Merge pull request #1103 from odin-lang/new-thread-pool

Improved Thread Pool implementation for the Compiler

9 files changed, 346 insertions, 188 deletions

diff --git a/src/checker.cpp b/src/checker.cpp
index 7e9b2c672..e3af570f8 100644
--- a/src/checker.cpp
+++ b/src/checker.cpp
@@ -4140,23 +4140,20 @@ struct ThreadProcCheckerSection {
 };
 
 
-void check_with_workers(Checker *c, ThreadProc *proc, isize total_count) {
+void check_with_workers(Checker *c, WorkerTaskProc *proc, isize total_count) {
 	isize thread_count = gb_max(build_context.thread_count, 1);
 	isize worker_count = thread_count-1; // NOTE(bill): The main thread will also be used for work
 	if (!build_context.threaded_checker) {
 		worker_count = 0;
 	}
-
+	
 	semaphore_post(&c->info.collect_semaphore, cast(i32)thread_count);
-
 	if (worker_count == 0) {
 		ThreadProcCheckerSection section_all = {};
 		section_all.checker = c;
 		section_all.offset = 0;
 		section_all.count = total_count;
-		Thread dummy_main_thread = {};
-		dummy_main_thread.user_data = &section_all;
-		proc(&dummy_main_thread);
+		proc(&section_all);
 		return;
 	}
 
@@ -4173,33 +4170,18 @@ void check_with_workers(Checker *c, ThreadProc *proc, isize total_count) {
 	}
 	GB_ASSERT(remaining_count <= 0);
 
-	Thread *threads = gb_alloc_array(permanent_allocator(), Thread, worker_count);
-	for (isize i = 0; i < worker_count; i++) {
-		thread_init(threads+i);
-	}
-
-	for (isize i = 0; i < worker_count; i++) {
-		thread_start(threads+i, proc, thread_data+i);
+	
+	for (isize i = 0; i < thread_count; i++) {
+		global_thread_pool_add_task(proc, thread_data+i);
 	}
-	Thread dummy_main_thread = {};
-	dummy_main_thread.user_data = thread_data+worker_count;
-	proc(&dummy_main_thread);
-
+	global_thread_pool_wait();
 	semaphore_wait(&c->info.collect_semaphore);
-
-	for (isize i = 0; i < worker_count; i++) {
-		thread_join(threads+i);
-	}
-
-	for (isize i = 0; i < worker_count; i++) {
-		thread_destroy(threads+i);
-	}
 }
 
 
-THREAD_PROC(thread_proc_collect_entities) {
-	auto *data = cast(ThreadProcCheckerSection *)thread->user_data;
-	Checker *c = data->checker;
+WORKER_TASK_PROC(thread_proc_collect_entities) {
+	auto *cs = cast(ThreadProcCheckerSection *)data;
+	Checker *c = cs->checker;
 	CheckerContext collect_entity_ctx = make_checker_context(c);
 	defer (destroy_checker_context(&collect_entity_ctx));
 
@@ -4208,8 +4190,8 @@ THREAD_PROC(thread_proc_collect_entities) {
 	UntypedExprInfoMap untyped = {};
 	map_init(&untyped, heap_allocator());
 
-	isize offset = data->offset;
-	isize file_end = gb_min(offset+data->count, c->info.files.entries.count);
+	isize offset = cs->offset;
+	isize file_end = gb_min(offset+cs->count, c->info.files.entries.count);
 
 	for (isize i = offset; i < file_end; i++) {
 		AstFile *f = c->info.files.entries[i].value;
@@ -4246,9 +4228,9 @@ void check_export_entities_in_pkg(CheckerContext *ctx, AstPackage *pkg, UntypedE
 	}
 }
 
-THREAD_PROC(thread_proc_check_export_entities) {
-	auto data = cast(ThreadProcCheckerSection *)thread->user_data;
-	Checker *c = data->checker;
+WORKER_TASK_PROC(thread_proc_check_export_entities) {
+	auto cs = cast(ThreadProcCheckerSection *)data;
+	Checker *c = cs->checker;
 
 	CheckerContext ctx = make_checker_context(c);
 	defer (destroy_checker_context(&ctx));
@@ -4256,8 +4238,8 @@ THREAD_PROC(thread_proc_check_export_entities) {
 	UntypedExprInfoMap untyped = {};
 	map_init(&untyped, heap_allocator());
 
