https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65485
Bug ID: 65485
Summary: Use openmp in dynamic library but not in calling
program causes segfault at the end of execution
Product: gcc
Version: 4.9.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: thomas.izard at silkan dot com
Consider:
#include "omp.h"
#include <iostream>
extern "C" {
int test() {
int N = omp_get_max_threads();
#pragma omp parallel num_threads(N)
{
std::cout << omp_get_thread_num() << std::endl;
}
return 0;
}
};
Compiled with g++ -fopenmp -shared -fPIC -Wall -Wextra -o shared.so shared.cpp
(no warnings/no errors).
and :
#include <iostream>
#include <dlfcn.h>
int main() {
void* handle = dlopen("./shared.so", RTLD_NOW);
if (!handle) {
std::cerr << "can not open shared.so" << std::endl;
return 1;
}
int(*f)() = (int(*)()) dlsym(handle,"test");
if (!f) {
std::cerr << "can not find 'test' symbol in shared.so" << std::endl;
return 1;
}
(*f)();
if (dlclose(handle)) {
std::cerr << "can not close shared.so" << std::endl;
return 1;
}
return 0;
}
compiled with:
g++ -Wall -Wextra -o main main.cpp -ldl (no warnings/no errors).
The main program loads the dynamic library and calls the "test" function. It
does not directly use openmp.
The problem is that a segmentation fault happen at the very end of the program.
According to valgrind, at this point some threads are still active.
Adding a -fopenmp flag while compiling the main program does not fix the
problem. It seems that in this case the compiled code is not linked to GOMP
(confirmed by ldd).
The only workaround is to actually add an OpenMP directive (or a call to an
OpenMP runtime function) in the main.cpp file.
Same issue with (at least): g++-4.6.4, g++-4.7, g++-4.8.2
Systems used :
Linux 3.2.0-77-generic #114-Ubuntu SMP (Ubuntu 64 12.04)
Linux 3.13.0-43-generic #72-Ubuntu SMP (Ubuntu 64 server 14.04)
Thomas
