Swish算子在TB RK1808S计算棒上推理出错

Ubuntu版本: 18.04
python版本: 3.6.9
rknn-toolkit版本: 1.6
TB RK1808S 计算棒固件版本: 1.5.2

帖子下边是只有一个Swish的rknn模型在netron中的网络图。

这个rknn模型可以在PC的模拟器正常运行，但是在init_runtime(target='rk1808')指定device='rk1808'后，运行就会出错：

1. (m1808_rknn_yolo_demo) fjc@utdnn:/data/home/fjc/workspace/rknn_demo$ python rknn_test.py multiply_sigmoid.rknn
   
2. W Verbose file path is invalid, debug info will not dump to file.
   
3. 2021-03-31 16:43:25.875993: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
   
4. Loading model +++
   
5. Loading model ---
   
6. Initializing runtime +++
   
7. D target set by user is: rk1808
   
8. D Starting ntp or adb, target is RK1808, host is None
   
9. D Start ntp...
   
10. I npu_transfer_proxy pid: 9790, status: sleeping
    
11. I NPUTransfer: Starting NPU Transfer Client, Transfer version 2.1.0 (b5861e7@2020-11-23T11:50:36)
    
12. D NPUTransfer: Transfer spec = local:transfer_proxy
    
13. D NPUTransfer: Transfer interface successfully opened, fd = 3
    
14. E RKNNAPI: rknn_init, msg_load_ack fail, ack = 1(ACK_FAIL), expect 0(ACK_SUCC)!
    
15. E RKNNAPI: ==============================================
    
16. E RKNNAPI: RKNN VERSION:
    
17. E RKNNAPI: API: 1.6.0 (79320de build: 2020-12-29 10:56:36)
    
18. E RKNNAPI: DRV: 1.4.0 (b4a8096 build: 2020-09-14 11:15:57)
    
19. E RKNNAPI: ==============================================
    
20. D NPUTransfer: Transfer client closed, fd = 3
    
21. E Catch exception when init runtime!
    
22. E Traceback (most recent call last):
    
23. E File "rknn/api/rknn_base.py", line 844, in rknn.api.rknn_base.RKNNBase.init_runtime
    
24. E File "rknn/api/rknn_runtime.py", line 340, in rknn.api.rknn_runtime.RKNNRuntime.build_graph
    
25. E Exception: RKNN init failed. error code: RKNN_ERR_MODEL_INVALID
    
26. E Current device id is: None
    
27. E Devices connected:
    
28. E ['TS018083201102007']
    
29. Init runtime environment failed

复制代码
测试例在压缩包test.rar，解压后运行

1. python rknn_test.py multiply_sigmoid.rknn

复制代码测试。

升级到新固件后可以成功推理单个Swish算子 ，但是被动模式NTP方式的YOLOv5推理失败了。

1. (m1808_rknn_yolo_demo) fjc@utdnn:/data/home/fjc/workspace/yolov5 (v4.0-rknn)$ python detect.py --weights weights/yolov5s.rknn --img 640 --target rk1808
   
2. Namespace(agnostic_nms=False, augment=False, cfg='./models/yolov5s.yaml', classes=None, conf_thres=0.25, device='', exist_ok=False, img_size=640, iou_thres=0.45, name='exp', project='runs/detect', save_conf=False, save_txt=False, source='data/images', target='rk1808', update=False, view_img=False, weights=['weights/yolov5s.rknn'])
   
3. Using torch 1.8.0 CUDA:0 (GeForce RTX 2080 Ti, 11019.4375MB)
   
4.                   CUDA:1 (GeForce RTX 2080 Ti, 11019.4375MB)
   
5.                   CUDA:2 (GeForce RTX 2080 Ti, 11019.4375MB)
   
6.                   CUDA:3 (GeForce RTX 2080 Ti, 11016.3125MB)
   
7. 
   
8. 
   
9. W Verbose file path is invalid, debug info will not dump to file.
   
10. 2021-04-02 17:57:04.835380: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
    
11. Generating grammar tables from /usr/lib/python3.6/lib2to3/Grammar.txt
    
12. Generating grammar tables from /usr/lib/python3.6/lib2to3/PatternGrammar.txt
    
13. Loading model +++
    
14. Loading model ---
    
15. Initializing runtime +++
    
16. D target set by user is: rk1808
    
17. D Starting ntp or adb, target is RK1808, host is None
    
18. D Start ntp...
    
19. I npu_transfer_proxy pid: 9296, status: sleeping
    
20. I NPUTransfer: Starting NPU Transfer Client, Transfer version 2.1.0 (b5861e7@2020-11-23T11:50:36)
    
21. D NPUTransfer: Transfer spec = local:transfer_proxy
    
22. D NPUTransfer: Transfer interface successfully opened, fd = 19
    
23. D NPUTransfer: The connection has been closed, socket read fd = 19
    
24. D NPUTransfer: Transfer client closed, fd = 19
    
25. E RKNNAPI: rknn_init,  recv(MsgLoadAck) fail, -9(ERROR_PIPE) != 368!
    
26. E Catch exception when init runtime!
    
27. E Traceback (most recent call last):
    
28. E   File "rknn/api/rknn_base.py", line 844, in rknn.api.rknn_base.RKNNBase.init_runtime
    
29. E   File "rknn/api/rknn_runtime.py", line 340, in rknn.api.rknn_runtime.RKNNRuntime.build_graph
    
30. E Exception: RKNN init failed. error code: RKNN_ERR_DEVICE_UNAVAILABLE
    
31. E Current device id is: None
    
32. E Devices connected:
    
33. E ['TS018083201102007']
    
34. Init runtime environment failed

复制代码

看出错信息（ERROR_PIPE）应该是连接的原因。