-	isize end = gb_min(data->offset + data->count, c->info.packages.entries.count);
-	for (isize i = data->offset; i < end; i++) {
+	isize end = gb_min(cs->offset + cs->count, c->info.packages.entries.count);
+	for (isize i = cs->offset; i < end; i++) {
 		AstPackage *pkg = c->info.packages.entries[i].value;
 		check_export_entities_in_pkg(&ctx, pkg, &untyped);
 	}
@@ -4575,15 +4557,19 @@ void calculate_global_init_order(Checker *c) {
 }
 
 
-void check_proc_info(Checker *c, ProcInfo *pi, UntypedExprInfoMap *untyped, ProcBodyQueue *procs_to_check_queue) {
+bool check_proc_info(Checker *c, ProcInfo *pi, UntypedExprInfoMap *untyped, ProcBodyQueue *procs_to_check_queue) {
 	if (pi == nullptr) {
-		return;
+		return false;
 	}
 	if (pi->type == nullptr) {
-		return;
+		return false;
 	}
+	Entity *e = pi->decl->entity;
 	if (pi->decl->proc_checked) {
-		return;
+		if (e != nullptr) {
+			GB_ASSERT(e->flags & EntityFlag_ProcBodyChecked);
+		}
+		return true;
 	}
 
 	CheckerContext ctx = make_checker_context(c);
@@ -4601,14 +4587,13 @@ void check_proc_info(Checker *c, ProcInfo *pi, UntypedExprInfoMap *untyped, Proc
 			token = ast_token(pi->poly_def_node);
 		}
 		error(token, "Unspecialized polymorphic procedure '%.*s'", LIT(name));
-		return;
+		return false;
 	}
 
 	if (pt->is_polymorphic && pt->is_poly_specialized) {
-		Entity *e = pi->decl->entity;
 		if ((e->flags & EntityFlag_Used) == 0) {
 			// NOTE(bill, 2019-08-31): It was never used, don't check
-			return;
+			return false;
 		}
 	}
 
@@ -4622,16 +4607,17 @@ void check_proc_info(Checker *c, ProcInfo *pi, UntypedExprInfoMap *untyped, Proc
 		ctx.state_flags |= StateFlag_no_bounds_check;
 		ctx.state_flags &= ~StateFlag_bounds_check;
 	}
-	if (pi->body != nullptr && pi->decl->entity != nullptr) {
-		GB_ASSERT((pi->decl->entity->flags & EntityFlag_ProcBodyChecked) == 0);
+	if (pi->body != nullptr && e != nullptr) {
+		GB_ASSERT((e->flags & EntityFlag_ProcBodyChecked) == 0);
 	}
 
 	check_proc_body(&ctx, pi->token, pi->decl, pi->type, pi->body);
-	if (pi->body != nullptr && pi->decl->entity != nullptr) {
-		pi->decl->entity->flags |= EntityFlag_ProcBodyChecked;
+	if (e != nullptr) {
+		e->flags |= EntityFlag_ProcBodyChecked;
 	}
 	pi->decl->proc_checked = true;
 	add_untyped_expressions(&c->info, ctx.untyped);
+	return true;
 }
 
 GB_STATIC_ASSERT(sizeof(isize) == sizeof(void *));
@@ -4681,9 +4667,10 @@ void check_unchecked_bodies(Checker *c) {
 	ProcInfo *pi = nullptr;
 	while (mpmc_dequeue(q, &pi)) {
 		Entity *e = pi->decl->entity;
-		consume_proc_info_queue(c, pi, q, &untyped);
-		add_dependency_to_set(c, e);
-		GB_ASSERT(e->flags & EntityFlag_ProcBodyChecked);
+		if (consume_proc_info_queue(c, pi, q, &untyped)) {
+			add_dependency_to_set(c, e);
+			GB_ASSERT(e->flags & EntityFlag_ProcBodyChecked);
+		}
 	}
 
 }
@@ -4728,15 +4715,15 @@ bool consume_proc_info_queue(Checker *c, ProcInfo *pi, ProcBodyQueue *q, Untyped
 		// NOTE(bill): In single threaded mode, this should never happen
 		if (parent->kind == Entity_Procedure && (parent->flags & EntityFlag_ProcBodyChecked) == 0) {
 			mpmc_enqueue(q, pi);
-			return true;
+			return false;
 		}
 	}
 	if (untyped) {
 		map_clear(untyped);
 	}
-	check_proc_info(c, pi, untyped, q);
+	bool ok = check_proc_info(c, pi, untyped, q);
 	total_bodies_checked.fetch_add(1, std::memory_order_relaxed);
-	return false;
+	return ok;
 }
 
 struct ThreadProcBodyData {
@@ -4747,11 +4734,11 @@ struct ThreadProcBodyData {
 	ThreadProcBodyData *all_data;
 };
 
