The need for high level language support for parallel computers is universally accepted. This is illustrated by the intense interest in High Performance Fortran (HPF) - data parallel Fortran stemming directly from Arpa funded technology. Although HPF is now adopted by most HPCC vendors, much research remains for parallel compilers - this includes extensions of data parallelism to other languages such support for a wider range of applications than those included in HPF and study of compilers for machines with new and different hardware capabilities. Within this project a basic public domain infrastructure to provide runtime support for high level parallel languages will be developed.
The project combine the expertise of eight institutions for producing a common runtime software base which exploits the experience and expertise of the groups. Syracuse University coordinates this effort. A single software system will support six existing and one new Arpa funded compiler development activities. A special attention will be payed to software engineering issues for this new runtime system. Cooperating Systems and University of Texas, Austin would be responsible for this part of the project. This task would involve engineering the runtime system for reuse and extensibility as well as testing and validation.
In the first phase of the project, a simple and compact standard for creating and accessing distributed data in a parallel environment is being developed. The specification will provide a standard API for low-level compiler-generated code, runtime systems, and parallel toolkits that can interoperate and share layers.
Interoperability is a fundamental requirement for this specification. The specification will be available to both HPC++ and HPF compilers. This will permit programs to be constructed from HPF and HPC++ libraries and object files. For example, a C++ program will be able to create distributed data structures that are later passed to a library created with HPF.
Interoperability is also important at the tool level. For example, a tool builder should be able to provide source-language independent visualization tools to view the placement and contents of data structures by using a run-time layer that implements this specification. A debugging tool that permits the user to modify or determine the current owner of a data element during a break is another example of a user for such a run-time layer. Similarly, compilers for HPF, HPC++, C, and pC++ can share the layer of a run-time system built using this specification.