Programming Based on Data Parallelism
In a data parallel programming model, the focus is on data distribution. Every processor works with a portion of data. We shall discuss one of the most popular languages for parallel programming based on data parallel model.
High Performance FORTRAN
In 1993 the High Performance FORTRAN Forum, a group of a lot of leading hardware and software vendors and academicians in the field of parallel processing, discovered an informal language standard known as High Performance Fortran (HPF). It was based on Fortran 90, and then it extended the set of parallel features, and gives extensive support for computation on distributed memory parallel computers. The standard was supported by a majority of vendors of parallel hardware.
HPF is a highly suitable language for data parallel programming models on SIMD and MIMD architecture. It let programmers to add a number of compiler directives that minimize inter-process communication overhead and utilize the load-balancing techniques.
We shall not discuss the total HPF here; rather we shall discuss only on augmenting features like:
- Processor Arrangements,
- Data Distribution,
- Data Alignment,
- FORALL Statement,
- INDEPENDENT loops, and
- Intrinsic Functions.