F29H850TU: C29 code efficiency improvement against C28 is not as expected

Hang,

What type of code are they benchmarking? A real algorithm, or something close to a real algorithm, is best. I ask because I've seen customers try to benchmark one or two operations repeated multiple times. That doesn't take advantage of all the parallel processing units. 

There is an optimization guide for F29 here: https://www.ti.com/lit/pdf/sprujg0

Please check the suggestions against their code and let us know if you have questions.  The post link optimization is still in work. 

Regards

Lori

Hang,

Also, who is the customer?

Hi Lori,

It's tested on actual application code, including logic and math operations.

Hi Sira,

It's Eaton

Hang,

With qualitative details like this, it's impossible to figure out what is going on.

Have they looked at sprujg0 and the optimizations listed there?

A few key examples:

- In addition to O3, is -ffast-math enabled?

- Do they have atan, atan2 and other RTS functions like fmod, roundf, etc. in their code? In the upcoming 2.0.0.STS, these are efficiently implemented in the compiler (provided the -ffast-math setting is used).

Can they share the code with us to analyze?

Thanks,

Sira

Hi Hang, Any update on this benchmarking?

Hi Lori,

Still contacting with customer to acquire the test condition.

Regards,

Hang

Hi Lori,

Code is sent via other channel

Hang,

Apologies for the delayed reply. I've responded via email.

Thanks,

Sira