[mpich-discuss] MPI I/O using large integers in file type

[Nick Stokes]

Dear all,

I have an application where a large field data is computed in pieces.
The pieces are stored contiguously in process's local memory, but
correspond to an arbitrary collection of  (monotonically increasing)
indexed locations in the final (large) field data.

The natural choice seemed to be to define a File view using
MPI_Type_index filetype, and READ & WRITE local data using an
MPI_Vector (or contiguous) type. This works fine for  small enough
fields, but it seems it is going to fail for  large fields (>4GB),
since MPI_Type_indexed takes int for displacements.

Is there an indexed type which can take 64-bit integers and can be
used for defining file views? Or is there an alternative?

Thanks
- Nick