-
Notifications
You must be signed in to change notification settings - Fork 241
Prime Number Functions #400
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: develop
Are you sure you want to change the base?
Changes from 5 commits
4015fbc
3ee737b
9512bb6
d762398
5375a1d
a684dbd
3e4db8a
dd8a61c
4debd0d
2a7e031
a68910e
386cc4d
d79eddb
3fdf917
6ca245b
7d3a520
a1ac504
35d2aa1
243a299
2eadeff
173ce0d
23fba36
f7b45fd
d687d5e
e51d727
1245d27
2052081
9f81e0d
55ea045
31a2105
f545928
d780db5
2639bed
94fc1ac
6ebb906
0521854
b233a80
f95c2cf
8e2e29a
1a24f16
c63f1f1
b7d4256
fbc38c8
2e46b81
6759ede
1b40403
97244be
fa04133
c4a89c8
91836f6
6d6b19f
81e4a6c
0b8f1d5
5ba0a1d
8e240f7
7c0cbbf
3d9b77c
302fb5f
9792a23
84a69f0
5dc3523
0dbe69c
eee2c86
f5d789a
f2277e3
1d2f03c
0d9d31b
c361cde
66c2642
de1f331
eaea5f9
e8196f3
cb5d978
830ccc4
4dc3eb2
90be100
a772782
2d1461f
e7cdb32
29eef88
b5a28b5
6c26b53
980bfe7
07e6f58
86b9e5a
f19149e
1ad0d51
5f6d06a
670b06d
7c7a491
e507ba5
c9cb41c
e8b71a0
d649f65
ed98892
c3b3934
e9aa05d
7c5d792
2bbfad2
ddc5c8a
f517c00
a356b47
2212e45
b12e2dc
f39a4d8
eafbefc
cb36b89
16c2354
0b1a690
8c883d1
f357e0f
a4f2d89
d16e562
91be2c3
d9ae8c2
7d22010
b541987
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,108 @@ | ||
// Copyright 2020 Matt Borland | ||
// | ||
// Use, modification and distribution are subject to 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) | ||
|
||
#ifndef BOOST_MATH_SPECIAL_FUNCTIONS_PRIME_FUNCTIONS_HPP | ||
#define BOOST_MATH_SPECIAL_FUNCTIONS_PRIME_FUNCTIONS_HPP | ||
|
||
#include <boost/math/special_functions/prime.hpp> | ||
#include <boost/assert.hpp> | ||
#include <deque> | ||
#include <vector> | ||
#include <iterator> | ||
#include <iostream> | ||
|
||
namespace boost { namespace math | ||
{ | ||
|
||
// https://mathworld.wolfram.com/SieveofEratosthenes.html | ||
// https://www.cs.utexas.edu/users/misra/scannedPdf.dir/linearSieve.pdf | ||
template<class Z, class OutputIterator> | ||
auto prime_sieve(Z lower_bound, Z upper_bound, OutputIterator output) -> decltype(output) | ||
{ | ||
static_assert(std::is_integral<Z>::value, "No primes for floating point types"); | ||
BOOST_ASSERT_MSG(upper_bound + 1 < std::numeric_limits<Z>::max(), "Type Overflow"); | ||
mborland marked this conversation as resolved.
Show resolved
Hide resolved
|
||
std::vector<Z> least_divisors(upper_bound + 1, 0); | ||
std::deque<Z> primes; | ||
mborland marked this conversation as resolved.
Show resolved
Hide resolved
|
||
|
||
for (Z i{2}; i <= upper_bound; ++i) | ||
{ | ||
if (least_divisors[i] == 0) | ||
{ | ||
least_divisors[i] = i; | ||
primes.emplace_back(i); | ||
} | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I would wager that this is a factor in the performance difference: Kim's implementation estimates the number of primes and reserves that number of elements upfront. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Yes; I would agree-I thought we had this earlier . . . There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I did. The snippets in question are below:
and this was in the main
I thought making it a little less generic to gain performance was worthwhile. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Call me a purist, but you'll make it faster by making it more generic. Allocating memory for the output is not the responsibility of the algorithm. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I initially was returning via output iterator instead of using a provided container. Profiling showed that more than 10% of my runtime was the call to There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I am not convinced this interface change is worth ~5x more CPU runtime. Care to weigh in @pabristow @jzmaddock ? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. I think we have a number of issues here:
Question: how far can we go with a range-based output? As in "put the next N primes here". The range could either be a pair of iterators, an iterator and the value N, or a boost::range. If that's not faster than a container based interface (no memory allocation) then there's something wrong somewhere. But... I don't know how you would divide that up for multithreading? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. @mborland I would test it something like this. Create the vector to the size required outside the loop.