-THREAD_PROC(thread_proc_body) {
-	ThreadProcBodyData *data = cast(ThreadProcBodyData *)thread->user_data;
-	Checker *c = data->checker;
+WORKER_TASK_PROC(thread_proc_body) {
+	ThreadProcBodyData *bd = cast(ThreadProcBodyData *)data;
+	Checker *c = bd->checker;
 	GB_ASSERT(c != nullptr);
-	ProcBodyQueue *this_queue = data->queue;
+	ProcBodyQueue *this_queue = bd->queue;
 
 	UntypedExprInfoMap untyped = {};
 	map_init(&untyped, heap_allocator());
@@ -4774,12 +4761,18 @@ void check_procedure_bodies(Checker *c) {
 	u32 worker_count = thread_count-1; // NOTE(bill): The main thread will also be used for work
 	if (!build_context.threaded_checker) {
 		worker_count = 0;
-
-		auto *q = &c->procs_to_check_queue;
-		ProcInfo *pi = nullptr;
-		while (mpmc_dequeue(q, &pi)) {
-			consume_proc_info_queue(c, pi, q, nullptr);
+	}
+	if (worker_count == 0) {
+		auto *this_queue = &c->procs_to_check_queue;
+		
+		UntypedExprInfoMap untyped = {};
+		map_init(&untyped, heap_allocator());
+		
+		for (ProcInfo *pi = nullptr; mpmc_dequeue(this_queue, &pi); /**/) {
+			consume_proc_info_queue(c, pi, this_queue, &untyped);
 		}
+		
+		map_destroy(&untyped);
 
 		debugf("Total Procedure Bodies Checked: %td\n", total_bodies_checked.load(std::memory_order_relaxed));
 		return;
@@ -4821,31 +4814,14 @@ void check_procedure_bodies(Checker *c) {
 	}
 	GB_ASSERT(total_queued == original_queue_count);
 
-
 	semaphore_post(&c->procs_to_check_semaphore, cast(i32)thread_count);
-
-	Thread *threads = gb_alloc_array(permanent_allocator(), Thread, worker_count);
-	for (isize i = 0; i < worker_count; i++) {
-		thread_init(threads+i);
-	}
-
-	for (isize i = 0; i < worker_count; i++) {
-		thread_start(threads+i, thread_proc_body, thread_data+i);
+	
+	for (isize i = 0; i < thread_count; i++) {
+		global_thread_pool_add_task(thread_proc_body, thread_data+i);
 	}
-	Thread dummy_main_thread = {};
-	dummy_main_thread.user_data = thread_data+worker_count;
-	thread_proc_body(&dummy_main_thread);
-
+	global_thread_pool_wait();
 	semaphore_wait(&c->procs_to_check_semaphore);
 
-	for (isize i = 0; i < worker_count; i++) {
-		thread_join(threads+i);
-	}
-
-	for (isize i = 0; i < worker_count; i++) {
-		thread_destroy(threads+i);
-	}
-
 	isize global_remaining = c->procs_to_check_queue.count.load(std::memory_order_relaxed);
 	GB_ASSERT(global_remaining == 0);
 
diff --git a/src/common.cpp b/src/common.cpp
index af0a195a8..b132c26b0 100644
--- a/src/common.cpp
+++ b/src/common.cpp
@@ -44,11 +44,9 @@ void debugf(char const *fmt, ...);
 #include "queue.cpp"
 #include "common_memory.cpp"
 #include "string.cpp"
-
-
-
 #include "range_cache.cpp"
 
+
 u32 fnv32a(void const *data, isize len) {
 	u8 const *bytes = cast(u8 const *)data;
 	u32 h = 0x811c9dc5;
diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp
index d00883bd9..034814bf1 100644
--- a/src/llvm_backend.cpp
+++ b/src/llvm_backend.cpp
@@ -1104,9 +1104,6 @@ void lb_generate_code(lbGenerator *gen) {
 
 	LLVMBool do_threading = (LLVMIsMultithreaded() && USE_SEPARATE_MODULES && MULTITHREAD_OBJECT_GENERATION && worker_count > 0);
 
-	thread_pool_init(&lb_thread_pool, heap_allocator(), worker_count, "LLVMBackend");
-	defer (thread_pool_destroy(&lb_thread_pool));
-
 	lbModule *default_module = &gen->default_module;
 	CheckerInfo *info = gen->info;
 
