Introduction

In this report, we introduce HPJava language, a programming language extended from Java for parallel programming on message passing systems, from multiprocessor systems to workstation clusters.

Although it has a close relationship with HPF[1], the design of HPJava does not follow HPF directly. Instead it introduces a high level structured SPMD programming style, HPspmd, which can be summarized as follows:

• Structured SPMD programming. Programs written in the programming language presented here explicitly coordinate a well-organized process group. As in a conventional distributed-memory SPMD program, only a process owning a data item is allowed to access the item directly. The language provides special constructs that allow programmers to meet this constraint conveniently.

• Global name space. Besides the normal local variables of the sequential base language, the language provides classes of variables accompanied by non-trivial data descriptors, providing a global name space in the form of globally subscripted arrays, with assorted distributed patterns. This helps to relieve programmers of error-prone activities such as the local-to-global, global-to-local address translations which occur in data parallel applications.

• Hybrid of data and task parallel programming. The language also provides special constructs to facilitate both data parallel and task parallel programming. Through language constructs, different processors can either simultaneously work on global addressed data, or independently execute complex procedures on their own local data. The conversion between these phases is seamless.

• Communication libraries. In the traditional SPMD model, the language itself does not provide implicit data movement semantics. Different communication patterns are implemented as library functions. This greatly simplifies the task of the compiler, and should encourage programmers to use algorithms that exploit locality. Data on remote processors are accessed exclusively through explicit library calls. In particular, the initial HPJava implementation relies on a library of powerful collective communication routines. Other distributed-array oriented communication libraries may be bound to the language later. The low level MPI communication is always available as a fall-back. Since the language itself only provides basic concepts to organize data arrays and process groups, it allows the possibility that when a new communication pattern is needed, it should be relatively easy to integrate through new libraries.

In our earlier work on HPF compilation [2] the role of runtime support was emphasized. Difficulties in compiling HPF efficiently suggested to make the runtime communication library directly visible in the programming model. Since Java language is simple, elegant language, we implemented our prototype based upon this language.