libtakum

This ISO C99 library is a reference implementation of takum arithmetic, a new tapered precision machine number format.

Eight data types are defined, namely takum8, takum16, takum32, takum64 for the (floating-point) takums and takum_log8, takum_log16, takum_log32 and takum_log64 for the logarithmic takums, a logarithmic number system.

A wide range of functions is implemented, including a full set of trigonometric functions and a selection of constants. Please refer to libtakum(7) for an overview.

Using this package one is able to evaluate takums for real-world applications in terms of precision. Given it is a software implementation the performance is overall worse than the respective usual IEEE 754 floating-point hardware implementations, but it is sufficient to evaluate the formats for reasonably sized numerical benchmarks.

Getting started

You can automatically configure and install libtakum (including manuals) via

./configure
make -j install

The -j flag is optional and enables the parallel processing of the make rules, yielding a significant speedup. You can also specify it as -j N (where N is, for example, the number of CPU threads) for more fine-grained control.

The suite of unit tests is run with

make -j test

The manual pages are accessed via man(1). Every function is fully documented.

man libtakum
man takum16_from_float64
man takum_log32_csch

The library also comes with examples. One example is a full-fledged multi-precision RPN calculator which is run via

make -j example
./example/calculator

Example

As follows a simple example is given:

#include <stdio.h>
#include <takum.h>

int
main(void)
{
	takum16 a, b;

	a = takum16_from_float64(4.5);
	b = takum16_from_float64(6.5);

	printf("4.5 + 6.5 = %f\n",
	       takum16_to_float64(takum16_addition(a, b)));

	return 0;
}

The example is compiled and run with

cc -o example example.c -ltakum
./example

Author and License

libtakum is developed by Laslo Hunhold and licensed under the ISC license. See LICENSE for copyright and license details.