关于 orinx编译DLA_standalone出现Safety certified DLA should only have one graph for the whole network. 问题

自主机器 Jetson 开发者专区
,
1

关于 orinx编译DLA_standalone出现Safety certified DLA should only have one graph for the whole network. 问题

  • trt版本：8.6.15.17-1+cuda11.4

  • driverOS版本： 6.0.11.0-1

  • 模型的backbone是CSPDarknet，三个特征图输出经过三个yolo类head输出

  • trt编译：

/usr/src/tensorrt/bin/trtexec --onnx=sim.onnx --buildDLAStandalone --useDLACore=0 --verbose --fp16 --inputIOFormats=fp16:chw16 --outputIOFormats=fp16:chw16

  • 问题现象：

[03/29/2023-14:51:16] [V] [TRT] ---------- Layers Running on GPU ----------

[03/29/2023-14:51:16] [V] [TRT] No layer is running on GPU

[03/29/2023-14:51:16] [E] Error[2]: [foreignNode.cpp::determineCandidateForeignNodes::895] Error Code 2: Internal Error (Safety certified DLA should only have one graph for the whole network.)

[03/29/2023-14:51:16] [E] Engine could not be created from network

[03/29/2023-14:51:16] [E] Building engine failed

[03/29/2023-14:51:16] [E] Failed to create engine from model or file.

[03/29/2023-14:51:16] [E] Engine set up failed

问题在于如何找出为什么会在没有GPU层的情况下出现如上报错，可以通过什么手段排查吗