Tensorrt can not set explicit shape when build engine to use DLA

Autonomous Machines Jetson & Embedded Systems Jetson Xavier NX
,
1

trtexec --onnx=model.onnx --shapes=‘fts_input_images:0’:3x384x576x3 --fp16 --useDLACore=0 --allowGPUFallback

Input build shape: fts_input_images:0=3x384x576x3+3x384x576x3+3x384x576x3

[TRT] DLA Layer model/7_3/BiasAdd does not support dynamic shapes in any dimension

→ all layer run on GPU, no have layer run on DLA

Shape in model.onnx: fts_input_images:0 = -1x-1x-1x3
Have way to set explicit shape when build engine to use DLA ?

2

Hi,

Please refer to the following document, which may help you.

Thank you.

3
  • DLA does not support dynamic dimensions. Thus, for wildcard dimensions, the min, max, and opt values of the profile must be equal.

In my case the min, max, and opt values of the profile equal, why all layer can not use DLA with error “does not support dynamic shapes in any dimension.”
my model file

I test on Xavier NX, Jetpack 5.0.2, TensorRT 8.4 and 8.5

6

We are moving this post to the Jetson Xavier NX forum to get better help.

Thank you.

9

Dear @anhtu812,
Is it possible to test the issue on latest release?

10

i tested on nvcr.io/nvidia/l4t-tensorrt:r8.4.1.5-devel

11

Dear @anhtu812,
Is it possible to test with Jetpack 5.1.1(TRT 8.5.2)?

12

my machine is product, i can not up to Jetpack 5.1.1. Did you tested on Jetpack 5.1.1(TRT 8.5.2)? it word ?

17

Dear @anhtu812 ,
Dynamic shapes are not supported on DLA.