Parallel Compiler Runtime Consortium
PCRC Home Page
Partners
NPAC 
Maryland
Indiana
Rochester
Texas, Austin
CSC
Rice
Florida |
PCRC at NPAC |
Kernel runtime
HPF compiler
HPJava/mpiJava
WebTop computing
Team |
Documents
Software Approach
The Parallel Compiler Runtime Consortium includes several leading research groups working on parallel runtime systems and compilers. Most partners brought into the project pre-existing software for run-time management of communication and computation in data parallel languages. Such libraries have disparate representations of distributed data, but all provide methods for partitioning and accessing array elements or remapping arrays. They all make some optimizations for reducing data movement costs, such as message-blocking, collective data movement, message coallescing, aggregration, and so on.In the course of the project new runtime software packages have been developed. The NPAC kernel runtime is built on technologies pioneered in the earlier packages, founded on a common descriptor for distributed arrays, and captured in an object-oriented framework. The kernel library can be called from various languages (currently Fortran, C++ and Java) through language-specific interfaces. Three high-performance language translators that use this runtime software exist or are under active development. A separate package developed at Maryland supports interoperability between pre-existing libraries by allowing distributed arrays to be transferred seamlessly from one package to another. Other partners have developed run-time libraries for more specific problem areas.
In the later phases of the project the workplan was strongly influenced by the emergence of Java. Early in 1996, participants of the project issued a draft white paper on the implications of Java for HPCC, and subsequently organized a series of workshops on the theme of Java for Computational Science and Engineering. This series spawned the Java Grande Forum -- a group of academic and industrial partners aiming to promote Java standards for communication and compute intensive applications.