Memory Management NVDLA and GPU

Hi,

i’m using an Jetson AGX Xavier and want to use the NVDLA and GPU for inference.
I’m unsure of the SOC memory management.

  1. Does the NVDLA and GPU have a unified shared memory?
  2. Can data be passed from GPU to NVDLA without any copies and vice versa?

Thanks.

Hi,

DLA has it’s own SRAM.

But the data transfer between GPU <-> DLA is handled by an internal driver.
You will only need to prepare a GPU buffer for TensorRT as input.

Thanks.

Thanks for your reply.
Could you please provide me any sample code?

Hi,

You can check following for the DLA sample:

Thanks.

1 Like