what is the performance impact of using int64_t instead of int32_t on 32-bit systems?

Question

Our C++ library currently uses time_t for storing time values. I\'m beginning to need sub-second precision in some places, so a larger data type will be necessary there anyw

Answer 1

Your question sounds pretty weird in its environment. You use time_t that uses up 32 bits. You need additional info, what means more bits. So you are forced to use something bigger than int32. It doesn't matter what the performance is, right? Choices will go between using just say 40 bits or go ahead to int64. Unless millions of instances must be stored of it, the latter is a sensible choice.

As others pointed out the only way to know the true performance is to measure it with profiler, (in some gross samples a simple clock will do). so just go ahead and measure. It must not be hard to globalreplace your time_t usage to a typedef and redefine it to 64 bit and patch up the few instances where real time_t was expected.

My bet would be on "unmeasurable difference" unless your current time_t instances take up at least a few megs of memory. on current Intel-like platforms the cores spend most of the time waiting for external memory to get into cache. A single cache miss stalls for hundred(s) of cycles. What makes calculating 1-tick differences on instructions infeasible. Your real performance may drop due yo things like your current structure just fits a cache line and the bigger one needs two. And if you never measured your current performance you might discover that you could gain extreme speedup of some funcitons just by adding some alignment or exchange order of some members in a structure. Or pack(1) the structure instead of using the default layout...