Not all, only those that can be proven to be passed to atomic calls.

Given the rejuvenated emphasis on compile performance, it's probably not worthwhile to figure that out :) Overall Go's "don't sugarcoat everything with abstractions" attitude suits me fine for now.

I have a need for quickly aggregating counters across multiple goroutines. `atomic.AddUint64(&central, threadLocal)` works quite well for that, better than pushing to a channel, better than deferring the aggregation until a sync point.

Any needs i have for alignment are derived solely from the requirements of the atomic package.