Merge pull request #387 from wheremyfoodat/idle-skeep

Add idle skipping

include/cpu_dynarmic.hpp

@@ -179,6 +179,8 @@ class CPU {
 		return scheduler;
 	}
 
+    void addTicks(u64 ticks) { env.AddTicks(ticks); }
+
     void clearCache() { jit->ClearCache(); }
     void runFrame();
 };

include/kernel/kernel.hpp

@@ -95,6 +95,7 @@ public:
 	void releaseMutex(Mutex* moo);
 	void cancelTimer(Timer* timer);
 	void signalTimer(Handle timerHandle, Timer* timer);
+	u64 getWakeupTick(s64 ns);
 
 	// Wake up the thread with the highest priority out of all threads in the waitlist
 	// Returns the index of the woken up thread

include/kernel/kernel_types.hpp

@@ -83,56 +83,53 @@ struct Port {
 };
 
 struct Session {
-    Handle portHandle; // The port this session is subscribed to
+	Handle portHandle;  // The port this session is subscribed to
-    Session(Handle portHandle) : portHandle(portHandle) {}
+	Session(Handle portHandle) : portHandle(portHandle) {}
 };
 
 enum class ThreadStatus {
-    Running,     // Currently running
+	Running,      // Currently running
-    Ready,       // Ready to run
+	Ready,        // Ready to run
-    WaitArbiter, // Waiting on an address arbiter
+	WaitArbiter,  // Waiting on an address arbiter
-    WaitSleep,   // Waiting due to a SleepThread SVC
+	WaitSleep,    // Waiting due to a SleepThread SVC
-    WaitSync1,   // Waiting for the single object in the wait list to be ready
+	WaitSync1,    // Waiting for the single object in the wait list to be ready
-    WaitSyncAny, // Wait for one object of the many that might be in the wait list to be ready
+	WaitSyncAny,  // Wait for one object of the many that might be in the wait list to be ready
-    WaitSyncAll, // Waiting for ALL sync objects in its wait list to be ready
+	WaitSyncAll,  // Waiting for ALL sync objects in its wait list to be ready
-    WaitIPC,     // Waiting for the reply from an IPC request
+	WaitIPC,      // Waiting for the reply from an IPC request
-    Dormant,     // Created but not yet made ready
+	Dormant,      // Created but not yet made ready
-    Dead         // Run to completion, or forcefully terminated
+	Dead          // Run to completion, or forcefully terminated
 };
 
 struct Thread {
-    u32 initialSP;  // Initial r13 value
+	u32 initialSP;   // Initial r13 value
-    u32 entrypoint; // Initial r15 value
+	u32 entrypoint;  // Initial r15 value
-    u32 priority;
+	u32 priority;
-    u32 arg;
+	u32 arg;
-    ProcessorID processorID;
+	ProcessorID processorID;
-    ThreadStatus status;
+	ThreadStatus status;
-    Handle handle;  // OS handle for this thread
+	Handle handle;  // OS handle for this thread
-    int index; // Index of the thread. 0 for the first thread, 1 for the second, and so on
+	int index;      // Index of the thread. 0 for the first thread, 1 for the second, and so on
 
-    // The waiting address for threads that are waiting on an AddressArbiter
+	// The waiting address for threads that are waiting on an AddressArbiter
-    u32 waitingAddress;
+	u32 waitingAddress;
 
-    // The nanoseconds until a thread wakes up from being asleep or from timing out while waiting on an arbiter
+	// For WaitSynchronization(N): A vector of objects this thread is waiting for
-    u64 waitingNanoseconds;
+	std::vector<Handle> waitList;
-    // The tick this thread went to sleep on
+	// For WaitSynchronizationN: Shows whether the object should wait for all objects in the wait list or just one
-    u64 sleepTick;
+	bool waitAll;
-    // For WaitSynchronization(N): A vector of objects this thread is waiting for
+	// For WaitSynchronizationN: The "out" pointer
-    std::vector<Handle> waitList;
+	u32 outPointer;
-    // For WaitSynchronizationN: Shows whether the object should wait for all objects in the wait list or just one
+	u64 wakeupTick;
-    bool waitAll;
-    // For WaitSynchronizationN: The "out" pointer
-    u32 outPointer;
 
-    // Thread context used for switching between threads
+	// Thread context used for switching between threads
-    std::array<u32, 16> gprs;
+	std::array<u32, 16> gprs;
-    std::array<u32, 32> fprs; // Stored as u32 because dynarmic does it
+	std::array<u32, 32> fprs;  // Stored as u32 because dynarmic does it
-    u32 cpsr;
+	u32 cpsr;
-    u32 fpscr;
+	u32 fpscr;
-    u32 tlsBase; // Base pointer for thread-local storage
+	u32 tlsBase;  // Base pointer for thread-local storage
 
-    // A list of threads waiting for this thread to terminate. Yes, threads are sync objects too.
+	// A list of threads waiting for this thread to terminate. Yes, threads are sync objects too.
-    u64 threadsWaitingForTermination;
+	u64 threadsWaitingForTermination;
 };
 
 static const char* kernelObjectTypeToString(KernelObjectType t) {

src/core/kernel/events.cpp

@@ -126,8 +126,7 @@ void Kernel::waitSynchronization1() {
 		auto& t = threads[currentThreadIndex];
 		t.waitList.resize(1);
 		t.status = ThreadStatus::WaitSync1;
-		t.sleepTick = cpu.getTicks();
+		t.wakeupTick = getWakeupTick(ns);
-		t.waitingNanoseconds = ns;
 		t.waitList[0] = handle;
 
