boost/move/algo/adaptive_merge.hpp
////////////////////////////////////////////////////////////////////////////// // // (C) Copyright Ion Gaztanaga 2015-2016. // Distributed under the Boost Software License, Version 1.0. // (See accompanying file LICENSE_1_0.txt or copy at // http://www.boost.org/LICENSE_1_0.txt) // // See http://www.boost.org/libs/move for documentation. // ////////////////////////////////////////////////////////////////////////////// #ifndef BOOST_MOVE_ADAPTIVE_MERGE_HPP #define BOOST_MOVE_ADAPTIVE_MERGE_HPP #include <boost/move/detail/config_begin.hpp> #include <boost/move/algo/detail/adaptive_sort_merge.hpp> #include <cassert> #if defined(BOOST_CLANG) || (defined(BOOST_GCC) && (BOOST_GCC >= 40600)) #pragma GCC diagnostic push #pragma GCC diagnostic ignored "-Wsign-conversion" #pragma GCC diagnostic ignored "-Wconversion" #endif namespace boost { namespace movelib { ///@cond namespace detail_adaptive { template<class RandIt, class Compare, class XBuf> inline void adaptive_merge_combine_blocks( RandIt first , typename iter_size<RandIt>::type len1 , typename iter_size<RandIt>::type len2 , typename iter_size<RandIt>::type collected , typename iter_size<RandIt>::type n_keys , typename iter_size<RandIt>::type l_block , bool use_internal_buf , bool xbuf_used , Compare comp , XBuf & xbuf ) { typedef typename iter_size<RandIt>::type size_type; size_type const len = size_type(len1+len2); size_type const l_combine = size_type(len-collected); size_type const l_combine1 = size_type(len1-collected); if(n_keys){ RandIt const first_data = first+collected; RandIt const keys = first; BOOST_MOVE_ADAPTIVE_SORT_PRINT_L2(" A combine: ", len); if(xbuf_used){ if(xbuf.size() < l_block){ xbuf.initialize_until(l_block, *first); } assert(xbuf.size() >= l_block); size_type n_block_a, n_block_b, l_irreg1, l_irreg2; combine_params( keys, comp, l_combine , l_combine1, l_block, xbuf , n_block_a, n_block_b, l_irreg1, l_irreg2); //Outputs op_merge_blocks_with_buf (keys, comp, first_data, l_block, l_irreg1, n_block_a, n_block_b, l_irreg2, comp, move_op(), xbuf.data()); BOOST_MOVE_ADAPTIVE_SORT_PRINT_L1(" A mrg xbf: ", len); } else{ size_type n_block_a, n_block_b, l_irreg1, l_irreg2; combine_params( keys, comp, l_combine , l_combine1, l_block, xbuf , n_block_a, n_block_b, l_irreg1, l_irreg2); //Outputs if(use_internal_buf){ op_merge_blocks_with_buf ( keys, comp, first_data, l_block, l_irreg1, n_block_a, n_block_b , l_irreg2, comp, swap_op(), first_data-l_block); BOOST_MOVE_ADAPTIVE_SORT_PRINT_L2(" A mrg buf: ", len); } else{ merge_blocks_bufferless (keys, comp, first_data, l_block, l_irreg1, n_block_a, n_block_b, l_irreg2, comp); BOOST_MOVE_ADAPTIVE_SORT_PRINT_L1(" A mrg nbf: ", len); } } } else{ xbuf.shrink_to_fit(l_block); if(xbuf.size() < l_block){ xbuf.initialize_until(l_block, *first); } size_type *const uint_keys = xbuf.template aligned_trailing<size_type>(l_block); size_type n_block_a, n_block_b, l_irreg1, l_irreg2; combine_params( uint_keys, less(), l_combine , l_combine1, l_block, xbuf , n_block_a, n_block_b, l_irreg1, l_irreg2, true); //Outputs BOOST_MOVE_ADAPTIVE_SORT_PRINT_L2(" A combine: ", len); assert(xbuf.size() >= l_block); op_merge_blocks_with_buf (uint_keys, less(), first, l_block, l_irreg1, n_block_a, n_block_b, l_irreg2, comp, move_op(), xbuf.data()); xbuf.clear(); BOOST_MOVE_ADAPTIVE_SORT_PRINT_L1(" A mrg buf: ", len); } } template<class RandIt, class Compare, class XBuf> inline void adaptive_merge_final_merge( RandIt first , typename iter_size<RandIt>::type len1 , typename iter_size<RandIt>::type len2 , typename iter_size<RandIt>::type collected , typename iter_size<RandIt>::type l_intbuf , typename iter_size<RandIt>::type //l_block , bool //use_internal_buf , bool xbuf_used , Compare comp , XBuf & xbuf ) { typedef typename iter_size<RandIt>::type size_type; size_type n_keys = size_type(collected-l_intbuf); size_type len = size_type(len1+len2); if (!xbuf_used || n_keys) { xbuf.clear(); const size_type middle = xbuf_used && n_keys ? n_keys: collected; unstable_sort(first, first + middle, comp, xbuf); BOOST_MOVE_ADAPTIVE_SORT_PRINT_L2(" A k/b srt: ", len); stable_merge(first, first + middle, first + len, comp, xbuf); } BOOST_MOVE_ADAPTIVE_SORT_PRINT_L1(" A fin mrg: ", len); } template<class SizeType> inline static SizeType adaptive_merge_n_keys_without_external_keys(SizeType l_block, SizeType len1, SizeType len2, SizeType l_intbuf) { typedef SizeType size_type; //This is the minimum number of keys to implement the ideal algorithm size_type n_keys = size_type(len1/l_block + len2/l_block); const size_type second_half_blocks = size_type(len2/l_block); const size_type first_half_aux = size_type(len1 - l_intbuf); while(n_keys >= ((first_half_aux-n_keys)/l_block + second_half_blocks)){ --n_keys; } ++n_keys; return n_keys; } template<class SizeType> inline static SizeType adaptive_merge_n_keys_with_external_keys(SizeType l_block, SizeType len1, SizeType len2, SizeType l_intbuf) { typedef SizeType size_type; //This is the minimum number of keys to implement the ideal algorithm size_type n_keys = size_type((len1-l_intbuf)/l_block + len2/l_block); return n_keys; } template<class SizeType, class Xbuf> inline SizeType adaptive_merge_n_keys_intbuf(SizeType &rl_block, SizeType len1, SizeType len2, Xbuf & xbuf, SizeType &l_intbuf_inout) { typedef SizeType size_type; size_type l_block = rl_block; size_type l_intbuf = xbuf.capacity() >= l_block ? 0u : l_block; if (xbuf.capacity() > l_block){ l_block = xbuf.capacity(); } //This is the minimum number of keys to implement the ideal algorithm size_type n_keys = adaptive_merge_n_keys_without_external_keys(l_block, len1, len2, l_intbuf); assert(n_keys >= ((len1-l_intbuf-n_keys)/l_block + len2/l_block)); if(xbuf.template supports_aligned_trailing<size_type> ( l_block , adaptive_merge_n_keys_with_external_keys(l_block, len1, len2, l_intbuf))) { n_keys = 0u; } l_intbuf_inout = l_intbuf; rl_block = l_block; return n_keys; } // Main explanation of the merge algorithm. // // csqrtlen = ceil(sqrt(len)); // // * First, csqrtlen [to be used as buffer] + (len/csqrtlen - 1) [to be used as keys] => to_collect // unique elements are extracted from elements to be sorted and placed in the beginning of the range. // // * Step "combine_blocks": the leading (len1-to_collect) elements plus trailing len2 elements // are merged with a non-trivial ("smart") algorithm to form an ordered range trailing "len-to_collect" elements. // // Explanation of the "combine_blocks" step: // // * Trailing [first+to_collect, first+len1) elements are divided in groups of cqrtlen elements. // Remaining elements that can't form a group are grouped in front of those elements. // * Trailing [first+len1, first+len1+len2) elements are divided in groups of cqrtlen elements. // Remaining elements that can't form a group are grouped in the back of those elements. // * In parallel the following two steps are performed: // * Groups are selection-sorted by first or last element (depending whether they are going // to be merged to left or right) and keys are reordered accordingly as an imitation-buffer. // * Elements of each block pair are merged using the csqrtlen buffer taking into account // if they belong to the first half or second half (marked by the key). // // * In the final merge step leading "to_collect" elements are merged with rotations // with the rest of merged elements in the "combine_blocks" step. // // Corner cases: // // * If no "to_collect" elements can be extracted: // // * If more than a minimum number of elements is extracted // then reduces the number of elements used as buffer and keys in the // and "combine_blocks" steps. If "combine_blocks" has no enough keys due to this reduction // then uses a rotation based smart merge. // // * If the minimum number of keys can't be extracted, a rotation-based merge is performed. // // * If auxiliary memory is more or equal than min(len1, len2), a buffered merge is performed. // // * If the len1 or len2 are less than 2*csqrtlen then a rotation-based merge is performed. // // * If auxiliary memory is more than csqrtlen+n_keys*sizeof(std::size_t), // then no csqrtlen need to be extracted and "combine_blocks" will use integral // keys to combine blocks. template<class RandIt, class Compare, class XBuf> void adaptive_merge_impl ( RandIt first , typename iter_size<RandIt>::type len1 , typename iter_size<RandIt>::type len2 , Compare comp , XBuf & xbuf ) { typedef typename iter_size<RandIt>::type size_type; if(xbuf.capacity() >= min_value<size_type>(len1, len2)){ buffered_merge( first, first+len1 , first + len1+len2, comp, xbuf); } else{ const size_type len = size_type(len1+len2); //Calculate ideal parameters and try to collect needed unique keys size_type l_block = size_type(ceil_sqrt(len)); //One range is not big enough to extract keys and the internal buffer so a //rotation-based based merge will do just fine if(len1 <= l_block*2 || len2 <= l_block*2){ merge_bufferless(first, first+len1, first+len1+len2, comp); return; } //Detail the number of keys and internal buffer. If xbuf has enough memory, no //internal buffer is needed so