boost/multiprecision
2
0
Fork 0
mirror of https://github.com/boostorg/multiprecision.git synced 2026-01-19 04:22:11 +00:00
Code Issues Projects Releases Wiki Activity

Fix a typo and get even more cover dec-float

Browse Source
This commit is contained in:
ckormanyos
2025-07-21 15:00:23 +02:00
parent 01f516eed3
commit 76224e2466
2 changed files with 20 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/exercise_threading_log_agm.cpp
View File

@@ -71,7 +71,7 @@
//#define BOOST_MP_EXERCISE_THREADING_BACKEND_TYPE BOOST_MP_EXERCISE_THREADING_BACKEND_MPFR_FLOAT
#endif
constexpr unsigned local_mp_digits { 101U };
constexpr unsigned local_mp_digits { 301U };
#if (BOOST_MP_EXERCISE_THREADING_BACKEND_TYPE == BOOST_MP_EXERCISE_THREADING_BACKEND_CPP_DEC_FLOAT)
#include <boost/multiprecision/cpp_dec_float.hpp>

27
include/boost/multiprecision/cpp_dec_float.hpp
View File

@@ -1246,7 +1246,11 @@ cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, Expone
exponent_type ne;
x.extract_parts(dd, ne);
// Do the inverse estimate using double precision estimates of mantissa and exponent.
// Setup the iteration.
// Calculate the initial estimate of the inverse using double
// precision approximations of mantissa and exponent.
operator=(cpp_dec_float<Digits10, ExponentType, Allocator>(1.0 / dd, -ne));
// Compute the inverse of *this. Quadratically convergent Newton-Raphson iteration
@@ -1316,7 +1320,10 @@ cpp_dec_float<Digits10, ExponentType, Allocator>& cpp_dec_float<Digits10, Expone
}
// Setup the iteration.
// Estimate the square root using simple manipulations.
// Calculate the initial estimate of the root using double
// precision approximations of mantissa and exponent.
const double sqd = std::sqrt(dd);
*this = cpp_dec_float<Digits10, ExponentType, Allocator>(sqd, static_cast<ExponentType>(ne / static_cast<ExponentType>(2)));
@@ -1535,7 +1542,8 @@ void cpp_dec_float<Digits10, ExponentType, Allocator>::extract_parts(double& man
mantissa /= p10;
// Extract the rest of the mantissa piecewise from the limbs.
// This loop does not round.
// This loop does not round. For finite values, the absolute
// value of the mantissa is scaled and between 1 and 10.
auto itr_data = data.cbegin() + static_cast<std::size_t>(UINT8_C(1));
@@ -2423,11 +2431,14 @@ cpp_dec_float<Digits10, ExponentType, Allocator>::cpp_dec_float(const double man
{
exponent_type exp10_val { exp10 };
while (d_mant > 10.0)
{
d_mant /= 10.0;
++exp10_val;
}
// This subroutine is used after a call to extract_parts().
// The corresponding scaling loop for (d_mant > 10.0) is not
// needed here since extract_parts() returns a mantissa value
// between 1 and 10. The loop for (d_mant < 1.0) is needed
// since the initial guess of the square root in tghe subroutine
// calculate_root() might have been divided by 10 in order
// to obtain an exponent value that is a multiple of 2.
while (d_mant < 1.0)
{
d_mant *= 10.0;