amull@Morgan.COM (Andrew P. Mullhaupt) (08/20/90)
Mr Felscher has asked me to post this for him, and I can't see any
reason why not. I would ask that e-mail replies be directed to him,
or perhaps posted. (He is able to read comp.lang.apl, but evidently
not able to post.)
-----
Subject: Long Arithmetic in APL
Background: It obviously is tempting to use APL for algorithms transforming
matrices into their various normal forms. Everything works well over
fields GF(p). Working over integers, already intermeadiary results
easily surpass APL's precise 15 digits. Working over rationals is
quite hopeless since the need to form common denominators leads to
large numbers very fast.
Question: Has anyone implemented a well working integer multiple precision
arithmetic for APL ?
I did, some time ago, write a multiple precision arithmetic (MPA)
_within_ APL , but this, understandably, will work fairly slow.
And of course, it isn't vectorial in the way APL's basic operations
are, so that using it will require additional lines in a program.
A much better solution would be an AP (auxiliary processor)
written in assembler and containing the MPA functions. That then
could be associated with help of shared variables . [The hardware
I am using are PS2's with either the 286 or the 386 processor, plus
coprocessor; the particular APL is IBM's APL2-PC. ]
Walter Felscher / Math. Institut / Univ. Tuebingen
mife001@mailserv.zdv.uni-tuebingen.de