.#
.#
.#
.ch "Conclusions"
It appears as if the standard libraries under Primos have been
implemented without anything other than a cursory check of accuracy.
A number of the library routines return incorrect results that
are mathematically absurd.  Other routines trigger errors on
values which should be well within range.
.pp
Although the single precision arithmetic is acceptable for most
calculations,
the double precision floating point arithmetic on Prime 400/550 machines (and possibly
on the new 2250, as well) is seriously flawed.  Critical calculations
should not be performed on any of these machines since the error
induced by certain unstable operations can completely ruin
the accuracy of the results.  Bizarre behavior of programs which
work on other machines may also be noted due to some of the odd
quirks in the floating point structure.
Users should run their own tests to determine if their applications
will be affected adversely by these problems.
.pp
An increase in accuracy may very well be obtained in some programs
by recoding the standard functions.  It has been shown that the
SWT Math Library significantly outperforms the standard Prime
libraries in virtually every instance; it is possible that the encoding
of different algorithms might also result in increased precision.
.pp
This paper has also presented differences in the architecture of
the Prime 400/550 computer and the 750 which violate the claim
of strict upward compatibility of software.  Programs which directly
access the register structure or make specific assumptions about
precision should be coded with these differences in mind.