Minimizing Communication/Running Time in Parallel Executions

[Doruk Gür]

Hello all, 

I am trying to minimize the running time for elementwise operations I am doing. Basically there are multiple arrays of n elements which do the same operation on each element. To sum up, the order of operations look something like this:

Array 3  = Array 1 >= 0
Array 4 = Array 1 + Array 2 >= 0 
Array 5 = Several multiplications and additions based on Arrays 1,2,3,4
Array 6 = Array 1 - Array 5

The immediate solution seems to be using vectorized operations, which I did and seems to be working fine, compiling without the "-n" flag and hence having constant number of rounds as expected. However I was wondering if this could be faster using multiple threads instead. I have tried to convert this into a simple multithread version using @for_range_multithread() however it doesn't seem to work without the "-n" flag, which slows down the execution in general. 

I have read somewhere that only thread 0 has IO capabilities since synchronizing different threads in different players would be a nightmare to deal with so it is best to use threads in local operations that do not require communication. But I have seen an example with multiplication of secret integers within the thread. 

So I am curious if there is a proper way to do elementwise comparisons and multiplications within different threads with constant number of rounds that I am not aware of or threading is useful for mostly local operations? Or alternatively, would the vectorized operations be as fast as it will get?

Best regards and thanks,