aboutsummaryrefslogtreecommitdiff
path: root/src/ARMJIT_A64
diff options
context:
space:
mode:
Diffstat (limited to 'src/ARMJIT_A64')
-rw-r--r--src/ARMJIT_A64/ARMJIT_Branch.cpp10
-rw-r--r--src/ARMJIT_A64/ARMJIT_Compiler.cpp74
-rw-r--r--src/ARMJIT_A64/ARMJIT_Compiler.h7
-rw-r--r--src/ARMJIT_A64/ARMJIT_LoadStore.cpp21
4 files changed, 70 insertions, 42 deletions
diff --git a/src/ARMJIT_A64/ARMJIT_Branch.cpp b/src/ARMJIT_A64/ARMJIT_Branch.cpp
index eeabfb0..2f640c8 100644
--- a/src/ARMJIT_A64/ARMJIT_Branch.cpp
+++ b/src/ARMJIT_A64/ARMJIT_Branch.cpp
@@ -27,7 +27,7 @@ namespace ARMJIT
{
template <typename T>
-void jumpToTrampoline(T* cpu, u32 addr, bool changeCPSR)
+void JumpToTrampoline(T* cpu, u32 addr, bool changeCPSR)
{
cpu->JumpTo(addr, changeCPSR);
}
@@ -301,7 +301,7 @@ void Compiler::Comp_JumpTo(Arm64Gen::ARM64Reg addr, bool switchThumb, bool resto
bool cpsrDirty = CPSRDirty;
SaveCPSR();
SaveCycles();
- PushRegs(restoreCPSR);
+ PushRegs(restoreCPSR, true);
if (switchThumb)
MOV(W1, addr);
@@ -315,11 +315,11 @@ void Compiler::Comp_JumpTo(Arm64Gen::ARM64Reg addr, bool switchThumb, bool resto
MOV(X0, RCPU);
MOVI2R(W2, restoreCPSR);
if (Num == 0)
- QuickCallFunction(X3, jumpToTrampoline<ARMv5>);
+ QuickCallFunction(X3, JumpToTrampoline<ARMv5>);
else
- QuickCallFunction(X3, jumpToTrampoline<ARMv4>);
+ QuickCallFunction(X3, JumpToTrampoline<ARMv4>);
- PopRegs(restoreCPSR);
+ PopRegs(restoreCPSR, true);
LoadCycles();
LoadCPSR();
if (CurInstr.Cond() < 0xE)
diff --git a/src/ARMJIT_A64/ARMJIT_Compiler.cpp b/src/ARMJIT_A64/ARMJIT_Compiler.cpp
index 4fbb804..7dc854a 100644
--- a/src/ARMJIT_A64/ARMJIT_Compiler.cpp
+++ b/src/ARMJIT_A64/ARMJIT_Compiler.cpp
@@ -58,9 +58,14 @@ namespace ARMJIT
template <>
const ARM64Reg RegisterCache<Compiler, ARM64Reg>::NativeRegAllocOrder[] =
- {W19, W20, W21, W22, W23, W24, W25, W26};
+{
+ W19, W20, W21, W22, W23, W24, W25,
+ W8, W9, W10, W11, W12, W13, W14, W15
+};
template <>
-const int RegisterCache<Compiler, ARM64Reg>::NativeRegsAvailable = 8;
+const int RegisterCache<Compiler, ARM64Reg>::NativeRegsAvailable = 15;
+
+const BitSet32 CallerSavedPushRegs({W8, W9, W10, W11, W12, W13, W14, W15});
const int JitMemSize = 16 * 1024 * 1024;
#ifndef __SWITCH__
@@ -164,44 +169,55 @@ void Compiler::A_Comp_MSR()
MOV(W2, RCPSR);
MOV(X0, RCPU);
- PushRegs(true);
-
- QuickCallFunction(X3, (void*)&UpdateModeTrampoline);
-
- PopRegs(true);
+ PushRegs(true, true);
+ QuickCallFunction(X3, UpdateModeTrampoline);
+ PopRegs(true, true);
}
}
}
-void Compiler::PushRegs(bool saveHiRegs)
