ABI_CallFunction(InterpretTHUMB[CurInstr.Info.Kind]);
}
else
(this->*comp)();
@ -434,7 +452,9 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
}
}
else if (cond == 0xF)
{
Comp_AddCycles_C();
}
else
{
IrregularCycles = false;
@ -443,25 +463,36 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
if (cond < 0xE)
skipExecute = CheckCondition(cond);
u32 icode = ((CurInstr.Instr >> 4) & 0xF) | ((CurInstr.Instr >> 16) & 0xFF0);
if (comp == NULL)
{
MOV(64, R(ABI_PARAM1), R(RCPU));
ABI_CallFunction(ARMInterpreter::ARMInstrTable[icode]);
ABI_CallFunction(InterpretARM[CurInstr.Info.Kind]);
}
else
(this->*comp)();
Comp_SpecialBranchBehaviour();
if (CurInstr.Cond() < 0xE)
{
if (IrregularCycles)
if (IrregularCycles || (CurInstr.BranchFlags & branch_FollowCondTaken))
{
FixupBranch skipFailed = J();
SetJumpTarget(skipExecute);
Comp_AddCycles_C(true);
if (CurInstr.BranchFlags & branch_FollowCondTaken)
{
RegCache.PrepareExit();
SaveCPSR(false);
MOV(32, R(RAX), Imm32(ConstantCycles));
ABI_PopRegistersAndAdjustStack(BitSet32(ABI_ALL_CALLEE_SAVED & ABI_ALL_GPRS & ~BitSet32({RSP})), 8);
RET();
}
SetJumpTarget(skipFailed);
}
else
@ -483,6 +514,12 @@ CompiledBlock Compiler::CompileBlock(ARM* cpu, FetchedInstr instrs[], int instrs
ABI_PopRegistersAndAdjustStack(BitSet32(ABI_ALL_CALLEE_SAVED & ABI_ALL_GPRS & ~BitSet32({RSP})), 8);
RET();
/*FILE* codeout = fopen("codeout", "a");
fprintf(codeout, "beginning block argargarg__ %x!!!", instrs[0].Addr);
fwrite((u8*)res, GetWritableCodePtr() - (u8*)res, 1, codeout);
fclose(codeout);*/
return res;
}
@ -528,4 +565,89 @@ void Compiler::Comp_AddCycles_CI(Gen::X64Reg i, int add)
}
}
void Compiler::Comp_AddCycles_CDI()
{
if (Num == 0)
Comp_AddCycles_CD();
else
{
IrregularCycles = true;
s32 cycles;
s32 numC = NDS::ARM7MemTimings[CurInstr.CodeCycles][Thumb ? 0 : 2];
s32 numD = CurInstr.DataCycles;
if (CurInstr.DataRegion == 0x02) // mainRAM
{
if (CodeRegion == 0x02)
cycles = numC + numD;
else
{
numC++;
cycles = std::max(numC + numD - 3, std::max(numC, numD));
}
}
else if (CodeRegion == 0x02)
{
numD++;
cycles = std::max(numC + numD - 3, std::max(numC, numD));
}
else
{
cycles = numC + numD + 1;
}
printf("%x: %d %d cycles cdi (%d)\n", CurInstr.Instr, Num, CurInstr.DataCycles, cycles);
if (!Thumb && CurInstr.Cond() < 0xE)
ADD(32, MDisp(RCPU, offsetof(ARM, Cycles)), Imm8(cycles));
else
ConstantCycles += cycles;
}
}
void Compiler::Comp_AddCycles_CD()
{
u32 cycles = 0;
if (Num == 0)
{
s32 numC = (R15 & 0x2) ? 0 : CurInstr.CodeCycles;
s32 numD = CurInstr.DataCycles;
//if (DataRegion != CodeRegion)
cycles = std::max(numC + numD - 6, std::max(numC, numD));
IrregularCycles = cycles != numC;
}
else
{
s32 numC = NDS::ARM7MemTimings[CurInstr.CodeCycles][Thumb ? 0 : 2];
s32 numD = CurInstr.DataCycles;
if (CurInstr.DataRegion == 0x02)
{
if (CodeRegion == 0x02)
cycles += numC + numD;
else
cycles += std::max(numC + numD - 3, std::max(numC, numD));
}
else if (CodeRegion == 0x02)
{
cycles += std::max(numC + numD - 3, std::max(numC, numD));
}
else
{
cycles += numC + numD;
}
IrregularCycles = true;
}
if (!Thumb && CurInstr.Cond() < 0xE)
ADD(32, MDisp(RCPU, offsetof(ARM, Cycles)), Imm8(cycles));
else
ConstantCycles += cycles;
}
}