[Enhance] support TensorRT engine for onnxruntime (#1739)

* Support trt engine for onnxruntime

* Apply lint

* Check trt execution provider

* Fix typo

* Fix provider order

* Check device

mmdeploy/backend/onnxruntime/wrapper.py

@@ -50,9 +50,16 @@ class ORTWrapper(BaseWrapper):
             logger.warning('The library of onnxruntime custom ops does '
                            f'not exist: {ort_custom_op_path}')
         device_id = parse_device_id(device)
-        providers = ['CPUExecutionProvider'] \
-            if device == 'cpu' else \
-            [('CUDAExecutionProvider', {'device_id': device_id})]
+        providers = []
+        if 'cuda' in device:
+            if 'TensorrtExecutionProvider' in ort.get_available_providers():
+                providers.append(('TensorrtExecutionProvider', {
+                    'device_id': device_id
+                }))
+            providers.append(('CUDAExecutionProvider', {
+                'device_id': device_id
+            }))
+        providers.append('CPUExecutionProvider')
         sess = ort.InferenceSession(
             onnx_file, session_options, providers=providers)
         if output_names is None: