vk/gl: Fix flush synchronization to be kinder to weaker CPUs but not harm higher end CPUs

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -1557,12 +1557,8 @@ bool GLGSRender::on_access_violation(u32 address, bool is_writing)
 		work_item &task = post_flush_request(address, result);
 
 		vm::temporary_unlock();
-		{
+		task.producer_wait();
-			std::unique_lock<std::mutex> lock(task.guard_mutex);
-			task.cv.wait(lock, [&task] { return task.processed; });
-		}
 
-		task.received = true;
 		return true;
 	}
 
@@ -1594,13 +1590,8 @@ void GLGSRender::do_local_task(rsx::FIFO_state state)
 		{
 			if (q.processed) continue;
 
-			std::unique_lock<std::mutex> lock(q.guard_mutex);
 			q.result = m_gl_texture_cache.flush_all(q.section_data);
 			q.processed = true;
-
-			//Notify thread waiting on this
-			lock.unlock();
-			q.cv.notify_one();
 		}
 	}
 	else if (!in_begin_end && state != rsx::FIFO_state::lock_wait)

rpcs3/Emu/RSX/GL/GLGSRender.h

@@ -35,15 +35,23 @@ namespace gl
 
 struct work_item
 {
-	std::condition_variable cv;
-	std::mutex guard_mutex;
-
 	u32  address_to_flush = 0;
 	gl::texture_cache::thrashed_set section_data;
 
 	volatile bool processed = false;
 	volatile bool result = false;
 	volatile bool received = false;
+
+	void producer_wait()
+	{
+		while (!processed)
+		{
+			_mm_lfence();
+			std::this_thread::yield();
+		}
+
+		received = true;
+	}
 };
 
 struct driver_state

rpcs3/Emu/RSX/VK/VKGSRender.h

@@ -241,7 +241,7 @@ struct flush_request_task
 		while (pending_state.load())
 		{
 			_mm_lfence();
-			_mm_pause();
+			std::this_thread::yield();
 		}
 	}
 };