[A92] Re: e^x
[Prev][Next][Index][Thread]
[A92] Re: e^x
Yes i reversed engineered the rom of tis and hps, but the constants aren't
thouse in a taylor polynomial.
They do something like :
y = x/c1 , the quotient (just the 1st digit) to digit 8
y1 = y/c2, again the quotint to digit 1
y2 = y1/c3 , blah blah...
later this number, y10, is divided by itself divided by powers of 10, and
builds the result. I don't understand very well..
c1 = Ln 10
c2 = Ln 1024
c3 = 9.53.... ???
c4 = 9.953.... ???
c5 = 9.995...
c6 = 9.999....
do you have any clue ?
Besides that a taylor poly has a accuracy problem, and you only can
calculate it between 0 and 1 with 19 constants (for 16 digits...), so you
need more constants..
and so.
David Kuehling wrote:
> >>>>> "Alejandro" == Alejandro Paz <psys@cotelco.com.ar> writes:
>
> > Anybody knows how the e^x calculation algorithm works ?
>
> > Note : ti calcs don't use a taylor polynomial to calc it.
>
> Why are you so sure that it's not a taylor polynomial? Did you
> reverse-engineer the ROM? I once implemented a floating point e^x using
> the Taylor polynomial, and it just worked great.
>
> David
References: