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.

TDA4VL-Q1: tda4

Part Number: TDA4VL-Q1
Other Parts Discussed in Thread: TDA4VL

hi

i have successfully reasoned the unet.onnx in Ubuntu platform ,but when i migrated to tda4vl, it occured some mistakes,as the below fig shows.

especially kernel and node fail to createdCry.

could you please give me some suggestion to solve the problem!

  • Hi,

    It seems that the memory is not sufficient for your model. Have you changed the memory map in vision_apps from the default SDK?

    Regards,

    Nikhil

  • as the figs shows:

    iwe fixed  the ddr spaces to 2 Gbs,and  made ddr_shared_mem_addr  and ddr_shared_mem_addr_phys have same address in vision_apps。

    how can i solve the problem? please give me some detailsGrinning 

  • Hi,

    The issue here is the size of the either c7x_1 local heap or scratch. Can you increase them by few more MBs and you can decrease the same amount from C7x_2 local heap and scratch region (as they are not used for TIDL in J721s2)

    Could you try this?

    Regards,

    Nikhil

  • Okay, I'll add some memory first and give it a try. Also, can you explain the difference in the use of local heap and scratch? Because I see that tidlObj was created at the TIVX-MEM-EXTERNAL position, does it mean that the scratch space is not used? I have always had this question here

  • Hi,

    TIVX-MEM-EXTERNAL refers to the Local heap region, But the error is coming from the TIDL layer, where it is trying to allocate mem-records in the Local heap and scratch region.  

    i.e. From the function, tivxAlgiVisionAllocMem() in the file tivx_alg_vision.c, you could see the error log "Failed to Allocate memory record...".

    Local heap region is used to put the weights and model related contents, whereas scratch region is used (as the name suggests) as a scratchpad region for intermediate calculations. 

    Here, it would be most probably the local heap region that needs to be increased and this region is a persistant memory and there are chances that more of this region is needed by the model, depending on how big the model is.

    Regards,

    Nikhil