The Programming Guide says that
[indent]if a non-atomic instruction executed by a warp writes to the same location in global
or shared memory for more than one of the threads of the warp, the number of
serialized writes that occur to that location and the order in which they occur is
undefined, but one of the writes is guaranteed to succeed.[/indent]
I wonder that how do a non-atomic instruction executed by other warps
write to the same location in global or shared memory.
edit: oops, misunderstood your question
In practice, colliding memory writes from other warps also seem to follow the rule that at least one write succeeds. But as the programming guide does not guarantee such behavior, one should not write an algorithm depending on it.
Since the WARP scheduling is not deterministic, we can safely assume that the programmer should not assume about what will be there in that memory location. - also one should avoid writing code that depends on such behaviour.