l_intbuf will remain 0. size_type l_intbuf = 0; size_type n_keys = adaptive_merge_n_keys_intbuf(l_block, len1, len2, xbuf, l_intbuf); size_type const to_collect = size_type(l_intbuf+n_keys); //Try to extract needed unique values from the first range size_type const collected = collect_unique(first, first+len1, to_collect, comp, xbuf); BOOST_MOVE_ADAPTIVE_SORT_PRINT_L1("\n A collect: ", len); //Not the minimum number of keys is not available on the first range, so fallback to rotations if(collected != to_collect && collected < 4){ merge_bufferless(first, first+collected, first+len1, comp); merge_bufferless(first, first + len1, first + len1 + len2, comp); return; } //If not enough keys but more than minimum, adjust the internal buffer and key count bool use_internal_buf = collected == to_collect; if (!use_internal_buf){ l_intbuf = 0u; n_keys = collected; l_block = lblock_for_combine(l_intbuf, n_keys, len, use_internal_buf); //If use_internal_buf is false, then then internal buffer will be zero and rotation-based combination will be used l_intbuf = use_internal_buf ? l_block : 0u; } bool const xbuf_used = collected == to_collect && xbuf.capacity() >= l_block; //Merge trailing elements using smart merges adaptive_merge_combine_blocks(first, len1, len2, collected, n_keys, l_block, use_internal_buf, xbuf_used, comp, xbuf); //Merge buffer and keys with the rest of the values adaptive_merge_final_merge (first, len1, len2, collected, l_intbuf, l_block, use_internal_buf, xbuf_used, comp, xbuf); } } } //namespace detail_adaptive { ///@endcond //! <b>Effects</b>: Merges two consecutive sorted ranges [first, middle) and [middle, last) //! into one sorted range [first, last) according to the given comparison function comp. //! The algorithm is stable (if there are equivalent elements in the original two ranges, //! the elements from the first range (preserving their original order) precede the elements //! from the second range (preserving their original order). //! //! <b>Requires</b>: //! - RandIt must meet the requirements of ValueSwappable and RandomAccessIterator. //! - The type of dereferenced RandIt must meet the requirements of MoveAssignable and MoveConstructible. //! //! <b>Parameters</b>: //! - first: the beginning of the first sorted range. //! - middle: the end of the first sorted range and the beginning of the second //! - last: the end of the second sorted range //! - comp: comparison function object which returns true if the first argument is is ordered before the second. //! - uninitialized, uninitialized_len: raw storage starting on "uninitialized", able to hold "uninitialized_len" //! elements of type iterator_traits<RandIt>::value_type. Maximum performance is achieved when uninitialized_len //! is min(std::distance(first, middle), std::distance(middle, last)). //! //! <b>Throws</b>: If comp throws or the move constructor, move assignment or swap of the type //! of dereferenced RandIt throws. //! //! <b>Complexity</b>: Always K x O(N) comparisons and move assignments/constructors/swaps. //! Constant factor for comparisons and data movement is minimized when uninitialized_len //! is min(std::distance(first, middle), std::distance(middle, last)). //! Pretty good enough performance is achieved when uninitialized_len is //! ceil(sqrt(std::distance(first, last)))*2. //! //! <b>Caution</b>: Experimental implementation, not production-ready. template<class RandIt, class Compare> void adaptive_merge( RandIt first, RandIt middle, RandIt last, Compare comp , typename iterator_traits<RandIt>::value_type* uninitialized = 0 , typename iter_size<RandIt>::type uninitialized_len = 0) { typedef typename iter_size<RandIt>::type size_type; typedef typename iterator_traits<RandIt>::value_type value_type; if (first == middle || middle == last){ return; } //Reduce ranges to merge if possible do { if (comp(*middle, *first)){ break; } ++first; if (first == middle) return; } while(1); RandIt first_high(middle); --first_high; do { --last; if (comp(*last, *first_high)){ ++last; break; } if (last == middle) return; } while(1); ::boost::movelib::adaptive_xbuf<value_type, value_type*, size_type> xbuf(uninitialized, size_type(uninitialized_len)); ::boost::movelib::detail_adaptive::adaptive_merge_impl(first, size_type(middle - first), size_type(last - middle), comp, xbuf); } } //namespace movelib { } //namespace boost { #if defined(BOOST_CLANG) || (defined(BOOST_GCC) && (BOOST_GCC >= 40600)) #pragma GCC diagnostic pop #endif #include <boost/move/detail/config_end.hpp> #endif //#define BOOST_MOVE_ADAPTIVE_MERGE_HPP