ceph/src/boost/libs/fiber/performance/fiber/skynet_shared_join.cpp

   1
   2 //          Copyright Oliver Kowalke 2015.
   3 // Distributed under the Boost Software License, Version 1.0.
   4 //    (See accompanying file LICENSE_1_0.txt or copy at
   5 //          http://www.boost.org/LICENSE_1_0.txt)
   6
   7 // based on https://github.com/atemerev/skynet from Alexander Temerev
   8
   9 #include <algorithm>
  10 #include <cassert>
  11 #include <chrono>
  12 #include <cmath>
  13 #include <condition_variable>
  14 #include <cstddef>
  15 #include <cstdint>
  16 #include <cstdlib>
  17 #include <deque>
  18 #include <iostream>
  19 #include <memory>
  20 #include <mutex>
  21 #include <numeric>
  22 #include <vector>
  23
  24 #include <boost/fiber/all.hpp>
  25 #include <boost/predef.h>
  26
  27 #include "barrier.hpp"
  28
  29 using allocator_type = boost::fibers::fixedsize_stack;
  30 using channel_type = boost::fibers::buffered_channel< std::uint64_t >;
  31 using clock_type = std::chrono::steady_clock;
  32 using duration_type = clock_type::duration;
  33 using lock_type = std::unique_lock< std::mutex >;
  34 using time_point_type = clock_type::time_point;
  35
  36 static bool done = false;
  37 static std::mutex mtx{};
  38 static boost::fibers::condition_variable_any cnd{};
  39
  40 // microbenchmark
  41 void skynet( allocator_type & salloc, channel_type & c, std::size_t num, std::size_t size, std::size_t div) {
  42     if ( 1 == size) {
  43         c.push( num);
  44     } else {
  45         channel_type rc{ 16 };
  46         std::vector< boost::fibers::fiber > fibers;
  47         for ( std::size_t i = 0; i < div; ++i) {
  48             auto sub_num = num + i * size / div;
  49             fibers.emplace_back( boost::fibers::launch::dispatch,
  50                                  std::allocator_arg, salloc,
  51                                  skynet,
  52                                  std::ref( salloc), std::ref( rc), sub_num, size / div, div);
  53         }
  54         for ( auto & f: fibers) {
  55             f.join();
  56         }
  57         std::uint64_t sum{ 0 };
  58         for ( std::size_t i = 0; i < div; ++i) {
  59             sum += rc.value_pop();
  60         }
  61         c.push( sum);
  62     }
  63 }
  64
  65 void thread( unsigned int idx, barrier * b) {
  66     boost::fibers::numa::pin_thread( idx);
  67     boost::fibers::use_scheduling_algorithm< boost::fibers::algo::shared_work >();
  68     b->wait();
  69     lock_type lk( mtx);
  70     cnd.wait( lk, [](){ return done; });
  71     BOOST_ASSERT( done);
  72 }
  73
  74 int main() {
  75     try {
  76         boost::fibers::use_scheduling_algorithm< boost::fibers::algo::shared_work >();
  77         unsigned int n = std::thread::hardware_concurrency();
  78         barrier b( n);
  79         boost::fibers::numa::pin_thread( n - 1);
  80         std::size_t size{ 1000000 };
  81         std::size_t div{ 10 };
  82         std::vector< std::thread > threads;
  83         for ( unsigned int i = 1; i < n; ++i) {
  84             threads.emplace_back( thread, i - 1, & b);
  85         };
  86         // Windows 10 and FreeBSD require a fiber stack of 8kb
  87         // otherwise the stack gets exhausted
  88         // stack requirements must be checked for other OS too
  89 #if BOOST_OS_WINDOWS || BOOST_OS_BSD
  90         allocator_type salloc{ 2*allocator_type::traits_type::page_size() };
  91 #else
  92         allocator_type salloc{ allocator_type::traits_type::page_size() };
  93 #endif
  94         std::uint64_t result{ 0 };
  95         channel_type rc{ 2 };
  96         b.wait();
  97         time_point_type start{ clock_type::now() };
  98         skynet( salloc, rc, 0, size, div);
  99         result = rc.value_pop();
 100         if ( 499999500000 != result) {
 101             throw std::runtime_error("invalid result");
 102         }
 103         auto duration = clock_type::now() - start;
 104         lock_type lk( mtx);
 105         done = true;
 106         lk.unlock();
 107         cnd.notify_all();
 108         for ( std::thread & t : threads) {
 109             t.join();
 110         }
 111         std::cout << "duration: " << duration.count() / 1000000 << " ms" << std::endl;
 112         return EXIT_SUCCESS;
 113     } catch ( std::exception const& e) {
 114         std::cerr << "exception: " << e.what() << std::endl;
 115     } catch (...) {
 116         std::cerr << "unhandled exception" << std::endl;
 117     }
 118         return EXIT_FAILURE;
 119 }