20.1.16. Optimization Problems, an Overview

MPI provides operations that may be hidden from the user code and run concurrently with it, accessing the same memory as user code. Examples include the data transfer for an MPI_IRECV. The optimizer of a compiler will assume that it can recognize periods when a copy of a variable can be kept in a register without reloading from or storing to memory. When the user code is working with a register copy of some variable while the hidden operation reads or writes the memory copy, problems occur. These problems are independent of the Fortran support method; i.e., they occur with the mpi_f08 module, the mpi module, and the (deprecated) mpif.h include file.

This section shows four problematic usage areas (the abbrevations in parentheses are used in the table below):

• Use of nonblocking routines or persistent requests (Nonbl.).
• Use of one-sided routines (1-sided).
• Use of MPI parallel file I/O split collective operations (Split).
• Use of MPI_BOTTOM together with absolute displacements in MPI datatypes, or relative displacements between two variables in such datatypes (Bottom).

• Code movement and register optimization problems; see Section Problems with Code Movement and Register Optimization.
• Temporary data movement and temporary memory modifications; see Section Temporary Data Movement and Temporary Memory Modification.
• Permanent data movement (e.g., through garbage collection); see Section Permanent Data Movement.

Table 29: Occurrence of Fortran optimization problems in several usage areas

• to minimize the burden for the application programmer, e.g., as shown in Sections Solutions through The (Poorly Performing) Fortran VOLATILE Attribute on pages Solutions--The (Poorly Performing) Fortran VOLATILE Attribute,
• to minimize the drawbacks on compiler based optimization, and
• to minimize the requirements defined in Section Requirements on Fortran Compilers.