MPIAint

MPI_Aint is a signed integer type defined by the Message Passing Interface (MPI) standard used to hold memory addresses and displacements in MPI operations. It provides a portable way to represent pointers or offsets when constructing derived datatypes or performing address arithmetic, across platforms where native pointer sizes may vary.

The size and exact representation of MPI_Aint are implementation-defined, but it is guaranteed to be large enough to hold a pointer on the target architecture. In C, MPI_Aint is defined in the MPI header (mpi.h) and is distinct from standard C integer types to ensure portability of address-related computations. In Fortran, address-length values are typically represented using a corresponding kind such as MPI_ADDRESS_KIND.

A common use of MPI_Aint is in the construction of derived datatypes, where displacements of struct members or array elements are required. The typical workflow involves obtaining the memory address of a base object and of each relevant member with MPI_Get_address, subtracting the base address to obtain displacements in MPI_Aint, and supplying these displacements to functions like MPI_Type_create_struct or MPI_Type_create_hindexed. MPI_Aint ensures that displacement values are represented consistently regardless of the platform’s pointer size.

MPI_Aint also supports portable address arithmetic through utilities such as MPI_Aint_add and MPI_Aint_diff, which perform addition and subtraction on MPI_Aint values in a portable way.

See also: MPI_Get_address, MPI_Type_create_struct, MPI_Type_create_hindexed, MPI_Aint_add, MPI_Aint_diff.