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. @jzmaddock What about building a wrapper class that would support range-based output and other functions like last/next value, random value, etc? I think that is currently the best way to go about supporting a wider variety of data structures. As for removing dynamic memory allocation? I think it is in the realm of the possible, but would require massive retooling. You would become highly dependent on mutex locking which is not a trivial operation. There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more.
(Btw, I'm going to use x for prime numbers and It's funny, but the output iterator interface is almost a range interface. As you say, either an (iterator, iterator) or an (iterator, |
||
|
||
for (size_t j{}; j < least_divisors.size(); ++j) | ||
{ | ||
if (j >= primes.size()) | ||
{ | ||
break; | ||
} | ||
|
||
else if (primes[j] > least_divisors[i]) | ||
{ | ||
break; | ||
} | ||
|
||
else if (i * primes[j] > upper_bound) | ||
{ | ||
break; | ||
} | ||
|
||
else | ||
{ | ||
least_divisors[i * primes[j]] = primes[j]; | ||
} | ||
} | ||
mborland marked this conversation as resolved.
Show resolved
Hide resolved
|
||
} | ||
|
||
auto it{primes.begin()}; | ||
while (*it < lower_bound && it != primes.end()) | ||
{ | ||
primes.pop_front(); | ||
++it; | ||
} | ||
|
||
return std::move(primes.begin(), primes.end(), output); | ||
} | ||
|
||
template<class Z, class OutputIterator> | ||
auto prime_range(Z lower_bound, Z upper_bound, OutputIterator output) -> decltype(output) | ||
{ | ||
if (upper_bound <= 104729) | ||
{ | ||
Z i{2}; | ||
unsigned counter {}; | ||
std::deque<Z> primes; | ||
while (i <= upper_bound) | ||
{ | ||
if (i >= lower_bound) | ||
{ | ||
primes.emplace_back(i); | ||
} | ||
|
||
++counter; | ||
i = static_cast<Z>(boost::math::prime(counter)); | ||
} | ||
|
||
return std::move(primes.begin(), primes.end(), output); | ||
} | ||
|
||
else | ||
{ | ||
return prime_sieve(lower_bound, upper_bound, output); | ||
} | ||
} | ||
|
||
template<class Z, class OutputIterator> | ||
inline auto prime_range(Z upper_bound, OutputIterator output) -> decltype(output) | ||
{ | ||
return prime_range(static_cast<Z>(2), upper_bound, output); | ||
} | ||
}} | ||
|
||
#endif //BOOST_MATH_SPECIAL_FUNCTIONS_PRIME_FUNCTIONS_HPP |
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,66 @@ | ||
// Copyright 2020 Matt Borland | ||
// | ||
// Use, modification and distribution are subject to 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) | ||
|
||
#include <boost/math/special_functions/prime_sieve.hpp> | ||
#include <benchmark/benchmark.h> | ||
|
||
template <class Z> | ||
void prime_sieve(benchmark::State& state) | ||
{ | ||
Z upper = static_cast<Z>(state.range(0)); | ||
for(auto _ : state) | ||
{ | ||
std::vector<Z> primes; | ||
benchmark::DoNotOptimize(boost::math::prime_sieve(static_cast<Z>(2), upper, std::back_inserter(primes))); | ||
} | ||
state.SetComplexityN(state.range(0)); | ||
} | ||
|
||
template <typename Z> | ||
void prime_range(benchmark::State& state) | ||
{ | ||
Z upper = static_cast<Z>(state.range(0)); | ||
for(auto _ : state) | ||
{ | ||
std::vector<Z> primes; | ||
benchmark::DoNotOptimize(boost::math::prime_range(static_cast<Z>(2), upper, std::back_inserter(primes))); | ||
} | ||
state.SetComplexityN(state.range(0)); | ||
} | ||
|
||
template <class Z> | ||
void prime_sieve_partial_range(benchmark::State& state) | ||
{ | ||
Z upper = static_cast<Z>(state.range(0)); | ||
Z lower = static_cast<Z>(state.range(0)) > 2 ? static_cast<Z>(state.range(0)) : 2; | ||
for(auto _ : state) | ||
{ | ||
std::vector<Z> primes; | ||
benchmark::DoNotOptimize(boost::math::prime_sieve(lower, upper, std::back_inserter(primes))); | ||
} | ||
state.SetComplexityN(state.range(0)); | ||
} | ||
|
||
BENCHMARK_TEMPLATE(prime_sieve, int32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_sieve, int64_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_sieve, uint32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_sieve_partial_range, int32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_sieve_partial_range, int64_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_sieve_partial_range, uint32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_range, int32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_range, int64_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_range, uint32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 22)->Complexity(); | ||
|
||
// Direct comparison of lookup vs sieve using only range of lookup | ||
