Troubles with IEEE 754 floating point numbers

[ms292606]

Hi,

I am currently working on a program to decode data that is collected on a gps receiver and i've ran into a problem. Some of the data is encoded in IEEE 754. I wrote the following functions to decode the the binary:

/**************************************************************************************************
*Converts binary into IEEE 754 single percision
***************************************************************************************************/
float bin2float(unsigned long binary){
      float fp;
      unsigned int s=0,e=0,f=0;
      s=binary>>31;/*sign bit*/
      e=(binary>>23&0xff);/*exponent*/
      f=binary&0x7fffff;/*fraction*/
      fp=(sign_bit(s)<<(e-127))*(1+f*(pow(2,-23)));

      return(fp);
}
/**************************************************************************************************
*Converts binary into IEEE 754 double percision
***************************************************************************************************/
double bin2double(int binary1, int binary2){//binary1=bits 0-31 binary2=bits 32-63
      double fp;
      unsigned int s=0,e=0,f=0;
      s=binary2>>31;/*sign bit*/
      e=(binary2>>20&0x7ff);/*exponent*/
      f=(binary1*pow(2,-51)+((binary2&0x7ffff)*pow(2,-21)));/*fraction*/
      fp=(sign_bit(s)*pow(2,(e-1023)))*(1+f);

      return(fp);

}

these return the correct values but I do not have enough processing power available to use the pow function. Any ideas on a way to decrease cycles? Is there a predefined way in C to convert the binary to IEEE 754? Any help would be appreciated I don't work with C too often. Oh almost forgot, I am running this on a linux platform. Thanks
-Matt

[weaknessforcats]

Is there some reason you can't just:

[code=c]
unsigned long binary = 1234;
float f = binary;
double d = binary;
[/code]

??

[AdrianH]

Is there some reason you can't just:

[code=c]
unsigned long binary = 1234;
float f = binary;
double d = binary;
[/code]

??

True. But the data is coming from somewhere else, so a cast is really what is needed.

[code=c]
void * address; // contains the address of the double
double value = (double*)addres s;
[/code]


Adrian

[AdrianH]

Since all of your pow() calls are using a base of 2 use the >> or << operators instead. But your algorithm is probably slower then a cast.


Adrian

[ms292606]

thanks for the help guys, I got it working fast enough. My problem was i was using a 4 byte int. once i changed it to a long long that could store 8 bytes of data then i was able to do everything by using bit shifting.

[AdrianH]

thanks for the help guys, I got it working fast enough. My problem was i was using a 4 byte int. once i changed it to a long long that could store 8 bytes of data then i was able to do everything by using bit shifting.

But why don't you just let the compiler/processor interpret the floating point number? Why do the manual translation?


Adrian