vk: Clean up around vkQueueSubmit handling

- Explicitly declare one version for CB flush and the other for Async flush
- Always flush descriptors on CB flush in case of page fault handling.
  Other threads other than offloader can also enter the method and require normal flow.
- Fix overlapping interrupt IDs.
- Minor formatting fixes

rpcs3/Emu/RSX/Common/bitfield.hpp

@@ -87,7 +87,7 @@ namespace rsx
 
 		void clear()
 		{
-			m_data.store(0);
+			m_data.release(0);
 		}
 	};
 }

rpcs3/Emu/RSX/VK/VKCommandStream.cpp

@@ -23,19 +23,8 @@ namespace vk
 		g_submit_mutex.unlock();
 	}
 
-	void queue_submit(VkQueue queue, const VkSubmitInfo* info, fence* pfence, VkBool32 flush)
+	FORCE_INLINE
-	{
+	static void queue_submit_impl(VkQueue queue, const VkSubmitInfo* info, fence* pfence)
-		if (rsx::get_current_renderer()->is_current_thread())
-		{
-			vk::descriptors::flush();
-		}
-
-		if (!flush && g_cfg.video.multithreaded_rsx)
-		{
-			auto packet = new submit_packet(queue, pfence, info);
-			g_fxo->get<rsx::dma_manager>().backend_ctrl(rctrl_queue_submit, packet);
-		}
-		else
 	{
 		acquire_global_submit_lock();
 		vkQueueSubmit(queue, 1, info, pfence->handle);
@@ -44,5 +33,27 @@ namespace vk
 		// Signal fence
 		pfence->signal_flushed();
 	}
+
+	void queue_submit(VkQueue queue, const VkSubmitInfo* info, fence* pfence, VkBool32 flush)
+	{
+		// Access to this method must be externally synchronized.
+		// Offloader is guaranteed to never call this for async flushes.
+		vk::descriptors::flush();
+
+		if (!flush && g_cfg.video.multithreaded_rsx)
+		{
+			auto packet = new submit_packet(queue, pfence, info);
+			g_fxo->get<rsx::dma_manager>().backend_ctrl(rctrl_queue_submit, packet);
+		}
+		else
+		{
+			queue_submit_impl(queue, info, pfence);
+		}
+	}
+
+	void queue_submit(const vk::submit_packet* packet)
+	{
+		// Flush-only version used by asynchronous submit processing (MTRSX)
+		queue_submit_impl(packet->queue, &packet->submit_info, packet->pfence);
 	}
 }

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -760,7 +760,7 @@ bool VKGSRender::on_access_violation(u32 address, bool is_writing)
 		}
 
 		bool has_queue_ref = false;
-		if (!is_current_thread())
+		if (!is_current_thread()) [[likely]]
 		{
 			// Always submit primary cb to ensure state consistency (flush pending changes such as image transitions)
 			vm::temporary_unlock();
@@ -2340,8 +2340,8 @@ void VKGSRender::renderctl(u32 request_code, void* args)
 	{
 	case vk::rctrl_queue_submit:
 	{
-		auto packet = reinterpret_cast<vk::submit_packet*>(args);
+		const auto packet = reinterpret_cast<vk::submit_packet*>(args);
-		vk::queue_submit(packet->queue, &packet->submit_info, packet->pfence, VK_TRUE);
+		vk::queue_submit(packet);
 		free(packet);
 		break;
 	}

rpcs3/Emu/RSX/VK/VKHelpers.h

@@ -32,13 +32,13 @@ namespace vk
 	class image;
 	class instance;
 	class render_device;
+	struct submit_packet;
 
-	//VkAllocationCallbacks default_callbacks();
 	enum runtime_state
 	{
 		uninterruptible = 1,
 		heap_dirty      = 2,
-		heap_changed = 3
+		heap_changed    = 4,
 	};
 
 	const vk::render_device *get_current_renderer();
@@ -54,7 +54,7 @@ namespace vk
 	// Sync helpers around vkQueueSubmit
 	void acquire_global_submit_lock();
 	void release_global_submit_lock();
-	void queue_submit(VkQueue queue, const VkSubmitInfo* info, fence* pfence, VkBool32 flush = VK_FALSE);
+	void queue_submit(const vk::submit_packet* packet);
 
 	template<class T>
 	T* get_compute_task();

rpcs3/Emu/RSX/VK/vkutils/commands.cpp

@@ -6,7 +6,7 @@
 namespace vk
 {
 	// This queue flushing method to be implemented by the backend as behavior depends on config
-	void queue_submit(VkQueue queue, const VkSubmitInfo* info, fence* pfence, VkBool32 flush = VK_FALSE);
+	void queue_submit(VkQueue queue, const VkSubmitInfo* info, fence* pfence, VkBool32 flush);
 
 	void command_pool::create(vk::render_device& dev, u32 queue_family_id)
 	{