SK-AM64B: Processing Time Measurement Method

Hello Yoshiki,

Yoshiki said:

Is ClockP_getTimeUsec() the elapsed time since the CPU reset?

Yes you are correct. Have you experimented this?

Allow me sometime to comment on your other questions.

Regards,

Vaibhav

Hello Vaibhav,

Just what I suspected.
Thank you.
Regards,
Yoshiki

Hi,

Yoshiki said:

Q4.In case of Linux, cpu occupancy can be checked with the top command.
 In case of MCU+SDK, how can I measure cpu occupancy?

To address this there is a recently created FAQ: https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1392509/faq-how-to-profile-cpu-usage-with-mcu-sdk

Regards,

Vaibhav

Hello Vaibhav,

As always, thank you!!!! I will check it out.

Regards,
Yoshiki

Hello,

Yoshiki said:

Q2. How do you measure the difference time between COREs?
  t1 = RCV0-1 - SND0-0
  t2 = RCV0-0 - RCV0-1

Q3 If return value of ClockP_getTimeUsec() returns the elapsed time since CPU reset

I have discussed this internally and the logic can be as follows.

So lets say there are two cores on which NOTIFY has been used to make sure operations are synced between the two cores.

On this note, you can try to perform the following:

Core 1 application:

Note timestamp here t1 = lets say 1 s

NOTIFY call

Note timestamp here t2 = lets say 10 s

Core 2 application:

Note timestamp here t3 = lets say 4 s

NOTIFY call

So the time taken would be simply, t2 - t3 = 10 - 4 = 6 seconds.

Let me know if this makes sense.

Regards,

Vaibhav

Hello Vaibhav,

I have plotted t2 - t3. Is my understanding correct?

Hi Yoshiki,

The expert is currently out of office for a week. Responses to this thread will be delayed.

Thanks for your patience.

Regards,

Tushar

Hello Tushar,

Thank you for contacting us.

Regards,

Yoshiki

Yoshiki said:

I have plotted t2 - t3. Is my understanding correct?

Yes the understanding is correct. You can proceed with it. Let me know in case of any other doubts.

Thank you, Kumar.

I used the IPC notify evcho application to measure the processing time.
I believe the time returned by ClockP_getTimeUsec() is incorrect.
I believe that the time retrieved by the A53 core and the time retrieved by the R5F core are incorrect.

I would like to post the processing times here.

Regards,

Yoshiki

Hello,

Yoshiki said:

I believe that the time retrieved by the A53 core and the time retrieved by the R5F core are incorrect.

I would like to post the processing times here.

Can you post the timings here?

Do you see inconsistent values?

Regards,

Vaibhav

Hello Kumar.
Sorry I'm late.

PROT 1017-1.

This chart assumes that the pon time is simultaneous.
I don't think they are exactly simultaneous.

1017-Q1.

  Can I find out the blank time?
  How long does it take for the blank time?

  A53==>R5F communication interval
    The communication interval for A53 was approximately 30 usec.
    R5F also had a communication interval of about 30 usec.

    I am not sure about the communication interval of A53 ==> R5F. Maybe it is about 15 usec.

Below is a detailed explanation.
My IPC notify echo project has a main core of A53.

TABLE. ClockP_getTimeUsec

PROT 1017-1."t2 - t3 = 10 - 4 = 6"
  Using this formula, it is 114511(usec).
  A53=>R5F communication interval of 114511 (usec) is incorrect.

  A53=>R5F Communication Interval:
    echo3(t3) ==> echo2(t2)      114511(usec)

  A53=>A53 Communication Interval:.
    echo2(t2) ==> echo4(t4)    29(usec)
    29/2 = 14.5(usec)          A53=>R5F communication interval

TABLE. ClockP_getTimeUsec "sync"
  The completion time of SyncAll() for R5F is 129062 (usec).
  The completion time of SyncAll() for A53 is 18816 (usec).

  If SyncAll() completes at the same time for both A53 and R5F, then the time returned by ClockP_getTimeUsec() is not correct.

1017-Q2.
  If the time returned by ClockP_getTimeUsec() is correct, then IpcNotify_SyncAll() in R5F should complete before A53.
  A53=>R5F notification latency is 4273usec.
  TABLE. ClockP_getTimeUsec "sync ==> t1"

 R5F SyncAll() => A53 SyncAll() => A53 sendMsg() => R5F echo1
    The first IpcNotify_sendMsg() “t0” in A53 is executed immediately after SyncAll().
     A53 IpcNotify_sendMsg() is delayed because SyncAll() completes later than R5F.
     In R5F, after SyncAll, wait for reception.
     The receive latency is 4273 usec.

  Probably the first R5F application was launched after the SBL was completed.
  Then, after about 4000usec, I believe the A53 application was launched.
  Is this conjecture correct?

1017-Q3.
  What is the order in which applications are launched?
  OSPI Boot Loader:                 R5F 0-0 => R5F 0-1 => R5F 1-0 => R5F 1-1 => M4F 0-0 => A53 0-0 => A53 0-1 ?
  OSPI multi-partition boot loader: R5F 0-0 => R5F 0-1 => R5F 1-0 => R5F 1-1 => M4F 0-0 => A53 0-0 => A53 0-1 ?

Regards,
Yoshiki

Hello Yoshiki,

Thanks for your patience.

This thread has been assigned to the correct expert.

Please expect responses in few business days.

Regards,

Vaibhav