@@ -1691,10 +1688,10 @@ void lb_generate_code(lbGenerator *gen) {
 			wd->code_gen_file_type = code_gen_file_type;
 			wd->filepath_obj = filepath_obj;
 			wd->m = m;
-			thread_pool_add_task(&lb_thread_pool, lb_llvm_emit_worker_proc, wd);
+			global_thread_pool_add_task(lb_llvm_emit_worker_proc, wd);
 		}
 
-		thread_pool_wait(&lb_thread_pool);
+		thread_pool_wait(&global_thread_pool);
 	} else {
 		for_array(j, gen->modules.entries) {
 			lbModule *m = gen->modules.entries[j].value;
diff --git a/src/llvm_backend_general.cpp b/src/llvm_backend_general.cpp
index 94fe1e562..2348ac523 100644
--- a/src/llvm_backend_general.cpp
+++ b/src/llvm_backend_general.cpp
@@ -1,7 +1,5 @@
 void lb_add_debug_local_variable(lbProcedure *p, LLVMValueRef ptr, Type *type, Token const &token);
 
-gb_global ThreadPool lb_thread_pool = {};
-
 gb_global Entity *lb_global_type_info_data_entity   = {};
 gb_global lbAddr lb_global_type_info_member_types   = {};
 gb_global lbAddr lb_global_type_info_member_names   = {};
diff --git a/src/main.cpp b/src/main.cpp
index dab1257bd..f94b6bd4f 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -7,6 +7,20 @@
 #include "exact_value.cpp"
 #include "build_settings.cpp"
 
+gb_global ThreadPool global_thread_pool;
+void init_global_thread_pool(void) {
+	isize thread_count = gb_max(build_context.thread_count, 1);
+	isize worker_count = thread_count-1; // NOTE(bill): The main thread will also be used for work
+	thread_pool_init(&global_thread_pool, permanent_allocator(), worker_count, "ThreadPoolWorker");
+}
+bool global_thread_pool_add_task(WorkerTaskProc *proc, void *data) {
+	return thread_pool_add_task(&global_thread_pool, proc, data);
+}
+void global_thread_pool_wait(void) {
+	thread_pool_wait(&global_thread_pool);
+}
+
+
 void debugf(char const *fmt, ...) {
 	if (build_context.show_debug_messages) {
 		gb_printf_err("[DEBUG] ");
@@ -2160,6 +2174,9 @@ int main(int arg_count, char const **arg_ptr) {
 	// 	return 1;
 	// }
 
+	init_global_thread_pool();
+	defer (thread_pool_destroy(&global_thread_pool));
+	
 	init_universal();
 	// TODO(bill): prevent compiling without a linker
 
diff --git a/src/parser.cpp b/src/parser.cpp
index a98259454..d8f3a9e33 100644
--- a/src/parser.cpp
+++ b/src/parser.cpp
@@ -4810,6 +4810,7 @@ bool init_parser(Parser *p) {
 	string_set_init(&p->imported_files, heap_allocator());
 	array_init(&p->packages, heap_allocator());
 	array_init(&p->package_imports, heap_allocator());
+	mutex_init(&p->wait_mutex);
 	mutex_init(&p->import_mutex);
 	mutex_init(&p->file_add_mutex);
 	mutex_init(&p->file_decl_mutex);
@@ -4837,6 +4838,7 @@ void destroy_parser(Parser *p) {
 	array_free(&p->packages);
 	array_free(&p->package_imports);
 	string_set_destroy(&p->imported_files);
+	mutex_destroy(&p->wait_mutex);
 	mutex_destroy(&p->import_mutex);
 	mutex_destroy(&p->file_add_mutex);
 	mutex_destroy(&p->file_decl_mutex);
@@ -4870,7 +4872,7 @@ void parser_add_file_to_process(Parser *p, AstPackage *pkg, FileInfo fi, TokenPo
 	auto wd = gb_alloc_item(heap_allocator(), ParserWorkerData);
 	wd->parser = p;
 	wd->imported_file = f;
-	thread_pool_add_task(&parser_thread_pool, parser_worker_proc, wd);
+	global_thread_pool_add_task(parser_worker_proc, wd);
 }
 
