>Displacements in bytes are one option (I probably got that wrong in my
>example). Another possibility is to specify a "least common denominator"
>datatype, and require that all other datatypes be built out of this datatype.
>Thus, if we scatter from a matic of reals, the lcd datatype is MPI_REAL, all
>other datatypes are built of reals, and displacements are in size_of_real.
If we use the "least common denominator" the advantages are:
1) User simplicity - no need to use "sizeof(datatype)" [and the corresponding
well known error "sizeof(MPI_FLOAT)" !!!]
2) Possible implementation optimizations in not having to handle a myriad of
unaligned data conditions.
The disadvantage seems to me:
1) The collective functions will STILL not be as general as the point-to-point
functions. Someone eventually will want to move unaligned data, they will
have a very valid reason, and the MPI functions will be insufficient. Do we
then need to raise this issue again, to define the most general functions?
Lloyd Lewins
Hughes Aircraft Co.
llewins@msmail4.hac.com