Core: Revise clock timing (WIP)

include/mgba/internal/arm/isa-inlines.h

@@ -88,7 +88,8 @@ static inline void _ARMSetMode(struct ARMCore* cpu, enum ExecutionMode execution
 	case MODE_THUMB:
 		cpu->cpsr.t = 1;
 	}
-	cpu->nextEvent = cpu->cycles;
+	cpu->nextEvent -= cpu->cycles;
+	cpu->cycles = 0;
 }
 
 static inline void _ARMReadCPSR(struct ARMCore* cpu) {

src/arm/arm.c

@@ -166,7 +166,7 @@ void ARMRaiseIRQ(struct ARMCore* cpu) {
 	_ARMSetMode(cpu, MODE_ARM);
 	cpu->spsr = cpsr;
 	cpu->cpsr.i = 1;
-	cpu->cycles += currentCycles;
+	cpu->cycles -= currentCycles;
 }
 
 void ARMRaiseSWI(struct ARMCore* cpu) {
@@ -186,7 +186,7 @@ void ARMRaiseSWI(struct ARMCore* cpu) {
 	_ARMSetMode(cpu, MODE_ARM);
 	cpu->spsr = cpsr;
 	cpu->cpsr.i = 1;
-	cpu->cycles += currentCycles;
+	cpu->cycles -= currentCycles;
 }
 
 void ARMRaiseUndefined(struct ARMCore* cpu) {
@@ -206,7 +206,7 @@ void ARMRaiseUndefined(struct ARMCore* cpu) {
 	_ARMSetMode(cpu, MODE_ARM);
 	cpu->spsr = cpsr;
 	cpu->cpsr.i = 1;
-	cpu->cycles += currentCycles;
+	cpu->cycles -= currentCycles;
 }
 
 static inline void ARMStep(struct ARMCore* cpu) {
@@ -265,7 +265,7 @@ static inline void ARMStep(struct ARMCore* cpu) {
 			break;
 		}
 		if (!conditionMet) {
-			cpu->cycles += ARM_PREFETCH_CYCLES;
+			cpu->cycles -= ARM_PREFETCH_CYCLES;
 			return;
 		}
 	}
@@ -288,18 +288,18 @@ void ARMRun(struct ARMCore* cpu) {
 	} else {
 		ARMStep(cpu);
 	}
-	if (cpu->cycles >= cpu->nextEvent) {
+	if (cpu->cycles <= 0) {
 		cpu->irqh.processEvents(cpu);
 	}
 }
 
 void ARMRunLoop(struct ARMCore* cpu) {
 	if (cpu->executionMode == MODE_THUMB) {
-		while (cpu->cycles < cpu->nextEvent) {
+		while (cpu->cycles > 0) {
 			ThumbStep(cpu);
 		}
 	} else {
-		while (cpu->cycles < cpu->nextEvent) {
+		while (cpu->cycles > 0) {
 			ARMStep(cpu);
 		}
 	}

src/arm/debugger/debugger.c

@@ -139,7 +139,9 @@ void ARMDebuggerDeinit(struct mDebuggerPlatform* platform) {
 static void ARMDebuggerEnter(struct mDebuggerPlatform* platform, enum mDebuggerEntryReason reason, struct mDebuggerEntryInfo* info) {
 	struct ARMDebugger* debugger = (struct ARMDebugger*) platform;
 	struct ARMCore* cpu = debugger->cpu;
-	cpu->nextEvent = cpu->cycles;
+	cpu->nextEvent -= cpu->cycles;
+	cpu->cycles = 0;
+
 	if (reason == DEBUGGER_ENTER_BREAKPOINT) {
 		struct ARMDebugBreakpoint* breakpoint = _lookupBreakpoint(&debugger->swBreakpoints, _ARMPCAddress(cpu));
 		if (breakpoint && breakpoint->isSw) {

src/arm/isa-arm.c

@@ -288,7 +288,7 @@ ATTRIBUTE_NOINLINE static void _neutralS(struct ARMCore* cpu, int32_t d) {
 	static void _ARMInstruction ## NAME (struct ARMCore* cpu, uint32_t opcode) { \
 		int currentCycles = ARM_PREFETCH_CYCLES; \
 		BODY; \
-		cpu->cycles += currentCycles; \
+		cpu->cycles -= currentCycles; \
 	}
 
 #define DEFINE_ALU_INSTRUCTION_EX_ARM(NAME, S_BODY, SHIFTER, BODY) \

src/arm/isa-thumb.c

@@ -51,7 +51,7 @@
 	static void _ThumbInstruction ## NAME (struct ARMCore* cpu, uint16_t opcode) {  \
 		int currentCycles = THUMB_PREFETCH_CYCLES; \
 		BODY; \
-		cpu->cycles += currentCycles; \
+		cpu->cycles -= currentCycles; \
 	}
 
