feat: improved task counting thread safety

src/simulation.cpp

@@ -17,7 +17,7 @@
 #define likely(cond_) (cond_)
 #endif
 
-#define SIM_MULTITHREADING 0
+#define SIM_MULTITHREADING 1
 
 namespace simulation {
 bool drawDebugInfo{ true };
@@ -35,6 +35,7 @@ static std::vector<std::shared_ptr<ThreadWorkload>> underpopulated{};
 static std::vector<std::shared_ptr<ThreadWorkload>> born{};
 static uint64_t generationStartTime{ 0 };
 static unsigned tasks{ 0 }; std::mutex tasksMutex{};
+static std::condition_variable tasksChanged{};
 
 CellIterator::CellIterator(Cell begin, Cell end)
 : state{ begin } , begin{ begin }, end{ end } {}
@@ -77,14 +78,28 @@ static size_t CountNeighbors(Cell const &cell) {
 	return count;
 }
 
+static void BlockUntilTasksDone() {
+	std::unique_lock lock{ tasksMutex };
+	while (tasks > 0) {
+		tasksChanged.wait(lock);
+	}
+}
+
+static void TaskBegin() {
+	std::scoped_lock lock{ tasksMutex };
+	tasks++;
+}
+
 static void TaskComplete() {
 	std::scoped_lock lock{ tasksMutex };
 	if (--tasks == 0) {
 		SDL_Log("Generation Complete %lfs", (double)(SDL_GetTicksNS() - generationStartTime) * 0.0000000001);
 	}
+	tasksChanged.notify_all();
 }
 
 static void FindOverpopulated(std::shared_ptr<ThreadWorkload> wl) {
+	TaskBegin();
 	std::scoped_lock lock{ wl->mtx };
 	wl->changes.clear();
 
@@ -102,6 +117,7 @@ static void FindOverpopulated(std::shared_ptr<ThreadWorkload> wl) {
 }
 
 static void FindUnderpopulated(std::shared_ptr<ThreadWorkload> wl) {
+	TaskBegin();
 	std::scoped_lock lock{ wl->mtx };
 	wl->changes.clear();
 
@@ -119,6 +135,7 @@ static void FindUnderpopulated(std::shared_ptr<ThreadWorkload> wl) {
 }
 
 static void FindBorn(std::shared_ptr<ThreadWorkload> wl) {
+	TaskBegin();
 	std::scoped_lock lock{ wl->mtx };
 	wl->changes.clear();
 
@@ -139,6 +156,7 @@ static void FindBorn(std::shared_ptr<ThreadWorkload> wl) {
 static void PopulateChanges() {
 	static bool first_run{ true };
 #if SIM_MULTITHREADING
+	BlockUntilTasksDone();
 	constexpr size_t split{ 4 };
 	if (first_run) {
 		first_run = false;
@@ -148,8 +166,6 @@ static void PopulateChanges() {
 	}
 
 	SDL_Log("Multithreading ON");
-	{ std::scoped_lock lock{ tasksMutex };
-		tasks = 3 * split; }
 	generationStartTime = SDL_GetTicksNS();
 	size_t const seg_length{ living.size() / split };
 	for (size_t i{ 0 }; i < split; ++i) {
@@ -196,6 +212,7 @@ static void PopulateChanges() {
 }
 
 void InitializeRandom(size_t livingChance, int64_t fillArea) {
+	BlockUntilTasksDone();
 	living.clear();
 
 	Cell itr{ 0, 0 };
@@ -213,6 +230,7 @@ void InitializeRandom(size_t livingChance, int64_t fillArea) {
 }
 
 void Step() {
+	BlockUntilTasksDone();
 	for (std::shared_ptr<ThreadWorkload> wl : underpopulated) {
 		if (wl == nullptr) continue;
 		std::scoped_lock l{ wl->mtx };
@@ -258,6 +276,7 @@ void Draw(SDL_Renderer *renderer, double cellSizePercent) {
 	if (!drawDebugInfo) {
 		return;
 	}
+	BlockUntilTasksDone();
 	for (std::shared_ptr<ThreadWorkload> wl : overpopulated) {
 		std::scoped_lock l{ wl->mtx };
 		SDL_SetRenderDrawColor(renderer, 255, 0, 0, 255);