• State TI Thinks Resolved
  • Locked Locked
  • Replies 5 replies
  • Answers 1 answer
  • Subscribers 99 subscribers
  • Views 1387 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TDA4VM: TIDL Facing the Segmentation Fault

Tsurumoto
Genius 16375 points

Part Number: TDA4VM

HI

I'm currently facing the Segmentation Fault.

【Verification environemnt】

　I'm referring at  TFLite Runtime + TIDL Heterogeneous Execution.

　- The network we use are based on CenternNet=HarDNet（https://github.com/PingoLH/CenterNet-HarDNet）, and modified by ourselves.

   - Model trains on Pytorch

   - Convert Pytorch to Onnx files using torch.onnx built into pytorch

   - Convert onnyx into Tensorflow, and Quantize with Tensorflow and convert to tflite.

   - Modify tflrt_delegate.py by referring to TFLite Runtime + TIDL Heterogeneous Execution, and run the model

 

【Source code, tflite file】

　    will send these files to the BOX

 

【Error log】

＜Log when the model is run on PC＞

xxxx:~/Ti/ti-processor-sdk-rtos-j721e-evm-07_02_00_06/tidl_j7_01_04_00_08/ti_dl/test/tflrt$ python3 tflrt_compile.py

  • Running model         CenterNet-HarDNet-exp022-930-nofeature-simple-int8-5image
  • 0.0s:         VX_ZONE_INIT:Enabled
  • 0.8s:         VX_ZONE_ERROR:Enabled
  • 0.9s:         VX_ZONE_WARNING:Enabled

 

  • Number of subgraphs:32 , 153 nodes delegated out of 495 nodes

  • Unsupported Tensor->type 9
  • ********** Frame Index 1 Running float import and float inference **********
  • WARNING: [TIDL_E_DATAFLOW_INFO_NULL] ti_cnnperfsim.out fails to allocate memory in MSMC. Please look into perfsim log. This model can only be used on PC emulation, it will get fault on target.
  • ****************************************************
  • **         1 WARNINGS         0 ERRORS         **
  • ****************************************************
  • Unsupported Tensor->type 9
  • Unsupported Tensor->type 9 /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_src_j7_01_04_00_08/ti_dl/tfl_delegate/src/pc/../tfl_delegate.cpp 627
  • Unsupported Tensor->type 9 /opt/Jenkins_OUT/workspace/J7_TIDL_Parent/c7x-mma-tidl/ti_dl/release/tidl_src_j7_01_04_00_08/ti_dl/tfl_delegate/src/pc/../tfl_delegate.cpp 627
  • Segmentation fault (core dumped)

 

＜Log when the model is run on EVM＞ 

root@j7-evm:/opt/test/tflrt# python3 tflrt_compile.py

  • Running model         CenterNet-HarDNet-exp022-930-nofeature-simple-int8-5image
  • start interpreter creation
  • APP: Init ... !!!
  • MEM: Init ... !!!
  • MEM: Initialized DMA HEAP (fd=4) !!!
  • MEM: Init ... Done !!!
  • IPC: Init ... !!!
  • IPC: Init ... Done !!!
  • REMOTE_SERVICE: Init ... !!!
  • REMOTE_SERVICE: Init ... Done !!!
  • APP: Init ... Done !!!
  • 1768.994824 s: VX_ZONE_INIT:Enabled
  • 1768.994836 s: VX_ZONE_ERROR:Enabled
  • 1768.994840 s: VX_ZONE_WARNING:Enabled
  • 1768.995538 s: VX_ZONE_INIT:[tivxInit:71] Initialization Done !!!
  • 1768.995799 s: VX_ZONE_INIT:[tivxHostInit:48] Initialization Done for HOST !!!
  • Segmentation fault (core dumped)

【Questions】

 #1: Why the execution was stopped at  Segmentation Fault  on PC environment?

 #2: Why the execution was stopped at  Segmentation Fault  on EVM environment?

#3: About the log "Unsupported Tensor -> type 9", type 9 (=kTfin8??) is not supported?

#4: What's the reason of the following message,

    WARNING: [TIDL_E_DATAFLOW_INFO_NULL] ti_cnnperfsim.out fails to allocate memory in MSMC. Please look into perfsim log. This model can only be used on PC emulation, it will get fault on target.

 

 Thanks and Best regards,

HaTa