 #define DEFINE_IMMEDIATE_5_INSTRUCTION_THUMB(NAME, BODY) \

src/core/timing.c

@@ -23,9 +23,10 @@ void mTimingClear(struct mTiming* timing) {
 }
 
 void mTimingSchedule(struct mTiming* timing, struct mTimingEvent* event, int32_t when) {
-	int32_t nextEvent = when + *timing->relativeCycles;
+	int32_t nextEvent = when + *timing->nextEvent - *timing->relativeCycles;
 	event->when = nextEvent + timing->masterCycles;
-	if (nextEvent < *timing->nextEvent) {
+	if (when < *timing->relativeCycles) {
+		*timing->relativeCycles = when;
 		*timing->nextEvent = nextEvent;
 	}
 	struct mTimingEvent** previous = &timing->root;
@@ -91,7 +92,7 @@ int32_t mTimingTick(struct mTiming* timing, int32_t cycles) {
 }
 
 int32_t mTimingCurrentTime(const struct mTiming* timing) {
-	return timing->masterCycles + *timing->relativeCycles;
+	return timing->masterCycles + *timing->nextEvent - *timing->relativeCycles;
 }
 
 int32_t mTimingNextEvent(struct mTiming* timing) {
@@ -99,9 +100,9 @@ int32_t mTimingNextEvent(struct mTiming* timing) {
 	if (!next) {
 		return INT_MAX;
 	}
-	return next->when - timing->masterCycles - *timing->relativeCycles;
+	return next->when - timing->masterCycles - *timing->nextEvent + *timing->relativeCycles;
 }
 
 int32_t mTimingUntil(const struct mTiming* timing, const struct mTimingEvent* event) {
-	return event->when - timing->masterCycles - *timing->relativeCycles;
+	return event->when - timing->masterCycles - *timing->nextEvent + *timing->relativeCycles;
 }

src/gb/gb.c

@@ -639,10 +639,10 @@ void GBUpdateIRQs(struct GB* gb) {
 void GBProcessEvents(struct LR35902Core* cpu) {
 	struct GB* gb = (struct GB*) cpu->master;
 	do {
-		int32_t cycles = cpu->cycles;
+		int32_t cycles = cpu->nextEvent - cpu->cycles;
 		int32_t nextEvent;
 
-		cpu->cycles = 0;
+		cpu->cycles = INT_MAX;
 		cpu->nextEvent = INT_MAX;
 
 		nextEvent = cycles;
@@ -650,9 +650,10 @@ void GBProcessEvents(struct LR35902Core* cpu) {
 			nextEvent = mTimingTick(&gb->timing, nextEvent);
 		} while (gb->cpuBlocked);
 		cpu->nextEvent = nextEvent;
+		cpu->cycles = nextEvent;
 
 		if (cpu->halted) {
-			cpu->cycles = cpu->nextEvent;
+			cpu->cycles = 0;
 			if (!gb->memory.ie || !gb->memory.ime) {
 				break;
 			}
@@ -660,7 +661,7 @@ void GBProcessEvents(struct LR35902Core* cpu) {
 		if (gb->earlyExit) {
 			break;
 		}
-	} while (cpu->cycles >= cpu->nextEvent);
+	} while (cpu->cycles <= 0);
 	gb->earlyExit = false;
 }
 
@@ -713,7 +714,7 @@ static void _enableInterrupts(struct mTiming* timing, void* user, uint32_t cycle
 void GBHalt(struct LR35902Core* cpu) {
 	struct GB* gb = (struct GB*) cpu->master;
 	if (!(gb->memory.ie & gb->memory.io[REG_IF])) {
-		cpu->cycles = cpu->nextEvent;
+		cpu->cycles = 0;
 		cpu->halted = true;
 	} else if (gb->model < GB_MODEL_CGB) {
 		mLOG(GB, STUB, "Unimplemented HALT bug");

src/gb/memory.c

@@ -476,8 +476,9 @@ void GBMemoryDMA(struct GB* gb, uint16_t base) {
 	}
 	mTimingDeschedule(&gb->timing, &gb->memory.dmaEvent);
 	mTimingSchedule(&gb->timing, &gb->memory.dmaEvent, 8);
-	if (gb->cpu->cycles + 8 < gb->cpu->nextEvent) {
+	if (gb->cpu->cycles < 8) {
-		gb->cpu->nextEvent = gb->cpu->cycles + 8;
+		gb->cpu->nextEvent += 8 - gb->cpu->cycles;
+		gb->cpu->cycles = 8;
 	}
 	gb->memory.dmaSource = base;
 	gb->memory.dmaDest = 0;

src/gb/serialize.c

@@ -43,7 +43,8 @@ void GBSerialize(struct GB* gb, struct GBSerializedState* state) {
 	STORE_16LE(gb->cpu->sp, 0, &state->cpu.sp);
 	STORE_16LE(gb->cpu->pc, 0, &state->cpu.pc);
 
