[mpiwg-p2p] Next meeting: 19th January 2015 11am Central US

Daniel Holmes dholmes at epcc.ed.ac.uk
Fri Jan 16 06:34:36 CST 2015

Hi All,

The next point-to-point teleconference will be on Monday 19th January 
2015 at 11am Central US via webex.

Connection details are on the point-to-point wiki page:

1) New INFO keys - specifically the "can INFO keys change the semantic 
behaviour of MPI?" issue (see summary from Jim below)
2) Progress on Arecv and Fsend
3) Other stuff?


On 12/01/2015 15:39, Jim Dinan wrote:
> Hi All,
> Apologies for missing the meeting.  There was a lot of discussion at 
> the Forum about the validity of restricting the application's behavior 
> using info keys (or, if you prefer, the application asserting its 
> behavior through info keys).  Folks indicated that this qualifies as a 
> semantic change for the given communicator, which is forbidden by 
> Section 6.4.4 and made more complicated because MPI_Comm_dup 
> propagates info keys.
> In response to this feedback, I started working on a straw man 
> communicator assertions interface that would allow the application to 
> assert what its behavior will be.  I will prepare this for discussion 
> at a future WG meeting.
> In parallel, I have had discussions with others who feel that the 
> opinion expressed at the Forum meeting is incorrect; that it should be 
> valid to provide hints on application behavior through info keys.
> At the same time, the straw vote at the December meeting indicated 
> that the Forum as a whole would like to pursue this capability for the 
> MPI standard.  So, much more discussion and debate is needed.  Looking 
> forward to it,
>  ~Jim.

Dan Holmes
Applications Consultant in HPC Research
EPCC, The University of Edinburgh
James Clerk Maxwell Building
The Kings Buildings
Peter Guthrie Tait Road
T: +44(0)131 651 3465
E: dholmes at epcc.ed.ac.uk

*Please consider the environment before printing this email.*

The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.

More information about the mpiwg-p2p mailing list