+
+void Compiler::PushRegs(bool saveHiRegs, bool saveRegsToBeChanged, bool allowUnload)
{
+ BitSet32 loadedRegs(RegCache.LoadedRegs);
+
if (saveHiRegs)
{
- if (Thumb || CurInstr.Cond() == 0xE)
+ BitSet32 hiRegsLoaded(RegCache.LoadedRegs & 0x7F00);
+ for (int reg : hiRegsLoaded)
{
- BitSet16 hiRegsLoaded(RegCache.LoadedRegs & 0x7F00);
- for (int reg : hiRegsLoaded)
+ if (Thumb || CurInstr.Cond() == 0xE)
RegCache.UnloadRegister(reg);
+ else
+ SaveReg(reg, RegCache.Mapping[reg]);
+ // prevent saving the register twice
+ loadedRegs[reg] = false;
}
- else
+ }
+
+ for (int reg : loadedRegs)
+ {
+ if (CallerSavedPushRegs[RegCache.Mapping[reg]]
+ && (saveRegsToBeChanged || !((1<<reg) & CurInstr.Info.DstRegs && !((1<<reg) & CurInstr.Info.SrcRegs))))
{
- BitSet16 hiRegsDirty(RegCache.LoadedRegs & 0x7F00);
- for (int reg : hiRegsDirty)
+ if ((Thumb || CurInstr.Cond() == 0xE) && !((1 << reg) & (CurInstr.Info.DstRegs|CurInstr.Info.SrcRegs)) && allowUnload)
+ RegCache.UnloadRegister(reg);
+ else
SaveReg(reg, RegCache.Mapping[reg]);
}
}
}
-void Compiler::PopRegs(bool saveHiRegs)
+void Compiler::PopRegs(bool saveHiRegs, bool saveRegsToBeChanged)
{
- if (saveHiRegs)
+ BitSet32 loadedRegs(RegCache.LoadedRegs);
+ for (int reg : loadedRegs)
{
- if (!Thumb && CurInstr.Cond() != 0xE)
+ if ((saveHiRegs && reg >= 8 && reg < 15)
+ || (CallerSavedPushRegs[RegCache.Mapping[reg]]
+ && (saveRegsToBeChanged || !((1<<reg) & CurInstr.Info.DstRegs && !((1<<reg) & CurInstr.Info.SrcRegs)))))
{
- BitSet16 hiRegsLoaded(RegCache.LoadedRegs & 0x7F00);
-
- for (int reg : hiRegsLoaded)
- LoadReg(reg, RegCache.Mapping[reg]);
+ LoadReg(reg, RegCache.Mapping[reg]);
}
}
}
@@ -267,6 +283,7 @@ Compiler::Compiler()
}
/*
+ W4 - whether the register was written to
W5 - mode
W1 - reg num
W3 - in/out value of reg
@@ -358,7 +375,7 @@ Compiler::Compiler()
{
for (int reg = 0; reg < 32; reg++)
{
- if (!(reg == W4 || (reg >= W19 && reg <= W26)))
+ if (!(reg == W4 || (reg >= W8 && reg <= W15) || (reg >= W19 && reg <= W25)))
continue;
ARM64Reg rdMapped = (ARM64Reg)reg;
PatchedStoreFuncs[consoleType][num][size][reg] = GetRXPtr();
@@ -371,7 +388,7 @@ Compiler::Compiler()
{
MOV(W1, rdMapped);
}
- ABI_PushRegisters({30});
+ ABI_PushRegisters(BitSet32({30}) | CallerSavedPushRegs);
if (consoleType == 0)
{
switch ((8 << size) | num)
@@ -397,7 +414,7 @@ Compiler::Compiler()
}
}
- ABI_PopRegisters({30});
+ ABI_PopRegisters(BitSet32({30}) | CallerSavedPushRegs);
RET();
for (int signextend = 0; signextend < 2; signextend++)
