How to control double precision computing to avoid rounding errors in C++ linux x86_64?

Question

In a C++ code on linux x86_64, I need to double precision computing (+ or -).

26.100000000000001 - 26 + 0.10000000000000001

I got:

0.20000000000000143

I want to get 0.2.

here, the display format is not import, the computing results will be used for some if-else branch conditions. So, I only want the if-else conditions compare the 4 digits after the decimal digit.

It seems a rounding error ?

How to restrict the computing precision to 4 digits after decimal point ?

I do not want to call any functions in order to avoid overhead.

I do not want to use stringstream due to transformation overhead.

Any better ideas ?

thanks

Mike DeSimone · Accepted Answer · 2011-12-25 19:34:04Z

8

The computing precision is fine. It's the display precision you're trying to control. You can do that with setprecision() if using <iostream> or a formatter like "%.4f" if using <stdio.h>.

You are already calling functions since you are displaying the result as decimal!

P.S. 0.1 cannot be exactly represented by a float, double, or any binary-mantissa format. It factors into (1/2) * (1/5), and decimal 1/5 is an infinitely-repeating digit sequence in binary.

P.P.S. Go look at GMP. It might be your best hope.

edited Dec 25, 2011 at 19:34

answered Dec 25, 2011 at 18:48

Mike DeSimone

43k10 gold badges79 silver badges97 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

user1002288 Over a year ago

here, the display format is not import, the computing results will be used for some if-else branch conditions. So, I only want the if-else conditions compare the 4 digits after the decimal digit.

Mysticial Over a year ago

@user1002288 What you're trying to do is not possible with binary floating-point. No amount of rounding hacks will make decimal arithmetic on binary floating-point work like magic..

aland Over a year ago

@user1002288 You can just add some magic epsilon value to all you floating-precision comparisons. Like if (a==b) {} transforms into if (abs(a-b) < 1e-4) {}. This is the best thing you can do using floating-point arithmetics.

Neo · Accepted Answer · 2011-12-25 18:47:38Z

2

If you just want to print it it, you can use printf("%10.4lf"). You can alter the precision to anything you want, of course.

answered Dec 25, 2011 at 18:47

Neo

1,5542 gold badges15 silver badges28 bronze badges

2 Comments

user1002288 Over a year ago

here, the display format is not import, the computing results will be used for some if-else branch conditions. So, I only want the if-else conditions compare the 4 digits after the decimal digit.

Neo Over a year ago

oh you mean display format is not importANT! ok.. you can try multiplying with 10^(precision) [precision=4 in your case].

Ted Hopp · Accepted Answer · 2011-12-26 01:40:49Z

1

If you are only interested in equality up to four decimal places, multiply everything by 10,000 and use integer arithmetic. (You might need to round after multiplying by 10,000.)

answered Dec 26, 2011 at 1:40

Ted Hopp

235k48 gold badges412 silver badges533 bronze badges

Collectives™ on Stack Overflow

How to control double precision computing to avoid rounding errors in C++ linux x86_64?

3 Answers 3

3 Comments

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

3 Comments

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related