Scientific computing by using HPF

Parallel programs usually contain three parts: pre-loop communication, loop computation, and post-loop communication. When you parallelize your program by using HPF, you have to direct HPF how to make pre/post communication and how to distribute loop computation by using HPF DIRECTIVE's. Since there are some, but not many, programming patterns of pre/post-loop communications, you need 'not so many' HPF DIRECTIVES. When you describe parallelization of finite difference method with structured mesh, you use DISTRIBUTE, SHADOW, REFLECT, and REDISTRIBUTE directives. However, particle simulations contain patterns which you describe with very complicated programmings and directives. This is because HPF has not enough directives to direct pre/post-communications which are used in particles simulations. (author)