Welcome to Arm Community

In contrast to the MDK Community Edition, the "MDK for STMicroelectronic Cortex/M0/M0+" edition can be used for commercial use. As you already found out, the license is paid by STMicroelectronics.

The "right answer" is going to be determined by your algorithm, so it's very hard to give a generic answer that will be useful.

At a very high level ...

Multi-threading is usually a good thing if the algorithm allows it. Using more cores can give better performance for burst workloads, as well as better energy efficiency for sustained workloads. 

Coarse division of work across threads is usually the most efficient way to do things. The smaller your pieces of work are, the more overhead you have synchronizing the threads, so fine-grained division has more overhead.

... but there are always exceptions to this ...

hi,

in my case i remarks that big function run faster, there may be a limit of size but it is out of the scope. Or smal function using INLINE at compile time for outside function. So big function looh to be the best. Same function on 4 core is better than 2 function on 2 core. 

One more question. In case of the 1 function on 4 core does the instruction are copied to the for core or one core send instruction to the 4 core.

;))

this is the direct link to the FVP download page 

https://developer.arm.com/Tools%20and%20Software/Fixed%20Virtual%20Platforms/IoT%20FVPs

Use this link to access the latest ML Embedded Evaluation Kit repository: https://gitlab.arm.com/artificial-intelligence/ethos-u/ml-embedded-evaluation-kit