@@ -405,7 +422,7 @@ Compiler::Compiler()
PatchedLoadFuncs[consoleType][num][size][signextend][reg] = GetRXPtr();
if (num == 0)
MOV(X1, RCPU);
- ABI_PushRegisters({30});
+ ABI_PushRegisters(BitSet32({30}) | CallerSavedPushRegs);
if (consoleType == 0)
{
switch ((8 << size) | num)
@@ -430,7 +447,7 @@ Compiler::Compiler()
case 9: QuickCallFunction(X3, SlowRead7<u8, 1>); break;
}
}
- ABI_PopRegisters({30});
+ ABI_PopRegisters(BitSet32({30}) | CallerSavedPushRegs);
if (size == 32)
MOV(rdMapped, W0);
else if (signextend)
@@ -673,7 +690,7 @@ void Compiler::Comp_BranchSpecialBehaviour(bool taken)
}
}
-JitBlockEntry Compiler::CompileBlock(ARM* cpu, bool thumb, FetchedInstr instrs[], int instrsCount)
+JitBlockEntry Compiler::CompileBlock(ARM* cpu, bool thumb, FetchedInstr instrs[], int instrsCount, bool hasMemInstr)
{
if (JitMemMainSize - GetCodeOffset() < 1024 * 16)
{
@@ -695,6 +712,9 @@ JitBlockEntry Compiler::CompileBlock(ARM* cpu, bool thumb, FetchedInstr instrs[]
RegCache = RegisterCache<Compiler, ARM64Reg>(this, instrs, instrsCount, true);
CPSRDirty = false;
+ if (hasMemInstr)
+ MOVP2R(RMemBase, Num == 0 ? ARMJIT_Memory::FastMem9Start : ARMJIT_Memory::FastMem7Start);
+
for (int i = 0; i < instrsCount; i++)
{
CurInstr = instrs[i];
diff --git a/src/ARMJIT_A64/ARMJIT_Compiler.h b/src/ARMJIT_A64/ARMJIT_Compiler.h
index d18da93..24e730b 100644
--- a/src/ARMJIT_A64/ARMJIT_Compiler.h
+++ b/src/ARMJIT_A64/ARMJIT_Compiler.h
@@ -32,6 +32,7 @@
namespace ARMJIT
{
+const Arm64Gen::ARM64Reg RMemBase = Arm64Gen::X26;
const Arm64Gen::ARM64Reg RCPSR = Arm64Gen::W27;
const Arm64Gen::ARM64Reg RCycles = Arm64Gen::W28;
const Arm64Gen::ARM64Reg RCPU = Arm64Gen::X29;
@@ -99,8 +100,8 @@ public:
Compiler();
~Compiler();
- void PushRegs(bool saveHiRegs);
- void PopRegs(bool saveHiRegs);
+ void PushRegs(bool saveHiRegs, bool saveRegsToBeChanged, bool allowUnload = true);
+ void PopRegs(bool saveHiRegs, bool saveRegsToBeChanged);
Arm64Gen::ARM64Reg MapReg(int reg)
{
@@ -108,7 +109,7 @@ public:
return RegCache.Mapping[reg];
}
- JitBlockEntry CompileBlock(ARM* cpu, bool thumb, FetchedInstr instrs[], int instrsCount);
+ JitBlockEntry CompileBlock(ARM* cpu, bool thumb, FetchedInstr instrs[], int instrsCount, bool hasMemInstr);
bool CanCompile(bool thumb, u16 kind);
diff --git a/src/ARMJIT_A64/ARMJIT_LoadStore.cpp b/src/ARMJIT_A64/ARMJIT_LoadStore.cpp
index 4f05d4d..5ac629b 100644
--- a/src/ARMJIT_A64/ARMJIT_LoadStore.cpp
+++ b/src/ARMJIT_A64/ARMJIT_LoadStore.cpp
@@ -194,13 +194,11 @@ void Compiler::Comp_MemAccess(int rd, int rn, Op2 offset, int size, int flags)
