Reshape test: variable used for polling should be volatile

[devreal]

Signed-off-by: Joseph Schuchart schuchart@icl.utk.edu

[abouteiller]

Don't think that fix is entirely correct https://www.kernel.org/doc/Documentation/process/volatile-considered-harmful.rst

Should have the memory yield instruction in the loop instead.

@ICLDisco/staff

[devreal]

I agree that volatile is not the perfect choice here, I am not worried about performance in this case though. And we do not have something like cpu_relax() in PaRSEC, nor can we use _Atomic as a replacement for volatile. PaRSEC uses volatile in many places where _Atomic would be a better choice (lifo, list, taskpool, arena, barrier, rwlock, even in the C code generated from JDF). If we want to go down the route of adding cpu_relax() I'm all for it, but it likely requires architecture-specific implementations. We can peek at the kernel code for that, of course. Or just bite the bullet and transition to requiring C11 atomics and allow for _Atomic to be used directly.