You're definitely right. An MPI_UB is also required.
Therefore, I am asking Bill to make this slight modification to the
view definition. The mention of the necessity of an MPI_LB did
not mean to be exclusive of the necessity of an MPI_UB.
It is probably better to mention the necessity of both an MPI-LB
and MPI_UB to remove any ambiguity.
Jean-Pierre
linda @ k2.llnl.gov
06/09/97 06:22 PM
To: mpi-io @ mcs.anl.gov
cc: (bcc: Jean-Pierre Prost/Watson/IBM Research)
Subject: Request for clarification
My interpretation of the filetype tiling (after corrections from previous
discussions here) is that holes at the beginning or end of a filetype
will be repeated in the filetype tiling only if explicit MPI_LB and
MPI_UB are used to set the extent appropriately to include the holes.
The words describing the tiling for the filetype in figure 10.1 only
mention the need for an MPI_LB to repeat the 1-unit hole at the beginning,
leaving me to wonder about how the 3-unit hole at the end is repeated.
(See page 216, the definition of view, in the May 8 text.)
In the example, with explicit MPI_LB and MPI_UB, I believe the extent of
the filetype is 6 units, and without, the extent is only 2 units. Thus to
get the tiling shown in the figure, I expect an MPI_UB would be needed.
Do others agree with this assessment? If not, can someone explain the
error in my reasoning?
Linda Stanberry
lstanberry@llnl.gov