summaryrefslogtreecommitdiff
path: root/matching/include/phat/representations/bit_tree_pivot_column.h
diff options
context:
space:
mode:
Diffstat (limited to 'matching/include/phat/representations/bit_tree_pivot_column.h')
-rw-r--r--matching/include/phat/representations/bit_tree_pivot_column.h165
1 files changed, 165 insertions, 0 deletions
diff --git a/matching/include/phat/representations/bit_tree_pivot_column.h b/matching/include/phat/representations/bit_tree_pivot_column.h
new file mode 100644
index 0000000..4d48e88
--- /dev/null
+++ b/matching/include/phat/representations/bit_tree_pivot_column.h
@@ -0,0 +1,165 @@
+/* Copyright 2013 IST Austria
+ Contributed by: Hubert Wagner
+
+ This file is part of PHAT.
+
+ PHAT is free software: you can redistribute it and/or modify
+ it under the terms of the GNU Lesser General Public License as published by
+ the Free Software Foundation, either version 3 of the License, or
+ (at your option) any later version.
+
+ PHAT is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public License
+ along with PHAT. If not, see <http://www.gnu.org/licenses/>. */
+
+#pragma once
+
+#include <phat/helpers/misc.h>
+#include <phat/representations/abstract_pivot_column.h>
+
+namespace phat {
+
+ // This is a bitset indexed with a 64-ary tree. Each node in the index
+ // has 64 bits; i-th bit says that the i-th subtree is non-empty.
+ // Supports practically O(1), inplace, zero-allocation: insert, remove, max_element
+ // and clear in O(number of ones in the bitset).
+ // 'add_index' is still the real bottleneck in practice.
+ class bit_tree_column
+ {
+ protected:
+
+ size_t offset; // data[i + offset] = ith block of the data-bitset
+ typedef uint64_t block_type;
+ std::vector< block_type > data;
+
+
+ size_t debrujin_magic_table[ 64 ];
+
+ enum { block_size_in_bits = 64 };
+ enum { block_shift = 6 };
+
+ // Some magic: http://graphics.stanford.edu/~seander/bithacks.html
+ // Gets the position of the rightmost bit of 'x'. 0 means the most significant bit.
+ // (-x)&x isolates the rightmost bit.
+ // The whole method is much faster than calling log2i, and very comparable to using ScanBitForward/Reverse intrinsic,
+ // which should be one CPU instruction, but is not portable.
+ size_t rightmost_pos( const block_type value ) const {
+ return 64 - 1 - debrujin_magic_table[ ( (value & (-(int64_t)value) ) * 0x07EDD5E59A4E28C2 ) >> 58 ];
+ }
+
+ public:
+
+ void init( index num_cols ) {
+ int64_t n = 1; // in case of overflow
+ int64_t bottom_blocks_needed = ( num_cols + block_size_in_bits - 1 ) / block_size_in_bits;
+ int64_t upper_blocks = 1;
+
+ // How many blocks/nodes of index needed to index the whole bitset?
+ while( n * block_size_in_bits < bottom_blocks_needed ) {
+ n *= block_size_in_bits;
+ upper_blocks += n;
+ }
+
+ offset = upper_blocks;
+ data.resize( upper_blocks + bottom_blocks_needed, 0 );
+
+ std::size_t temp_array[ 64 ] = {
+ 63, 0, 58, 1, 59, 47, 53, 2,
+ 60, 39, 48, 27, 54, 33, 42, 3,
+ 61, 51, 37, 40, 49, 18, 28, 20,
+ 55, 30, 34, 11, 43, 14, 22, 4,
+ 62, 57, 46, 52, 38, 26, 32, 41,
+ 50, 36, 17, 19, 29, 10, 13, 21,
+ 56, 45, 25, 31, 35, 16, 9, 12,
+ 44, 24, 15, 8, 23, 7, 6, 5 };
+
+ std::copy( &temp_array[ 0 ], &temp_array[ 64 ], &debrujin_magic_table[ 0 ] );
+ }
+
+ index get_max_index() const {
+ if( !data[ 0 ] )
+ return -1;
+
+ size_t n = 0;
+ size_t newn = 0;
+ size_t index = 0;
+ while( newn < data.size() ) {
+ n = newn;
+ index = rightmost_pos( data[ n ] );
+ newn = ( n << block_shift ) + index + 1;
+ }
+
+ return ( ( n - offset ) << block_shift ) + index;
+ }
+
+ bool is_empty() const {
+ return data[ 0 ] == 0;
+ }
+
+ void add_index( const size_t entry ) {
+ const block_type ONE = 1;
+ const block_type block_modulo_mask = ( ONE << block_shift ) - 1;
+ size_t index_in_level = entry >> block_shift;
+ size_t address = index_in_level + offset;
+ size_t index_in_block = entry & block_modulo_mask;
+
+ block_type mask = ( ONE << ( block_size_in_bits - index_in_block - 1 ) );
+
+ data[ address ] ^= mask;
+
+ // Check if we reached the root. Also, if anyone else was in this block, we don't need to update the path up.
+ while( address && !( data[ address ] & ~mask ) ) {
+ index_in_block = index_in_level & block_modulo_mask;
+ index_in_level >>= block_shift;
+ --address;
+ address >>= block_shift;
+ mask = ( ONE << ( block_size_in_bits - index_in_block - 1 ) );
+ data[ address ] ^= mask;
+ }
+ }
+
+ void get_col_and_clear( column &out ) {
+ index mx = this->get_max_index();
+ while( mx != -1 ) {
+ out.push_back( mx );
+ add_index( mx );
+ mx = this->get_max_index();
+ }
+
+ std::reverse( out.begin(), out.end() );
+ }
+
+ void add_col(const column &col) {
+ for( size_t i = 0; i < col.size(); ++i )
+ add_index(col[i]);
+ }
+
+ void clear() {
+ index mx = this->get_max_index();
+ while( mx != -1 ) {
+ add_index( mx );
+ mx = this->get_max_index();
+ }
+ }
+
+ void remove_max() {
+ add_index( get_max_index() );
+ }
+
+ void set_col( const column& col ) {
+ clear();
+ add_col( col );
+ }
+
+ void get_col( column& col ) {
+ get_col_and_clear( col );
+ add_col( col );
+ }
+ };
+
+ typedef abstract_pivot_column<bit_tree_column> bit_tree_pivot_column;
+}