Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uwm.edu!ogicse!ucsd!hub.ucsb.edu!ucsbuxa!6600m00n
From: 6600m00n@ucsbuxa.ucsb.edu (Steelworker)
Newsgroups: comp.sys.intel
Subject: Re: 80387 computation (how to question)
Message-ID: <12121@hub.ucsb.edu>
Date: 19 Jun 91 18:56:41 GMT
References: <1991Jun19.023041.25008@metro.ucc.su.OZ.AU>
Sender: news@hub.ucsb.edu
Lines: 24

In article <1991Jun19.023041.25008@metro.ucc.su.OZ.AU>
glenn@suphys.physics.su.OZ.AU (Glenn Geers) writes:

]	I've written most of the UNIX math library routines in 386/387 assembler
]in order to get more bang for my buck. However, there is one that has eluded me
]so far (at least efficiently) and thats exp(x)-1. The 387 has an instruction
]for doing 2^x - 1 but I can't see a neat way of scaling the result or
]prescaling
]the input. Anyone know how?

]Thanks in adavnce,
]	Glenn

One way is to compute 2^( x * log e) -1  , where the log is in
base two.   That involves only a bit more work. ( precalculating log e 
of course).
one line proof:
ln(2^(x * log e)) = x * log e * ln 2 = x

Have fun with the math, 
Rob Blair
6600m00n@ucsbuxa.ucsb.edu
( One other task you might want to do is to make a library that takes
high precision 80 bit reals, for that extra accuracy.)