BENCHMARK_TEMPLATE(prime_sieve, int32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 16)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_range, int32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 16)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_sieve, int64_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 16)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_range, int64_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 16)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_sieve, uint32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 16)->Complexity(); | ||
BENCHMARK_TEMPLATE(prime_range, uint32_t)->RangeMultiplier(2)->Range(1 << 1, 1 << 16)->Complexity(); | ||
|
||
BENCHMARK_MAIN(); |
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,127 @@ | ||
// Copyright 2020 Matt Borland | ||
// | ||
// Use, modification and distribution are subject to 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) | ||
|
||
#include <boost/math/special_functions/prime_sieve.hpp> | ||
#include <boost/core/lightweight_test.hpp> | ||
#include <boost/multiprecision/cpp_int.hpp> | ||
#include <list> | ||
#include <deque> | ||
#include <array> | ||
|
||
template<typename Z> | ||
void test_prime_sieve() | ||
{ | ||
std::vector<Z> primes; | ||
Z ref {168}; // Calculated with wolfram-alpha | ||
|
||
// Does the function work with a vector | ||
boost::math::prime_sieve(2, 1000, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), ref); | ||
|
||
// Tests for correctness | ||
// 100 | ||
primes.clear(); | ||
boost::math::prime_sieve(2, 100, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), 25); | ||
|
||
// 10'000 | ||
primes.clear(); | ||
boost::math::prime_sieve(2, 10000, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), 1229); | ||
|
||
// 100'000 | ||
primes.clear(); | ||
boost::math::prime_sieve(2, 100000, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), 9592); | ||
|
||
// 1'000'000 | ||
primes.clear(); | ||
boost::math::prime_sieve(2, 1000000, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), 78498); | ||
|
||
// Does the function work with a list? | ||
std::list<Z> l_primes; | ||
boost::math::prime_sieve(2, 1000, std::back_inserter(l_primes)); | ||
BOOST_TEST_EQ(l_primes.size(), ref); | ||
|
||
// Does the function work with a deque? | ||
std::deque<Z> d_primes; | ||
boost::math::prime_sieve(2, 1000, std::back_inserter(d_primes)); | ||
BOOST_TEST_EQ(d_primes.size(), ref); | ||
} | ||
|
||
template<typename Z> | ||
void test_prime_range() | ||
{ | ||
std::vector<Z> primes; | ||
Z ref {168}; // Calculated with wolfram-alpha | ||
|
||
// Does the upper and lower bound call work | ||
boost::math::prime_range(2, 1000, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), ref); | ||
|
||
// Does the upper bound call work | ||
primes.clear(); | ||
boost::math::prime_range(1000, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), ref); | ||
|
||
// Does it work with a deque? | ||
std::deque<Z> d_primes; | ||
boost::math::prime_range(1000, std::back_inserter(d_primes)); | ||
BOOST_TEST_EQ(d_primes.size(), ref); | ||
|
||
// Does it work with a list? | ||
std::list<Z> l_primes; | ||
boost::math::prime_range(1000, std::front_inserter(l_primes)); | ||
BOOST_TEST_EQ(l_primes.size(), ref); | ||
|
||
// Does the lower bound change the results? | ||
ref = 143; // Calculated with wolfram-alpha | ||
primes.clear(); | ||
boost::math::prime_range(100, 1000, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), ref); | ||
|
||
// Does it work with 0 difference? | ||
primes.clear(); | ||
boost::math::prime_range(2, 2, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), 1); | ||
|
||
// Will it call the sieve for large input | ||
ref = 78498; // Calculated with wolfram-alpha | ||
primes.clear(); | ||
boost::math::prime_range(1000000, std::back_inserter(primes)); | ||
BOOST_TEST_EQ(primes.size(), ref); | ||
} | ||
|
||
template<typename Z> | ||
void test_prime_sieve_overflow() | ||
{ | ||
std::vector<Z> primes; | ||
|
||
// Should die with call to BOOST_ASSERT | ||
boost::math::prime_sieve(static_cast<Z>(2), static_cast<Z>(std::numeric_limits<Z>::max()), | ||
std::back_inserter(primes)); | ||
} | ||
|
||
int main() | ||
{ | ||
test_prime_sieve<int>(); | ||
test_prime_sieve<int32_t>(); | ||
test_prime_sieve<int64_t>(); | ||
test_prime_sieve<uint32_t>(); | ||
|
||
test_prime_range<int>(); | ||
test_prime_range<int32_t>(); | ||
test_prime_range<int64_t>(); | ||
test_prime_range<uint32_t>(); | ||
|
||
test_prime_sieve<boost::multiprecision::cpp_int>(); | ||
|
||
//test_prime_sieve_overflow<int16_t>(); | ||
|
||
boost::report_errors(); | ||
} |
Uh oh!
There was an error while loading. Please reload this page.