Problem regarding sum of digits of floating point number in C

[Shreyas Kulkarni]

hi there,

recently i have got a problem regarding calculation of sum of digits in
a floating point or precision number. the weird behaviour of
compiler/language is preventing me from calculating the sum of all
digits. the compiler doesnt store the number as given by the user. some
of the precision digits are lost or changed (at will!). so by any
method, i m unable to reliably calculate the sum of provided number;
except i input the number as a string!

i have tried 2 types of logic -

first i use the intuitive logic of multiplying by 10 and adding the
integer part to the 'sum' in each iteration. all goes well until final
precision digit where this final digit is expanded. e.g. if 0.7 is
remaining, it is expanded to 0.69999999998! why dear, why r u doing
those things that we don want u to do?


/***************************************************/
width=strlen(ltoa((int)a,0,10));
sprintf(str,"%*.*f",width,MAXLEN-width,a);
do{
if(*s=='.') // ignore '.'
continue;
if(!isdigit(*s)) //we have got a char in num, so break away!
break;
else sum+=*s-0x30; //convert to digit from ASCII
}while(*++s);
/***************************************************/
in this case, the sprintf truncates or even expands(reverse of
rounding) long precision numbers.

even the good old gcc is truncating the precision digits at will.

this problem proved to be harder than what i thought initially.

any suggestions?

btw, sorry for this long post, but i m hurt u know? ;-)
TIA
Shreyas Kulkarni

 

[Gregory Toomey]

hi there,

recently i have got a problem regarding calculation of sum of digits in
a floating point or precision number. the weird behaviour of
compiler/language is preventing me from calculating the sum of all
digits. the compiler doesnt store the number as given by the user. some
of the precision digits are lost or changed (at will!). so by any
method, i m unable to reliably calculate the sum of provided number;
except i input the number as a string!

i have tried 2 types of logic -

first i use the intuitive logic of multiplying by 10 and adding the
integer part to the 'sum' in each iteration. all goes well until final
precision digit where this final digit is expanded. e.g. if 0.7 is
remaining, it is expanded to 0.69999999998! why dear, why r u doing
those things that we don want u to do?


/***************************************************/
width=strlen(ltoa((int)a,0,10));
sprintf(str,"%*.*f",width,MAXLEN-width,a);
do{
if(*s=='.') // ignore '.'
continue;
if(!isdigit(*s)) //we have got a char in num, so break away!
break;
else sum+=*s-0x30; //convert to digit from ASCII
}while(*++s);
/***************************************************/
in this case, the sprintf truncates or even expands(reverse of
rounding) long precision numbers.

even the good old gcc is truncating the precision digits at will.

this problem proved to be harder than what i thought initially.

any suggestions?

btw, sorry for this long post, but i m hurt u know? ;-)
TIA
Shreyas Kulkarni

You need a course in computer arithmetic!

Floats are generally represented using the IEEE standard #754.
http://en.wikipedia.org/wiki/IEEE_floating-point_standard
as 32 bits, 64 bits. Bog standard Intel x86 CPUs do all their floating point
arithmetic using the IEEE representation in hardware.

But this is base 2, and when you convert to/from base 10 rounding errors
occur and you get the errors you see.

Single precision (float in C) has about 6 to 7 significant digits; double
precision (double in C) has about 15 significant digits.

You can try
- using double instead of float
- using double, and rounding to say 10 significant digits.

gtoomey