Segmentation fault on FFI call containing OpenMP clause
- From: Ciprian Tomoiaga <ciprian.tomoiaga@xxxxxxxxx>
- To: luajit@xxxxxxxxxxxxx
- Date: Tue, 9 Dec 2014 20:51:58 +0100
Hi all,
I am writing to ask for methods of debugging the following scenario.
I have a C function containing OpenMP clause which I call with ffi. On my
machine, the program terminates with a segmentation fault. However, it
works without a problem on a Mac and on another machine. Here is my small
example:
//---- test.c -------------
#include <stdio.h>
#include <omp.h>
// gcc -std=c99 -Wall -pedantic -fopenmp -shared -fPIC test.c -o libtest.so
void parallel_test() {
#pragma omp parallel num_threads(2)
{
printf("In Parallel\n");
}
}
--------------------------------
--- test.lua --------
local ffi = require "ffi"
local testLib = ffi.load("./libtest.so")
ffi.cdef[[
void parallel_test();
]]
testLib.parallel_test()
-----------------------
Now call
:
luajit test.lua
My machine, Linux Mint 17, based on Ubuntu 14.04, compiled with gcc (v4.7,
v4.8, v4.9) results in Segmentation fault when num_threads is greater than
1.
A Mac machine, compiled with gcc v4.9 works fine.
Another Linux machine, SLC 6.6, based on RHEL6, compiled with gcc v4.4
works fine. Moreover, I did a crazy (stupid?) thing and ran it with the .so
compiled for my machine, and still worked fine (intel CPUs both, different
versions).
Can anyone please point me towards what tools I can use to debug this
problem? I expect valgrind would be the choice, but I'm not sure how to use
it with LuaJIT.
Many thanks everyone!
Regards,
Ciprian
Other related posts:
