On Thu, Nov 13, 2003 at 09:46:17AM -0800, Praveen Agrawal wrote:
>I have been wondering why should this be. Below are my
>reasons to beleive so:(corresponding sizes)
> ILP32 ILP64
>long 32 64<different>
>pointer 32 64<different>
>So, in our mpi pgm if we just don't use long then
>there shouldn't be any problem. As sharing address
>across machines doen't make sence.
if you have structures with pointers or longs in them, then the
structures will be different sizes on each machine. An MPI_Datatype
created from these structures will then be incorrect on one
architecture.
Ummmm - I can't think of anything else.
Endian-ness of the data should be taken care of by LAM ok...
cheers,
robin
|
