libfixmath
| Developer(s) | Ben Brewer (aka flatmush) |
|---|---|
| Stable release |
r64
/ February 2, 2012 |
| Repository |
code |
| Development status | active |
| Written in | C99 |
| Operating system | Cross-platform |
| Type | fixed point math library |
| License | MIT |
| Website | https://code.google.com/p/libfixmath |
libfixmath is a platform-independent fixed point maths library aimed at developers wanting to perform fast non-integer maths on platforms lacking a (or with a low performance) FPU. It offers developers a similar interface to the standard math.h functions for use on Q16.16 fixed point numbers. libfixmath has no external dependencies other than stdint.h and a compiler which supports 64-bit integer arithmetic (such as GCC).[1] Conditional compilation options exist to remove the requirement for a 64-bit capable compiler as many compilers for microcontrollers and DSPs do not support 64-bit arithmetic.[2]
History
libfixmath was originally developed by Ben Brewer (aka flatmush) and first released publicly as part of the Dingoo SDK.[3] It has since been used to implement a software 3D graphics library called FGL.[4]
Q16.16 Functions
| Name | Description |
|---|---|
fix16_acos | inverse cosine |
fix16_asin | inverse sine |
fix16_atan | one-parameter inverse tangent |
fix16_atan2 | two-parameter inverse tangent |
fix16_cos | cosine |
fix16_exp | exponential function |
fix16_sin | sine |
fix16_sqrt | square root |
fix16_tan | tangent |
fix16_mul | multiplication |
fix16_div | division |
fix16_sadd | saturated addition |
fix16_smul | saturated multiplication |
fix16_sdiv | saturated division |
Other Functions
| Name | Description |
|---|---|
fix16_to_dbl | Convert Q16.16 to a double |
fix16_to_float | Convert Q16.16 to a float |
fix16_to_int | Convert Q16.16 to an integer |
fix16_from_dbl | Convert double to a Q16.16 |
fix16_from_float | Convert float to a Q16.16 |
fix16_from_int | Convert integer to a Q16.16 |
Performance
For the most intensive function (atan2) benchmark results show the following results:
| Name | Time Compared to Float |
|---|---|
| ARM Cortex-M0 | 26.3% |
| Marvell PXA270 (ARM) @ 312 MHz | 58.45% |
| Intel T5500 | 120% |
| Intel Atom N280 | 141% |
Note: These results were calculated using fixtest with caching optimizations turned off.[5]
Licensing
libfixmath is released under the MIT License, a permissive free software licence, and is free software.