Boost C++ Libraries of the most highly regarded and expertly designed C++ library projects in the world. Herb Sutter and Andrei Alexandrescu, C++ Coding Standards


Using Boost.Multiprecision

All new projects are recommended to use Boost.Multiprecision.

Using Boost.Multiprecision to generate a high-precision array of sine coefficients for use with FFT.

The Boost.Multiprecision library can be used for computations requiring precision exceeding that of standard built-in types such as float, double and long double. For extended-precision calculations, Boost.Multiprecision supplies a template data type called cpp_bin_float. The number of decimal digits of precision is fixed at compile-time via a template parameter.

One often needs to compute tables of numbers in mathematical software. To avoid the Table-maker's dilemma it is necessary to use a higher precision type to compute the table values so that they have the nearest representable bit-pattern for the type, say double, of the table value.

This example is a program fft_since_table.cpp that writes a header file sines.hpp containing an array of sine coefficients for use with a Fast Fourier Transform (FFT), that can be included by the FFT program.

To use Boost.Multiprecision's high-precision floating-point types and constants, we need some includes:

#include <boost/math/constants/constants.hpp>
// using boost::math::constants::pi;

#include <boost/multiprecision/cpp_bin_float.hpp> // for
// using boost::multiprecision::cpp_bin_float and
// using boost::multiprecision::cpp_bin_float_50;
// using boost::multiprecision::cpp_bin_float_quad;

#include <boost/array.hpp> // or <array> for std::array

#include <iostream>
#include <limits>
#include <vector>
#include <algorithm>
#include <iomanip>
#include <iterator>
#include <fstream>

First, this example defines a prolog text string which is a C++ comment with the program licence, copyright etc. (You would of course, tailor this to your needs, including your copyright claim). This will appear at the top of the written header file sines.hpp.

using boost::multiprecision::cpp_bin_float_50;
using boost::math::constants::pi;

A fast Fourier transform (FFT), for example, may use a table of the values of sin((π/2n) in its implementation details. In order to maximize the precision in the FFT implementation, the precision of the tabulated trigonometric values should exceed that of the built-in floating-point type used in the FFT.

The sample below computes a table of the values of sin(π/2n) in the range 1 <= n <= 31.

This program makes use of, among other program elements, the data type boost::multiprecision::cpp_bin_float_50 for a precision of 50 decimal digits from Boost.Multiprecision, the value of constant π retrieved from Boost.Math, guaranteed to be initialized with the very last bit of precision for the type, here cpp_bin_float_50, and a C++11 lambda function combined with std::for_each().

define the number of values (32) in the array of sines.

std::size_t size = 32U;
//cpp_bin_float_50 p = pi<cpp_bin_float_50>();
cpp_bin_float_50 p = boost::math::constants::pi<cpp_bin_float_50>();

std::vector <cpp_bin_float_50> sin_values (size);
unsigned n = 1U;
// Generate the sine values.
  sin_values.begin (),
  sin_values.end (),
  [&n](cpp_bin_float_50& y)
    y = sin( pi<cpp_bin_float_50>() / pow(cpp_bin_float_50 (2), n));

Define the floating-point type for the generated file, either built-in double, float, or long double, or a user defined type like cpp_bin_float_50.

std::string fp_type = "double";

std::cout << "Generating an `std::array` or `std::array` for floating-point type: "
  << fp_type << ". " << std::endl;

By default, output would only show the standard 6 decimal digits, so set precision to show enough significant digits for the chosen floating-point type. For cpp_bin_float_50 is 50. (50 decimal digits should be ample for most applications).

std::streamsize precision = std::numeric_limits<cpp_bin_float_50>::digits10;

std::cout << "Sines table precision is " << precision << " decimal digits. " << std::endl;

Of course, one could also choose a lower precision for the table values, for example,

std::streamsize precision = std::numeric_limits<cpp_bin_float_quad>::max_digits10;

128-bit 'quad' precision of 36 decimal digits would be sufficient for the most precise current long double implementations using 128-bit. In general, it should be a couple of decimal digits more (guard digits) than std::numeric_limits<RealType>::max_digits10 for the target system floating-point type. (If the implementation does not provide max_digits10, the the Kahan formula std::numeric_limits<RealType>::digits * 3010/10000 + 2 can be used instead).

The compiler will read these values as decimal digits strings and use the nearest representation for the floating-point type.

Now output all the sine table, to a file of your chosen name.

const char sines_name[] = "sines.hpp";  // Assuming in same directory as .exe

std::ofstream fout(sines_name, std::ios_base::out);  // Creates if no file exists,
// & uses default overwrite/ ios::replace.
if (fout.is_open() == false)
{  // failed to open OK!
  std::cout << "Open file " << sines_name << " failed!" << std::endl;
  return EXIT_FAILURE;
{ // Write prolog etc as a C++ comment.
  std::cout << "Open file " << sines_name << " for output OK." << std::endl;
  fout << prolog
  << "// Table of " << sin_values.size() << " values with "
    << precision << " decimal digits precision,\n"
    "// generated by program fft_sines_table.cpp.\n" << std::endl;

fout << "#include <array> // std::array" << std::endl;

// Write the table of sines as a C++ array.
  fout <<  "\nstatic const std::array<double, " << size << "> sines =\n"
  "{{\n"; // 2nd { needed for some old GCC compiler versions.

  for (unsigned int i = 0U; ;)
    fout << "  " << sin_values[i];
    if (i == sin_values.size()-1)
    { // next is last value.
      fout << "\n}};  // array sines\n"; // 2nd } needed for some old GCC compiler versions.
      fout << ",\n";
  } // for

  std::cout << "Closed file " << sines_name << " for output." << std::endl;

The output file generated can be seen at ../../example/sines.hpp

The table output is:

The printed table is:


The output can be copied as text and readily integrated into a given source code. Alternatively, the output can be written to a text or even be used within a self-written automatic code generator as this example.

A computer algebra system can be used to verify the results obtained from Boost.Math and Boost.Multiprecision. For example, the Wolfram Mathematica computer algebra system can obtain a similar table with the command:

Table[N[Sin[Pi / (2^n)], 50], {n, 1, 31, 1}]

The Wolfram Alpha computational knowledge engine can also be used to generate this table. The same command can be pasted into the compute box.

The full source of this example is at fft_sines_table.cpp