It is desirable that a library be scalable. This section shows that Fortran 90D/HPF library provides a corresponding performance improvement as the number of processors increases.
Table presents a sample of performance numbers
for a subset of the intrinsic functions on iPSC/860.
A detailed performance study is presented in [51].
The times in the table include both the computation and
communication times for each function.
For most of the functions we were able to obtain almost linear speedups.
In the case of the TRANSPOSE function, going from one processor
to two or four actually results in an increase in the time due to
the communication requirements. However, for larger size multiprocessors
the times decrease as expected.