ARMNN : ACL : NEon backend : Intermediate tensor memory dump

[U1m2]

Hi ,
Is there a way to dump the intermediate tensors allocated by ACL from ARMNN during model inference. Will any compilation options helps here?
ARMNN_DUMP_INTERMEDIATE_TENSORS=1

or
ARM_COMPUTE_DEBUG_MEMORY=1

Thanks,
Uma

[morgolock]

Hi @U1m2

Your best option is using ExecuteNetwork with these options

-p, --print-intermediate-layers
                                If this option is enabled, the output of 
                                every graph layer will be printed.
  -F, --print-intermediate-layers-to-file
                                If this option is enabled, the output of 
                                every graph layer will be printed within 
                                separate files.

Hope this helps