CRPC-TR 92204

> >1. Abstract
> >-----------
> >
> >Static performance estimation seeks to determine at compile time how
> >long a given program construct, such as a loop or a subroutine call,
> >will take to execute.  Performance estimation provides useful
> >information to parallelizing compilers, particularly compilers which
> >aggressively use code transformations to improve parallelism.  In an
> >interactive parallel programming tool, performance estimation can
> >direct users to the most important and computation-intensive portions
> >of their programs.  This paper describes the design and implementation
> >of a performance estimator developed to assist in the parallelization
> >and optimization of scientific Fortran programs for shared-memory
> >multiprocessors in the ParaScope programming environment.
> >
> >2. Keywords
> >-----------
> >
> >performance estimation
> >performance prediction
> >training sets
> >parallelizing compilers