tags:

views:

39

answers:

1
+1  Q: 

Practices to limit floating point accuracy problems

As programmers, most (if not all of us) know that floating point numbers have a tendency to not be very accurate. I know that this problem can't be avoided entirely, but I am wondering if there are any certain practices, patterns, etc. that can be used to at least reduce floating point errors.

Thanks in advance.

+2  A: 

Use fixed-point mathematics where you can deal with a known limited precision.

As an example, the Rockbox music player firmware uses almost entirely fixed-point media codecs.

If you must be perfectly accurate, use an infinite-length storage type like those provided by the GMP library.

If you're just trying to cut down on your errors, try to work as close to zero as possible, where the IEEE FP numbers are more precise. Reorder your operations to avoid letting your absolute values get too large.

Borealid  2010-08-20 03:53:37
I imagine that Rockbox uses fixed-point codecs by preference because some of its targets don't have FPUs so performance would be inadequate. I don't think accuracy is a major concern - if it were, we'd be using Tremor on desktops instead of libvorbis.
Peter  2010-08-20 04:25:53
Example: summing a vector, sort by absolute value first. This way, you don't suffer from loss of precision as the size of your sum grows. However, this adds some overhead, so if speed is more important than precision, don't even worry about it.
Ian  2010-08-28 06:40:02

related questions

Why do I see a double variable initialized to some value like 21.4 as 21.399999618530273?
ParseFloat function in JavaScript
Haskell FFI / C MPFR library wrapper woes
When to use Fixed Point these days
Confusing return statement
How to alter a float by its smallest increment (or close to it)?
Significant figures in the decimal module
Test if a floating point number is an integer
SQL server 2005 numeric precision loss
Why would I use 2's complement to compare two doubles instead of comparing their differences against an epsilon value?
Double and floats in C#
Converting floating point exceptions into C++ exceptions
How to manually parse a floating point number from a string
What is the fastest way to convert float to int on x86
Determine if a string is an integer or a float in ANSI C
cout prints "-0" instead of "0"
how to use "%f" to populate a double value into a string with the right precision
Convert from scientific notation string to float in C#
How is floating point stored? When does it matter?
round() in Python doesn't seem to be rounding properly
How do you generate a random number in C#?
Comparing IEEE floats and doubles for equality
Most effective way for float and double comparison
Convert Bytes to Floating Point Numbers in Python
A little diversion into floating point (im)precision, part 1