Determine which store to memory that will happen first on a multi-core system?

Question

Assuming I have two seperate threads, each running on their own CPU.

Both attempt to write to a shared piece of memory at the same time, how do I know which value will be stored into memory first/will get the most up-to-date verison?

Or will certain memory consistency models make sure this is impossible?

If so how?

Answer 1

"At the same time" does not really have meaning in multi-core systems. Do you mean the moment when the store instruction is executed (actually a span of time), the moment the value is put into the core-local store cache, the moment the store cache is flushed to the CPU-shared cache (note that there may be CPUs where a subset of cores has a shared cache), or when the cache is flushed to memory? Which core's view of the system do you want to use for observing the result? (In quad-core systems, it could well be that when core 1 and 2 both write values, cores 3 and 4 observe the writes in different orders.)

CPUs have ways of ensuring that two writes have a guaranteed global order (memory barriers and interlocked load/store operations) - unless you use those, the results are unspecified and guaranteed to be surprising. And in the case of the C11 and C++11 memory model, they're fully undefined.

Answer 2

If two threads are writing to the same memory at the same time, there's no way to know the order the writes happen in, by definition (otherwise it wouldn't be "at the same time").

Even if there was, what would happen if one of them was delayed by a nanosecond?