Compiler/TMS320F28388D: Optimization level 0 or 1 remove function that was called

Strong,

Did the compiler inline the function?

Regards

Lori

Hi Lori,

The compiler they used is V18.12.4，these function are not inline function.

Strong,

You're likely thinking of the COFF ABI. With the EABI format, it will remove unused code.

Read more here:

https://software-dl.ti.com/ccs/esd/documents/C2000_c28x_migration_from_coff_to_eabi.html#conditional-linking-section-removal-or-retention

I'm not sure what stack variant you are using, but the C2000 examples with v5.12 stack using CoE in C2000Ware have been tested up to OP level 2 and everything worked as expected.

Best regards

Chris

Hi Loir, Chris,

Just have a update of the issue and the question.

Did the compiler inline the function?

--> Yes, customer double check and find it does be the inline function, also find that the code can run to these inline function.

Issue is that when OP level set to 1, the code will jump to hard fault ISR. when turn off the OP, then everything will be ok. it is not able to find which code executed that lead to jump to hard fault ISR by single step debug, after jump to hard fault ISR customer read the PC address which point to a function, so they try to add a breakpoint in this function, and run again the code after reset, find that the code did not stop at this function but jump to the hard fault ISR directly.

Any method to find out the root cause that jump to hard fault ISR?

The hard fault ISR is what the below link talking about. 

https://www.freertos.org/Debugging-Hard-Faults-On-Cortex-M-Microcontrollers.html

Strong,

Are you using freeRTOS?

Have you checked out the CM sys control SCB and CSFR registers? Those provide some error status information that may be helpful.

I'll sync with the team if there are any other useful methods.

Best regards

Chris

Hi Chris,

No, they did not use any RTOS. control SCB and CSFR registers are as below:

Hello

UFSR indicates it a coprocessor usage fault.

Are you maybe trying to do a operation that requires float? Might have to include a a floating point library in the CCS project properties -> processor options.

Best regards

Chris 

Hi Chris,

Yes, they use float support, it seem there is no option to inclue a floating point library in the CCS project properties -> processor options. customer did the configuration as below:

For C28x core they use the below configuration:

Strong,

They look to already be including FP option on CM which is good.

Are they trying to use any peripherals or memories from CM that require allocate to CM by CPU1?

Are they using the EtherCAT SSC configuration files from C2000Ware? 

Does the issue appear if the EtherCAT stack isn't included? If they can try to strip down their example, this may help narrow down the issue.

Best regards

Chris

Hi Chris,

Please see the input:

Are they trying to use any peripherals or memories from CM that require allocate to CM by CPU1?

----> Only use Ethercat module that require allocate to CM by CPU1.

Are they using the EtherCAT SSC configuration files from C2000Ware? 

----> Yes, they use  EtherCAT SSC configuration files from C2000Ware,

They have use the below file, if Optimization level 0 or off everything is ok.

Strong,

I've seen no issue with optimization running the EtherCAT examples provided. What code has been modified? Are they running CPU1 first?

As I mentioned, please have them look into the following:

• Does the issue appear if the EtherCAT stack isn't included? If they can try to strip down their example, this may help narrow down the issue.

Best regards

Chris

Chris,

They only implement EtherCAT function in CM core, if remove  EtherCAT stack, it may not be able to reproduce the issue.

there are a lot of parameters and struct defined in their code, they try to remove some of them to see if has the same issue.

Another question is that can they use GCC compiler? which they have verify their code with other platform.

Strong

I only recommend to remove stack, temporarily, as a means to try to narrow down the issue.

Yes, let me know once they have more debugging after removing added codes.

For the CM core, GCC should be usable. However, note that ALL of our validation and testing is done with the TI ARM compiler. Our will be better if using the TI ARM compiler.

There is some GCC ARM recommendations on this page: http://www.ti.com/tool/TI-CGT 

Best regards

Chris

Hi Chris,

Customer change the floating point support to none,  then it is ok even the Optimization level is set to 2.

they want to know why set the floating point support to none that can fix the issue? do you have any idea to explain?

Hi Chris,

Customer change the floating point support to none,  then it is ok even the Optimization level is set to 2.

they want to know why set the floating point support to none that can fix the issue? do you have any idea to explain?

Hi Chris,

Customer change the floating point support to none,  then it is ok even the Optimization level is set to 2.

they want to know why set the floating point support to none that can fix the issue? do you have any idea to explain?

Hi Chris,

Yes choose "none" for float support can fix this issue, and set the compiler on is also ok.

but there have another issue that the run the math code very slow since there is no hardware FP support.

is there any DSP lib that can be used to increase the running speed? like increase the efficiency of math function like sqrt.

and the compare logic like if(a>2.0F) will take up 80 cycle, customer have set the compiler level to 2 and --opt_for_speed to 5, but have a little improvement, any other way to improve?

Strong

Besides sending the data to the C28 core, I'm unsure of a FP solution. I suggestion creating a new forum post on that inquiry.

Best regards

Chris