JIT implementations of lwzux and stwux. Unfortunately, not really a

visible performance increase.



git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@1672 8ced0084-cf51-0410-be5f-012b33b47a6e
This commit is contained in:
magumagu9 2008-12-26 08:53:40 +00:00
parent ba58962d24
commit 32402b465d
3 changed files with 57 additions and 2 deletions

View file

@ -275,6 +275,10 @@ public:
void lbzx(UGeckoInstruction inst);
void lwzx(UGeckoInstruction inst);
void lhax(UGeckoInstruction inst);
void lwzux(UGeckoInstruction inst);
void stwux(UGeckoInstruction inst);
void lmw(UGeckoInstruction inst);
void stmw(UGeckoInstruction inst);

View file

@ -230,6 +230,31 @@
return;
}
void Jit64::lwzux(UGeckoInstruction inst)
{
if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITLoadStoreOff)
{Default(inst); return;} // turn off from debugger
INSTRUCTION_START;
int a = inst.RA, b = inst.RB, d = inst.RD;
if (!a || a == d || a == b)
{
Default(inst);
return;
}
gpr.Lock(a, b, d);
gpr.LoadToX64(d, b == d, true);
gpr.LoadToX64(a, true, true);
ADD(32, gpr.R(a), gpr.R(b));
MOV(32, R(EAX), gpr.R(a));
SafeLoadRegToEAX(EAX, 32, 0, false);
MOV(32, gpr.R(d), R(EAX));
gpr.UnlockAll();
return;
}
// Zero cache line.
void Jit64::dcbz(UGeckoInstruction inst)
{
@ -390,6 +415,32 @@
}
}
void Jit64::stwux(UGeckoInstruction inst)
{
if(Core::g_CoreStartupParameter.bJITOff || Core::g_CoreStartupParameter.bJITLoadStoreOff)
{Default(inst); return;} // turn off from debugger
INSTRUCTION_START;
int a = inst.RA, b = inst.RB, s = inst.RS;
if (!a || a == s || a == b)
{
Default(inst);
return;
}
gpr.Lock(a, b, s);
gpr.FlushLockX(ABI_PARAM1, ABI_PARAM2);
gpr.LoadToX64(a, true, true);
ADD(32, gpr.R(a), gpr.R(b));
MOV(32, R(ABI_PARAM2), gpr.R(a));
MOV(32, R(ABI_PARAM1), gpr.R(s));
SafeWriteRegToReg(ABI_PARAM1, ABI_PARAM2, 32, 0);
gpr.UnlockAll();
gpr.UnlockAllX();
return;
}
// A few games use these heavily in video codecs.
void Jit64::lmw(UGeckoInstruction inst)
{

View file

@ -311,7 +311,7 @@ static GekkoOPTemplate table31[] =
//load word
{23, Interpreter::lwzx, &Jit64::lwzx, {"lwzx", OPTYPE_LOAD, FL_OUT_D | FL_IN_A0 | FL_IN_B}},
{55, Interpreter::lwzux, &Jit64::Default, {"lwzux", OPTYPE_LOAD, FL_OUT_D | FL_OUT_A | FL_IN_A | FL_IN_B}},
{55, Interpreter::lwzux, &Jit64::lwzux, {"lwzux", OPTYPE_LOAD, FL_OUT_D | FL_OUT_A | FL_IN_A | FL_IN_B}},
//load halfword
{279, Interpreter::lhzx, &Jit64::Default, {"lhzx", OPTYPE_LOAD, FL_OUT_D | FL_IN_A0 | FL_IN_B}},
@ -339,7 +339,7 @@ static GekkoOPTemplate table31[] =
//store word
{151, Interpreter::stwx, &Jit64::Default, {"stwx", OPTYPE_STORE, FL_IN_A0 | FL_IN_B}},
{183, Interpreter::stwux, &Jit64::Default, {"stwux", OPTYPE_STORE, FL_OUT_A | FL_IN_A | FL_IN_B}},
{183, Interpreter::stwux, &Jit64::stwux, {"stwux", OPTYPE_STORE, FL_OUT_A | FL_IN_A | FL_IN_B}},
//store halfword
{407, Interpreter::sthx, &Jit64::Default, {"sthx", OPTYPE_STORE, FL_IN_A0 | FL_IN_B}},