 		// Add the current thread to the object's wait list
@@ -220,8 +219,7 @@ void Kernel::waitSynchronizationN() {
 		t.waitList.resize(handleCount);
 		t.status = ThreadStatus::WaitSyncAny;
 		t.outPointer = outPointer;
-		t.waitingNanoseconds = ns;
+		t.wakeupTick = getWakeupTick(ns);
-		t.sleepTick = cpu.getTicks();
 
 		for (s32 i = 0; i < handleCount; i++) {
 			t.waitList[i] = waitObjects[i].first; // Add object to this thread's waitlist

src/core/kernel/idle_thread.cpp

@@ -8,13 +8,6 @@
 	The code for our idle thread looks like this
 
 idle_thread_main:
-	mov r0, #4096            @ Loop counter
-	
-	.loop:
-	nop; nop; nop; nop       @ NOP 4 times to waste some cycles
-	subs r0, #1              @ Decrement counter by 1, go back to looping if loop counter != 0
-	bne .loop
-
 	// Sleep for 0 seconds with the SleepThread SVC, which just yields execution
 	mov r0, #0
 	mov r1, #0
@@ -24,14 +17,10 @@ idle_thread_main:
 */
 
 static constexpr u8 idleThreadCode[] = {
-	0x01, 0x0A, 0xA0, 0xE3, // mov r0, #4096
-	0x00, 0xF0, 0x20, 0xE3, 0x00, 0xF0, 0x20, 0xE3, 0x00, 0xF0, 0x20, 0xE3, 0x00, 0xF0, 0x20, 0xE3, // nop (4 times)
-	0x01, 0x00, 0x50, 0xE2,  // subs r0, #1
-	0xF9, 0xFF, 0xFF, 0x1A,  // bne loop
 	0x00, 0x00, 0xA0, 0xE3,  // mov r0, #0
 	0x00, 0x10, 0xA0, 0xE3,  // mov r1, #0
 	0x0A, 0x00, 0x00, 0xEF,  // svc SleepThread
-	0xF4, 0xFF, 0xFF, 0xEA   // b idle_thread_main
+	0xFB, 0xFF, 0xFF, 0xEA   // b idle_thread_main
 };
 
 // Set up an idle thread to run when no thread is able to run

src/core/kernel/threads.cpp

@@ -52,14 +52,8 @@ bool Kernel::canThreadRun(const Thread& t) {
 		return true;
 	} else if (t.status == ThreadStatus::WaitSleep || t.status == ThreadStatus::WaitSync1
 		|| t.status == ThreadStatus::WaitSyncAny || t.status == ThreadStatus::WaitSyncAll) {
-		const u64 elapsedTicks = cpu.getTicks() - t.sleepTick;
-
-		constexpr double ticksPerSec = double(CPU::ticksPerSec);
-		constexpr double nsPerTick = ticksPerSec / 1000000000.0;
-
 		// TODO: Set r0 to the correct error code on timeout for WaitSync{1/Any/All}
-		const s64 elapsedNs = s64(double(elapsedTicks) * nsPerTick);
+		return cpu.getTicks() >= t.wakeupTick;
-		return elapsedNs >= t.waitingNanoseconds;
 	}
 
 	// Handle timeouts and stuff here
@@ -82,6 +76,15 @@ std::optional<int> Kernel::getNextThread() {
 	return std::nullopt;
 }
 
+u64 Kernel::getWakeupTick(s64 ns) {
+	// Timeout == -1 means that the thread doesn't plan on waking up automatically
+	if (ns == -1) {
+		return std::numeric_limits<u64>::max();
+	}
+
+	return cpu.getTicks() + Scheduler::nsToCycles(ns);
+}
+
 // See if there is a higher priority, ready thread and switch to that
 void Kernel::rescheduleThreads() {
 	Thread& current = threads[currentThreadIndex];  // Current running thread
@@ -368,13 +371,30 @@ void Kernel::sleepThread(s64 ns) {
 			if (index != idleThreadIndex) {
 				switchThread(index);
 			}
+		} else {
+			if (currentThreadIndex == idleThreadIndex) {
+				const Scheduler& scheduler = cpu.getScheduler();
+				u64 timestamp = scheduler.nextTimestamp;
+
+				for (auto i : threadIndices) {
+					const Thread& t = threads[i];
+					if (t.status == ThreadStatus::WaitSleep || t.status == ThreadStatus::WaitSync1 || t.status == ThreadStatus::WaitSyncAny ||
+						t.status == ThreadStatus::WaitSyncAll) {
+						timestamp = std::min<u64>(timestamp, t.wakeupTick);
+					}
+				}
+
+				if (timestamp > scheduler.currentTimestamp) {
+					u64 idleCycles = timestamp - scheduler.currentTimestamp;
+					cpu.addTicks(idleCycles);
+				}
+			}
 		}
 	} else {  // If we're sleeping for >= 0 ns
 		Thread& t = threads[currentThreadIndex];
 
 		t.status = ThreadStatus::WaitSleep;
-		t.waitingNanoseconds = ns;
+		t.wakeupTick = getWakeupTick(ns);
-		t.sleepTick = cpu.getTicks();
 
 		requireReschedule();
 	}