 WORKER_TASK_PROC(foreign_file_worker_proc) {
@@ -4909,7 +4911,7 @@ void parser_add_foreign_file_to_process(Parser *p, AstPackage *pkg, AstForeignFi
 	wd->parser = p;
 	wd->imported_file = f;
 	wd->foreign_kind = kind;
-	thread_pool_add_task(&parser_thread_pool, foreign_file_worker_proc, wd);
+	global_thread_pool_add_task(foreign_file_worker_proc, wd);
 }
 
 
@@ -5619,10 +5621,6 @@ ParseFileError process_imported_file(Parser *p, ImportedFile imported_file) {
 ParseFileError parse_packages(Parser *p, String init_filename) {
 	GB_ASSERT(init_filename.text[init_filename.len] == 0);
 
-	isize thread_count = gb_max(build_context.thread_count, 1);
-	isize worker_count = thread_count-1; // NOTE(bill): The main thread will also be used for work
-	thread_pool_init(&parser_thread_pool, heap_allocator(), worker_count, "ParserWork");
-
 	String init_fullpath = path_to_full_path(heap_allocator(), init_filename);
 	if (!path_is_directory(init_fullpath)) {
 		String const ext = str_lit(".odin");
@@ -5631,38 +5629,45 @@ ParseFileError parse_packages(Parser *p, String init_filename) {
 			return ParseFile_WrongExtension;
 		}
 	}
+	
 
-	TokenPos init_pos = {};
-	{
-		String s = get_fullpath_core(heap_allocator(), str_lit("runtime"));
-		try_add_import_path(p, s, s, init_pos, Package_Runtime);
-	}
+	{ // Add these packages serially and then process them parallel
+		mutex_lock(&p->wait_mutex);
+		defer (mutex_unlock(&p->wait_mutex));
+		
+		TokenPos init_pos = {};
+		{
+			String s = get_fullpath_core(heap_allocator(), str_lit("runtime"));
+			try_add_import_path(p, s, s, init_pos, Package_Runtime);
+		}
 
-	try_add_import_path(p, init_fullpath, init_fullpath, init_pos, Package_Init);
-	p->init_fullpath = init_fullpath;
+		try_add_import_path(p, init_fullpath, init_fullpath, init_pos, Package_Init);
+		p->init_fullpath = init_fullpath;
 
-	if (build_context.command_kind == Command_test) {
-		String s = get_fullpath_core(heap_allocator(), str_lit("testing"));
-		try_add_import_path(p, s, s, init_pos, Package_Normal);
-	}
-
-	for_array(i, build_context.extra_packages) {
-		String path = build_context.extra_packages[i];
-		String fullpath = path_to_full_path(heap_allocator(), path); // LEAK?
-		if (!path_is_directory(fullpath)) {
-			String const ext = str_lit(".odin");
-			if (!string_ends_with(fullpath, ext)) {
-				error_line("Expected either a directory or a .odin file, got '%.*s'\n", LIT(fullpath));
-				return ParseFile_WrongExtension;
-			}
+		if (build_context.command_kind == Command_test) {
+			String s = get_fullpath_core(heap_allocator(), str_lit("testing"));
+			try_add_import_path(p, s, s, init_pos, Package_Normal);
 		}
-		AstPackage *pkg = try_add_import_path(p, fullpath, fullpath, init_pos, Package_Normal);
-		if (pkg) {
-			pkg->is_extra = true;
+		
+
+		for_array(i, build_context.extra_packages) {
+			String path = build_context.extra_packages[i];
+			String fullpath = path_to_full_path(heap_allocator(), path); // LEAK?
+			if (!path_is_directory(fullpath)) {
+				String const ext = str_lit(".odin");
+				if (!string_ends_with(fullpath, ext)) {
+					error_line("Expected either a directory or a .odin file, got '%.*s'\n", LIT(fullpath));
+					return ParseFile_WrongExtension;
+				}
+			}
+			AstPackage *pkg = try_add_import_path(p, fullpath, fullpath, init_pos, Package_Normal);
+			if (pkg) {
+				pkg->is_extra = true;
+			}
 		}
 	}
-
-	thread_pool_wait(&parser_thread_pool);
+	
+	global_thread_pool_wait();
 
 	for (ParseFileError err = ParseFile_None; mpmc_dequeue(&p->file_error_queue, &err); /**/) {
 		if (err != ParseFile_None) {
diff --git a/src/parser.hpp b/src/parser.hpp
index 082415d16..ddcb27322 100644
--- a/src/parser.hpp
+++ b/src/parser.hpp
@@ -191,6 +191,7 @@ struct Parser {
 	isize                     file_to_process_count;
 	isize                     total_token_count;
 	isize                     total_line_count;
+	BlockingMutex             wait_mutex;
 	BlockingMutex             import_mutex;
 	BlockingMutex             file_add_mutex;
 	BlockingMutex             file_decl_mutex;
@@ -198,9 +199,6 @@ struct Parser {
 	MPMCQueue<ParseFileError> file_error_queue;
 };
 
