Short answer: The standard says so.
Longer answer: You need only commit types that are being used for
communication. Committing gives the library the opportunity to
"compile" the derived datatype representation into a more efficient
form. Thus, if you build a derived datatype in steps based on other
derived datatypes the library need not spend the time putting the
intermediate datatypes into this more efficient representation.
I have always thought this was silly as I feel if your code is doing
enough datatype constructions for the time spent in the construction
calls to impact your overall performance you should probably look at
redesigning your code. Maybe some people have experience with codes
where this is actually an issue. I would love to hear about it.
Dave.
Angel Tsankov wrote:
> Can anyone tell me why the library does not incorporate the function of
> MPI_Type_commit into MPI_Type_contiguous and the other type conntructors?
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
--
Dr. David Cronk, Ph.D. phone: (865) 974-3735
Research Leader fax: (865) 974-8296
Innovative Computing Lab http://www.cs.utk.edu/~cronk
University of Tennessee, Knoxville
|