-	STORE_32LE(gb->cpu->cycles, 0, &state->cpu.cycles);
+	int32_t cycles = gb->cpu->nextEvent - gb->cpu->cycles;
+	STORE_32LE(cycles, 0, &state->cpu.cycles);
 	STORE_32LE(gb->cpu->nextEvent, 0, &state->cpu.nextEvent);
 
 	STORE_16LE(gb->cpu->index, 0, &state->cpu.index);
@@ -163,8 +164,10 @@ bool GBDeserialize(struct GB* gb, const struct GBSerializedState* state) {
 	gb->doubleSpeed = GBSerializedCpuFlagsGetDoubleSpeed(flags);
 	gb->audio.timingFactor = gb->doubleSpeed + 1;
 
-	LOAD_32LE(gb->cpu->cycles, 0, &state->cpu.cycles);
+	int32_t cycles;
+	LOAD_32LE(cycles, 0, &state->cpu.cycles);
 	LOAD_32LE(gb->cpu->nextEvent, 0, &state->cpu.nextEvent);
+	gb->cpu->cycles = gb->cpu->nextEvent - cycles;
 	gb->timing.root = NULL;
 
 	uint32_t when;

src/gba/gba.c

@@ -251,23 +251,20 @@ static void GBAProcessEvents(struct ARMCore* cpu) {
 	}
 
 	int32_t nextEvent = cpu->nextEvent;
-	while (cpu->cycles >= nextEvent) {
+	while (cpu->cycles <= 0) {
+		int32_t cycles = cpu->nextEvent - cpu->cycles;
 		cpu->nextEvent = INT_MAX;
-		nextEvent = 0;
+		cpu->cycles = INT_MAX;
+		nextEvent = cycles;
 		do {
-			int32_t cycles = cpu->cycles;
+			nextEvent = mTimingTick(&gba->timing, nextEvent);
-			cpu->cycles = 0;
-#ifndef NDEBUG
-			if (cycles < 0) {
-				mLOG(GBA, FATAL, "Negative cycles passed: %i", cycles);
-			}
-#endif
-			nextEvent = mTimingTick(&gba->timing, nextEvent + cycles);
 		} while (gba->cpuBlocked);
 
 		cpu->nextEvent = nextEvent;
+		cpu->cycles = nextEvent;
+
 		if (cpu->halted) {
-			cpu->cycles = nextEvent;
+			cpu->cycles = 0;
 			if (!gba->memory.io[REG_IME >> 1] || !gba->memory.io[REG_IE >> 1]) {
 				break;
 			}
@@ -499,12 +496,13 @@ void GBATestIRQ(struct ARMCore* cpu) {
 	struct GBA* gba = (struct GBA*) cpu->master;
 	if (gba->memory.io[REG_IME >> 1] && gba->memory.io[REG_IE >> 1] & gba->memory.io[REG_IF >> 1]) {
 		gba->springIRQ = gba->memory.io[REG_IE >> 1] & gba->memory.io[REG_IF >> 1];
-		gba->cpu->nextEvent = gba->cpu->cycles;
+		gba->cpu->nextEvent -= gba->cpu->cycles;
+		gba->cpu->cycles = 0;
 	}
 }
 
 void GBAHalt(struct GBA* gba) {
-	gba->cpu->nextEvent = gba->cpu->cycles;
+	gba->cpu->cycles = 0;
 	gba->cpu->halted = 1;
 }
 
@@ -516,7 +514,7 @@ void GBAStop(struct GBA* gba) {
 			callbacks->sleep(callbacks->context);
 		}
 	}
-	gba->cpu->nextEvent = gba->cpu->cycles;
+	gba->cpu->cycles = 0;
 }
 
 void GBADebug(struct GBA* gba, uint16_t flags) {

src/gba/memory.c

@@ -1613,7 +1613,7 @@ int32_t GBAMemoryStall(struct ARMCore* cpu, int32_t wait) {
 	memory->lastPrefetchedPc = cpu->gprs[ARM_PC] + WORD_SIZE_THUMB * (loads + previousLoads - 1);
 
