boost/multiprecision
2
0
Fork 0
mirror of https://github.com/boostorg/multiprecision.git synced 2026-02-20 02:42:26 +00:00
Code Issues Projects Releases Wiki Activity

Fix rounding bug in cpp_bin_float - in the case that the remainder is zero, we may still have a tie if the extra bit in the quotient is one.

Browse Source
This commit is contained in:
jzmaddock
2014-06-29 10:34:45 +01:00
parent 6ae735d8d5
commit 3b9399c521
1 changed files with 4 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
include/boost/multiprecision/cpp_bin_float.hpp
View File

@@ -855,11 +855,12 @@ inline void eval_divide(cpp_bin_float<Digits, DigitBase, Allocator, Exponent, Mi
//
// OK we have cpp_bin_float<Digits, DigitBase, Allocator, Exponent, MinE, MaxE>::bit_count+1 bits,
// so we already have rounding info,
// we just need to changes things if the last bit is 1 and the
// remainder is non-zero (ie we do not have a tie).
// we just need to changes things if the last bit is 1 and either the
// remainder is non-zero (ie we do not have a tie) or the quotient would
// be odd if it were shifted to the correct number of bits (ie a tiebreak).
//
BOOST_ASSERT((eval_msb(q) == cpp_bin_float<Digits, DigitBase, Allocator, Exponent, MinE, MaxE>::bit_count));
if((q.limbs()[0] & 1u) && eval_get_sign(r))
if((q.limbs()[0] & 1u) && (eval_get_sign(r) || (q.limbs()[0] & 2u)))
{
eval_increment(q);
}