Re: [avr-gcc-list] floating point calculations - How many cycles

[Bjarne Laursen]

intiha Ho gai wrote:

> currTime = (uint32_t)(skew_*(double)currTime + offset_);//compensation
> over here!!
>
> here the currTime and offset are 32 bit integers, while the skew is a
> floating point with value very close to zero (0.999987 therof).
>  
Keep in mind that 'double' has less precision that uint32_t. If you need 
the full range for currTime you will run in to problems.
Consitter doing something like this


uint32 currTime;
int32 offset;
int16 gain;
currTime =  (((int64_t)currTime*gain)>>16) + offset;

The shift determines the resolution for gain,.
(Yes, It will take a long time)

-Bjarne Laursen, RoseTechnology A/S