Ravi Ponnusamy, Joel Saltz, Alok Choudhary, Yuan-Shin Hwang, Geoffrey Fox.
This paper describes two new ideas by which an HPF compiler can deal with irregular computations effectively. The first mechanism invokes a user specified mapping procedure via a set of compiler directives. The directives allow use of program arrays to describe graph connectivity, spatial location of array elements and computational load. The second mechanism is a simple conservative method that in many cases enables a compiler to recognize that it is possible to reuse previously computed information from inspectors (e.g. communication schedules, loop iteration partitions, information that associates off-processor data copies with on-processor buffer locations). We present performance results for these mechanisms from a Fortran 90D compiler implementation.
Ravi Ponnusamy, Yuan-Shin Hwang, Joel Saltz, Alok Choudhary, Geoffrey Fox.
We present methods that make it possible to efficiently support an important subclass of irregular problems using data parallel languages. The approach we describe involves the use of a portable, compiler-independent, runtime support library called CHAOS. The CHAOS runtime support library contains procedures that support static and dynamic distributed array partitioning, partition loop iterations and indirection arrays, remap arrays from one distribution to another, and carry out index translation, buffer allocation and communication schedule generation.
The CHAOS runtime procedures are used by a prototype Fortran 90D compiler as runtime support for irregular problems. We present performance results of compiler-generated and hand-parallelized versions of two stripped down applications codes. The first code is derived from an unstructured mesh computational fluid dynamics flow solver and the second is derived from the molecular dynamics code CHARMM.
A method is described that makes it possible to emulate irregular distributions in HPF by reordering elements of data arrays and renumbering indirection arrays. We present results that suggest that an HPF compiler could use reordering and renumbering extrinsic functions to obtain performance comparable to that achieved by a compiler for a language (such as Fortran 90D) that directly supports irregular distributions.