Skip to content

Improve performance and rounding of 128-bit mul #1310

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mborland merged 5 commits into from
Jan 22, 2026
Merged

Improve performance and rounding of 128-bit mul #1310

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
ad17c31
Add reproducer rounding error
mborland Jan 21, 2026
ec90aa6
Increase number of digits saved for rounding in constructor
mborland Jan 21, 2026
deaeac9
Round directly instead of chopping digits
mborland Jan 21, 2026
ea75802
Fix incomplete type compile failure on older compilers
mborland Jan 22, 2026
ad4173e
Fix type conflict on embedded
mborland Jan 22, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions include/boost/decimal/detail/fenv_rounding.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -202,7 +202,7 @@ constexpr auto coefficient_rounding(T1& coeff, T2& exp, T3& biased_exp, const bo
using demoted_integer_type = std::conditional_t<std::numeric_limits<T1>::digits10 < std::numeric_limits<sig_type>::digits10, T1, sig_type>;

// How many digits need to be shifted?
const auto shift_for_large_coeff {(coeff_digits - detail::precision_v<TargetDecimalType>) - 1};
const int shift_for_large_coeff {(coeff_digits - detail::precision_v<TargetDecimalType>) - 1};
int shift {};
BOOST_DECIMAL_IF_CONSTEXPR (is_fast_type_v<TargetDecimalType>)
{
Expand All @@ -212,7 +212,7 @@ constexpr auto coefficient_rounding(T1& coeff, T2& exp, T3& biased_exp, const bo
}
else
{
const auto shift_for_small_exp {(-biased_exp) - 1};
const auto shift_for_small_exp {static_cast<int>((-biased_exp) - 1)};
shift = std::max(shift_for_small_exp, shift_for_large_coeff);
}

Expand Down
6 changes: 3 additions & 3 deletions include/boost/decimal/detail/mul_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -162,11 +162,11 @@ constexpr auto d128_mul_impl(const T1& lhs_sig, const U1 lhs_exp, const bool lhs

// 34 is the number of digits in the d128 significand
// this way we can skip rounding in the constructor a second time
const auto digit_delta {sig_dig - 34};
const auto digit_delta {sig_dig - std::numeric_limits<sig_type>::digits10};
if (BOOST_DECIMAL_LIKELY(digit_delta > 0))
{
res_sig /= pow10<sig_type>(digit_delta);
res_exp += digit_delta;
auto biased_exp {res_exp + detail::bias_v<ReturnType>};
detail::coefficient_rounding<ReturnType>(res_sig, res_exp, biased_exp, sign, sig_dig);
}

BOOST_DECIMAL_ASSERT((res_sig[3] | res_sig[2]) == 0U);
Expand Down
2 changes: 2 additions & 0 deletions test/github_issue_1026.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,8 @@ int main()
BOOST_TEST_EQ("9999999999999999999999999992345678.49"_DL, "9999999999999999999999999992345678"_DL);
BOOST_TEST_EQ("9999999999999999999999999992345678.50"_DL, "9999999999999999999999999992345678"_DL);
BOOST_TEST_EQ("9999999999999999999999999992345678.51"_DL, "9999999999999999999999999992345679"_DL);
BOOST_TEST_EQ("145433.2908011933696719165119928295655062562131932287426051970822"_DL, "145433.2908011933696719165119928296"_DL);
BOOST_TEST_EQ("30269.587755640502150977251770554"_DL * "4.8046009735990873395936309640543"_DL, "145433.2908011933696719165119928296"_DL);

BOOST_TEST_EQ(("0"_DF + "8.4e-96"_DF), "8.4e-96"_DF);
BOOST_TEST_EQ(("0"_DF + std::numeric_limits<decimal32_t>::denorm_min()), std::numeric_limits<decimal32_t>::denorm_min());
Expand Down
Loading