[Custom Device] Add singleton to custom device (#46963)

* add singleton to custom device

* Update custom_device.cc

Init device_init_flag_ in default

paddle/phi/backends/custom/custom_device.cc

@@ -53,11 +53,14 @@ class CustomDevice : public DeviceInterface {
   ~CustomDevice() override { Finalize(); }
 
   size_t GetDeviceCount() override {
-    size_t count;
-    if (pimpl_->get_device_count(&count) != C_SUCCESS) {
-      count = 0;
+    if (!device_init_flag_) {
+      if (pimpl_->get_device_count(&device_count_) != C_SUCCESS) {
+        device_count_ = 0;
+      } else {
+        device_init_flag_ = true;
+      }
     }
-    return count;
+    return device_count_;
   }
 
   std::vector<size_t> GetDeviceList() override {
@@ -882,6 +885,8 @@ class CustomDevice : public DeviceInterface {
   std::unique_ptr<C_DeviceInterface> pimpl_;
   void* dso_handle_;
   std::unordered_map<size_t, C_Device_st> devices_pool;
+  bool device_init_flag_ = false;
+  size_t device_count_;
 };
 
 bool ValidCustomCustomRuntimeParams(const CustomRuntimeParams* params) {

python/paddle/fluid/tests/custom_runtime/test_custom_cpu_plugin.py

@@ -140,6 +140,7 @@ class TestCustomCPUPlugin(unittest.TestCase):
 
         import paddle
         paddle.set_device('custom_cpu')
+        paddle.device.get_available_device()
         x_tensor = paddle.to_tensor(x, stop_gradient=False)
         y_tensor = paddle.to_tensor(y)
         z1_tensor = paddle.matmul(x_tensor, y_tensor)