Implementation of concurrency primitives

[Andrey Lomakin]

Good morning, guys.

I am considering creating an application that will use a
thread-per-core approach. Data mostly will not be shared between cores and
I am going to use a single thread executor to schedule fibres.

My question is may I use my own concurrency primitives which do not inject
memory barriers unlike the standard concurrency primitives to
achieve maximum performance and if that is not possible could you explain
why?

-- 
Best regards,
Andrey Lomakin.