Advanced compiler optimizations for supercomputers
Communications of the ACM - Special issue on parallelism
Automatic translation of FORTRAN programs to vector form
ACM Transactions on Programming Languages and Systems (TOPLAS)
Compiler transformations for high-performance computing
ACM Computing Surveys (CSUR)
A programmer's guide to ZPL
Automatic intra-register vectorization for the Intel architecture
International Journal of Parallel Programming
The design and implementation of a region-based parallel programming language
The design and implementation of a region-based parallel programming language
| Hi-index | 0.00 |
In this paper, we investigate the issues of compiling high-level languages for vector architectures. Vector architectures have regained popularity in recent years, from simple desktop computers with small vector units motivated by multimedia applications to large-scale vector multiprocessing machines motivated by ever-growing computational demands. We show that generating code for various types of vector architectures can be done using several idioms, and that the best idiom is not what a programmer would normally do. Using a set of benchmark programs, we also show that the benefits of vectorization can be significant and must not be ignored. Our results show that high-level languages are an attractive means of programming vector architectures since their compilers can generate code using the specific idioms that are most effective for the low-level vectorizing compiler. This leads to source code that is clearer and more maintainable, has excellent performance across the full spectrum of vector architectures, and therefore improves programmer productivity.