<div class="gmail_quote">On Thu, Nov 10, 2011 at 21:14, Kai Germaschewski <span dir="ltr"><<a href="mailto:kai.germaschewski@unh.edu">kai.germaschewski@unh.edu</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div>Just a (kinda arbitrary) data point: I introduced MPI_IN_PLACE in my code a year or so ago (not realizing that it requires MPI-2.0 at the time) and within a year, I hit two cases of machines that were still running MPI 1.x -- and the code doesn't get to run in all that many places in the first place.<br>
</div></blockquote><div><br></div><div>What sort of configurations were these?</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;"><div>
<br>Having said that, I think it's generally a good idea to not keep backwards compatibility cruft around forever. OTOH, requiring MPI2 just for MPI_IN_PLACE isn't a real convincing case, since it's so easy to do the same with MPI1.</div>
</blockquote></div><br><div>I agree, this is mostly cosmetic, but there are more MPI-2 features that would also be useful. There are places where having MPI 1-sided would greatly simplify, for example.</div>