Loss precision on floating point calculation

Question

I am using RCM6760 with Dynamic C 10.72B. I have created the following test code:

void main()
{
float tempOnTime;
float tempPeriod;
float tempOffTime;
float tempDutyCycle;

tempOnTime = 0.001;
tempOffTime = 0.0;
tempPeriod = tempOnTime + tempOffTime;
tempDutyCycle = tempOnTime / tempPeriod;
tempDutyCycle = tempDutyCycle * 100;
printf("%.6f\n", tempDutyCycle);
}

The printf() gave 99.999992 which I expect to have the value of 100. The Watches show the tempDutyCycle is "float 100". Can you suggest a way to avoid such precision lost?

Thanks,

Answer 1

How much precision do you require? 99.999992 is only 0.000008% off from 100.

If it isn't precise enough, you'll need to switch to fixed-point math where you use integral units. For example, if 0.001 is seconds, switch to doing your math with milliseconds or microseconds and integer math.

In general, avoid float unless absolutely necessary for your algorithm. And with integer math, do all of your multiplication first (ensuring that you won't overflow a 32-bit value) and then do division (which will lose any fractional results).

Code:

tempOnTime = 1;
tempOffTime = 0
tempPeriod = tempOnTime + tempOffTime;
tempDutyCycle = (100 * tempOnTime) / tempPeriod;