How accurate is omap at cos functions, C - Math.h

Hi everyone,

I was wondering how good the OMAP3 is at floating point calculations?

I've written a little program to calculate the cos() of a value, but
it is far from what a calculator computes.

#include <stdio.h>
#include <math.h>

int main (void)
{
  double a;
        a = 111.109*cos(14.892116);
  printf("\nA = %f\n", a);
}

The answer should be 107.377701.

the printf displays: -76.137617.

Does anyone have any ideas about this?

Hurdy <robert@roberthurd.com> writes:

Hi everyone,

I was wondering how good the OMAP3 is at floating point calculations?

It's not very fast.

I've written a little program to calculate the cos() of a value, but
it is far from what a calculator computes.

#include <stdio.h>
#include <math.h>

int main (void)
{
  double a;
        a = 111.109*cos(14.892116);
  printf("\nA = %f\n", a);
}

The answer should be 107.377701.

the printf displays: -76.137617.

The argument to cos() should be in radians, not degrees.

thank you mans for your response. I've just realised what the problem
is.

My calculator is set to give the result in degrees, whereas the maths
library for cos returns a value in radians.

I like the CORDIC (COordinate Rotation DIgital Computer) algorithm for
computing sin, cos, and arctan. It just uses shifts and adds, with
each iteration improving accuracy by about a bit.

http://en.wikipedia.org/wiki/Cordic

For representing angles in a computer, I like Binary Angular
Measurement (BAM). Wikipedia describes the 8-bit version of this in
http://en.wikipedia.org/wiki/Angle, where you split the circle into
256 angles calling each or them a "binary degree" or "binary radian".
Personally I like to call them "octal grads", since there are 400
(octal) of them in a circle.

BAM can get you any resolution. For example, with a 32-bit integer
you get a resolution of one 4 billionth of a circle or about one
billionth of a radian. BAMs are fixed-point numbers, so you don't
need those awful floats and doubles. Calculating modulo 360 degrees
is really easy: you just mask off the high bits. If you use an entire
word, you don't need to worry about whether you're using signed or
unsigned numbers. With signed numbers you get 0 to just less than
360. With signed numbers you get -180 to just less than +180.
Totally equivalent, and uses every bit combination. When you add and
subtract full-word angles you do it as normal integer arithmetic. The
overflows and underflows do the modulo 360 for you.

Why they're so rare in computing is something I don't understand. I
guess it's because the numerical side of computing developed in math
departments and they love their radians. One of the most common uses
of BAMs is for optical shaft encoders which use a disk with black and
white arcs to show position using a linear array of optical sensors.
In this case they use a Gray code version of BAM.