DLA cores do not support DIV ElementWise operation

[insup.choi]

Hello, There

I am trying to convert onnx to trt file with DLA option.
$ trtexec --onnx=onnx_sim.onnx --useDLACore=0
but I am faced to the problems like the below and Div_80 node is fallbacked to GPU layer
[TRT] Div_80: DLA cores do not support DIV ElementWise operation.
Could you tell me what ElementWise operation means and how it can be converted to DLA layer?
In addition, both of the input tensors of Div_80 node are “type: float32[1,3,750,1333]”.

Thanks.

[AastaLLL]

Hi,

Could you try it with fp16 or int8 mode?

Developer Guide :: NVIDIA Deep Learning TensorRT Documentation

Thanks.

[insup.choi]

Helllo,

I have already tried it with --fp16, --int8 but it still has same problems.

Isn’t DIV operator supported in DLA?

Thanks.

[AastaLLL]

Hi,

Sorry for the missing.

We don’t support the Div operator in the DLA ElementWise layer.
Please fallback it to the GPU mode.

Thanks.

[insup.choi]

Hello,

How about POW operator?
POW may not be well composed with Sum, Sub, Product, Max and Min.

Thanks.

[AastaLLL]

Hi,

Unfortunately, no.
Currently, DLA only supports sum, sub, product, max, and min operators.

Thanks.