-
-gb_global ThreadPool parser_thread_pool = {};
-
 struct ParserWorkerData {
 	Parser *parser;
 	ImportedFile imported_file;
diff --git a/src/thread_pool.cpp b/src/thread_pool.cpp
index b29fcc5d9..a8bc327e5 100644
--- a/src/thread_pool.cpp
+++ b/src/thread_pool.cpp
@@ -4,85 +4,156 @@
 typedef WORKER_TASK_PROC(WorkerTaskProc);
 
 struct WorkerTask {
-	WorkerTask *next_task;
+	WorkerTask *    next;
 	WorkerTaskProc *do_work;
-	void *data;
+	void *          data;
 };
 
 struct ThreadPool {
-	std::atomic<isize> outstanding_task_count;
-	WorkerTask *volatile next_task;
-	BlockingMutex task_list_mutex;
-	isize thread_count;
+	gbAllocator   allocator;
+	BlockingMutex mutex;
+	Condition     task_cond;
+	
+	Slice<Thread> threads;
+	
+	WorkerTask *task_queue;
+	
+	std::atomic<isize> ready;
+	std::atomic<bool>  stop;
+	
 };
 
-void thread_pool_thread_entry(ThreadPool *pool) {
-	while (pool->outstanding_task_count) {
-		if (!pool->next_task) {
-			yield(); // No need to grab the mutex.
-		} else {
-			mutex_lock(&pool->task_list_mutex);
-
-			if (pool->next_task) {
-				WorkerTask *task = pool->next_task;
-				pool->next_task = task->next_task;
-				mutex_unlock(&pool->task_list_mutex);
-				task->do_work(task->data);
-				pool->outstanding_task_count.fetch_sub(1);
-				gb_free(heap_allocator(), task);
-			} else {
-				mutex_unlock(&pool->task_list_mutex);
-			}
-		}
+THREAD_PROC(thread_pool_thread_proc);
+
+void thread_pool_init(ThreadPool *pool, gbAllocator const &a, isize thread_count, char const *worker_name) {
+	pool->allocator = a;
+	pool->stop = false;
+	mutex_init(&pool->mutex);
+	condition_init(&pool->task_cond);
+	
+	slice_init(&pool->threads, a, thread_count);
+	for_array(i, pool->threads) {
+		Thread *t = &pool->threads[i];
+		thread_init(t);
+	}
+	
+	for_array(i, pool->threads) {
+		Thread *t = &pool->threads[i];
+		thread_start(t, thread_pool_thread_proc, pool);
 	}
 }
 
-#if defined(GB_SYSTEM_WINDOWS)
-	DWORD __stdcall thread_pool_thread_entry_platform(void *arg) {
-		thread_pool_thread_entry((ThreadPool *) arg);
-		return 0;
-	}
+void thread_pool_destroy(ThreadPool *pool) {
+	pool->stop = true;
+	condition_broadcast(&pool->task_cond);
 
-	void thread_pool_start_thread(ThreadPool *pool) {
-		CloseHandle(CreateThread(NULL, 0, thread_pool_thread_entry_platform, pool, 0, NULL));
+	for_array(i, pool->threads) {
+		Thread *t = &pool->threads[i];
+		thread_join(t);
 	}
-#else
-	void *thread_pool_thread_entry_platform(void *arg) {
-		thread_pool_thread_entry((ThreadPool *) arg);
-		return NULL;
+	
+	for_array(i, pool->threads) {
+		Thread *t = &pool->threads[i];
+		thread_destroy(t);
 	}
+	
+	gb_free(pool->allocator, pool->threads.data);
+	mutex_destroy(&pool->mutex);
+	condition_destroy(&pool->task_cond);
+}
 
-	void thread_pool_start_thread(ThreadPool *pool) {
-		pthread_t handle;
-		pthread_create(&handle, NULL, thread_pool_thread_entry_platform, pool);
-		pthread_detach(handle);
-	}
-#endif
+bool thread_pool_queue_empty(ThreadPool *pool) {
+	return pool->task_queue == nullptr;
+}
 
