Abstract
This research presents the design and implementation of a prototype programming tool for vectorization and parallelization assistance, called the Workstation Vectorization and Parallelization Assistance Environment (WVPAE). The proposed working environment for WVPAE is the workstation. The WVPAE is designed to function as an experimentation facility for interactive vectorization and parallelization assistance during the implementation or maintenance of parallel computing applications. The target high-level language for implementing parallel computing applications is chosen to be a Fortran-like language, and the target parallel machine has to be specified by the user. The target parallel machine architecture can be either a vector or a multiprocessor machine. The assistance provided by the WVPAE is based on analyzing user programs to discover all barriers that may cause either vectorization blocking or parallelization blocking. Vectorization and parallelization barriers are categorized in this research into barriers due to sequential language constructs and those due to dependence relationships. The WVPAE tool provides diagnostic messages and advice for most of the barriers defined in this research. User interaction is inquired by the WVPAE tool to avoid some vectorization or parallelization barriers. The user's response has an effect on modifying or redefining parts of the analyzed program information bases. The analyzed program is represented in the WVPAE tool by a program flow graph and information bases for program syntactic and semantic knowledge, program loop structures, symbol tables, and array subscript information. To facilitate matching user programs to parallel machines, the WVPAE tool provides a software tool for Interactive Loop Interchanging, called ILI. The ILI facility allows, incrementally, tuning the nested DO-loop ordering to a particular machine using a machine dependent criterion. The execution time of vectorized operations has been adopted for tuning nested DO-loop ordering to vector machines, and the induced delay interval for the Doacross schedule has been adopted for tuning nested DO-loop ordering to multi-processor machines...
Arafeh, Bassel Rashad (1986). Parallelizing assistant for parallel architectures. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -18413.