[mod.computers.gould] Forwarded: improved UTX mathlib

gouldbugs@ucbvax.UUCP (11/25/85)

From ncr-sd!sdcc6!sdcc3!sdcsvax!sdcrdcf!trwrb!scgvaxd!wlbr!voder!nsc!pyramid!pesnta!amd!amdcad!amdimage!prls!philabs!ron1!polyof!john Tue Nov 19 08:22:38 1985
Relay-Version: version B 2.10.3 alpha 4/15/85; site gould9.UUCP
Posting-Version: version B 2.10.2 9/5/84; site polyof.UUCP
Path: gould9!ncr-sd!sdcc6!sdcc3!sdcsvax!sdcrdcf!trwrb!scgvaxd!wlbr!voder!nsc!pyramid!pesnta!amd!amdcad!amdimage!prls!philabs!ron1!polyof!john
From: john@polyof.UUCP ( John Buck )
Newsgroups: net.sources
Subject: Gould Concept 32 libm routines
Message-ID: <171@polyof.UUCP>
Date: 19 Nov 85 16:22:38 GMT
Date-Received: 24 Nov 85 10:50:49 GMT
Distribution: net
Organization: Polytechnic Inst. of NY, Farmingdale
Lines: 11

We have re-written several of the heavily used routines in libm.a
in assembler for the Gould 32/67, 32/97, PN60xx PN90xx machines.

Specifically, sqrt(), ldexp() and frexp().
sqrt() is about 10 times faster, and ldexp(), frexp() are MUCH faster
than that.  If you do lots of f77 stuff (exp, log, etc), these call
frexp and ldexp, so their times will be much better too.

If you are interested, send E-mail.  I would have just posted them,
but they are about 4K ofsource, and comments, and the target group
is quite small.