ptrdiff_t memopStart = GetCodeOffset();
LoadStorePatch patch;
- assert((rdMapped >= W19 && rdMapped <= W26) || rdMapped == W4);
+ assert((rdMapped >= W8 && rdMapped <= W15) || (rdMapped >= W19 && rdMapped <= W25) || rdMapped == W4);
patch.PatchFunc = flags & memop_Store
? PatchedStoreFuncs[NDS::ConsoleType][Num][__builtin_ctz(size) - 3][rdMapped]
: PatchedLoadFuncs[NDS::ConsoleType][Num][__builtin_ctz(size) - 3][!!(flags & memop_SignExtend)][rdMapped];
- MOVP2R(X7, Num == 0 ? ARMJIT_Memory::FastMem9Start : ARMJIT_Memory::FastMem7Start);
-
// take a chance at fastmem
if (size > 8)
ANDI2R(W1, W0, addressMask);
@@ -208,11 +206,11 @@ void Compiler::Comp_MemAccess(int rd, int rn, Op2 offset, int size, int flags)
ptrdiff_t loadStorePosition = GetCodeOffset();
if (flags & memop_Store)
{
- STRGeneric(size, rdMapped, size > 8 ? X1 : X0, X7);
+ STRGeneric(size, rdMapped, size > 8 ? X1 : X0, RMemBase);
}
else
{
- LDRGeneric(size, flags & memop_SignExtend, rdMapped, size > 8 ? X1 : X0, X7);
+ LDRGeneric(size, flags & memop_SignExtend, rdMapped, size > 8 ? X1 : X0, RMemBase);
if (size == 32 && !addrIsStatic)
{
UBFIZ(W0, W0, 3, 2);
@@ -230,12 +228,16 @@ void Compiler::Comp_MemAccess(int rd, int rn, Op2 offset, int size, int flags)
if (addrIsStatic)
func = ARMJIT_Memory::GetFuncForAddr(CurCPU, staticAddress, flags & memop_Store, size);
+ PushRegs(false, false);
+
if (func)
{
if (flags & memop_Store)
MOV(W1, rdMapped);
QuickCallFunction(X2, (void (*)())func);
+ PopRegs(false, false);
+
if (!(flags & memop_Store))
{
if (size == 32)
@@ -314,6 +316,8 @@ void Compiler::Comp_MemAccess(int rd, int rn, Op2 offset, int size, int flags)
}
}
+ PopRegs(false, false);
+
if (!(flags & memop_Store))
{
if (size == 32)
@@ -515,8 +519,7 @@ s32 Compiler::Comp_MemAccessBlock(int rn, BitSet16 regs, bool store, bool preinc
ptrdiff_t fastPathStart = GetCodeOffset();
ptrdiff_t loadStoreOffsets[8];
- MOVP2R(X1, Num == 0 ? ARMJIT_Memory::FastMem9Start : ARMJIT_Memory::FastMem7Start);
- ADD(X1, X1, X0);
+ ADD(X1, RMemBase, X0);
u32 offset = 0;
BitSet16::Iterator it = regs.begin();
@@ -655,6 +658,8 @@ s32 Compiler::Comp_MemAccessBlock(int rn, BitSet16 regs, bool store, bool preinc
}
}
+ PushRegs(false, false, !compileFastPath);
+
ADD(X1, SP, 0);
MOVI2R(W2, regsCount);
@@ -680,6 +685,8 @@ s32 Compiler::Comp_MemAccessBlock(int rn, BitSet16 regs, bool store, bool preinc
}
}
+ PopRegs(false, false);
+
if (!store)
{
if (usermode && !regs[15] && (regs & BitSet16(0x7f00)))