C - pow() strange behavior

Question

I am facing a strange issue with the below code,

 uint32 number = 1000000000;
 number = number * pow(10, -9);
 printf("number is %d\n", number);

I see the number getting printed as "0", while I expect the number to be "1". Can anyone point what is possibly causing this? This is on MPC8248 platform.

Thanks in advance.

mk

1) There is no problem compiling or linking. @MitchWheat can you pls elaborate on what could be wrong with the precision? — iammk, Jan 11 '14 at 02:16
There is no such floating point number as 10^-9, so the result of `pow` will necessarily be an approximation. It's unlikely that multiplying this approximation by 1000000000 will yield exactly 1. — R.. GitHub STOP HELPING ICE, Jan 11 '14 at 02:25
@R.. I get the expected result when using the same code (same libraries) on a ARM based platform. I also get '1' when using gcc (on SunOS). I am quite interested to know why it is different in MPC. — iammk, Jan 11 '14 at 02:53
Try `number = round(number * pow(10, -9));` to avoid fraction truncation and get round to nearest. — chux - Reinstate Monica, Jan 11 '14 at 03:30
[What Every Computer Scientist Should Know About Floating-Point Arithmetic](http://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html) — Adam Rosenfield, Jan 11 '14 at 04:00
@chux: Adding random rounding operations is rarely the right way to fix bugs in your floating point code. — R.. GitHub STOP HELPING ICE, Jan 11 '14 at 15:35

score 3 · Answer 1 · edited May 23 '17 at 12:03

pow returns a double, and with doubles you must always worry about precision. The product number * pow(...) may not return exactly 1. It could be 0.99 or something else that's almost 1 but not quite. When this value gets assigned back to the unsigned integer number, it would get demoted to become an int, and rounded down to 0.

To get around this, you can always implement your own pow function with integers. I'd recommend implementing with longs, though, because your integers can overflow fast (e.g. base = 50, exponent = 6 easily exceeds 32-bits typically allotted to ints).

See answers to this SO question for more details on the pow issue.

Wait, what's precision?

Many numbers cannot be expressed with a finite binary representation. The float type is one such finite binary representation. A float can only store a certain number of bits of information (which you can think of as precision), so any bits that don't fit are discarded. This loss of information is why floating point numbers can have errors like these. You can read more about this here.

score 0 · Answer 2 · answered Jan 30 '14 at 13:47

number = number * SomeFloat

integer multiplied by floating numbers ? Should give a floating point number. Result assigned in a integer variable ? Float gets implicitely converted to an integer ??? Danger !

Use the round function, at least, as suggested by comments.

Don't expect number artihmetics in your computer being the same that numbers in math as you were taught in school (at least, not exactly)

C - pow() strange behavior

2 Answers2