[ofiwg] the problems with atomics

[Jason Gunthorpe]

On Mon, Jul 13, 2015 at 08:31:19PM +0000, Sur, Sayantan wrote:
> >Isn't it the other way around? libfabric needs to know what the app
> >was compiled with?
> 
> I assumed that if we punt, then all we need the app to do is to check
> whether its definition of long double matched OFI’s, and if not, bail
> (which may mean emulate atomics itself).

If you want to punt then app needs to make sure the local app, local
libfabric, local HCA, remote HCA, remote libfabric and remote app all
use the same type system.

As I suggested, the best way to do this is with some kind of header
file logic that figures out what format the compiler is using for all
of the floating point types.

Jason