BitMagic-C++
sample11.cpp

Example of how to use various bit counting techniques

See also
bm::bvector<>::count()
bm::bvector<>::count_range()
bm::bvector<>::count_to()
bm::count_and()
bm::bvector<>::counted_enumerator
/*
Copyright(c) 2002-2017 Anatoliy Kuznetsov(anatoliy_kuznetsov at yahoo.com)
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
For more information please visit: http://bitmagic.io
*/
/** \example sample11.cpp
Example of how to use various bit counting techniques
\sa bm::bvector<>::count()
\sa bm::bvector<>::count_range()
\sa bm::bvector<>::count_to()
\sa bm::count_and()
\sa bm::bvector<>::counted_enumerator
*/
/*! \file sample11.cpp
\brief Example: bvector<> bit-counting techniques analysis
*/
#include <iostream>
#include <random>
#include <memory>
#include "bm.h"
#include "bmalgo.h"
#include "bmtimer.h"
using namespace std;
// timing storage for benchmarking
const unsigned benchmark_count = 10000;
unsigned vector_max = 400000000;
std::random_device rand_dev;
std::mt19937 gen(rand_dev()); // mersenne_twister_engine
std::uniform_int_distribution<> rand_dis(1, int(vector_max)); // generate uniform numebrs for [1, vector_max]
/// generate pseudo-random bit-vector, mix of blocks
///
static
{
for (i = 0; i < vector_max;)
{
// generate bit-blocks
for (j = 0; j < 65535*8; i += 10, j++)
{
bv.set(i);
}
if (i > vector_max)
break;
// generate GAP (compressed) blocks
for (j = 0; j < 65535; i += 120, j++)
{
unsigned len = rand() % 64;
bv.set_range(i, i + len);
i += len;
if (i > vector_max)
break;
}
}
// compress vector
bv.optimize(tb);
// compute bit-vector statistics
bv.calc_stat(&st);
std::cout << "Bit-vector statistics: GAP (compressed blocks)=" << st.gap_blocks
<< ", BIT (uncompressed blocks)=" << st.bit_blocks
<< std::endl << std::endl;
}
/// "pre-heat" CPU to minimize dynamic overclocking effects
///
static
{
for (unsigned i = 0; i < benchmark_count; ++i)
{
cnt += bv.count();
m+=cnt*cnt;
}
return m;
}
/// simple population count for the whole vector
///
static
{
{
bm::chrono_taker tt1("1. bvector<>::count()", benchmark_count / 2, &timing_map);
for (unsigned i = 0; i < benchmark_count / 2; ++i)
{
cnt += bv.count();
}
}
// this is mostly to prevent compiler to optimize loop away
std::cout << "Count test finished." << cnt << "\r";
}
/// count_range() test
///
static
{
{
bm::chrono_taker tt1("2. bvector<>::count_range()", benchmark_count, &timing_map);
for (unsigned i = 0; i < benchmark_count; ++i)
{
unsigned from = unsigned(rand_dis(gen));
unsigned to = unsigned(rand_dis(gen));
if (from > to)
swap(from, to);
cnt += bv.count_range(from, to);
}
}
// this is mostly to prevent compiler to optimize loop away
std::cout << "Count range test finished." << cnt << "\r";
}
/// count_range() test using pre-calculated blocks bit count
///
static
{
std::unique_ptr<bm::bvector<>::rs_index_type> rs(new bm::bvector<>::rs_index_type());
bv.build_rs_index(rs.get());
{
bm::chrono_taker tt1("3. bvector<>::count_range() with rs_index", benchmark_count, &timing_map);
cnt = 0;
for (unsigned i = 0; i < benchmark_count; ++i)
{
unsigned from = unsigned(rand_dis(gen));
unsigned to = unsigned(rand_dis(gen));
if (from > to)
swap(from, to);
cnt += bv.count_range(from, to, *rs); // use rs index for acceleration
} // for i
}
// this is mostly to prevent compiler to optimize loop away
std::cout << "Count range with blocks test finished." << cnt << "\r";
}
/// count_to() test using pre-calculated rank-select index
///
static
{
// build a block population count list, used for count_to() acceleration
std::unique_ptr<bm::bvector<>::rs_index_type> rs(new bm::bvector<>::rs_index_type());
bv.build_rs_index(rs.get());
{
bm::chrono_taker tt1("4. bvector<>::count_to() with rs_index", benchmark_count, &timing_map);
for (unsigned i = 0; i < benchmark_count; ++i)
{
unsigned to = unsigned(rand_dis(gen));
cnt += bv.count_to(to, *rs); // use rank-select index for acceleration
}
}
// this is mostly to prevent compiler to optimize loop away
std::cout << "Count to with blocks test finished." << cnt << "\r";
}
/// count_range implemented via two count_to() calls using pre-calculated
/// rank-select index
///
static
{
// build a block population count list, used for count_to() acceleration
std::unique_ptr<bm::bvector<>::rs_index_type> rs(new bm::bvector<>::rs_index_type());
bv.build_rs_index(rs.get());
{
bm::chrono_taker tt1("5. bvector<>::count_to to simulate count_range()", benchmark_count, &timing_map);
for (unsigned i = 0; i < benchmark_count; ++i)
{
unsigned from = unsigned(rand_dis(gen));
