• State
  • Locked Locked
  • Replies 2 replies
  • Subscribers 100 subscribers
  • Views 642 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.

Slow QDMA in one core

Thomas Rakesh
Intellectual 390 points

Hi,

We are using TI6608 DSP processor.

It is observed that QDMA in core id 2 is slower than core id 3 .some bytes are transfered from DDR to L2 RAM(we also tried L2 to L2).

Core id 3 has 2 edma transfers of X bytes ,where has core id 2 has 1 edma transfer of X bytes.still core 2 takes more time.

This is observed sometimes like once in 10 ms or 20 ms.

We have tried using different channel controller on both cores(EDMA3CC0 and EDMA3CC2),still it did not help.We use Qdma Link Channel Id =0 in all Controllers.

Can you please help us in this regard.

Thanks,

Thomas

  • 0
    TI__Guru**** 161600 points
    Hi Thomas,

    I've notified the EDMA experts. The feedback should be posted directly here.

    Best Regards,
    Yordan
  • 0 in reply to Yordan Kovachev
    TI__Genius 12805 points
    Thomas you are right.

    The speed of QDMA as well as DMA channels does not depend on which core initiates the transfer.

    So let's do some investigation:

    1. Make sure that you have not changed the SES and the SMS registers in the system. The SES and the SMS registers are like the MPAX registers. They translate logical memory to physical memory. First step to verify that these registers are in the default state. The MSMC User Guide gives information about the default state. (Some information may be in the device User guide though)

    2. Modify your code to run only on one core, without dependency on the core number, and then move the same buffer to the same buffer using the same transfer controller and the same channel number and PaRam. Run the code on core 2 and on Core 3 (or any other cores) and verify that the move takes the same time. Please report to the posting what are your observations,

    3. Go back to your original code and exchange core 2 with core 3 in terms of CC and TC. Record the results and post them here.

    Please post the results of your experiments and we will continue from there

    Ran