Add destroy_gpu_instance() function new parameter need_wait about it will wait for all devices to be idle before destroy

src/gpu.cpp

@@ -2091,6 +2091,19 @@ void destroy_gpu_instance()
     if (g_instance.created == 0)
         return;
 
+    for (int i = 0; i < NCNN_MAX_GPU_COUNT; i++)
+    {
+        VulkanDevice* vulkan_device = g_default_vkdev[i];
+        if (vulkan_device)
+        {
+            VkDevice vkdev = g_default_vkdev[i]->vkdevice();
+            if (vkdev)
+            {
+                vkDeviceWaitIdle(vkdev);
+            }
+        }
+    }
+
     // NCNN_LOGE("destroy_gpu_instance");
 
     if (g_instance.glslang_initialized)

src/gpu.h

@@ -37,6 +37,7 @@ NCNN_EXPORT VkInstance get_gpu_instance();
 
 // Destroy VkInstance object and free the memory of the associated object
 // Usually called in the destructor of the main program exit
+// The function will internally ensure that all vulkan devices are idle before proceeding with destruction.
 NCNN_EXPORT void destroy_gpu_instance();
 
 // vulkan core