unsigned to = unsigned(rand_dis(gen));
if (from > to)
swap(from, to);
bm::bvector<>::size_type cnt_to = bv.count_to(to, *rs);
bm::bvector<>::size_type cnt_from = bv.count_to(from - 1, *rs);
bm::bvector<>::size_type cnt_r = cnt_to - cnt_from;
cnt += cnt_r;
}
}
// this is mostly to prevent compiler to optimize loop away
std::cout << "Count range via count_to test finished." << cnt << "\r";
}
/// count_range implemented via bm::count_and
///
/// this method can be used, when we need co compute multiple ranges in one call
///
static
void bv_count_and(const bm::bvector<>& bv)
{
{
bm::chrono_taker tt1("6. bm::count_and with mask vector", benchmark_count, &timing_map);
bm::bvector<> mask_bv(bm::BM_GAP); // use compressed mask, better seluts on long ranges
for (unsigned i = 0; i < benchmark_count; ++i)
{
unsigned from = unsigned(rand_dis(gen));
unsigned to = unsigned(rand_dis(gen));
if (from > to)
swap(from, to);
mask_bv.set_range(from, to, true); // set mask vector
cnt += bm::count_and(bv, mask_bv);
mask_bv.clear(true); // clear and free memory (faster)
}
}
// this is mostly to prevent compiler to optimize loop away
std::cout << "count AND finished." << cnt << "\r";
}
/// count_to implemented via bm::bvector<>::counted_enumerator
///
/// Counted enumerator is an iterator automata, which counts the running population count
/// along the iteration sequence
///
static
{
{
// This is a slow method so we use less iterators
bm::chrono_taker tt1("7. bm::bvector<>::counted_enumerator", benchmark_count/20, &timing_map);
for (unsigned i = 0; i < benchmark_count/20; ++i)
{
unsigned to = unsigned(rand_dis(gen));
for (; en.valid(); ++en)
{
if (*en > to)
break;
}
cnt += en.count();
}
}
std::cout << "counted_enumerator finished." << cnt << "\r";
}
int main(void)
{
try
{
/// pre-heat CPU to minimize dynamic overclocking
unsigned s = pre_heat(bv);
std::cout << s << "\r";
// Test 1.
// Uses plain bvector<>::count() to compute global population count
// This function would benefit from SIMD (SSE42 / AVX2) acceleration
//
// Test 2.
// Uses bvector<>::count_range() to compute population count in a randomly generated
// region of a bit-vector.
// This is should be naturally faster than Test 1, because it range is less than the whole
//
// Test 3.
// Uses bvector<>::count_range() together with bvector<>::count_blocks()
// (pre-calculated bit-count for each block).
// It make sense to use this method if bit-vector is constant (or chnages infrequently)
// and we need to do many range counting calculations
//
// Test 4.
// Uses bvector<>::count_to() to compute population count to a specified element.
// Equivalent of count_range(0, to);
// This method uses acceleration structure using bvector<>::running_count_blocks()
// It is similar to count_range acceleration, but uses a different (faster) algorithm
//
// Test 5.
// Uses bvector<>::count_to() twice to simulate count_range()
// using counting difference:
// count_r = count_to(0, from) - count_to(0, to-1)
// This method can actually be faster than count_range()
//
// Test 6.
// Compute range population count via a mask vector and logical AND operation.
// Not the fastest method, but can be useful, when multiple ranges needs to be computed
//
// Test 7.
// Compute cout using counted_enumerator iterator
// method combines iteratrion over bit vector and sliding population count
// print all test timing results
//
std::cout << " "
<< std::endl;
}
catch(std::exception& ex)
{
std::cerr << ex.what() << std::endl;
return 1;
}
return 0;
}
pre_heat
static bm::bvector ::size_type pre_heat(const bm::bvector<> &bv)
"pre-heat" CPU to minimize dynamic overclocking effects
Definition: sample11.cpp:96
bm::bvector::set_range
bvector< Alloc > & set_range(size_type left, size_type right, bool value=true)
Sets all bits in the specified closed interval [left,right] Interval must be inside the bvector's siz...
Definition: bm.h:2172
bm::bvector::calc_stat
void calc_stat(struct bm::bvector< Alloc >::statistics *st) const BMNOEXCEPT
Calculates bitvector statistics.
Definition: bm.h:3407
generate_bvector
static void generate_bvector(bm::bvector<> &bv)
generate pseudo-random bit-vector, mix of blocks
Definition: sample11.cpp:57
bm::bvector::rs_index_type
rs_index< allocator_type > rs_index_type
Definition: bm.h:845
benchmark_count
const unsigned benchmark_count
Definition: sample11.cpp:46
timing_map
bm::chrono_taker::duration_map_type timing_map
Definition: sample11.cpp:44
bm::chrono_taker
Utility class to collect performance measurements and statistics.