-void thread_pool_init(ThreadPool *pool, gbAllocator const &a, isize thread_count, char const *worker_prefix) {
-	memset(pool, 0, sizeof(ThreadPool));
-	mutex_init(&pool->task_list_mutex);
-	pool->thread_count = thread_count;
+WorkerTask *thread_pool_queue_pop(ThreadPool *pool) {
+	GB_ASSERT(pool->task_queue != nullptr);
+	WorkerTask *task = pool->task_queue;
+	pool->task_queue = task->next;
+	return task;
+}
+void thread_pool_queue_push(ThreadPool *pool, WorkerTask *task) {
+	GB_ASSERT(task != nullptr);
+	task->next = pool->task_queue;
+	pool->task_queue = task;
 }
 
-void thread_pool_destroy(ThreadPool *pool) {
-	mutex_destroy(&pool->task_list_mutex);
+bool thread_pool_add_task(ThreadPool *pool, WorkerTaskProc *proc, void *data) {
+	GB_ASSERT(proc != nullptr);
+	mutex_lock(&pool->mutex);
+	WorkerTask *task = gb_alloc_item(permanent_allocator(), WorkerTask);
+	if (task == nullptr) {
+		mutex_unlock(&pool->mutex);
+		GB_PANIC("Out of memory");
+		return false;
+	}
+	task->do_work = proc;
+	task->data = data;
+		
+	thread_pool_queue_push(pool, task);
+	GB_ASSERT(pool->ready >= 0);
+	pool->ready++;
+	mutex_unlock(&pool->mutex);
+	condition_signal(&pool->task_cond);
+	return true;
+}	
+
+
+void thread_pool_do_task(WorkerTask *task) {
+	task->do_work(task->data);
 }
 
 void thread_pool_wait(ThreadPool *pool) {
-	for (int i = 0; i < pool->thread_count; i++) {
-		thread_pool_start_thread(pool);
+	if (pool->threads.count == 0) {
+		while (!thread_pool_queue_empty(pool)) {
+			thread_pool_do_task(thread_pool_queue_pop(pool));
+			--pool->ready;
+		}
+		GB_ASSERT(pool->ready == 0);
+		return;
 	}
-	thread_pool_thread_entry(pool);
-}
+	for (;;) {
+		mutex_lock(&pool->mutex);
 
-void thread_pool_add_task(ThreadPool *pool, WorkerTaskProc *proc, void *data) {
-	WorkerTask *task = gb_alloc_item(heap_allocator(), WorkerTask);
-	task->do_work = proc;
-	task->data = data;
-	mutex_lock(&pool->task_list_mutex);
-	task->next_task = pool->next_task;
-	pool->next_task = task;
-	pool->outstanding_task_count.fetch_add(1);
-	mutex_unlock(&pool->task_list_mutex);
+		while (!pool->stop && pool->ready > 0 && thread_pool_queue_empty(pool)) {
+			condition_wait(&pool->task_cond, &pool->mutex);
+		}
+		if ((pool->stop || pool->ready == 0) && thread_pool_queue_empty(pool)) {
+			mutex_unlock(&pool->mutex);
+			return;
+		}
+
+		WorkerTask *task = thread_pool_queue_pop(pool);
+		mutex_unlock(&pool->mutex);
+	
+		thread_pool_do_task(task);
+		if (--pool->ready == 0) {
+			condition_broadcast(&pool->task_cond);
+		}
+	}
 }
+
+
+THREAD_PROC(thread_pool_thread_proc) {
+	ThreadPool *pool = cast(ThreadPool *)thread->user_data;
+	
+	for (;;) {
+		mutex_lock(&pool->mutex);
+
+		while (!pool->stop && thread_pool_queue_empty(pool)) {
+			condition_wait(&pool->task_cond, &pool->mutex);
+		}
+		if (pool->stop && thread_pool_queue_empty(pool)) {
+			mutex_unlock(&pool->mutex);
+			return 0;
+		}
+
+		WorkerTask *task = thread_pool_queue_pop(pool);
+		mutex_unlock(&pool->mutex);
+	
+		thread_pool_do_task(task);
+		if (--pool->ready == 0) {
+			condition_broadcast(&pool->task_cond);
+		}
+	}
+}
\ No newline at end of file
diff --git a/src/threading.cpp b/src/threading.cpp
index 349f234f4..61f9df2db 100644
--- a/src/threading.cpp
+++ b/src/threading.cpp
@@ -1,6 +1,7 @@
 struct BlockingMutex;
 struct RecursiveMutex;
 struct Semaphore;
+struct Condition;
 struct Thread;
 
