[mpiwg-rma] Memory model and source code portability
Dave Goodell (dgoodell)
dgoodell at cisco.com
Wed Oct 1 11:59:50 CDT 2014
On Oct 1, 2014, at 11:10 AM, Rolf Rabenseifner <rabenseifner at hlrs.de> wrote:
>> This is the same chicken little story that McClaren told us a while
>> back. No C compiler is going to do what you suggested. It's perverse
>> and not within the realm of possibility.
> It was my message that Compiler don't do this, because
> they implement C11 memory model.
> I other words, Boehms problems are historical for C.
> When I understood Boehm correctly, then it was a problem,
> i.e. not "perverse" and within "the realm of possibility".
To be precise, *some* compilers have begun to support the C11 memory model very recently (i.e., in the last year or two). Just as there are many MPI implementations out there with incomplete MPI-3 support, there are compilers out there with varying levels of C11 memory model support.
Prior to the availability of the C11 memory model, most compilers avoided the problems in the Boehm paper in various ways, some requiring non-portable attributes/pragmas.
More information about the mpiwg-rma