I know a lot can be optimized, but unfortunately it's not one of my
strong points. :) I tried to rewrite the formula, but none of the other
forms gave me anything that was more efficient computationally.

The formula calculates acceleration, measured by a digital
accelerometer. The T1's en T2's are intervals whose variation is a
measure of acceleration, relative to calibrated measurements as T1Cal
(ZCal) and T2Cal. So a division like T2/T2Cal will always be close to 1.
The output y is a measure for the acceleration. This will later be used
in arcsin, sin and cos calculations. The result of those are arctan'ed;
that's why I need precision.

The range of the variables is about 0-20000, so their product is 4x10^8
worst case. which is pretty much 32bit for my taste (ok it's 29, or 30
with sign; close 'nuff :)).

Anyway, y is always between -2 and +2. I do believe I need floats, but I
may be mistaken.

Martijn

Scott Dattalo wrote:
> On Thu, 5 Jun 2003, Martijn van Aartrijk wrote:
>
>
>>Unfortunately, yes, I need the precision for more trig math later on.
>
>
> Are you sure? The formula you posted can be optimized to get rid of 32-bit
> arithmetic. However, depending on relative magnitudes of the variables,
> the overall result could be less than 1. This may falsely lead you to
> believe that floating point arithmetic is necessary.
>
> I think before we can help, we need to know 1) the magnitude of the
> variables 2) what (exactly) the output of the formula is for. You mention
> trig stuff, but is it arctan, sine, or what?
>
> When it comes to math, there are many ways to optimize...
>
> Scott

--
http://www.piclist.com hint: To leave the PICList
mailto:piclist-unsubscribe-request@mitvma.mit.edu