Test as number of threads decreased : infinite loop?
pocdn opened this issue · 4 comments
I simplified the original test code to be the following
`#include "thread_pool.hpp"
#include <vector>
#include <iostream> // std::cout, std::endl
using namespace std;
/*
- Intel Xeon ES-1620 v3 3.5GHz
- git clone https://github.com/Tyler-Hardin/thread_pool.git
- git checkout -b cv origin/cv
- $ gcc --version
- gcc.exe (Rev3, Built by MSYS2 project) 6.3.0
- $ g++ --std=c++11 test.cpp thread_pool.cpp
- $ time ./a.exe
- ...
- real 5m23.883s for num_threads = 1 (aborted manually)
- real 3m55.198s for num_threads = 2 (aborted manually)
- why do above configurations never finish but the ones below do finish?
- Further, if add "-O3" to compiler option above, the run in 26s and 13s,why?
- real 0m12.690s for num_threads = 3
- real 0m9.199s for num_threads = 4
- real 0m7.971s for num_threads = 5
- real 0m7.654s for num_threads = 6
- real 0m7.304s for num_threads = 7
- real 0m7.740s for num_threads = 8
*/
bool func_bool_int(int n) {
if(n < 2)
return false;
for(int i = 2;i < n;i++)
if(n % i == 0)
return false;
return true;
}
int main() {
const int num_threads = 4;
thread_pool p(num_threads);
int n = 100000;
int max = 5 * n;
vector<future<bool>> f;
for( int i = n; i < max; i++ )
f.emplace_back(p.async(function<bool(int)>(func_bool_int), i));
std::cout << "..." << std::endl;
size_t accum = 0;
for(auto &i : f) {
accum += i.get();
}
std::cout << "accum = " << accum << std::endl;
return 0;
}
`
When I ran for num_threads = 8, 7, 6, 5, 4, 3 things went okay. But when I tried num_threads of 2 or 1 the runtime seemed to be in an infinite loop ie I was seeing about 7s to 12s runtimes above num_threads>2 but once I tried num_threads=2 or 1 I waited 3 minutes to 5 minutes (orders of magnitude longer than extrapolated runtime) before manually killing the process. Is this behavior to be expected? Note this was on a gcc 6.3 with "git checkout -b cv origin/cv" branch. Adding an "-O3" optimazation seemed to fix the issue; expected?
I was able to compile on Windows and successfully run with
gcc.exe (Rev3, Built by MSYS2 project) 6.3.0
only with the older ThreadPool revision 65879c9 since I didn't have the latest gcc comipler collection installed.
But on Ubuntu with the master revision using
g++-7 (Ubuntu 7.1.0-5ubuntu2~16.04) 7.1.0
ie g++-7 -std=c++17 -fconcepts test.cpp thread_pool.cpp -pthread
all worked fine.
It was my fault in using the cv branch originally.
Thanks
It's been helpful code. Thanks for sharing your work and efforts.