 #define THREAD_PROC(name) isize name(struct Thread *thread)
@@ -41,6 +42,14 @@ void semaphore_post   (Semaphore *s, i32 count);
 void semaphore_wait   (Semaphore *s);
 void semaphore_release(Semaphore *s) { semaphore_post(s, 1); }
 
+
+void condition_init(Condition *c);
+void condition_destroy(Condition *c);
+void condition_broadcast(Condition *c);
+void condition_signal(Condition *c);
+void condition_wait(Condition *c, BlockingMutex *m);
+void condition_wait_with_timeout(Condition *c, BlockingMutex *m, u32 timeout_in_ms);
+
 u32  thread_current_id(void);
 
 void thread_init            (Thread *t);
@@ -108,6 +117,27 @@ void yield_process(void);
 	void semaphore_wait(Semaphore *s) {
 		WaitForSingleObjectEx(s->win32_handle, INFINITE, FALSE);
 	}
+	
+	struct Condition {
+		CONDITION_VARIABLE cond;
+	};
+	
+	void condition_init(Condition *c) {
+	}
+	void condition_destroy(Condition *c) {	
+	}
+	void condition_broadcast(Condition *c) {
+		WakeAllConditionVariable(&c->cond);
+	}
+	void condition_signal(Condition *c) {
+		WakeConditionVariable(&c->cond);
+	}
+	void condition_wait(Condition *c, BlockingMutex *m) {
+		SleepConditionVariableSRW(&c->cond, &m->srwlock, INFINITE, 0);
+	}
+	void condition_wait_with_timeout(Condition *c, BlockingMutex *m, u32 timeout_in_ms) {
+		SleepConditionVariableSRW(&c->cond, &m->srwlock, timeout_in_ms, 0);
+	}
 
 #else
 	struct BlockingMutex {
@@ -170,9 +200,77 @@ void yield_process(void);
 		void semaphore_post   (Semaphore *s, i32 count) { while (count --> 0) sem_post(&s->unix_handle); }
 		void semaphore_wait   (Semaphore *s)            { int i; do { i = sem_wait(&s->unix_handle); } while (i == -1 && errno == EINTR); }
 	#else
-	#error
+	#error Implement Semaphore for this platform
 	#endif
+		
+		
+	struct Condition {
+		pthread_cond_t pthread_cond;
+	};
+	
+	void condition_init(Condition *c) {
+		pthread_cond_init(&c->pthread_cond, NULL);
+	}
+	void condition_destroy(Condition *c) {	
+		pthread_cond_destroy(&c->pthread_cond);
+	}
+	void condition_broadcast(Condition *c) {
+		pthread_cond_broadcast(&c->pthread_cond);
+	}
+	void condition_signal(Condition *c) {
+		pthread_cond_signal(&c->pthread_cond);
+	}
+	void condition_wait(Condition *c, BlockingMutex *m) {
+		pthread_cond_wait(&c->pthread_cond, &m->pthread_mutex);
+	}
+	void condition_wait_with_timeout(Condition *c, BlockingMutex *m, u32 timeout_in_ms) {
+		struct timespec abstime = {};
+		abstime.tv_sec = timeout_in_ms/1000;
+		abstime.tv_nsec = cast(long)(timeout_in_ms%1000)*1e6;
+		pthread_cond_timedwait(&c->pthread_cond, &m->pthread_mutex, &abstime);
+		
+	}
 #endif
+	
+	
+struct Barrier {
+	BlockingMutex mutex;
+	Condition cond;
+	isize index;
+	isize generation_id;
+	isize thread_count;	
+};
+
+void barrier_init(Barrier *b, isize thread_count) {
+	mutex_init(&b->mutex);
+	condition_init(&b->cond);
+	b->index = 0;
+	b->generation_id = 0;
+	b->thread_count = 0;
+}
+
+void barrier_destroy(Barrier *b) {
+	condition_destroy(&b->cond);
+	mutex_destroy(&b->mutex);
+}
+
+// Returns true if it is the leader
+bool barrier_wait(Barrier *b) {
+	mutex_lock(&b->mutex);
+	defer (mutex_unlock(&b->mutex));
+	isize local_gen = b->generation_id;
+	b->index += 1;
+	if (b->index < b->thread_count) {
+		while (local_gen == b->generation_id && b->index < b->thread_count) {
+			condition_wait(&b->cond, &b->mutex);
+		}
+		return false;
+	}
+	b->index = 0;
+	b->generation_id += 1;
+	condition_broadcast(&b->cond);
+	return true;
+}