This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TDA2EXEVM: Getting Max FAIL for some of the Convolution layers

Part Number: TDA2EXEVM

Hi,

I am able to build the source code for host emulation build now. With the default .BIN files for network and parameters, when I run "eve_test_dl_algo.out.exe" all the layers are passing(Max PASS).

But with our changed model, some of the layers are failing. I am getting FAIL for one of the layers(Max FAIL) when I execute "eve_test_dl_algo.out.exe".

Any reason why the TIDL Max and the reference output are not matching?

Regards,
Navinprashath.R.R

  • Hi,

    You can ignore Max FAIL print, they are for debugging.

    Thanks,
    Praveen
  • Hi Praveen,

    I am also facing this Max FAIL for few Conv Layers in my network model.
    As you said , We can ignore this. Just for my information, Can you give some more information about reference comparison.
    What is reference is TIDL ? What exactly it compare? Because , We are not providing any reference input for comparison.

    Regards,
    Sagar
  • Hi Sagar,

    We have a reference C implementation of all layers in TIDL to verify the EVE/DSP platform code flow changes for all TIDL layers.
    In the platform code flow, we may use DMA and other optimisations and to validate the changes done platform code flow, we use this reference code. So, platform code output is compared with output of reference code for each layer when "ENABLE_REF_COMPARISION" flag is defined.
    As the code flow is different in reference code and the platform code, some times MAX values cannot match between them which prints this Max FAIL and this can be ignored because the final outputs will match between them.

    Thanks,
    Praveen