JIT: implement timer support in mtspr

Faster, of course, since we avoid the interpreter, but also means we can
get more a more accurate timer in long blocks by adding the offset from the
start of the block to the retrieved timer. I don't know if this will actually
fix any issues, but it's more correct and a nearly-free improvement.

Source/Core/Core/HW/SystemTimers.cpp

@@ -70,31 +70,6 @@ namespace SystemTimers
 
 static u32 CPU_CORE_CLOCK  = 486000000u;             // 486 mhz (its not 485, stop bugging me!)
 
-/*
-GameCube                   MHz
-flipper <-> ARAM bus:      81 (DSP)
-gekko <-> flipper bus:     162
-flipper <-> 1T-SRAM bus:   324
-gekko:                     486
-
-These contain some guesses:
-Wii                             MHz
-hollywood <-> GDDR3 RAM bus:    ??? no idea really
-broadway <-> hollywood bus:     243
-hollywood <-> 1T-SRAM bus:      486
-broadway:                       729
-*/
-// Ratio of TB and Decrementer to clock cycles.
-// TB clk is 1/4 of BUS clk. And it seems BUS clk is really 1/3 of CPU clk.
-// So, ratio is 1 / (1/4 * 1/3 = 1/12) = 12.
-// note: ZWW is ok and faster with TIMER_RATIO=8 though.
-// !!! POSSIBLE STABLE PERF BOOST HACK THERE !!!
-
-enum
-{
-	TIMER_RATIO = 12
-};
-
 static int et_Dec;
 static int et_VI;
 static int et_SI;

Source/Core/Core/HW/SystemTimers.h

@@ -9,6 +9,31 @@
 namespace SystemTimers
 {
 
+/*
+GameCube                   MHz
+flipper <-> ARAM bus:      81 (DSP)
+gekko <-> flipper bus:     162
+flipper <-> 1T-SRAM bus:   324
+gekko:                     486
+
+These contain some guesses:
+Wii                             MHz
+hollywood <-> GDDR3 RAM bus:    ??? no idea really
+broadway <-> hollywood bus:     243
+hollywood <-> 1T-SRAM bus:      486
+broadway:                       729
+*/
+// Ratio of TB and Decrementer to clock cycles.
+// TB clk is 1/4 of BUS clk. And it seems BUS clk is really 1/3 of CPU clk.
+// So, ratio is 1 / (1/4 * 1/3 = 1/12) = 12.
+// note: ZWW is ok and faster with TIMER_RATIO=8 though.
+// !!! POSSIBLE STABLE PERF BOOST HACK THERE !!!
+
+enum
+{
+	TIMER_RATIO = 12
+};
+
 u32 GetTicksPerSecond();
 void PreInit();
 void Init();

Source/Core/Core/PowerPC/Jit64/Jit_SystemRegisters.cpp

@@ -166,10 +166,31 @@ void Jit64::mfspr(UGeckoInstruction inst)
 	int d = inst.RD;
 	switch (iIndex)
 	{
-	case SPR_WPAR:
-	case SPR_DEC:
 	case SPR_TL:
 	case SPR_TU:
+	{
+		// TODO: we really only need to call GetFakeTimeBase once per JIT block; this matters because
+		// typical use of this instruction is to call it three times, e.g. mftbu/mftbl/mftbu/cmpw/bne
+		// to deal with possible timer wraparound. This makes the second two (out of three) completely
+		// redundant for the JIT.
+		u32 registersInUse = CallerSavedRegistersInUse();
+		u32 offset = js.downcountAmount / SystemTimers::TIMER_RATIO;
+		ABI_PushRegistersAndAdjustStack(registersInUse, false);
+		ABI_CallFunction((void *)&SystemTimers::GetFakeTimeBase);
+		ABI_PopRegistersAndAdjustStack(registersInUse, false);
+		// The timer can change within a long block, so add in any difference
+		if (offset > 0)
+			ADD(64, R(RAX), Imm32(offset));
+		MOV(64, M(&TL), R(RAX));
+		gpr.Lock(d);
+		gpr.BindToRegister(d, false);
+		if (iIndex == SPR_TU)
+			SHR(64, R(RAX), Imm8(32));
+		MOV(32, gpr.R(d), R(EAX));
+		break;
+	}
+	case SPR_WPAR:
+	case SPR_DEC:
 	case SPR_PMC1:
 	case SPR_PMC2:
 	case SPR_PMC3:
@@ -179,9 +200,9 @@ void Jit64::mfspr(UGeckoInstruction inst)
 		gpr.Lock(d);
 		gpr.BindToRegister(d, false);
 		MOV(32, gpr.R(d), M(&PowerPC::ppcState.spr[iIndex]));
-		gpr.UnlockAll();
 		break;
 	}
+	gpr.UnlockAll();
 }
 
 void Jit64::mtmsr(UGeckoInstruction inst)