2D convolution on DLA and GPU at the same time

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

Hello,

I’m trying to deploy a model on Xavier AGX using DLA and GPU. All 2D convolutions are implemented on DLA, but two of them are implemented on GPU too. I don’t understand why is this happening. I have read this post: [[Xavier NX + DLA] does not support dynamic shapes, and CBUF size requirement], but I’m not sure that the problem is the same (my Jetpack version is 4.5.1).

I attach the warning with the deployment of layers on GPU and DLA.

image
image1920×312 26.6 KB

I want to use only DLA for this implementation, but I don’t know how to do that.

Thanks, Paula.

3

Hi,

The message indicates that these layers are not supported by the DLA.
Please note that there is some constraint in the DLA-supported convolution layer.

You can find the details below:

Thanks.

4

Hi,

I know that the message indicates that these layers are not supported by the DLA, but I mean that the same convolution layer is “divided” between DLA and GPU. Note that conv2d and conv2d_5 layers are on DLA and GPU at the same time.

Inkederror_LI
Inkederror_LI1920×312 101 KB

Thanks, Paula.

5

Hi,

The complete layer name is StatefulPartitionedCall/.../Conv2D.
So they are two different layers.

Thanks.