10.5 High performance fortran (hpf)

In March 1992, the High Performance Fortran Forum (HPFF) began meeting to discuss and define a set of additions to FORTRAN 90 to make it more practical for use in a scalable computing environment. The plan was to develop a specification within the calendar year so that vendors could quickly begin to implement the standard. The scope of the effort included the following:

• Identify scalars and arrays that will be distributed across a parallel machine.
• Say how they will be distributed. Will they be strips, blocks, or something else?
• Specify how these variables will be aligned with respect to one another.
• Redistribute and realign data structures at runtime.
• Add a FORALL control construct for parallel assignments that are difficult or impossible to construct using FORTRAN 90's array syntax.
• Make improvements to the FORTRAN 90 WHERE control construct.
• Add intrinsic functions for common parallel operations.

There were several sources of inspiration for the HPF effort. Layout directives were already part of the FORTRAN 90 programming environment for some SIMD computers (i.e., the CM-2). Also, PVM, the first portable message-passing environment, had been released a year earlier, and users had a year of experience trying to decompose by hand programs. They had developed some basic usable techniques for data decomposition that worked very well but required far too much bookkeeping. As we shall soon see.

The HPF effort brought together a diverse set of interests from all the major high performance computing vendors. Vendors representing all the major architectures were represented. As a result HPF was designed to be implemented on nearly all types of architectures.

There is an effort underway to produce the next FORTRAN standard: FORTRAN 95. FORTRAN 95 is expected to adopt some but not all of the HPF modifications.

Programming in hpf

At its core, HPF includes FORTRAN 90. If a FORTRAN 90 program were run through an HPF compiler, it must produce the same results as if it were run through a FORTRAN 90 compiler. Assuming an HPF program only uses FORTRAN 90 constructs and HPF directives, a FORTRAN 90 compiler could ignore the directives, and it should produce the same results as an HPF compiler.

As the user adds directives to the program, the semantics of the program are not changed. If the user completely misunderstands the application and inserts extremely ill-conceived directives, the program produces correct results very slowly. An HPF compiler doesn't try to "improve on" the user's directives. It assumes the programmer is omniscient. Always a safe assumption.

Once the user has determined how the data will be distributed across the processors, the HPF compiler attempts to use the minimum communication necessary and overlaps communication with computation whenever possible. HPF generally uses an "owner computes" rule for the placement of the computations. A particular element in an array is computed on the processor that stores that array element.

All the necessary data to perform the computation is gathered from remote processors, if necessary, to perform the computation. If the programmer is clever in decomposition and alignment, much of the data needed will be from the local memory rather then a remote memory. The HPF compiler is also responsible for allocating any temporary data structures needed to support communications at runtime.