Joe Orton wrote:

I am just looking at CuTifying testatomic - what is the thread_func_none
test for?  It appears to just test that z++ is atomic for a volatile
long z, which it isn't of course... am I missing something?

It isn't clear from the messages, but after playing with testatomic yesterday (and finding a platform where the none test actually fails) I think the value of it is this, given that z++ is not atomic and such a test can be expected to fail given an interesting level of concurrency:


if the z++ test works, then we probably haven't established the right level of concurrency to do a valid test of the atomic operations

if the z++ test fails, it is a strong suggestion that the other tests we did were meaningful (i.e., there was adequate concurrency)




Reply via email to