support multi npu partially

csrc/idle_offload/shm_worker.h

@@ -1,21 +1,27 @@
 #pragma once
 
+#include <vector>
+#include <atomic>
+#include <thread>
+
 #include "shm_helper.h"
 
+
 class ShmWorker {
  public:
   ShmWorker();
   ~ShmWorker();
 
-  bool register_worker(int32_t tgid, uint64_t *out_shareable_handle,
+  bool register_worker(int32_t tgid, int gpu_id, uint64_t *out_shareable_handle,
                        uint64_t *out_vmem_size);
 
-  bool lock_gpu();
+  bool try_lock_gpu(bool &out_self_hold);
+  bool lock_gpu(bool &out_self_hold);
   void unlock_gpu();
 
-  
  private:
   int32_t tgid;
+  int gpu_id;
   ShmHelper *shm_helper;
   std::thread heart_beat_thread;
   std::atomic<bool> stop_heart_beat;