A comparison study of automatically vectorizing Fortran compilers

The performance of current vector processing supercomputers is highly dependent on the capabilities of their automatically vectorizing Fortran compilers to generate efficient vector code. In this paper we shall focus on the performance of these compilers, and attempt to compare their efficiency on different types of Fortran loop constructs. Our analysis will involve a short analysis of previous compiler comparison studies and we will introduce a new technique for evaluating and comparing compiler performance.