• State Resolved
  • Locked Locked
  • Replies 3 replies
  • Subscribers 98 subscribers
  • Views 314 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.

DSP LINK performance

MH12441
Prodigy 80 points

Hi ,

I have an application that use dsp link message queue between the arm and dsp.

After having real-time issues, I observed that the call to the message queue put/get can take 0.15/0.2 msec.

Does it make sense?

Thanks,

  • 0
    TI__Expert 4930 points

    MH,

    On an average, the round trip time for MSGQ_put/MSGQ_get is 150 to 250 Usec.

    Which platform are you on? I will be able to give exact details then.

    Deepali

  • 0 in reply to Deepali Uppal
    Prodigy 80 points

    Thanks for the answer.

    I am using Davinci platform.

    Isn't it a long time for messages communication? I can implement a much simple and fater queue in

    the shared memory..

  • 0 in reply to MH12441
    TI__Expert 4930 points

    MH,

    DSPLink is primarily a generic porting kit. The code is written to allow easy multi-OS and multi-platform configurability. So the DSPLink has layers like HAL for hardware abstraction, OSAL i.e. OS adaptation layer, PMGR and LDRV for the user/kernel abstraction etc. But the layers are not the only reason for the numbers that you see.

    The time for messaging IPC on Linux for e.g. also include context switch time for OS, interrupts processing, cache coherency (for DSP) etc. The round trip time for messaging on Jacinto-Pros is much less, less than 100 usec for a round trip message. We also had a customer who profiled the IPC and reported the bulk of the delays in the Linux OS. Some TI engineers have also reported much lower numbers (~70 usec for round trip message time) using the PREMPT_RT patch for Linux on DM6446.

    I can share some options/suggestions to reduce the overhead. For e.g.:

    1) DSPLink gives two modes on DSP. Details here: http://processors.wiki.ti.com/index.php/DSPLink_FAQs#How_should_I_choose_between_TSK_mode_or_SWI_mode_in_the_--DspTskMode_option_when_I_am_trying_to_configure_DSPLink.3F

    Or

    2) If you are on DM6446, you can try PREMPT_RT

    Or

    3) If NOTIFY IPC with its lower cycle time (~50 usec) is useful for your use case

    Deepali