 	// The next |loads|S waitstates disappear entirely, so long as they're all in a row
-	cpu->cycles -= (s - 1) * loads;
+	cpu->cycles += (s - 1) * loads;
 	return wait;
 }
 

src/gba/serialize.c

@@ -44,7 +44,9 @@ void GBASerialize(struct GBA* gba, struct GBASerializedState* state) {
 	}
 	STORE_32(gba->cpu->cpsr.packed, 0, &state->cpu.cpsr.packed);
 	STORE_32(gba->cpu->spsr.packed, 0, &state->cpu.spsr.packed);
-	STORE_32(gba->cpu->cycles, 0, &state->cpu.cycles);
+
+	int32_t cycles = gba->cpu->nextEvent - gba->cpu->cycles;
+	STORE_32(cycles, 0, &state->cpu.cycles);
 	STORE_32(gba->cpu->nextEvent, 0, &state->cpu.nextEvent);
 	for (i = 0; i < 6; ++i) {
 		int j;
@@ -136,8 +138,12 @@ bool GBADeserialize(struct GBA* gba, const struct GBASerializedState* state) {
 	}
 	LOAD_32(gba->cpu->cpsr.packed, 0, &state->cpu.cpsr.packed);
 	LOAD_32(gba->cpu->spsr.packed, 0, &state->cpu.spsr.packed);
-	LOAD_32(gba->cpu->cycles, 0, &state->cpu.cycles);
+
+	int32_t cycles;
+	LOAD_32(cycles, 0, &state->cpu.cycles);
 	LOAD_32(gba->cpu->nextEvent, 0, &state->cpu.nextEvent);
+	gba->cpu->cycles = gba->cpu->nextEvent - cycles;
+
 	for (i = 0; i < 6; ++i) {
 		int j;
 		for (j = 0; j < 7; ++j) {

src/lr35902/debugger/debugger.c

@@ -123,7 +123,8 @@ static void LR35902DebuggerEnter(struct mDebuggerPlatform* platform, enum mDebug
 	UNUSED(info);
 	struct LR35902Debugger* debugger = (struct LR35902Debugger*) platform;
 	struct LR35902Core* cpu = debugger->cpu;
-	cpu->nextEvent = cpu->cycles;
+	cpu->nextEvent -= cpu->cycles;
+	cpu->cycles = 0;
 
 	if (debugger->d.p->entered) {
 		debugger->d.p->entered(debugger->d.p, reason, info);

src/lr35902/lr35902.c

@@ -102,7 +102,7 @@ static void _LR35902InstructionIRQ(struct LR35902Core* cpu) {
 }
 
 static void _LR35902Step(struct LR35902Core* cpu) {
-	++cpu->cycles;
+	--cpu->cycles;
 	enum LR35902ExecutionState state = cpu->executionState;
 	cpu->executionState = LR35902_CORE_IDLE_0;
 	switch (state) {
@@ -137,44 +137,44 @@ static void _LR35902Step(struct LR35902Core* cpu) {
 }
 
 void LR35902Tick(struct LR35902Core* cpu) {
-	while (cpu->cycles >= cpu->nextEvent) {
+	while (cpu->cycles <= 0) {
 		cpu->irqh.processEvents(cpu);
 	}
 	_LR35902Step(cpu);
-	if (cpu->cycles + 2 >= cpu->nextEvent) {
+	if (cpu->cycles <= 2) {
-		int32_t diff = cpu->nextEvent - cpu->cycles;
+		int32_t diff = cpu->cycles;
-		cpu->cycles = cpu->nextEvent;
+		cpu->cycles = 0;
 		cpu->executionState += diff;
 		cpu->irqh.processEvents(cpu);
-		cpu->cycles += LR35902_CORE_EXECUTE - cpu->executionState;
+		cpu->cycles -= LR35902_CORE_EXECUTE - cpu->executionState;
 	} else {
-		cpu->cycles += 2;
+		cpu->cycles -= 2;
 	}
 	cpu->executionState = LR35902_CORE_FETCH;
 	cpu->instruction(cpu);
-	++cpu->cycles;
+	--cpu->cycles;
 }
 
 void LR35902Run(struct LR35902Core* cpu) {
 	bool running = true;
 	while (running || cpu->executionState != LR35902_CORE_FETCH) {
-		if (cpu->cycles >= cpu->nextEvent) {
+		if (cpu->cycles <= 0) {
 			cpu->irqh.processEvents(cpu);
 			break;
 		}
 		_LR35902Step(cpu);
-		if (cpu->cycles + 2 >= cpu->nextEvent) {
+		if (cpu->cycles <= 2) {
-			int32_t diff = cpu->nextEvent - cpu->cycles;
+			int32_t diff = cpu->cycles;
-			cpu->cycles = cpu->nextEvent;
+			cpu->cycles = 0;
 			cpu->executionState += diff;
 			cpu->irqh.processEvents(cpu);
-			cpu->cycles += LR35902_CORE_EXECUTE - cpu->executionState;
+			cpu->cycles -= LR35902_CORE_EXECUTE - cpu->executionState;
 			running = false;
 		} else {
-			cpu->cycles += 2;
+			cpu->cycles -= 2;
 		}
 		cpu->executionState = LR35902_CORE_FETCH;
 		cpu->instruction(cpu);
-		++cpu->cycles;
+		--cpu->cycles;
 	}
 }