Firstly, N<=1e9 is quite rare and you would never have 1e9 elements in an array given input alone would probably exceed time limit. Secondly if it is used as sort of limit to a value e.g. values of an array a1,a2,a3... am where ai <= 1e9 then iterating over all values of [1,1e9] is rarely (never afaik) the intended solution and would likely require you to do very simple calculations per iteration and idk if anything more than a bitwise operation or two can run and often it would require several operations (and usually not bitwise ones) and thus isn't very practical. But if all condiitons are met, yes as maxrgby said it can be done with a VERY good constant factor.

If your code is incredibly tight, then I think so.

This can probably run >= 10^9 times in a second if your CPU is 4GHz:

a:
  add ...  # latency 1 cycle, throughput can run up to 4 independent ALU instructions in a single cycle
  add ...
  mul ...
  add ...
  jmp a  # generally 1 cycle

But, divisions can be 20+ cycles (now you only get maybe 2 * 10^8 of them in a second), and if n = 200,000 and you create a vector<int> that takes 800 KiB in memory, random accesses into it will generally hit L2 and each access takes 10-20 cycles. Sets and maps are worse, it does mallocs, which themselves traverse linked lists which are slow, and you will probably end up in L3 which takes far longer. If you have ifs, mispredicted branches can take 20+ cycles too.