\section*{Abstract.}

CODE is a parallel programming environment developed at the
University of Texas at Austin, and is based upon Unified
Computation Graphs (UCGs), a comprehensive high-level model
for specifying parallel computations.  CODE allows design
of parallel programs independently of the target
architecture and execution environment.  In this
paper we experimentally
evaluate the application of CODE to scientific
computations.  The evaluation measures are ease of
programming, architecture independence, and performance;
the results are better than expected for each
measure.

Two sequential wave-propagation application programs were
specified as parallel programs in the CODE framework.
Parallel Fortran programs were automatically generated for
a 2-CPU Cray XMP, an 8-CPU Cray YMP, and the IBM 3090.  The
ease of programming was reflected in the relatively modest
effort required for this conversion.  Architecture
independence was investigated by executing one of the
programs in both the Cray and the IBM 3090 environments; no
changes were required to the program specifications.
Finally, the performance measurements in the Cray
environments demonstrated the efficiency of parallel programs
generated by CODE.  Some limitations were found in the
current CODE 1.2 implementation of the UCG model, which are
being addressed in the design of a new version, CODE 2.