Need to add volatile to some globals to workaround gcc bug

[rolfv]

All the details are contained in this thread from the User's list. Looks like we planned on adding the workaround, but it never happened. I just had a second user run into this issue.
http://www.open-mpi.org/community/lists/users/2015/06/27039.php

I would like to make this a blocker for Open MPI 1.10.1.

[jsquyres]

Oy. Things really crawl out of the walls when we announce we want to release in the Very Near Future!

Can you make a PR for this ASAP (e.g., by tomorrow)?

[jsquyres]

BTW, I assume you're going to read the history about this on #625. Some solutions were proposed on there, IIRC.

[rolfv]

Yes, I read all the notes and both you and Dave Goodall already gave a thumbs up to the user of volatile. And the GCC guy also said to make it volatile. That is 3 yes votes! I will work on making it find its way through the 3 releases.

[rolfv]

I misread what you were pointing too. I was reading the email thread, not the thread from #625. I was not quite sure what your "softer" suggestion was.

[jsquyres]

See #625 (comment) and #625 (comment).

[rolfv]

Are suggesting making the whole structure volatile? Because then you are making things volatile that are unnecessary. Or am I missing something. @jsquyres

[jsquyres]

I think we never resolved which of the two approaches was better. 😄

[jsquyres]

Whichever you do, please submit a PR today

[rolfv]

Right, I am just trying to implement your solution to see if it works but I am not exactly sure what you mean. I have shown the volatile fixes it with the 4.9.2 compiler. @jsquyres

[jsquyres]

I don't know which way is better -- e.g., are there side effects of making the whole struct volatile that could impact performance, etc.