• State
  • Locked Locked
  • Replies 1 reply
  • Subscribers 99 subscribers
  • Views 105 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

AM62D-Q1: Does AM62D run TIDL Artifacts compiled for AM62A

Jai Guru
Prodigy 20 points

Part Number: AM62D-Q1

I want to run machine learning models on AM62D where there is no offical support for it...

But AM62A and AM62D are mostly similar... so what changes to be made to the linux sdk to run the model compiled for AM62A on AM62D..


when i run without any changes I am getting the following


****** In tidlDelegate::Init ****** ************ in TIDL_subgraphRtCreate ************ APP: Init ... !!! 2566.917397 s: APP_LOG: ERROR: Unable to map memory @ 0xa1000000 of size 262144 bytes !!! APP: ERROR: Log writer init failed !!! APP_LOG: ERROR: Unable to map memory @ 0xa2000000 of size 3288576 bytes !!! APP: ERROR: File IO writer init failed !!! APP: Init ... Done !!!

  • 0
    TI__Mastermind 18406 points

    Hello Jai,

    AM62D SDK software differs substantially from AM62A. You are correct that using TIDL software for neural network acceleration is not officially supported. 

    There are likely missing software components related to TI OpenVX (/usr/lib/libtivision_apps.so on AM62A), TIDL (multiple libraries in /usr/lib, like libvx_tidl_rt.so), firmware support for those softwares (/usr/lib/firmware/, and a memory map that identifies the core regions needed for TIOVX and the firmwares on the C7 NPU and on the ISP.

    The error you're seeing is a failure to allocate memory from a physical region in DDR that TI OpenVX was build to recognize. This is nominally controlled by a memory map that is agreed upon by all cores participating in TIOVX applications (A53, C7x, DM R5)

    We do provide support for TI Deep Learning on AM62D at this time. 

    Best Regards, 
    Reese