AM3357: CPU load of EtherCAT Slave stack

Kzk

Part Number: AM3357
Other Parts Discussed in Thread: SYSBIOS

Hello,

I'd like to know CPU load (or performance) of EtherCAT Slave Stack Code on AM3357. How much CPU load do you know even though I expect it is low? For example, in the case of the example software that receives LED pattern of several bytes in ethercat_slave of ISDK. Also, I found CPU performance of Master. Please give me some information.

EtherCAT Master Reference Design for AM335x CPSW
Table 1. AM 335x EC-Master CPU performance
www.ti.com/.../tiduaf8a.pdf

Regards,
Kazu

over 8 years ago

0 Biser Gatchev-XID over 8 years ago

TI__Guru**** 393215 points

Hi,

The EtherCAT team have been notified. They will respond here.

0 Kzk over 8 years ago in reply to Biser Gatchev-XID

Genius 5785 points

Hello,

Biser Gatchev-XID said:
The EtherCAT team have been notified. They will respond here.

I'm waiting for the reply from the EtherCAT team.

Regards,
Kazu

0 Biser Gatchev-XID over 8 years ago in reply to Kzk

TI__Guru**** 393215 points

Responses may be delayed at this time, due to the holidays.

0 Kzk over 8 years ago in reply to Biser Gatchev-XID

Genius 5785 points

Hello Biser,

I declared the PROFILE_ECAT_STACK definition in tiescutils.c, ran the ISDK example software and got the number of cycles of the HW_EcatIsr function.
C:\ti\sysbios_ind_sdk_02.01.03.02\sdk\examples\ethercat_slave\am335x_release\ethercat.out

The pdi_max variable was about 325,000. But, since Demo_Application function in tiescappl.c had time with I2C(DigOut) and SPI(DigInput), I commented out the processing of this function. At this time the pdi_max variable was about 8,500.

Is the result of cycle numbers above proper? It looks like the processing of EtherCAT slave stack needs 42.5 usec since 1 cycle is 200 MHz/PRU, in the case of data exchange of 4 bytes + 4 bytes like this example software. Is it correct?

Regards,
Kazu

0 Garrett Ding over 8 years ago in reply to Kzk

TI__Mastermind 43296 points

Hi Kazu,

You can verify the cycle numbers by bsp_get_local_sys_time() which uses IEP timer and provides nano seconds resolution. We don't have an updated cycle number for sysbios_ind_sdk_02.01.03.02 EtherCAT yet.

Regards,
Garrett

0 Kzk over 8 years ago in reply to Garrett Ding

Genius 5785 points

Hi Garrett,

Thank you for your reply. I checked bsp_get_local_sys_time in the wiki (PRU ICSS EtherCAT firmware API guide). Since pdi_max was 8,500, the time is 8,500 nsec.

Garrett Ding said:
We don't have an updated cycle number for sysbios_ind_sdk_02.01.03.02 EtherCAT yet.

Do you have a cycle number for previous ISDK? I'd like to confirm if this time is an appropriate value.

Regards,
Kazu

0 Garrett Ding over 8 years ago in reply to Kzk

TI__Mastermind 43296 points

Kazu,

Our measurement is out-of-date and based on Industrial SDK 01.00.00.02, can not be as a reference for your benchmark on the sysbios_ind_sdk_02.01.03.02 . There are major changes from ISDK 1.x to 2.x.

Regards,

Garrett

0 Kzk over 8 years ago in reply to Garrett Ding

Genius 5785 points

Hi Garrett,

Garrett Ding said:
Our measurement is out-of-date and based on Industrial SDK 01.00.00.02, can not be as a reference for your benchmark on the sysbios_ind_sdk_02.01.03.02 .

I don't mind even if that's the old information. I want to refer it so please give me the information.

Regards,
Kazu

0 Kzk over 8 years ago in reply to Kzk

Genius 5785 points

Hi Garrett,

Could you give me your benchmark of EtherCAT slave stack?

Regards,
Kazu

0 Garrett Ding over 8 years ago in reply to Kzk

TI__Mastermind 43296 points

Kazu,

Below is our Industrial SDK 01.00.00.02 EtherCAT slave benchmark:

ARM Frequency: A8@550MHz

% utilization: 15-20%

Tool chain: TI (*not* GCC)

Remarks: 5-byte process data (PD) output / 7-byte PD input application at 250 us cycle time; with cyclic EoE (ping) and CoE data update

Regards,

Garrett

0 Kzk over 8 years ago in reply to Garrett Ding

Genius 5785 points

Hello Garrett,

Thank you for your information. I really appreciate your help.

Regards,
Kazu

Processors

Processors forum

AM3357: CPU load of EtherCAT Slave stack