On Sun, 9 Dec 2001, Bora Ucar wrote:
> Before, using the MPI_Isend I just read the page describing Isend at
> www-unix.mcs.anl.gov/mpi/www/www3/MPI_Isend.html. At the page it is
> not stated that to complete the Isend some has to do some freeing
> steps, like Request_free, or Test, or Wait. Even if the message has
> been delivered to the destination there are somethings (that comsume
> time) to be freed. I later check the page for MPI_Wait and it does not
> state the "side-effects" of the function (freeing stuff). Then later I
> checked it in our system by typing man MPI_Isend and see the crucial
> note there("The communication started with this function is not
> guaranteed to progress or complete until req has been given to one of
> the test or wait functions MPI_Test, MPI_Testall,...").
Keep in mind that MPI is a standard -- there are many implementations of
this standard. Manual pages for a specific implementation may not provide
complete information about a given MPI function call -- it is likely that
they will only implementation-specific information and assume knowledge of
the actual standardized MPI call. From an MPI developer's point of view,
there is little use in duplicating the MPI standards on man pages.
Indeed, the LAM man pages for MPI function calls all state:
-----
MORE INFORMATION
For more information, please see the official MPI Forum
web site, which contains the text of both the MPI-1 and
MPI-2 standards. These documents contain detailed infor
mation about each MPI function (most of which is not
duplicated in these man pages).
http://www.mpi-forum.org/
-----
The official MPI Forum web site paired with the books that have been
mentioned on this list recently are excellent resources for MPI.
> By the way, is it more efficient (in terms of communication
> computation overlap) to use MPI_Isend(buf..) and immediately after use
> MPI_Request_free (like Mr. Lebeau) assuming that programmer has
> guarenteed that the "buf" is reusable.
It depends on the particular MPI implementation. With LAM, it's about the
same either way (freeing them immediately or explicitly testing the
requests later).
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
_______________________________________________
This list is archived at http://www.lam-mpi.org/MailArchives/lam/
|
