From: utzoo!decvax!cca!gwyn@Brl@sri-unix
Newsgroups: net.unix-wizards
Title: Re:  pdp11: C, f77, f4p compile study (long)
Article-I.D.: sri-unix.4084
Posted: Wed Oct 27 22:30:53 1982
Received: Thu Oct 28 07:02:50 1982

From:     Doug Gwyn <gwyn@Brl>
Date:     25 Oct 82 15:11:44-EDT (Mon)
I like the idea of generating in-line code for PDP-11 long multiplication
(it only takes a few instructions).

Whitesmiths (Mark Krieger, I believe) adjusted their PDP-11 code generator
so that it generated a fast (two-instruction) entry sequence instead of
calling csv() and jumped to crets instead of cret, whenever a procedure
didn't use the non-volatile registers (R2,R3,R4).

In my experience, most "real" heavy uses of floating-point computation
do require double precision to avoid loss of accuracy (e.g. when inverting
matrices, linear programming, solving D.E.s, etc.).  Therefore I think the
C default mode is reasonable.  Just use doubles in the first place to avoid
the conversion from/to float.