Definition: bmtimer.h:39
bv_counted_enumerator
static void bv_counted_enumerator(const bm::bvector<> &bv)
count_to implemented via bm::bvector<>::counted_enumerator
Definition: sample11.cpp:267
bm::bv_statistics::bit_blocks
size_t bit_blocks
Number of bit blocks.
Definition: bmfunc.h:56
BM_DECLARE_TEMP_BLOCK
#define BM_DECLARE_TEMP_BLOCK(x)
Definition: bm.h:47
bm::bvector::count_range
size_type count_range(size_type left, size_type right, const rs_index_type &rs_idx) const BMNOEXCEPT
Returns count of 1 bits in the given range [left..right] Uses rank-select index to accelerate the sea...
Definition: bm.h:2973
bv_count_to_acc
static void bv_count_to_acc(const bm::bvector<> &bv)
count_to() test using pre-calculated rank-select index
Definition: sample11.cpp:178
bv_count_test
static void bv_count_test(const bm::bvector<> &bv)
simple population count for the whole vector
Definition: sample11.cpp:113
bm::bvector::counted_enumerator::count
size_type count() const BMNOEXCEPT
Number of bits ON starting from the .
Definition: bm.h:771
vector_max
unsigned vector_max
Definition: sample11.cpp:47
bmalgo.h
Algorithms for bvector<> (main include)
bv_count_and
static void bv_count_and(const bm::bvector<> &bv)
count_range implemented via bm::count_and
Definition: sample11.cpp:237
bm::bv_statistics::gap_blocks
size_t gap_blocks
Number of GAP blocks.
Definition: bmfunc.h:57
bv_count_range_acc
static void bv_count_range_acc(const bm::bvector<> &bv)
count_range() test using pre-calculated blocks bit count
Definition: sample11.cpp:152
bm::bvector<>
bm::bvector::set
bvector< Alloc > & set(size_type n, bool val=true)
Sets bit n if val is true, clears bit n if val is false.
Definition: bm.h:3595
bm::bvector::count
size_type count() const BMNOEXCEPT
population cout (count of ON bits)
Definition: bm.h:2208
bm::bvector::iterator_base::valid
bool valid() const BMNOEXCEPT
Checks if iterator is still valid. Analog of != 0 comparison for pointers.
Definition: bm.h:280
gen
std::mt19937 gen(rand_dev())
bmtimer.h
Timing utilities for benchmarking (internal)
bm::chrono_taker::ct_ops_per_sec
@ ct_ops_per_sec
Definition: bmtimer.h:59
bm::BM_GAP
@ BM_GAP
GAP compression is ON.
Definition: bmconst.h:144
main
int main(void)
Definition: sample11.cpp:292
bm::chrono_taker::duration_map_type
std::map< std::string, statistics > duration_map_type
test name to duration map
Definition: bmtimer.h:65
bm::count_and
BV::size_type count_and(const BV &bv1, const BV &bv2) BMNOEXCEPT
Computes bitcount of AND operation of two bitsets.
Definition: bmalgo.h:49
bv_count_range
static void bv_count_range(const bm::bvector<> &bv)
count_range() test
Definition: sample11.cpp:131
bm::chrono_taker::print_duration_map
static void print_duration_map(const duration_map_type &dmap, format fmt=ct_time)
Definition: bmtimer.h:127
bm::bvector::build_rs_index
void build_rs_index(rs_index_type *rs_idx, bvector< Alloc > *bv_blocks=0) const
compute running total of all blocks in bit vector (rank-select index)
Definition: bm.h:2304
bm::bvector::first
enumerator first() const
Returns enumerator pointing on the first non-zero bit.
Definition: bm.h:1784
bm::bvector::size_type
bm::id_t size_type
Definition: bm.h:117
bm::bvector::counted_enumerator
Constant iterator designed to enumerate "ON" bits counted_enumerator keeps bitcount,...
Definition: bm.h:729
bv_count_to_range_acc
static void bv_count_to_range_acc(const bm::bvector<> &bv)
count_range implemented via two count_to() calls using pre-calculated rank-select index
Definition: sample11.cpp:204
bm.h
Compressed bit-vector bvector<> container, set algebraic methods, traversal iterators.
bm::bvector::optimize
void optimize(bm::word_t *temp_block=0, optmode opt_mode=opt_compress, statistics *stat=0)
Optimize memory bitvector's memory allocation.
Definition: bm.h:3085
rand_dis
std::uniform_int_distribution rand_dis(1, int(vector_max))
bm::bvector::count_to
size_type count_to(size_type n, const rs_index_type &rs_idx) const BMNOEXCEPT
Returns count of 1 bits (population) in [0..right] range.
Definition: bm.h:2547
bm::bvector::statistics
Statistical information about bitset's memory allocation details.
Definition: bm.h:121
rand_dev
std::random_device rand_dev
Definition: sample11.cpp:49