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.

Through put performance on msgcom on TI 6614

Hi,

I am working on syslib for ARM to DSP communication. I am using CCS Version: 6.0.1.00040, syslib_2_00_08_00 for DSP and DEV.SC-MCSDK-02.02.01.03A for ARM.

Do you have any benchmarking for msgcom transfer between ARM and DSP on TI 6614?

I am profiling the delay for the basic queue ring(test_queue_ring_basic_reader for Msgcom_ChannelMode_NON_BLOCKING) ARM writer DSP reader example for   1- iteration transferring 64bytes from ARM to DSP. The way I checked the delay is by checking the difference in CPU cycle after the Agent_push (agentHandle, channelName, ResMgr_NamedResourceType_INTERNAL_SYSLIB, Agent_NamedResource_CREATE, &errorCode) < 0) and after Msgcom_getMessage in Msgcom_ChannelMode_NON_BLOCKING. I find the latency to be of the order 10ms. Is such high latency expexted?

Thanks and regards,

Shabrin

  • Hi Ranjit,
    From my understanding, we do not have benchmarking for msgcom transfer. I'll ask some others to comment.
  • Hi Rajasekaran,

    Thanks for the quick reply.

    I am using Core0_msgCom_tmdxevm6614lxe_UnittestProject_little and Core1_msgCom_tmdxevm6614lxe_UnittestProject_little project for DSP to ARM communication.

    I got expected results for DSP write-ARM read(ie., in some micro sec) and I used functions "test_queue_ring_basic_writer()" & "test_queue_ring_basic_reader()" in DSP and ARM respectively. But for ARM write-DSP read I am getting in Milli sec which is not expected and I used functions "test_queue_ring_basic_writer()" & "test_queue_ring_basic_reader()" in ARM and DSP respectively.

    There was piece of code commented in the project for bi-directional communication. I executed enabling the function "test_queue_ring_bidirnl()", it is giving profiling results as sepected.

    Can you please tell me why uni-direction ARM write-DSP read is taking more cycles?

    Thanks and Regards,
    Shabrin
  • Hi Rajasekaran,

    Thanks for the quick reply.

    I am using Core0_msgCom_tmdxevm6614lxe_UnittestProject_little and Core1_msgCom_tmdxevm6614lxe_UnittestProject_little project for DSP to ARM communication.

    I got expected results for DSP write-ARM read(ie., in some micro sec) and I used functions "test_queue_ring_basic_writer()" & "test_queue_ring_basic_reader()" in DSP and ARM respectively. But for ARM write-DSP read I am getting in Milli sec which is not expected and I used functions "test_queue_ring_basic_writer()" & "test_queue_ring_basic_reader()" in ARM and DSP respectively.

    There was piece of code commented in the project for bi-directional communication. I executed enabling the function "test_queue_ring_bidirnl()", it is giving profiling results as expected.

    Can you please tell me why uni-direction ARM write-DSP read is taking more cycles?

    Thanks and Regards,
    Shabrin
  • The ARM-DSP communication uses UDMA which is know to be slow, the times you are seeing is in the expected range for ARM-DSP communication.