Skip to content
Fixed-Point Library for C++
Branch: master
Clone or download
gsarkis Merge pull request #9 from gsarkis/nearodd
Implement NearOdd rounding
Latest commit 284381f Feb 25, 2016
Type Name Latest commit message Commit time
Failed to load latest commit information.
examples rename example file Sep 16, 2013
include/fi Implement NearOdd rounding Feb 25, 2016
tests Implement NearOdd rounding Feb 25, 2016
tools add dependency on Boost Sep 16, 2013
.gitignore Add basic unit tests for min() and max() Sep 27, 2013
CMakeLists.txt add install make target Sep 16, 2013
LICENSE initial public release Jul 2, 2013

LibFi - A fixed-point arithmetic library.

What is LibFi?

LibFi is a binary fixed-point arithmetic library for C++, focusing on flexibility and correctness. It is intended to be used for modeling hardware systems. LibFi is template-based and is a header-only library.

Unlike most other C++ fixed-point libraries, the word length in Fi does not have to be a power of two, and the overflow behavior and rounding mode are customizable.

This is a header-only library so you include ./fi/Fixed.hpp, and any applicable overflow handlers and rounding modes from ./fi/overflow and ./fi/rounding.

Here is an example of multiplying 8-bit, signed fixed-point numbers with 4 fractional bits:

#include "fi/Fixed.hpp"
#include "fi/overflow/Wrap.hpp"

#include <iostream>

int main(int argc, char* argv[]) {

  Fi::Fixed<8, 4, Fi::SIGNED> a("3.14");
  Fi::Fixed<8, 4, Fi::SIGNED> b("1.14");

  std::cout << a*b << std::endl;

  return 0;



LibFi is documented using Doxygen. To build the documentation, you need Doxygen. In the library's root directory, run doxygen. The documentation will be available in doc/html/index.html.

Quick Reference

The Fi::Fixed<TOTAL_WIDTH, FRACTION_WIDTH, SIGNEDNESS, OVERFLOW, ROUNDING> datatype accepts five parameters:

Parameter Description Required
TOTAL_WIDTH Total number of bits in binary representation, including the sign for signed types. Yes
FRACTION_WIDTH Number of fractional bits. Yes
SIGNEDNESS Fi::SIGNED for signed values; Fi::UNSIGNED for unsigned values. Yes
OVERFLOW Behavior when a number overflows the range representable using the selected quantization parameters. See below for valid options. No (Default: Fi::Throw)
ROUNDING Behavior when a number is not representable using the selected quantization parameters. See below for valid options. No (Default: Fi::Fix)


Mode Description
Fi::Classic Round to nearest representable value. Ties are rounded away from zero.
Fi::Ceil Round towards positive infinity.
Fi::Floor Round towards negative infinity.
Fi::Fix Round towards zero.
Fi::NearEven Round to nearest representable value. Ties are rounded towards an LSB of 0.
Fi::NearOdd Round to nearest representable value. Ties are rounded towards an LSB of 1.


Mode Description
Fi::Saturate Saturate value to maximum or minimum value allowed by selected quantization parameters.
Fi::Wrap Wrap the value around when overflow occurs.
Fi::Throw When overflow occurs, throw a Fi::PositiveOverflow or a Fi::NegativeOverflow exception, depending on the direction of the overflow.
Fi::Undefined The behavior of overflow is undefined. Use when execution speed is more important than results.


Nightly test build results are available on the following page.

Supported Platforms

Tests are run on the following platforms:

  • 64-bit Linux using GCC.
  • 64-bit Linux using Clang.

The master branch always contains code which passes all the tests on the supported platforms.

External References


  • 2013-07-02 v1.0: First public release of LibFi.
You can’t perform that action at this time.