diff options
author | Josh Blum <josh@joshknows.com> | 2010-08-12 00:47:28 -0700 |
---|---|---|
committer | Josh Blum <josh@joshknows.com> | 2010-08-12 00:47:28 -0700 |
commit | ef6953024f1075a729e85f2511c75de337879888 (patch) | |
tree | 8c99087fc21372414d34a4f30905792089111e1b /host/lib/utils | |
parent | 836c6bcda0065ad5345e1b7545feefab5a81e566 (diff) | |
download | uhd-ef6953024f1075a729e85f2511c75de337879888.tar.gz uhd-ef6953024f1075a729e85f2511c75de337879888.tar.bz2 uhd-ef6953024f1075a729e85f2511c75de337879888.zip |
uhd: created floor_step to handle floating-point errors in gain group
Diffstat (limited to 'host/lib/utils')
-rw-r--r-- | host/lib/utils/gain_group.cpp | 31 |
1 files changed, 25 insertions, 6 deletions
diff --git a/host/lib/utils/gain_group.cpp b/host/lib/utils/gain_group.cpp index c113719c8..078fe56b2 100644 --- a/host/lib/utils/gain_group.cpp +++ b/host/lib/utils/gain_group.cpp @@ -35,6 +35,25 @@ static bool compare_by_step_size( return fcns.at(rhs).get_range().step > fcns.at(lhs).get_range().step; } +/*! + * Get a multiple of step with the following relation: + * result = step*floor(num/step) + * + * Due to small floating-point inaccuracies: + * num = n*step + e, where e is a small inaccuracy. + * When e is negative, floor would yeild (n-1)*step, + * despite that n*step is really the desired result. + * This function is designed to mitigate that issue. + * + * \param num the number to approximate + * \param step the step size to round with + * \param e the small inaccuracy to account for + * \return a multiple of step approximating num + */ +template <typename T> static T floor_step(T num, T step, T e = T(0.001)){ + return step*int(num/step + e); +} + /*********************************************************************** * gain group implementation **********************************************************************/ @@ -82,9 +101,9 @@ public: float gain_left_to_distribute = gain; BOOST_FOREACH(const gain_fcns_t &fcns, all_fcns){ const gain_range_t range = fcns.get_range(); - gain_bucket.push_back( - max_step*int(std::clip(gain_left_to_distribute, range.min, range.max)/max_step) - ); + gain_bucket.push_back(floor_step(std::clip( + gain_left_to_distribute, range.min, range.max + ), max_step)); gain_left_to_distribute -= gain_bucket.back(); } @@ -106,9 +125,9 @@ public: //fill in the largest step sizes first that are less than the remainder BOOST_FOREACH(size_t i, indexes_step_size_dec){ const gain_range_t range = all_fcns.at(i).get_range(); - float additional_gain = range.step*int( - std::clip(gain_bucket.at(i) + gain_left_to_distribute, range.min, range.max - )/range.step) - gain_bucket.at(i); + float additional_gain = floor_step(std::clip( + gain_bucket.at(i) + gain_left_to_distribute, range.min, range.max + ), range.step) - gain_bucket.at(i); gain_bucket.at(i) += additional_gain; gain_left_to_distribute -= additional_gain; } |